Working in any sort of large organization, you quickly come to recognize the signs of people trying to avoid blame or to just plain old scurry for cover. There’s one I always remember from my early days in the industry when we had a pretty useful fire in a fume hood across the hall. This was about 1990, in a building that doesn’t even exist any more, with a company that doesn’t exist either! I’m not sure how it got started, but someone from that group was walking towards a door and noticed the smoke filling the air, then went up to the window and could see the dancing flames in the hood. That gets attention pretty quickly! I was working across the hall; I had just been employed for a few months at that point. The shouting and commotion got my attention quickly, and I grabbed a fire extinguisher off the wall, pulling the pin as I headed across the way to see if there was anything to be done. I met up with one of the guys in that group, who was a volunteer firefighter in his own town in New Jersey, and he had an extinguisher ready to go, too. He motioned to crouch down as we entered the lab (I let him take the lead!) and once he got a good look at the hood he waved both of us back out. It was too large a blaze to be sure that we’d get it by that point; I think an oil bath was involved and there were reagent bottles nearby, too. The actual firefighters weren’t long in showing up, fortunately. One of the other group members had tried to call in the emergency using the red emergency phone on the corridor wall. She picked it up, but there was no response from the other end, which was supposed to be the main security office, so she shouted that there was a fire and its location, but there was no sound in response to that, either. She hung up the phone in frustration - I think someone else started calling various internal numbers to get the word out since that designated emergency system didn’t seem to be even working. It was around this time that it dawned on me that this was the radiochemistry lab that I’d gone blithely running into, but fortunately I found out shortly that it had been a cold reaction (and a cold reaction hood), which was a relief. The aftermath of the fire featured a good bit of agitated comment about the worthlessness of that red emergency phone - but the security department responded that they had gotten the message, that the phone had worked exactly as it was supposed to, that no, it wasn’t a two-way “phone” per se even if it looked exactly like one and that no response was to be expected when you picked it up and said something, and didn’t everyone already know that? No, no we didn’t, actually. Recriminations ensued, but what I remember most about those was the memo that came out after a bit from the security team “reminding” everyone about the workings of the emergency phone system. It pretended that we’d all been trained on this before - har, har - and finished up with a ringing declaration which I immediate committed to memory and have since never forgotten: “We will continue to monitor system functionality in order to maintain maximum operational availability”. Now that, I thought as I read it, is the sound of hot air being blown. I don’t think I’ve encountered a better once since in the workplace. Even now, it’s what I say to my wife if I unclog a drain or something.
Show full content
Working in any sort of large organization, you quickly come to recognize the signs of people trying to avoid blame or to just plain old scurry for cover. There’s one I always remember from my early days in the industry when we had a pretty useful fire in a fume hood across the hall. This was about 1990, in a building that doesn’t even exist any more, with a company that doesn’t exist either! I’m not sure how it got started, but someone from that group was walking towards a door and noticed the smoke filling the air, then went up to the window and could see the dancing flames in the hood.
That gets attention pretty quickly! I was working across the hall; I had just been employed for a few months at that point. The shouting and commotion got my attention quickly, and I grabbed a fire extinguisher off the wall, pulling the pin as I headed across the way to see if there was anything to be done. I met up with one of the guys in that group, who was a volunteer firefighter in his own town in New Jersey, and he had an extinguisher ready to go, too. He motioned to crouch down as we entered the lab (I let him take the lead!) and once he got a good look at the hood he waved both of us back out. It was too large a blaze to be sure that we’d get it by that point; I think an oil bath was involved and there were reagent bottles nearby, too.
The actual firefighters weren’t long in showing up, fortunately. One of the other group members had tried to call in the emergency using the red emergency phone on the corridor wall. She picked it up, but there was no response from the other end, which was supposed to be the main security office, so she shouted that there was a fire and its location, but there was no sound in response to that, either. She hung up the phone in frustration - I think someone else started calling various internal numbers to get the word out since that designated emergency system didn’t seem to be even working.
It was around this time that it dawned on me that this was the radiochemistry lab that I’d gone blithely running into, but fortunately I found out shortly that it had been a cold reaction (and a cold reaction hood), which was a relief. The aftermath of the fire featured a good bit of agitated comment about the worthlessness of that red emergency phone - but the security department responded that they had gotten the message, that the phone had worked exactly as it was supposed to, that no, it wasn’t a two-way “phone” per se even if it looked exactly like one and that no response was to be expected when you picked it up and said something, and didn’t everyone already know that?
No, no we didn’t, actually. Recriminations ensued, but what I remember most about those was the memo that came out after a bit from the security team “reminding” everyone about the workings of the emergency phone system. It pretended that we’d all been trained on this before - har, har - and finished up with a ringing declaration which I immediate committed to memory and have since never forgotten: “We will continue to monitor system functionality in order to maintain maximum operational availability”. Now that, I thought as I read it, is the sound of hot air being blown. I don’t think I’ve encountered a better once since in the workplace. Even now, it’s what I say to my wife if I unclog a drain or something.
The arXiv preprint server has become essential to several scientific fields, and has inspired similar services like BioRxiv and ChemRxiv. But the generative AI age has been hard on these sites, and you can see it by their steady imposition of new rules. Last November, arXiv announced that it was clamping down on the submission of review articles, particularly in computer science. These were becoming alarmingly easy to put together (often as not padded out with references that didn’t even quite exist), and the moderators were getting overwhelmed. No new reviews would be accepted in computer science, they ruled, unless they had already been through a peer-review process at a journal (or for a conference), which sort of turned the entire idea of the arXiv on its head in that manuscript category. But you can understand why. Then in late January came the announcement that first-time submitters would need an endorsement from another author who had published there before. The stated reason was a rise in outright fraudulent submissions. That article says that the moderator's rejection rates had tripled over 2025 and were apparently still rising, and something obviously had to be done. Now the chair of the site’s computer science editorial committee (Thomas Diettrich) has turned the flames up. “If the submission contains incontrovertible evidence that the authors did not check the results of LLM generation, it means we can’t trust anything in the paper. The penalty is a 1-year ban from arXiv followed by the requirement that subsequent arXiv submissions must first be accepted at a reputable peer-reviewed venue”. He went on to give examples of such evidence, including the above-mentioned hallucinated references (to papers that just don’t exist) or the presence of meta-comments from the LLM software ("Here is your summary" and so on). Those fake references are indeed becoming a plague, as illustrated by this manuscript at arXiv itself. Those authors set a lower bound of over 146,000 hallucinated citations appearing in 2025, and it's been rising sharply. This obviously threatens the integrity of the scientific literature itself, and that really didn’t need to take any more torpedos, thanks. If we don’t do something about this we run the risk of an eventual painful de-slopping process, and it would be a *lot* better for everyone if we blocked that crap from getting into the literature in the first place. It’ll be a race. . .
Show full content
The arXiv preprint server has become essential to several scientific fields, and has inspired similar services like BioRxiv and ChemRxiv. But the generative AI age has been hard on these sites, and you can see it by their steady imposition of new rules.
Last November, arXiv announced that it was clamping down on the submission of review articles, particularly in computer science. These were becoming alarmingly easy to put together (often as not padded out with references that didn’t even quite exist), and the moderators were getting overwhelmed. No new reviews would be accepted in computer science, they ruled, unless they had already been through a peer-review process at a journal (or for a conference), which sort of turned the entire idea of the arXiv on its head in that manuscript category. But you can understand why.
Then in late January came the announcement that first-time submitters would need an endorsement from another author who had published there before. The stated reason was a rise in outright fraudulent submissions. That article says that the moderator's rejection rates had tripled over 2025 and were apparently still rising, and something obviously had to be done.
Now the chair of the site’s computer science editorial committee (Thomas Diettrich) has turned the flames up. “If the submission contains incontrovertible evidence that the authors did not check the results of LLM generation, it means we can’t trust anything in the paper. The penalty is a 1-year ban from arXiv followed by the requirement that subsequent arXiv submissions must first be accepted at a reputable peer-reviewed venue”. He went on to give examples of such evidence, including the above-mentioned hallucinated references (to papers that just don’t exist) or the presence of meta-comments from the LLM software ("Here is your summary" and so on).
Those fake references are indeed becoming a plague, as illustrated by this manuscript at arXiv itself. Those authors set a lower bound of over 146,000 hallucinated citations appearing in 2025, and it's been rising sharply. This obviously threatens the integrity of the scientific literature itself, and that really didn’t need to take any more torpedos, thanks. If we don’t do something about this we run the risk of an eventual painful de-slopping process, and it would be a *lot* better for everyone if we blocked that crap from getting into the literature in the first place. It’ll be a race. . .
I definitely need to cover this recent work from Merck, because (1) it’s very interesting scientifically and (2) it has over 130 authors on the paper (!) It details the industrial synthesis of enlicitide, which is a beast of a macrocyclic peptide (see below!) Just looking at the structure tells you that this must be a ferociously active molecule with huge commercial potential, because there is just no way that anyone is going to make this on scale - or find a way to make this on scale! - otherwise. This drug hits the PCSK9 pathway, which has been quite a story over the years. This target was famously discovered by finding a few people with mutations in the underlying gene who had bizarrely low LDL concentrations and who seemed to have correspondingly robust cardiac health. The first drugs on the market to take action on this idea were antibodies to block the receptor, and those were approved by the FDA some years ago as injectables. They do indeed lower LDL, but (as that last blog post link notes) the results in humans have to be characterized as “good but not revolutionary”. I don’t know of any studies that show a definite advantage compared to statin (HMG-CoA reductase inhibitor) therapy, for example, although there are statin side effects in some patients that have to be taken into account. (On the other side of the question, statins seem to have some beneficial pleiotropic effects that we don’t quite understand, and whether these are shared by PCSK9 inhibition, I don’t know). And there are other people for whom statin therapy just comes up short, to be sure. Several companies have tried over the years to come up with a small-molecule (well, smallish-molecule) approach that could lead to an orally dosed therapy as opposed to an injectable, and Merck has apparently made it over the finish line with this one. Here are the results of a trial in over 1900 patients (compared to over 900 in the placebo group) treated with the drug over a year, 20mg once a day. Over that time, the placebo group’s LDL went up about 3%, while the treatment group’s went down about 57%, which is what we call “pretty darn significant”, statistically speaking, with no differences in adverse events. So now that you have a new cardiovascular drug, how do you make it for the hoped-for large patient population when it looks like, well, that thing to the right? That’s quite the multicyclic peptide, and while a lot of the key bond formations are good ol’ amide couplings, you have several that are not. The team divided up the molecules into “Western”, “Eastern”, and “Northern” pieces (based on the three macrocycles in the final structure) and demonstrated that they could make all of these in crystalline form (thus obviating the need for chromatographic purification). The Northern one was the toughest by far, with three unnatural amino acids and a choice of amine nucleophiles. The answer to putting all this together was harnessing amino acid ligase enzymes, which lets you couple unprotected amino acid partners when everything is working right. The Merck team looked over a list of AAL enzymes and found one from Bifidobacterium adolescentis that was accomodating enough to deal with their intermediate. All of the candidates turned out to be able to handle peptide chains as the nucleophilic partner while only accepting single amino acids as the electrophile. That’s too bad in a way, because you could imagine assembling larger fragments this way, but if you get the enzymatic process running smoothly enough you can just turn things through it several times in a row. And without having to do protection/deprotection steps! What looks like the nastiest traffic jam in the synthesis was the final stage of making that fragment, because you’re presented simultaneously with three amines and two carboxyls that need to be brought together in the proper pairings. The Merck group turned to esterase enzymes rather than proteases/amidases, not least because those can’t be tempted to run the amide formation in reverse because they don’t have the nucleophilic horsepower in their active sites. A previously-unreported enzyme from a Roseibacillus species showed promise, although I would certainly like to hear how many others got screened along the way, and even that one needed some engineering to increase its selectivity. They ended up being able to make the Northern piece with four enzymes and three separate building blocks in a single pot, which really is a tour de force. To elaborate on to the final macrocycle the team called on thioesterase enzymes, which are generally what come into play in biosynthesis pathways for natural products of this type. Another (no doubt wide-ranging) screen identified a likely one from Brevibacillus laterosporus, but this also needed artisinal modification in its sequence to get the yields and selectivity up. I am skipping over a lot of work when I write sentences like that one! The Northern and Eastern fragments also needed a reductive amination to bring them together, and a ketoreductase enzyme from a Kyrpidia species coupled with an imine reductase (from Pseudogymnoascus species) was used to make the requisite aldehyde, both after still more screening and protein engineering. This reaction mixture needed recycling of NADP/NADPH to run the synthesis enzymes, and still more enzymes were brought in for that cycle. In the end, the team achieved a one-pot five-enzyme process that did the overall transformation in 69% isolated yield. Coupling this to the Western fragment was done through good ol’ chemical means (diphenylphosphinic chloride) while protected a stray primary amine that needed to be ready for the final macrocyclization. There are many such syntheses that have come to grief at such final steps; these macro-ring closures do not always want to happen easily. Another engineered thioesterase (this time from a Streptomyces species) was found to do the job and without the traditional need for high dilution. You often have to do that dilution to encourage your molecule to bite its own tail and make the large ring as opposed to reacting with another one nearby (which makes useless dimers or oligomers), so this enzymatic route is a huge help. The transformation comes though in 84% yield and >99% purity after liquid-liquid extraction and salt formation! The overall yield, starting from 5-fluoro-N-aminohexyl tryptophan (you can see it hiding in there) is 39% and this has been run on multikilo scale with no chromatography. This is tremendously impressive, truly state-of-the-art process chemistry. I think we’re going to be seeing the reverberations of this work in macrocycle synthesis (and especially macrocyclic peptides) for a long time to come. Now to see if the drug itself performs up to its commercial and medical potential!
Show full content
I definitely need to cover this recent work from Merck, because (1) it’s very interesting scientifically and (2) it has over 130 authors on the paper (!) It details the industrial synthesis of enlicitide, which is a beast of a macrocyclic peptide (see below!) Just looking at the structure tells you that this must be a ferociously active molecule with huge commercial potential, because there is just no way that anyone is going to make this on scale - or find a way to make this on scale! - otherwise.
This drug hits the PCSK9 pathway, which has been quite a story over the years. This target was famously discovered by finding a few people with mutations in the underlying gene who had bizarrely low LDL concentrations and who seemed to have correspondingly robust cardiac health. The first drugs on the market to take action on this idea were antibodies to block the receptor, and those were approved by the FDA some years ago as injectables.
They do indeed lower LDL, but (as that last blog post link notes) the results in humans have to be characterized as “good but not revolutionary”. I don’t know of any studies that show a definite advantage compared to statin (HMG-CoA reductase inhibitor) therapy, for example, although there are statin side effects in some patients that have to be taken into account. (On the other side of the question, statins seem to have some beneficial pleiotropic effects that we don’t quite understand, and whether these are shared by PCSK9 inhibition, I don’t know). And there are other people for whom statin therapy just comes up short, to be sure.
Several companies have tried over the years to come up with a small-molecule (well, smallish-molecule) approach that could lead to an orally dosed therapy as opposed to an injectable, and Merck has apparently made it over the finish line with this one. Here are the results of a trial in over 1900 patients (compared to over 900 in the placebo group) treated with the drug over a year, 20mg once a day. Over that time, the placebo group’s LDL went up about 3%, while the treatment group’s went down about 57%, which is what we call “pretty darn significant”, statistically speaking, with no differences in adverse events.
So now that you have a new cardiovascular drug, how do you make it for the hoped-for large patient population when it looks like, well, that thing to the right? That’s quite the multicyclic peptide, and while a lot of the key bond formations are good ol’ amide couplings, you have several that are not. The team divided up the molecules into “Western”, “Eastern”, and “Northern” pieces (based on the three macrocycles in the final structure) and demonstrated that they could make all of these in crystalline form (thus obviating the need for chromatographic purification). The Northern one was the toughest by far, with three unnatural amino acids and a choice of amine nucleophiles.
The answer to putting all this together was harnessing amino acid ligase enzymes, which lets you couple unprotected amino acid partners when everything is working right. The Merck team looked over a list of AAL enzymes and found one from Bifidobacterium adolescentis that was accomodating enough to deal with their intermediate. All of the candidates turned out to be able to handle peptide chains as the nucleophilic partner while only accepting single amino acids as the electrophile. That’s too bad in a way, because you could imagine assembling larger fragments this way, but if you get the enzymatic process running smoothly enough you can just turn things through it several times in a row. And without having to do protection/deprotection steps!
What looks like the nastiest traffic jam in the synthesis was the final stage of making that fragment, because you’re presented simultaneously with three amines and two carboxyls that need to be brought together in the proper pairings. The Merck group turned to esterase enzymes rather than proteases/amidases, not least because those can’t be tempted to run the amide formation in reverse because they don’t have the nucleophilic horsepower in their active sites. A previously-unreported enzyme from a Roseibacillus species showed promise, although I would certainly like to hear how many others got screened along the way, and even that one needed some engineering to increase its selectivity. They ended up being able to make the Northern piece with four enzymes and three separate building blocks in a single pot, which really is a tour de force.
To elaborate on to the final macrocycle the team called on thioesterase enzymes, which are generally what come into play in biosynthesis pathways for natural products of this type. Another (no doubt wide-ranging) screen identified a likely one from Brevibacillus laterosporus, but this also needed artisinal modification in its sequence to get the yields and selectivity up. I am skipping over a lot of work when I write sentences like that one!
The Northern and Eastern fragments also needed a reductive amination to bring them together, and a ketoreductase enzyme from a Kyrpidia species coupled with an imine reductase (from Pseudogymnoascus species) was used to make the requisite aldehyde, both after still more screening and protein engineering. This reaction mixture needed recycling of NADP/NADPH to run the synthesis enzymes, and still more enzymes were brought in for that cycle. In the end, the team achieved a one-pot five-enzyme process that did the overall transformation in 69% isolated yield.
Coupling this to the Western fragment was done through good ol’ chemical means (diphenylphosphinic chloride) while protected a stray primary amine that needed to be ready for the final macrocyclization. There are many such syntheses that have come to grief at such final steps; these macro-ring closures do not always want to happen easily. Another engineered thioesterase (this time from a Streptomyces species) was found to do the job and without the traditional need for high dilution. You often have to do that dilution to encourage your molecule to bite its own tail and make the large ring as opposed to reacting with another one nearby (which makes useless dimers or oligomers), so this enzymatic route is a huge help. The transformation comes though in 84% yield and >99% purity after liquid-liquid extraction and salt formation!
The overall yield, starting from 5-fluoro-N-aminohexyl tryptophan (you can see it hiding in there) is 39% and this has been run on multikilo scale with no chromatography. This is tremendously impressive, truly state-of-the-art process chemistry. I think we’re going to be seeing the reverberations of this work in macrocycle synthesis (and especially macrocyclic peptides) for a long time to come. Now to see if the drug itself performs up to its commercial and medical potential!
I wrote here last year about the clinical trial results for the first “bifunctional degrader” molecule to make it that far, vepdegestrant from Arvinas and Pfizer. That post will send you to some background information about this class of molecule, but suffice it to say that they represent a completely new mode of action (destruction of a target protein in the living cell as opposed to chemical inhibition of it). As such, this program has been watched closely, and the results were. . .not spectacular. I mean, the drug worked, so that’s a good thing. But the hope was that this new mechanism would make it work even better than that traditional means. Vedegestrant targets the estrogen receptor, and there are of course a number of small-molecule ER agents out there. And it did not really demonstrate any advantages versus the comparison in the trial, fulvestrant. And to be sure, that was a pretty close comparison, because fulvestrant is also a degrader, although not by the same means as the bifunctionals. It binds to the receptor and destabilizes its conformation, marking it to the cellular quality-control machinery as defective/misfolded and targeting it for degradation. So this wasn’t really a head-to-head showdown between degradation and inhibition; it was more of a competition between one variety of degradation and another. And there didn’t seem to be much to choose from. Well, vepdegestrant got FDA approval earlier this month, the first bifunctional degrader to reach that milestone. And eleven days later, it was announced that Rigel Pharmaceuticals now has a global licensing agreement for the drug, which is being marketed as Veppanu. Their intent to out-license had already been announced last fall. This means that Pfizer and Arvinas are basically washing their hands of the drug and have offloaded it for $35 million to each company, and there are milestone payments in the deal that could pay out up to $320 million as the regulatory and commercial story unfolds. The details of those milestones do not seem to have been made public - myself, I’d be surprised if things get to that level, but who knows. But now let’s go back to 2021 when Pfizer and Arvinas announced their collaboration on the drug. Pfizer paid out $650 million upfront, and made a $350 million equity investment in Arvinas. So that’s a billion dollars out the door before that Phase III trial even started, and obviously more money has been spent since then on it. You do not need a calculator to work out that so far Pfizer has spent somewhere-more-than-a-billion and has received in turn $35 million in cash plus some promises of another hundred-and-something million if things should happen to go particularly well. This has not been a deal that they will look back on fondly, although God knows they’ve done worse. That’s the investigational drug business, though: you pays your money and you gets your clinical results, and good luck to you. Pfizer can certainly weather investments like this, although this sort of thing is not exactly a motivation to give anyone a break on the price of the things that actually do work out better. As for Arvinas, they received a very large (and much-needed) infusion of capital from the deal, but now have to reckon with the fact that that’s all they’re going to get. They’re laying off all the people they brought on to help with the commercial launch of vedegestrant, of course, and hoping for better next time. Clinical trials that show your drug works and lead to FDA approvals sound like just the thing, don’t they? But a further run of results like this could leave them in a tough position. My own thinking is that degraders really do have interesting possibilities, but there is a whole bunch of important stuff that we don’t know about the field yet. So it will not be dull.
Show full content
I wrote here last year about the clinical trial results for the first “bifunctional degrader” molecule to make it that far, vepdegestrant from Arvinas and Pfizer. That post will send you to some background information about this class of molecule, but suffice it to say that they represent a completely new mode of action (destruction of a target protein in the living cell as opposed to chemical inhibition of it). As such, this program has been watched closely, and the results were. . .not spectacular.
I mean, the drug worked, so that’s a good thing. But the hope was that this new mechanism would make it work even better than that traditional means. Vedegestrant targets the estrogen receptor, and there are of course a number of small-molecule ER agents out there. And it did not really demonstrate any advantages versus the comparison in the trial, fulvestrant. And to be sure, that was a pretty close comparison, because fulvestrant is also a degrader, although not by the same means as the bifunctionals. It binds to the receptor and destabilizes its conformation, marking it to the cellular quality-control machinery as defective/misfolded and targeting it for degradation. So this wasn’t really a head-to-head showdown between degradation and inhibition; it was more of a competition between one variety of degradation and another. And there didn’t seem to be much to choose from.
Well, vepdegestrant got FDA approval earlier this month, the first bifunctional degrader to reach that milestone. And eleven days later, it was announced that Rigel Pharmaceuticals now has a global licensing agreement for the drug, which is being marketed as Veppanu. Their intent to out-license had already been announced last fall. This means that Pfizer and Arvinas are basically washing their hands of the drug and have offloaded it for $35 million to each company, and there are milestone payments in the deal that could pay out up to $320 million as the regulatory and commercial story unfolds. The details of those milestones do not seem to have been made public - myself, I’d be surprised if things get to that level, but who knows.
But now let’s go back to 2021 when Pfizer and Arvinas announced their collaboration on the drug. Pfizer paid out $650 million upfront, and made a $350 million equity investment in Arvinas. So that’s a billion dollars out the door before that Phase III trial even started, and obviously more money has been spent since then on it. You do not need a calculator to work out that so far Pfizer has spent somewhere-more-than-a-billion and has received in turn $35 million in cash plus some promises of another hundred-and-something million if things should happen to go particularly well. This has not been a deal that they will look back on fondly, although God knows they’ve done worse.
That’s the investigational drug business, though: you pays your money and you gets your clinical results, and good luck to you. Pfizer can certainly weather investments like this, although this sort of thing is not exactly a motivation to give anyone a break on the price of the things that actually do work out better. As for Arvinas, they received a very large (and much-needed) infusion of capital from the deal, but now have to reckon with the fact that that’s all they’re going to get. They’re laying off all the people they brought on to help with the commercial launch of vedegestrant, of course, and hoping for better next time.
Clinical trials that show your drug works and lead to FDA approvals sound like just the thing, don’t they? But a further run of results like this could leave them in a tough position. My own thinking is that degraders really do have interesting possibilities, but there is a whole bunch of important stuff that we don’t know about the field yet. So it will not be dull.
I think that many synthetic organic chemists will be able to relate to the approach described in this paper, on software-aided route design. Its authors are trying to make such software take a viewpoint from higher over the synthesis, rather than working out every reaction. As noted in this commentary, for larger molecules that can leave you with a forest of rather-similar routes that differ in choice of protecting groups, relative oxidation states, order of reactions and other details that (in many cases) can well be adjusted when work is underway. Instead, the idea is to generalize more about functional groups and bond disconnections. What you get is a route that’s based somewhat more on abstractions, but ones that a synthetic chemist can recognize and work with: “OK, there’s going to need to be an aminomethylene group coming off here; we’ll assume that’s going to be derived from some sort of carbonyl - could be an amide formation followed by reduction, or a reductive amination off an aldehyde, whatever.” That sort of thing, as you can see at right. It’s more of a strategic approach than a tactics-based approach. This way the different routes that might be generated really do have a greater chance of being different from each other, rather than being lists of variations on a theme. Looking at the paper, you can tell that it took a lot of human-intensive curation to generate the general pathways, but I think it’s worth the effort. This is a lot closer to how most experienced chemists think, and the software hands off its suggestions in a form that a chemist is ready to evaluate and work with. Honestly, that’s what many of us do with a machine-generated route anyway - “It says here I need a methyl ester, but there are alternatives that I need to keep in mind”. The approach even avoids (a bit) the trap of having to depend on the huge shaggy mass of reported reactions, some of which may be real and some of which may not. “We’ll assume that there is a way to get this metal-catalyzed coupling step to work” is probably a lot closer to the truth than “Here, trust these exact conditions from this paper over here”. I mean, you might want to start with those, sure, but with only limited expectations that they will do the job for you. I well remember my first boss in my first job in this business sighing as yet another paper or patent reaction didn’t work as advertised, and saying “Lie, all lies”. There’s still a huge amount of work to do to make computer-based retrosynthesis realize its promise; the problems mentioned in this post haven’t gone away. In the end, I think we’re still going to have to recapitulate much of the literature under more controlled conditions (and, I would recommend, with more attention to reaction scope than the original papers may have provided!) It will be an interesting problem to figure out where to direct such efforts for maximum impact, i.e., which parts of the chemistry literature need the most shoring up for the biggest real-world effect? Better knowledge of the rules behind metal-catalyzed couplings are an obvious place to start (there are rules, right?), but nominations for other candidates are welcome below. . .
Show full content
I think that many synthetic organic chemists will be able to relate to the approach described in this paper, on software-aided route design. Its authors are trying to make such software take a viewpoint from higher over the synthesis, rather than working out every reaction. As noted in this commentary, for larger molecules that can leave you with a forest of rather-similar routes that differ in choice of protecting groups, relative oxidation states, order of reactions and other details that (in many cases) can well be adjusted when work is underway.
Instead, the idea is to generalize more about functional groups and bond disconnections. What you get is a route that’s based somewhat more on abstractions, but ones that a synthetic chemist can recognize and work with: “OK, there’s going to need to be an aminomethylene group coming off here; we’ll assume that’s going to be derived from some sort of carbonyl - could be an amide formation followed by reduction, or a reductive amination off an aldehyde, whatever.” That sort of thing, as you can see at right. It’s more of a strategic approach than a tactics-based approach. This way the different routes that might be generated really do have a greater chance of being different from each other, rather than being lists of variations on a theme. Looking at the paper, you can tell that it took a lot of human-intensive curation to generate the general pathways, but I think it’s worth the effort.
This is a lot closer to how most experienced chemists think, and the software hands off its suggestions in a form that a chemist is ready to evaluate and work with. Honestly, that’s what many of us do with a machine-generated route anyway - “It says here I need a methyl ester, but there are alternatives that I need to keep in mind”. The approach even avoids (a bit) the trap of having to depend on the huge shaggy mass of reported reactions, some of which may be real and some of which may not. “We’ll assume that there is a way to get this metal-catalyzed coupling step to work” is probably a lot closer to the truth than “Here, trust these exact conditions from this paper over here”. I mean, you might want to start with those, sure, but with only limited expectations that they will do the job for you. I well remember my first boss in my first job in this business sighing as yet another paper or patent reaction didn’t work as advertised, and saying “Lie, all lies”.
There’s still a huge amount of work to do to make computer-based retrosynthesis realize its promise; the problems mentioned in this post haven’t gone away. In the end, I think we’re still going to have to recapitulate much of the literature under more controlled conditions (and, I would recommend, with more attention to reaction scope than the original papers may have provided!) It will be an interesting problem to figure out where to direct such efforts for maximum impact, i.e., which parts of the chemistry literature need the most shoring up for the biggest real-world effect? Better knowledge of the rules behind metal-catalyzed couplings are an obvious place to start (there are rules, right?), but nominations for other candidates are welcome below. . .
I’ve written here many times about various post-translational protein modifications. That’s a huge field of study, because it has become more and more clear over the years that it’s not so much that living cells have huge numbers of different protein sequences in them (although they have plenty!) it’s that every protein seems to be modifiable by a long list of add-ons. These include (but are absolutely guaranteed not to be limited to!) phosphorylation, ubiquitination, farnesylation, glycosylation, acetylation, palmitoylation and so, so many more. We’re still discovering them! Just recently it’s been found that dopamine (of all things) is sometimes attached to glutamine residues of proteins, so you can add that to the list. These things have profound effects on protein activity - what partners they associate with, their cellular locations, their cellular half-lives and more. And it’s long been appreciated, in a hazy sort of way, that they also affect ligand binding. This new paper tries to put some order into this landscape by scanning the proteome with a variety of photoaffinity probes and controlling for the most common PTMs. They have discovered hundreds of proteins whose ligand binding is noticeably affected by phosphorylation or glycosylation state - and those are of course just two of the more common modifications. These proteins fall into all sorts of classes - kinase enzymes, GTPases, ubiquitin ligases, GPCRs, growth factors, cell adhesion proteins. It isn’t a small local effect. Interestingly, the proteins that didn’t show as much variability in PTM-dependent ligand binding were involved more in things like metabolism and protein translation, whereas the ones that did tended to bin into functions like protein localization, lysosomal processes, ionic state maintenance, and mRNA transport. Make of that what you will! These PTM states were altered by the use of the blunderbuss kinase inhibitor staurosporine or the equally wide-ranging tunicamycin for glycosylation. Those are definitely enough to show you the broad outlines and to demonstrate the point of the paper, but as the authors well note, they are not going to reveal the whole story to you even just for these two PTMs. There is just no telling how many cases like this are out there in the living cell - and given the often-transient nature of these modifications, it will be no small task to track them down. And keep in mind that we don’t have very complete knowledge of all the interactions and ligands involved here, either. This sort of thing could keep a lot of people busy for quite a long time. The authors put a lot of working into mapping some of these effects, and as you’d figure, some of them trace back to residues in or near ligand binding sites. But not all of them. There are a lot of allosteric sites from the look of things, and others where the PTM changes are upsteam of interactions with other protein partners that in turn change ligand-binding behavior. And as the paper details, these results have some direct drug discovery implications. KRAS, for example, has its ligand-binding affinity kicked around all over the place by phosphorylation state, which makes you wonder about what’s really going on in the cell with the ligands that people have spent so much time working on. (That’s in addition, of course, to the many changes in KRAS activity in general that are driven by those same modifications). There may be chances to optimize drug targeting to specific cellular states - well, once we work out what those are and what they mean!
Show full content
I’ve written here many times about various post-translational protein modifications. That’s a huge field of study, because it has become more and more clear over the years that it’s not so much that living cells have huge numbers of different protein sequences in them (although they have plenty!) it’s that every protein seems to be modifiable by a long list of add-ons. These include (but are absolutely guaranteed not to be limited to!) phosphorylation, ubiquitination, farnesylation, glycosylation, acetylation, palmitoylation and so, so many more. We’re still discovering them! Just recently it’s been found that dopamine (of all things) is sometimes attached to glutamine residues of proteins, so you can add that to the list.
These things have profound effects on protein activity - what partners they associate with, their cellular locations, their cellular half-lives and more. And it’s long been appreciated, in a hazy sort of way, that they also affect ligand binding. This new paper tries to put some order into this landscape by scanning the proteome with a variety of photoaffinity probes and controlling for the most common PTMs. They have discovered hundreds of proteins whose ligand binding is noticeably affected by phosphorylation or glycosylation state - and those are of course just two of the more common modifications. These proteins fall into all sorts of classes - kinase enzymes, GTPases, ubiquitin ligases, GPCRs, growth factors, cell adhesion proteins. It isn’t a small local effect.
Interestingly, the proteins that didn’t show as much variability in PTM-dependent ligand binding were involved more in things like metabolism and protein translation, whereas the ones that did tended to bin into functions like protein localization, lysosomal processes, ionic state maintenance, and mRNA transport. Make of that what you will!
These PTM states were altered by the use of the blunderbuss kinase inhibitor staurosporine or the equally wide-ranging tunicamycin for glycosylation. Those are definitely enough to show you the broad outlines and to demonstrate the point of the paper, but as the authors well note, they are not going to reveal the whole story to you even just for these two PTMs. There is just no telling how many cases like this are out there in the living cell - and given the often-transient nature of these modifications, it will be no small task to track them down. And keep in mind that we don’t have very complete knowledge of all the interactions and ligands involved here, either. This sort of thing could keep a lot of people busy for quite a long time.
The authors put a lot of working into mapping some of these effects, and as you’d figure, some of them trace back to residues in or near ligand binding sites. But not all of them. There are a lot of allosteric sites from the look of things, and others where the PTM changes are upsteam of interactions with other protein partners that in turn change ligand-binding behavior.
And as the paper details, these results have some direct drug discovery implications. KRAS, for example, has its ligand-binding affinity kicked around all over the place by phosphorylation state, which makes you wonder about what’s really going on in the cell with the ligands that people have spent so much time working on. (That’s in addition, of course, to the many changes in KRAS activity in general that are driven by those same modifications). There may be chances to optimize drug targeting to specific cellular states - well, once we work out what those are and what they mean!
Well, I know by now what happens when I bring up vaccines, but I’m going to do it anyway because we have a lot of news. As many will have heard, news broke recently of several large-scale studies of safety for the coronavirus and shingles vaccines. Here’s the New York Times on the story, and here’s the Guardian. These were done by evaluating millions of patient records, and seem to have concluded that these vaccines are in fact safe, with serious side effects being very rare across several populations examined. Here’s an abstract for one of them that remains online. These results are (1) expected and (2) still very good to see. And this is exactly the sort of work that should be done as newer vaccines are rolled out, because although clinical trials are extensive, nothing is as extensive as millions of patients out there in the real world. Reviewing safety on that scale is obviously good practice, and obviously money and effort well spent. But it turns out the the FDA has blocked publication of all of these studies, even though two of the coronavirus ones had already been accepted at a journal. The Times got a comment from HHS on this: “The studies were withdrawn because the authors drew broad conclusions that were not supported by the underlying data. The F.D.A. acted to protect the integrity of its scientific process and ensure that any work associated with the agency meets its high standards” And that is exactly the sort of bullshit that I would have expected. The higher-ups at HHS have been cloaking a lot of these decisions as issues of scientific integrity, reproducibility, and so on, when what’s really going on is obvious to any outside observer. There is broad, sustained opposition to vaccine development and deployment in this administration, from HHS Secretary Robert F. Kennedy Jr. on down, and there has been a series of decisions that all point in that same direction. Squashing publication of studies that help to confirm vaccine safety is absolutely on brand. The Times article mentions several instances of this, with a recent example being the cancellation by Jay Bhattacharya at CDC of a publication on the efficacy of coronavirus vaccines. (Howdy, Dr. B! Feel like taking a crack at answering the question I posed to you the other day?) Meanwhile, officials like Vinay Prasad, recently departed from the FDA and thank God, feel perfectly free to make statements about deaths from vaccines that they utterly refuse to back up with any data at all. It’s a scandal - a crime - and under any sort of normal circumstances careers would be ending over it. But here we are. This environment makes this news of great (and teeth-gritting) interest. Back earlier this year the FDA refused Moderna’s application outright for their mRNA-based flu vaccine. That came after the agency last year asked the company for more data about their combination flu-coronavirus mRNA vaccine shot, which I note has just been approved for use in Europe. But you can’t get it here, because RFK Jr. doesn’t want you to have it. The agency based its refusal-to-file for the flu vaccine on disagreements about the trial design in older patients. Well, Moderna has published its data on this population in the New England Journal of Medicine, and they have shown their vaccine to be superior to the standard-of-care comparison vaccine (GSK’s Fluarix). Vinay Prasad’s signed objection to accepting the FDA application said that the agency wanted to see a comparator to a different approved flu vaccine, but I have to classify that as more bullshit, cloaking a blanket opposition to vaccines (especially mRNA-based ones) as some sort of concern for scientific integrity. The safety signals for the Moderna flu vaccines showed more redness at the injection site and more fatigues, both of which were transient, and both of which are very much worth putting up with for better disease prevention. Again, under normal conditions these data would be very strong evidence for FDA approval of a vaccine candidate. But we’re a long way from normal here: RFK Jr. already refers to the existing mRNA coronavirus vaccines as some sort of deadly poisons and the agency has canceled hundreds of millions of dollars of funding for research in the area. I will be very surprised indeed if this vaccine approval goes through in any sort of decent fashion - i.e., without unfounded accusations, specious objections, and pointless foot-dragging. I hope I’m wrong. Watch it get approved in Europe, though, like the combination vaccine already mentioned above. Watch the rest of the world pull away from us in research that we used to lead the world in. Watch us slip further and further behind in medicine, in public health, and in science in general. And for what? For what? It’s all just so sad and stupid and pointless.
Show full content
Well, I know by now what happens when I bring up vaccines, but I’m going to do it anyway because we have a lot of news. As many will have heard, news broke recently of several large-scale studies of safety for the coronavirus and shingles vaccines. Here’s the New York Times on the story, and here’s the Guardian. These were done by evaluating millions of patient records, and seem to have concluded that these vaccines are in fact safe, with serious side effects being very rare across several populations examined. Here’s an abstract for one of them that remains online.
These results are (1) expected and (2) still very good to see. And this is exactly the sort of work that should be done as newer vaccines are rolled out, because although clinical trials are extensive, nothing is as extensive as millions of patients out there in the real world. Reviewing safety on that scale is obviously good practice, and obviously money and effort well spent. But it turns out the the FDA has blocked publication of all of these studies, even though two of the coronavirus ones had already been accepted at a journal.
The Times got a comment from HHS on this: “The studies were withdrawn because the authors drew broad conclusions that were not supported by the underlying data. The F.D.A. acted to protect the integrity of its scientific process and ensure that any work associated with the agency meets its high standards” And that is exactly the sort of bullshit that I would have expected. The higher-ups at HHS have been cloaking a lot of these decisions as issues of scientific integrity, reproducibility, and so on, when what’s really going on is obvious to any outside observer.
There is broad, sustained opposition to vaccine development and deployment in this administration, from HHS Secretary Robert F. Kennedy Jr. on down, and there has been a series of decisions that all point in that same direction. Squashing publication of studies that help to confirm vaccine safety is absolutely on brand. The Times article mentions several instances of this, with a recent example being the cancellation by Jay Bhattacharya at CDC of a publication on the efficacy of coronavirus vaccines. (Howdy, Dr. B! Feel like taking a crack at answering the question I posed to you the other day?) Meanwhile, officials like Vinay Prasad, recently departed from the FDA and thank God, feel perfectly free to make statements about deaths from vaccines that they utterly refuse to back up with any data at all. It’s a scandal - a crime - and under any sort of normal circumstances careers would be ending over it. But here we are.
This environment makes this news of great (and teeth-gritting) interest. Back earlier this year the FDA refused Moderna’s application outright for their mRNA-based flu vaccine. That came after the agency last year asked the company for more data about their combination flu-coronavirus mRNA vaccine shot, which I note has just been approved for use in Europe. But you can’t get it here, because RFK Jr. doesn’t want you to have it. The agency based its refusal-to-file for the flu vaccine on disagreements about the trial design in older patients.
Well, Moderna has published its data on this population in the New England Journal of Medicine, and they have shown their vaccine to be superior to the standard-of-care comparison vaccine (GSK’s Fluarix). Vinay Prasad’s signed objection to accepting the FDA application said that the agency wanted to see a comparator to a different approved flu vaccine, but I have to classify that as more bullshit, cloaking a blanket opposition to vaccines (especially mRNA-based ones) as some sort of concern for scientific integrity. The safety signals for the Moderna flu vaccines showed more redness at the injection site and more fatigues, both of which were transient, and both of which are very much worth putting up with for better disease prevention.
Again, under normal conditions these data would be very strong evidence for FDA approval of a vaccine candidate. But we’re a long way from normal here: RFK Jr. already refers to the existing mRNA coronavirus vaccines as some sort of deadly poisons and the agency has canceled hundreds of millions of dollars of funding for research in the area. I will be very surprised indeed if this vaccine approval goes through in any sort of decent fashion - i.e., without unfounded accusations, specious objections, and pointless foot-dragging. I hope I’m wrong.
Watch it get approved in Europe, though, like the combination vaccine already mentioned above. Watch the rest of the world pull away from us in research that we used to lead the world in. Watch us slip further and further behind in medicine, in public health, and in science in general. And for what? For what? It’s all just so sad and stupid and pointless.
When appropriate I usually wait until the end of one of these posts to draw attention to the way that its subject dovetails (or not) with the current fashion for AI/ML techniques. For one thing, I find it helpful to remember that really new results cannot generally be obtained by asking an LLM system that is trained on piles of existing text which it has blended and chopped and extruded back to you. That’s because those new results were not in there to start with, and LLMs are not in the business of reasoning for you about the gaps therein. I also find it useful to think about new results in terms of the AI/ML efforts to construct a “virtual cell” that (in theory) would allow you to do research on it rather than mess around with the persnickety live ones any more. I can’t say that that goal is impossible, but I think we are a long, long way away from it because there are just so many things that we don’t even realize that cells are doing. This new paper is a fine example. I’ve written here before about ubiquitin, that handy small protein that gets appended (in various forms) to so many other proteins for so many purposes (doubtless including many that we haven’t worked out yet). It’s often a waste disposal tag, signaling that the labeled protein is to be hauled off to the proteasome and demolished for parts. But that’s not always the case: there are proteins whose stability is enhanced by particular sites and modes of ubiquitination, and whose binding partners and functions can be tuned by it. So far, so complex. But it turns out that we didn’t realize the half of it. The reference above discusses a new assay workflow for general detection of ubiquitinated species, and it is revealing a lot of them. The paper focuses on one of these, ubiquitinated glycogen, and that phrase will probably be new to most people (it sure was for me). I had no idea that carbohydrate polymers like that could be ubiquitinated at all, much less that it’s apparently a major process for glycogen handling and disposal. Such reactions had been reported sporadically over the years, generally in vitro, but are poorly understood in cells. The authors find that this ubiquitination can deliver glycogen to lysosomes for destruction. That pathway has been recognized for a long time, but no one realized that this was a signal for it. Indeed, these ubiquitination pathways appear to be altered in the classic glycogen storage diseases, and their activity is elevated during fasting, when glycogen is consumed. How this fits into the existing knowledge of glycogen consumption is being worked out - for now, basic questions like what sort of ubiquitin chains get attached and what sorts of proteins recognize these species are yet to be answered. But it seems clear that a major aspect of glycogen handling has been missed up until now. That’s enough to keep anyone busy for quite a while, especially given the possible insights into therapies for the glycogen storage diseases. But there will be more results like this coming, because the authors say that “a surprisingly high abundance and diversity of non-proteinaceous material” turns out to be ubiquitinated. Who knew? Well, not us, to any great extent, that’s for sure. We’re going to have to rework some of our ideas in light of these results. I’m sure that any useful Virtual Cell model will incorporate these pathways, because they do seem to be important. But remember, they had to be discovered by good ol’ research at the lab bench for that to happen.
Show full content
When appropriate I usually wait until the end of one of these posts to draw attention to the way that its subject dovetails (or not) with the current fashion for AI/ML techniques. For one thing, I find it helpful to remember that really new results cannot generally be obtained by asking an LLM system that is trained on piles of existing text which it has blended and chopped and extruded back to you. That’s because those new results were not in there to start with, and LLMs are not in the business of reasoning for you about the gaps therein.
I also find it useful to think about new results in terms of the AI/ML efforts to construct a “virtual cell” that (in theory) would allow you to do research on it rather than mess around with the persnickety live ones any more. I can’t say that that goal is impossible, but I think we are a long, long way away from it because there are just so many things that we don’t even realize that cells are doing.
This new paper is a fine example. I’ve written here before about ubiquitin, that handy small protein that gets appended (in various forms) to so many other proteins for so many purposes (doubtless including many that we haven’t worked out yet). It’s often a waste disposal tag, signaling that the labeled protein is to be hauled off to the proteasome and demolished for parts. But that’s not always the case: there are proteins whose stability is enhanced by particular sites and modes of ubiquitination, and whose binding partners and functions can be tuned by it. So far, so complex.
But it turns out that we didn’t realize the half of it. The reference above discusses a new assay workflow for general detection of ubiquitinated species, and it is revealing a lot of them. The paper focuses on one of these, ubiquitinated glycogen, and that phrase will probably be new to most people (it sure was for me). I had no idea that carbohydrate polymers like that could be ubiquitinated at all, much less that it’s apparently a major process for glycogen handling and disposal. Such reactions had been reported sporadically over the years, generally in vitro, but are poorly understood in cells.
The authors find that this ubiquitination can deliver glycogen to lysosomes for destruction. That pathway has been recognized for a long time, but no one realized that this was a signal for it. Indeed, these ubiquitination pathways appear to be altered in the classic glycogen storage diseases, and their activity is elevated during fasting, when glycogen is consumed. How this fits into the existing knowledge of glycogen consumption is being worked out - for now, basic questions like what sort of ubiquitin chains get attached and what sorts of proteins recognize these species are yet to be answered. But it seems clear that a major aspect of glycogen handling has been missed up until now.
That’s enough to keep anyone busy for quite a while, especially given the possible insights into therapies for the glycogen storage diseases. But there will be more results like this coming, because the authors say that “a surprisingly high abundance and diversity of non-proteinaceous material” turns out to be ubiquitinated. Who knew? Well, not us, to any great extent, that’s for sure. We’re going to have to rework some of our ideas in light of these results. I’m sure that any useful Virtual Cell model will incorporate these pathways, because they do seem to be important. But remember, they had to be discovered by good ol’ research at the lab bench for that to happen.
OK, let’s talk helium today. A while back I wrote about the Haber-Bosch process for making ammonia (and thus making nitrogen fertilizer), and how that was already being impacted by the then fresh and new Iran war. Well here we are in the first week of May, and urea prices can best be described as “high and choppy”, jumping around (as does crude oil) on news of the war. Here’s the freight-on-board futures price for the Middle East, and if you’re a buyer in that market, it’s ugly. I mentioned in that post that there were other commodities that are affected in much the same way - energy-intensive materials that are coming out of the petrochemical zone of the Persian Gulf and have to make their way through the Strait of Hormuz somehow. Aluminum is one of those, because aluminum refining uses a ferocious amount of electricity, making it most feasible wherever there’s serious generating capacity. That one’s worth a post on its own, not least because we’re still basically using the method discovered in 1886. But helium is also on the list for some unique reasons. It’s an extremely uncommon gas in the atmosphere (down in the parts-per-million range) for obvious reasons - it’s so light that it goes straight into the upper atmosphere and escapes into space, primarily at the two magnetic poles. The best estimate on that is about 50g of helium per second being blown away from our planet entirely. This explains why no chemists or physicists even realized for many years that helium existed at all! It was the invention of emission/absorption spectroscopy that tipped everyone off. In 1868, several pioneering observers noted a bright yellow line in the solar spectrum that didn’t correspond to any assignment for a known element, but only Norman Lockyear was bold enough to hypothesize that it was indeed an element in the sun that we hadn’t encountered here on Earth. He thus chose “helium” as the name after the Greek “helios” (sun). It wasn’t detected in any terrestrial samples until 1881, and the most reliable sources were from uranium ores. It took until 1907 to prove that radioactive alpha particles (as given off by those ores) were in fact helium nuclei, and by that time large amounts of the gas had been found in natural gas reservoirs in the US. I put that incident into The Chemistry Book, where a gas strike in Kansas did not ignite, but rather extinguished all attempts to flame it off because of its high nitrogen and helium content (which startled onlookers at the time!) And that is indeed where underground helium seems to come from - radioactive decay from deep rocks, where it can accumulate with the right geology. You really need a solid nonporous rock layer on top, the stuff will seep right up through shale, for example. But now that you have it, you have to purify it, and that’s where the energy-intensive part comes in. The only feasible way to do that is through low-temperature distillation, where you get things down cold enough that only the helium can boil off. So you need big natural gas deposits (with reasonable helium concentrations, the higher the better), and you use some of that natural gas to generate the electricity needed for the serious chilling needed in production. The US is still the biggest helium producer in the world (about 43% of the global total) but Qatar is right behind it, with 33%. Next is Russia, followed by Algeria and Canada. So that means that one-third of global helium production is now literally bottled up behind the strait of Hormuz, a much higher percentage than world oil production (where the situation is bad enough already). On top of that, Qatari natural gas facilities have been taking Iranian drone and missile strikes, which is not going to help the supply chain much, either. So what does all this helium get used for? Well, it ain’t party balloons, not as makes any difference on a global scale. Chemists famously tend to cringe at the party-store helium tanks when they think about their own uses for the gas, but that's a minor effect. The biggest single use is cryogenics, since liquid helium is at such a ridiculously low temperature. That’s what all the big superconducting magnets in NMR and MRI machines use to keep operating. It’s also used as a totally inert shielding gas in high-temperature welding and in refining metals like titanium, and as a protective gas when growing silicon and germanium crystals for semiconductors. It's also used as an inert carrier/purge gas in rocketry and in gas chromatography. You will note that major chipmaking countries like Taiwan and South Korea have little or no domestic production, and both of them have had Qatar as their largest supplier. Helium is unique - there is no replacement gas in these applications, so the price is the price, and you pay it or you shut things down. NMR machines and the like, for example, now have helium-recovery equipment attached to them, whereas in my earlier years in the business that sort of equipment was rarely seen. This is not a commodity with easy-to-access trading prices, but the war has sent it sharply upwards. Things were able to run on storage reserves for a while, but that grace period, such as it was, is running out. I’m hearing reports of the major suppliers (Airgas, Linde, Air Liquide et al.) notifying industrial, medical, and academic customers of real shortages and of outright rationing. Semiconductors and medical MRIs seem to have first call on the available supplies, from what I can see, but the prices are going to hit everyone even if you can get deliveries. And because of the damage to Qatar’s infrastructure, even reopening the strait by magic (which seems to be the Trump administration’s main hope) will not return things to status quo ante. There are other places in the world with helium deposits - what appears to be a major one was recently found in Minnesota’s Iron Range, for example), but these are not going to come online soon enough to avoid major disruptions. We’re going to be in a world of expensive helium for some time, and we’re all going to get a chance to see what that means. Unfortunately.
Show full content
OK, let’s talk helium today. A while back I wrote about the Haber-Bosch process for making ammonia (and thus making nitrogen fertilizer), and how that was already being impacted by the then fresh and new Iran war. Well here we are in the first week of May, and urea prices can best be described as “high and choppy”, jumping around (as does crude oil) on news of the war. Here’s the freight-on-board futures price for the Middle East, and if you’re a buyer in that market, it’s ugly.
I mentioned in that post that there were other commodities that are affected in much the same way - energy-intensive materials that are coming out of the petrochemical zone of the Persian Gulf and have to make their way through the Strait of Hormuz somehow. Aluminum is one of those, because aluminum refining uses a ferocious amount of electricity, making it most feasible wherever there’s serious generating capacity. That one’s worth a post on its own, not least because we’re still basically using the method discovered in 1886.
But helium is also on the list for some unique reasons. It’s an extremely uncommon gas in the atmosphere (down in the parts-per-million range) for obvious reasons - it’s so light that it goes straight into the upper atmosphere and escapes into space, primarily at the two magnetic poles. The best estimate on that is about 50g of helium per second being blown away from our planet entirely. This explains why no chemists or physicists even realized for many years that helium existed at all!
It was the invention of emission/absorption spectroscopy that tipped everyone off. In 1868, several pioneering observers noted a bright yellow line in the solar spectrum that didn’t correspond to any assignment for a known element, but only Norman Lockyear was bold enough to hypothesize that it was indeed an element in the sun that we hadn’t encountered here on Earth. He thus chose “helium” as the name after the Greek “helios” (sun). It wasn’t detected in any terrestrial samples until 1881, and the most reliable sources were from uranium ores. It took until 1907 to prove that radioactive alpha particles (as given off by those ores) were in fact helium nuclei, and by that time large amounts of the gas had been found in natural gas reservoirs in the US. I put that incident into The Chemistry Book, where a gas strike in Kansas did not ignite, but rather extinguished all attempts to flame it off because of its high nitrogen and helium content (which startled onlookers at the time!)
And that is indeed where underground helium seems to come from - radioactive decay from deep rocks, where it can accumulate with the right geology. You really need a solid nonporous rock layer on top, the stuff will seep right up through shale, for example. But now that you have it, you have to purify it, and that’s where the energy-intensive part comes in. The only feasible way to do that is through low-temperature distillation, where you get things down cold enough that only the helium can boil off. So you need big natural gas deposits (with reasonable helium concentrations, the higher the better), and you use some of that natural gas to generate the electricity needed for the serious chilling needed in production.
The US is still the biggest helium producer in the world (about 43% of the global total) but Qatar is right behind it, with 33%. Next is Russia, followed by Algeria and Canada. So that means that one-third of global helium production is now literally bottled up behind the strait of Hormuz, a much higher percentage than world oil production (where the situation is bad enough already). On top of that, Qatari natural gas facilities have been taking Iranian drone and missile strikes, which is not going to help the supply chain much, either. So what does all this helium get used for?
Well, it ain’t party balloons, not as makes any difference on a global scale. Chemists famously tend to cringe at the party-store helium tanks when they think about their own uses for the gas, but that's a minor effect. The biggest single use is cryogenics, since liquid helium is at such a ridiculously low temperature. That’s what all the big superconducting magnets in NMR and MRI machines use to keep operating. It’s also used as a totally inert shielding gas in high-temperature welding and in refining metals like titanium, and as a protective gas when growing silicon and germanium crystals for semiconductors. It's also used as an inert carrier/purge gas in rocketry and in gas chromatography. You will note that major chipmaking countries like Taiwan and South Korea have little or no domestic production, and both of them have had Qatar as their largest supplier. Helium is unique - there is no replacement gas in these applications, so the price is the price, and you pay it or you shut things down.
NMR machines and the like, for example, now have helium-recovery equipment attached to them, whereas in my earlier years in the business that sort of equipment was rarely seen. This is not a commodity with easy-to-access trading prices, but the war has sent it sharply upwards. Things were able to run on storage reserves for a while, but that grace period, such as it was, is running out. I’m hearing reports of the major suppliers (Airgas, Linde, Air Liquide et al.) notifying industrial, medical, and academic customers of real shortages and of outright rationing. Semiconductors and medical MRIs seem to have first call on the available supplies, from what I can see, but the prices are going to hit everyone even if you can get deliveries.
And because of the damage to Qatar’s infrastructure, even reopening the strait by magic (which seems to be the Trump administration’s main hope) will not return things to status quo ante. There are other places in the world with helium deposits - what appears to be a major one was recently found in Minnesota’s Iron Range, for example), but these are not going to come online soon enough to avoid major disruptions. We’re going to be in a world of expensive helium for some time, and we’re all going to get a chance to see what that means. Unfortunately.
Dear Dr. Bhattacharya: I write to inquire about your state of mind. That may seem an unusual request, but these are unusual times, and God knows you have been appointed to serve in an extremely unusual administration. As Director of the NIH (a post - let’s be frank here - that for many years I’m sure you never seriously imagined that you might hold) you have of course a great many responsibilities, facing not only your direct reports and employees, but also the general public and of course your own superiors in HHS and above. And these are what lead me to write. Let me make clear, I’m actually not going to go into the state of the NIH under your tenure, nor into your actions while leading it. That’s not for any lack of questions on my part (where to begin?) but I note that Jeremy Berg among others have repeatedly asked about these topics and in great detail, and I think it’s safe to say that they probably do not feel as if they’ve had satisfactory answers from you. I don’t see how I would fare any differently - do you? And you probably already know what my point of view on these topics is; I haven’t been shy. Put simply, I think the Trump administration has been the single biggest disaster for US scientific research in this country’s history, and that what is being done to the NIH, NSF, and other science agencies is an historic crime. But you no doubt disagree. So as much as it pains me to walk past this topic, that’s what I’m going to do. No, I’d like to ask you about some of your statements regarding the administration and those that you report to. That’s because - try as I might - I have been unable to put myself into a frame of mind where I could imagine saying these things myself. That might sound like a foregone conclusion, but it isn’t. All of us have self-images and stories that we tell ourselves about our own behaviors, and I actually think it’s likely that you and I share a few of these ideas about our own behavior. Specifically, I think we probably tell ourselves that we are unafraid to speak unpopular truths and that we are unwilling to compromise our principles. Admirable stuff, am I right? Probably helps both of us sleep better. For example, I feel sure that you believed that you were doing just this when you spoke out during the coronavirus pandemic, for example. No worries, I’m going to bypass that topic, too. Many people have tried to pin you down on it over the last few years, and I won’t flatter myself to think that I would do any better than they have. As you can well imagine, I have a lot to say on this topic (and on your conduct) but I’ll spare you and spare the readership. But let me just ask you about one incident, because I think it illustrates some larger issues. You seem to spend a fair amount of time on Fox News for someone who’s as busy as you must be, but I recall you saying some months ago there that your own supervisor, Robert F. Kennedy Jr., is “devoted to the scientific method”. That has stuck in my head. Perhaps it’s odd that it should, given all the other public statements you’ve made that I’m in profound disagreement with, but as mentioned above I’m taking a lot of those topics off the table today. Let’s just think about that one for a minute. RFK Jr., as he makes plain in his awful 2021 book The Real Anthony Fauci, does not believe in the germ theory of disease. He seems to regard the idea of infectious pathogens spreading disease as a confidence trick played by the medical establishment and the pharmaceutical industry, and he advocates what he calls “miasma theory" instead. As noted here, he doesn’t quite land on the accepted definition of that term, either, but in general he believes that “fortifying the immune system” with healthy living and nutrition is the key to fighting disease. This fits in with his numerous past statements to the effect that children who die from infections are in fact malnourished and that vaccines themselves are at best useless and at worst harmful. I could continue. These are not one-offs or misstatements that the man has made in the heat of the moment. As much as one can be sure of another person’s beliefs, these certainly seem to be his. Dr. Bhattacharya, I’m not going to insult you by asking if you believe in the germ theory of disease or not. But what I am going to ask you is whether you think that your supervisor’s statements are, in fact, those of someone who is devoted to the scientific method. Because naturally enough, I can’t make these things add up. And this inability to square things is something that one runs into over and over when dealing with the current administration. It seems that every day of the week there are high-ranking officials out there telling us that oil and gasoline prices are dropping, that inflation is under control, that the President’s approval ratings are fantastically high, that the Strait of Hormuz is open/closed/ours/theirs/mined/blockaded/not a problem anyway, and that we’re entering a wonderful golden age in just about every respect. I could continue. But the only way to really describe these statements is that they are lies. Whoppers, blatant straight-faced falsehoods. Now I’m certainly not calling you to account for all of these! You have nothing to do with foreign policy, interest rates, and so on. But may I suggest that your statements at times have things in common with these others, probably more than you would feel comfortable with if you stopped to think about it. I’m giving you the benefit of the doubt here, as you may notice; you can make your own call as to how justified that is. I'll put it simply: as part of the Trump administration, you are surrounded by liars, Dr. Bhattacharya. It’s sad and unfortunate, but it’s true. This, like the rotting of a mackeral, works from the top down: our President lies constantly, widely, and vigorously about almost every topic that comes to his mind. How does working in this environment fit in with what I believe to be your own worldview, i.e. that you yourself are a truth-teller? Saying that Robert F. Kennedy is devoted to the scientific method does not help you make your case, in my own opinion. Is this something that bothers you in any way? I said earlier that I believed that some of our own self-images might have more similar features than one would think, but here is where that comparison might well break down. Because I don’t think that I could ever make peace with myself about that. Thoughts? You know, I'm going to shore up the disclaimer that appears with all these posts. Keep in mind that In the Pipeline is editorially independent from Science, and that everything that I write here is from me and reflects my own opinions. If they reflect yours, enjoy! If not, there's always the next post.
Show full content
Dear Dr. Bhattacharya:
I write to inquire about your state of mind. That may seem an unusual request, but these are unusual times, and God knows you have been appointed to serve in an extremely unusual administration. As Director of the NIH (a post - let’s be frank here - that for many years I’m sure you never seriously imagined that you might hold) you have of course a great many responsibilities, facing not only your direct reports and employees, but also the general public and of course your own superiors in HHS and above. And these are what lead me to write.
Let me make clear, I’m actually not going to go into the state of the NIH under your tenure, nor into your actions while leading it. That’s not for any lack of questions on my part (where to begin?) but I note that Jeremy Berg among others have repeatedly asked about these topics and in great detail, and I think it’s safe to say that they probably do not feel as if they’ve had satisfactory answers from you. I don’t see how I would fare any differently - do you? And you probably already know what my point of view on these topics is; I haven’t been shy. Put simply, I think the Trump administration has been the single biggest disaster for US scientific research in this country’s history, and that what is being done to the NIH, NSF, and other science agencies is an historic crime. But you no doubt disagree. So as much as it pains me to walk past this topic, that’s what I’m going to do.
No, I’d like to ask you about some of your statements regarding the administration and those that you report to. That’s because - try as I might - I have been unable to put myself into a frame of mind where I could imagine saying these things myself. That might sound like a foregone conclusion, but it isn’t. All of us have self-images and stories that we tell ourselves about our own behaviors, and I actually think it’s likely that you and I share a few of these ideas about our own behavior. Specifically, I think we probably tell ourselves that we are unafraid to speak unpopular truths and that we are unwilling to compromise our principles. Admirable stuff, am I right? Probably helps both of us sleep better.
For example, I feel sure that you believed that you were doing just this when you spoke out during the coronavirus pandemic, for example. No worries, I’m going to bypass that topic, too. Many people have tried to pin you down on it over the last few years, and I won’t flatter myself to think that I would do any better than they have. As you can well imagine, I have a lot to say on this topic (and on your conduct) but I’ll spare you and spare the readership.
But let me just ask you about one incident, because I think it illustrates some larger issues. You seem to spend a fair amount of time on Fox News for someone who’s as busy as you must be, but I recall you saying some months ago there that your own supervisor, Robert F. Kennedy Jr., is “devoted to the scientific method”. That has stuck in my head. Perhaps it’s odd that it should, given all the other public statements you’ve made that I’m in profound disagreement with, but as mentioned above I’m taking a lot of those topics off the table today. Let’s just think about that one for a minute.
RFK Jr., as he makes plain in his awful 2021 book The Real Anthony Fauci, does not believe in the germ theory of disease. He seems to regard the idea of infectious pathogens spreading disease as a confidence trick played by the medical establishment and the pharmaceutical industry, and he advocates what he calls “miasma theory" instead. As noted here, he doesn’t quite land on the accepted definition of that term, either, but in general he believes that “fortifying the immune system” with healthy living and nutrition is the key to fighting disease. This fits in with his numerous past statements to the effect that children who die from infections are in fact malnourished and that vaccines themselves are at best useless and at worst harmful. I could continue. These are not one-offs or misstatements that the man has made in the heat of the moment. As much as one can be sure of another person’s beliefs, these certainly seem to be his.
Dr. Bhattacharya, I’m not going to insult you by asking if you believe in the germ theory of disease or not. But what I am going to ask you is whether you think that your supervisor’s statements are, in fact, those of someone who is devoted to the scientific method. Because naturally enough, I can’t make these things add up. And this inability to square things is something that one runs into over and over when dealing with the current administration. It seems that every day of the week there are high-ranking officials out there telling us that oil and gasoline prices are dropping, that inflation is under control, that the President’s approval ratings are fantastically high, that the Strait of Hormuz is open/closed/ours/theirs/mined/blockaded/not a problem anyway, and that we’re entering a wonderful golden age in just about every respect. I could continue. But the only way to really describe these statements is that they are lies. Whoppers, blatant straight-faced falsehoods.
Now I’m certainly not calling you to account for all of these! You have nothing to do with foreign policy, interest rates, and so on. But may I suggest that your statements at times have things in common with these others, probably more than you would feel comfortable with if you stopped to think about it. I’m giving you the benefit of the doubt here, as you may notice; you can make your own call as to how justified that is.
I'll put it simply: as part of the Trump administration, you are surrounded by liars, Dr. Bhattacharya. It’s sad and unfortunate, but it’s true. This, like the rotting of a mackeral, works from the top down: our President lies constantly, widely, and vigorously about almost every topic that comes to his mind. How does working in this environment fit in with what I believe to be your own worldview, i.e. that you yourself are a truth-teller? Saying that Robert F. Kennedy is devoted to the scientific method does not help you make your case, in my own opinion. Is this something that bothers you in any way? I said earlier that I believed that some of our own self-images might have more similar features than one would think, but here is where that comparison might well break down. Because I don’t think that I could ever make peace with myself about that.
Thoughts?
You know, I'm going to shore up the disclaimer that appears with all these posts. Keep in mind that In the Pipeline is editorially independent from Science, and that everything that I write here is from me and reflects my own opinions. If they reflect yours, enjoy! If not, there's always the next post.
We have quite a regulatory situation developing around a drug called avacopan (Tavneos), which is given to patients with a particular type of vaculitis. That’s a complex disease area, and comes in several varieties, but a common theme in many of them is an autoimmune attack against various proteins found in neutrophils. The drug is an antagonist of the complement 5a receptor in the innate immune system, and it’s given along with other immunosuppressants. It was developed recently by a company called ChemoCentryx, who ran a pivotal trial in 330 patients. Half got the standard of care plus a placebo and half got that standard plus the drug over 52 weeks, and the endpoints were remission at 26 and 52 weeks. Both endpoints were similar, and both showed a real benefit to the patients. Tavneos was approved by both the FDA and the European CHMP in 2022, and Amgen went on to purchase the entire company. But earlier this year, the CHMP announced that they were starting an investigation based on reports of loss of data integrity in that trial. And the CDER at the FDA is proposing to have it withdrawn from the US market over the same issues. It’s very, very bad - here’s the FDA statement: . . .new information that only became known to CDER more than three years after approval shows that unblinded study personnel manipulated the results of the pivotal clinical study so the drug looked effective when the original analysis did not support that conclusion. The applicant also did not disclose the original analysis to FDA, in violation of FDA regulations. CDER can no longer conclude that there is, or has ever been, a valid demonstration that TAVNEOS is effective for its approved use. Ohhh boy. This is about as bad an accusation as you can make about a clinical trial, i.e. “The unblinded data were ugly, so we hocused the numbers until it looked like the drug worked”. I occasionally meet uninformed cynics who assume that this is how we always do things in the drug industry, but oh no, we don’t. We have an 85% failure rate in the clinic! Why would any clinical trial ever fail if we had constant recourse to bullshit like this? For more details, this FDA document at the Federal Register is the place to go. This all came to light due to a lawsuit against the company for securities fraud, which included a consultant’s report about the avacopan/Tavneos trial process. That all came about because during the initial approval process ChemoCentryx made numerous public statement about how straightforward the trial was and how uneventful their interactions with the FDA had been, but in May of 2021 the FDA review committee hearing instead detailed a whole list of pointed questions the agency had had about the trial design and the interpretability of the results. That sent the stock down about 80%, and that will get you a shareholder lawsuit every time. But as it turns out, the hapless shareholders don’t seem to have known the half of it. The consultant report introduced as evidence during the lawsuit claims that the initial analysis of the clinical trial showed it missing the primary endpoint, and that the company picked a number of cases for “readjudication”. Wouldntjaknowit, enough of these flipped over to positive during this reanalysis to cause the whole trial to meet its statistics. The report says that ChemoCentryx employees stated as much even before the reworking, calculating how many patient outcomes would need to be flipped. But none of this was disclosed in any way to the FDA, obviously, and yes, that is all flagrantly illegal if it’s what happened. The FDA says that it requested a detailed account of the data handling for the trial, and that Amgen’s response a month later “confirmed the key factual allegations” above. But the company went on to claim that the data in the NDA are accurate and that the patient readjudications were appropriate. (As it turns out, the lawsuit was later dismissed without ever addressing these accusations, so it doesn’t have any bearing on this situation). The case for the data changes being valid rests largely on patient glucocorticoid dosing or missing data, and I won’t get into the merits of that argument. But what seems beyond doubt is that ChemoCentryx made sure that the FDA never heard about it and made sure to submit only the freshly polished data set. Such readjudication-after-unblinding was, as you would imagine, absolutely not permitted under the study protocols. The FDA notes that one of the patients was initially marked down by ChemoCentryx as a non-responder due to missing data at week 26, but that same patient got helpfully moved to the “in remission” category after the unblinding. That’s precisely why you are not supposed to do that sort of thing. There’s even more bad news: not only are there doubts about the efficacy, the safety profile is looking bad, too. The FDA has received numerous reports of liver toxicity, which was a concern even during approval (Tavneos already has a label warning to that effect). But since getting on the market there have been several fatalities, some of which involve the extremely-alarmingly-named “vanishing bile duct syndrome”, which I had never heard of until now. It’s just what it sounds like: the bile ducts through the liver tissue deteriorate and disappear, and that is clearly just as bad as you think it is. Most of those have been reported in Japan, for reasons unknown, but man, you don’t want that showing up anywhere. I feel pretty sure that a “Whatever happened to the bile ducts” finding during the trials would have shut things down right there. One has the impression that Amgen might have been better off if they'd never heard of this drug at all instead of being shackled to defending it. The drug was approved with a requirement for a postmarketing study covering both safety and efficacy, but the FDA says that as of the most recent report, only 21 of the planned 300 patients had been enrolled. I’ve written about this problem before - too often, companies have an incentive to draaaaag their feet on these requirements and collect the data at glacial speed, and that may be just what we’re seeing here. But as it stands, I would agree with the FDA’s contention that the trial results “are uninterpretable and cannot be salvaged with further analysis”. What’s more, it would seem that the company made materially false statements to the agency during the application process. This is all really, really unfortunate - for vasculitis patients most of all, but for the integrity of the whole clinical trial and drug approval process as well. We really don’t need this sort of thing right now - hell, we never have, but especially not now.
Show full content
We have quite a regulatory situation developing around a drug called avacopan (Tavneos), which is given to patients with a particular type of vaculitis. That’s a complex disease area, and comes in several varieties, but a common theme in many of them is an autoimmune attack against various proteins found in neutrophils. The drug is an antagonist of the complement 5a receptor in the innate immune system, and it’s given along with other immunosuppressants.
It was developed recently by a company called ChemoCentryx, who ran a pivotal trial in 330 patients. Half got the standard of care plus a placebo and half got that standard plus the drug over 52 weeks, and the endpoints were remission at 26 and 52 weeks. Both endpoints were similar, and both showed a real benefit to the patients. Tavneos was approved by both the FDA and the European CHMP in 2022, and Amgen went on to purchase the entire company.
But earlier this year, the CHMP announced that they were starting an investigation based on reports of loss of data integrity in that trial. And the CDER at the FDA is proposing to have it withdrawn from the US market over the same issues. It’s very, very bad - here’s the FDA statement:
. . .new information that only became known to CDER more than three years after approval shows that unblinded study personnel manipulated the results of the pivotal clinical study so the drug looked effective when the original analysis did not support that conclusion. The applicant also did not disclose the original analysis to FDA, in violation of FDA regulations. CDER can no longer conclude that there is, or has ever been, a valid demonstration that TAVNEOS is effective for its approved use.
Ohhh boy. This is about as bad an accusation as you can make about a clinical trial, i.e. “The unblinded data were ugly, so we hocused the numbers until it looked like the drug worked”. I occasionally meet uninformed cynics who assume that this is how we always do things in the drug industry, but oh no, we don’t. We have an 85% failure rate in the clinic! Why would any clinical trial ever fail if we had constant recourse to bullshit like this?
For more details, this FDA document at the Federal Register is the place to go. This all came to light due to a lawsuit against the company for securities fraud, which included a consultant’s report about the avacopan/Tavneos trial process. That all came about because during the initial approval process ChemoCentryx made numerous public statement about how straightforward the trial was and how uneventful their interactions with the FDA had been, but in May of 2021 the FDA review committee hearing instead detailed a whole list of pointed questions the agency had had about the trial design and the interpretability of the results. That sent the stock down about 80%, and that will get you a shareholder lawsuit every time.
But as it turns out, the hapless shareholders don’t seem to have known the half of it. The consultant report introduced as evidence during the lawsuit claims that the initial analysis of the clinical trial showed it missing the primary endpoint, and that the company picked a number of cases for “readjudication”. Wouldntjaknowit, enough of these flipped over to positive during this reanalysis to cause the whole trial to meet its statistics. The report says that ChemoCentryx employees stated as much even before the reworking, calculating how many patient outcomes would need to be flipped. But none of this was disclosed in any way to the FDA, obviously, and yes, that is all flagrantly illegal if it’s what happened.
The FDA says that it requested a detailed account of the data handling for the trial, and that Amgen’s response a month later “confirmed the key factual allegations” above. But the company went on to claim that the data in the NDA are accurate and that the patient readjudications were appropriate. (As it turns out, the lawsuit was later dismissed without ever addressing these accusations, so it doesn’t have any bearing on this situation).
The case for the data changes being valid rests largely on patient glucocorticoid dosing or missing data, and I won’t get into the merits of that argument. But what seems beyond doubt is that ChemoCentryx made sure that the FDA never heard about it and made sure to submit only the freshly polished data set. Such readjudication-after-unblinding was, as you would imagine, absolutely not permitted under the study protocols. The FDA notes that one of the patients was initially marked down by ChemoCentryx as a non-responder due to missing data at week 26, but that same patient got helpfully moved to the “in remission” category after the unblinding. That’s precisely why you are not supposed to do that sort of thing.
There’s even more bad news: not only are there doubts about the efficacy, the safety profile is looking bad, too. The FDA has received numerous reports of liver toxicity, which was a concern even during approval (Tavneos already has a label warning to that effect). But since getting on the market there have been several fatalities, some of which involve the extremely-alarmingly-named “vanishing bile duct syndrome”, which I had never heard of until now. It’s just what it sounds like: the bile ducts through the liver tissue deteriorate and disappear, and that is clearly just as bad as you think it is. Most of those have been reported in Japan, for reasons unknown, but man, you don’t want that showing up anywhere. I feel pretty sure that a “Whatever happened to the bile ducts” finding during the trials would have shut things down right there. One has the impression that Amgen might have been better off if they'd never heard of this drug at all instead of being shackled to defending it.
The drug was approved with a requirement for a postmarketing study covering both safety and efficacy, but the FDA says that as of the most recent report, only 21 of the planned 300 patients had been enrolled. I’ve written about this problem before - too often, companies have an incentive to draaaaag their feet on these requirements and collect the data at glacial speed, and that may be just what we’re seeing here.
But as it stands, I would agree with the FDA’s contention that the trial results “are uninterpretable and cannot be salvaged with further analysis”. What’s more, it would seem that the company made materially false statements to the agency during the application process. This is all really, really unfortunate - for vasculitis patients most of all, but for the integrity of the whole clinical trial and drug approval process as well. We really don’t need this sort of thing right now - hell, we never have, but especially not now.
This article (open access) is the latest in a long, long series of implementations of an idea that is very simple to state and very difficult to achieve. That is, what if we (1) had a set of machines that could run organic chemistry reactions for us, ones that (2) could also analyze how well these reactions worked (yield, purity and so on), and that (3) could then use some sort of software evaluation of the data to set up another round of experiments based on those results, so as to (4) iteratively optimize the reactions themselves by successive rounds of improvement? This general scheme has been occurring to people for decades, naturally - I first saw a presentation on something like this at an ACS meeting in 1986, and it was not an original concept then. But it is not so easy to implement! I adduce blog posts here from 2009 here, 2013 here and here, 2014 here and here and here, 2015 here and here and here and here, 2016 here, 2018 here and here and here and here, 2019 here and here, 2020 here, along with 2021, 2022 here and here, 2024, and 2026. Man, I do go on. As both hardware and software have improved over the years the overall goal has come more and more into reach, though. Both of those had to become much more capable, but if I had to pick one hardware advance that really moved things closer it might be automated sample handling and data collection as applied to LC/MS analysis. That’s about as close as we have to a one-size-fits-all reaction monitoring method, and it’s done by microscale liquid handling, which is a great help. The new paper’s “RoboChem-Flex” system is the latest proposal for what are now being called by some “self-driving laboratories”. The authors here have done a lot of work in the area, and are good about noting some of the difficulties (sheer expense, the need for wide-ranging human expertise in even setting up such a system in the first place, and the possibility of exacerbating the “Matthew effect” of making high-powered labs even more high-powered at the expense of others. (That’s Matthew 13:12, “For whosoever hath, to him shall be given, and he shall have more abundance, but whosoever hath not, from him shall be taken away even that he hath”, or if you prefer, Matthew 25:29, which says darn near the same thing. Anyway. This work is trying to address these problems by making a lower-cost system that is easier to get off the ground, and that is really a difficult task. But it’s necessary, I’d say, and the sooner we get to some sort of standard data and operations format that a lot of people can use (and are using!) the better. Affordability and flexibility are achieved using three-dimensionally (3D)-printed or readily available subcomponents, which reduces costs considerably while allowing rapid customization and iterative development. Communication between hardware components is orchestrated by our open-source OmniPlatypus package, which ensures seamless modularity and enables a plug-and-play architecture with minimal coding effort required from the user. At the software level, RoboChem-Flex integrates a highly modular Bayesian optimization agent, allowing users to customize AI-driven optimization workflows to meet specific experimental goals. The platform also supports integration with a range of inline analytical instruments, including NMR spectroscopy, ultrahigh-performance liquid chromatography (UHPLC)–mass spectrometry and Raman spectroscopy, enabling fully closed-loop reaction optimization. Recognizing, however, that inline analytics may represent a sizeable investment, we have also developed a cost-effective, 3D-printed liquid sampling unit. This module enables the robot to collect reaction samples, which can then be analysed using existing, often departmentally shared, analytical equipment. This human-in-the-loop approach provides a practical and affordable entry point for laboratories, reducing the overall system cost to approximately US$5,000. Several really interesting examples are provided such as a photochemical trifluoromethylation, a photochemical deoxygenation/alkylation of a pyridine ring, several Buchwald-Hartwig aminations, and a 2+2 cycloaddition. There’s also an enzymatic reduction of a diketone, which uses the “human in the loop” workflow option as well. It’s notable that these tend to use several different optimization algorithms, and I would say that knowing which of these to use and when certainly falls in the “expertise” category. But in the end, the system seems to perform quite well. I am sure that there are the usual headaches along the way (leaks, clogged tubing, and so on) and the authors do mention some of this. We are not quite yet at the plug-n-play level of synthetic optimization. But neither are we (any more) quite in the bespoke mass-of-tubing-and-cables era; this paper is a deliberate attempt to get out of that. I applaud the initiative, and I will be watching with great interest to see where it goes. Putting this in a larger framework, I generally think of hardware and software advances in our field as “redefining grunt work” in the same way that it’s been redefined over and over for the last century or so. Here, the task that is heading towards the “automate-able grunt work” category is the repetitive optimization of reaction conditions - this solvent, that catalyst, at this temperature or that with a choice of these additives over here. Every organic chemist has done some of this sort of thing (and process chemists do a lot of it). And if you have had to do a lot of it, I don’t think you’re going to feel threatened or dehumanized by the prospect of some mechanical help. Thoughts?
Show full content
This article (open access) is the latest in a long, long series of implementations of an idea that is very simple to state and very difficult to achieve. That is, what if we (1) had a set of machines that could run organic chemistry reactions for us, ones that (2) could also analyze how well these reactions worked (yield, purity and so on), and that (3) could then use some sort of software evaluation of the data to set up another round of experiments based on those results, so as to (4) iteratively optimize the reactions themselves by successive rounds of improvement?
This general scheme has been occurring to people for decades, naturally - I first saw a presentation on something like this at an ACS meeting in 1986, and it was not an original concept then. But it is not so easy to implement! I adduce blog posts here from 2009 here, 2013 here and here, 2014 here and here and here, 2015 here and here and here and here, 2016 here, 2018 here and here and here and here, 2019 here and here, 2020 here, along with 2021, 2022 here and here, 2024, and 2026. Man, I do go on.
As both hardware and software have improved over the years the overall goal has come more and more into reach, though. Both of those had to become much more capable, but if I had to pick one hardware advance that really moved things closer it might be automated sample handling and data collection as applied to LC/MS analysis. That’s about as close as we have to a one-size-fits-all reaction monitoring method, and it’s done by microscale liquid handling, which is a great help.
The new paper’s “RoboChem-Flex” system is the latest proposal for what are now being called by some “self-driving laboratories”. The authors here have done a lot of work in the area, and are good about noting some of the difficulties (sheer expense, the need for wide-ranging human expertise in even setting up such a system in the first place, and the possibility of exacerbating the “Matthew effect” of making high-powered labs even more high-powered at the expense of others. (That’s Matthew 13:12, “For whosoever hath, to him shall be given, and he shall have more abundance, but whosoever hath not, from him shall be taken away even that he hath”, or if you prefer, Matthew 25:29, which says darn near the same thing. Anyway.
This work is trying to address these problems by making a lower-cost system that is easier to get off the ground, and that is really a difficult task. But it’s necessary, I’d say, and the sooner we get to some sort of standard data and operations format that a lot of people can use (and are using!) the better.
Affordability and flexibility are achieved using three-dimensionally (3D)-printed or readily available subcomponents, which reduces costs considerably while allowing rapid customization and iterative development. Communication between hardware components is orchestrated by our open-source OmniPlatypus package, which ensures seamless modularity and enables a plug-and-play architecture with minimal coding effort required from the user. At the software level, RoboChem-Flex integrates a highly modular Bayesian optimization agent, allowing users to customize AI-driven optimization workflows to meet specific experimental goals. The platform also supports integration with a range of inline analytical instruments, including NMR spectroscopy, ultrahigh-performance liquid chromatography (UHPLC)–mass spectrometry and Raman spectroscopy, enabling fully closed-loop reaction optimization. Recognizing, however, that inline analytics may represent a sizeable investment, we have also developed a cost-effective, 3D-printed liquid sampling unit. This module enables the robot to collect reaction samples, which can then be analysed using existing, often departmentally shared, analytical equipment. This human-in-the-loop approach provides a practical and affordable entry point for laboratories, reducing the overall system cost to approximately US$5,000.
Several really interesting examples are provided such as a photochemical trifluoromethylation, a photochemical deoxygenation/alkylation of a pyridine ring, several Buchwald-Hartwig aminations, and a 2+2 cycloaddition. There’s also an enzymatic reduction of a diketone, which uses the “human in the loop” workflow option as well. It’s notable that these tend to use several different optimization algorithms, and I would say that knowing which of these to use and when certainly falls in the “expertise” category. But in the end, the system seems to perform quite well.
I am sure that there are the usual headaches along the way (leaks, clogged tubing, and so on) and the authors do mention some of this. We are not quite yet at the plug-n-play level of synthetic optimization. But neither are we (any more) quite in the bespoke mass-of-tubing-and-cables era; this paper is a deliberate attempt to get out of that. I applaud the initiative, and I will be watching with great interest to see where it goes.
Putting this in a larger framework, I generally think of hardware and software advances in our field as “redefining grunt work” in the same way that it’s been redefined over and over for the last century or so. Here, the task that is heading towards the “automate-able grunt work” category is the repetitive optimization of reaction conditions - this solvent, that catalyst, at this temperature or that with a choice of these additives over here. Every organic chemist has done some of this sort of thing (and process chemists do a lot of it). And if you have had to do a lot of it, I don’t think you’re going to feel threatened or dehumanized by the prospect of some mechanical help. Thoughts?
Think about it: have you every heard of a case of heart cancer? It’s very rare indeed, and why that’s the case has been a longstanding puzzle. This new paper (here at Science) seems to have found a big piece of the answer, though. Cardiomyoctes are a rather special class of cell, but one of the fundamental things that make them special is that they are always in motion and under mechanical stress. If you raise them correctly in a cell culture dish, they will contract rhythmically even in there! And that is indeed the secret behind the lack of cancer in this tissue, apparently. The authors here made sure of the effect by overexpressing mutant K-Ras and deleting p53 in a mouse model, a terrible combination that resulting in multiple cancers in a range of tissues except the heart (even though it did indeed carry these mutations). Meanwhile, cells from various tumor types failed to proliferate as usual when implanted in cardiac tissue, but surgical remodeling of the heart vessels to decrease the mechanical load on the cardiac tissue allowed such tumors to take hold. Closer study of these tissues implicates chromatin remodeling, specifically methylation of the histone 3 lysine 9 (chromatin changes in general are a big story across oncology, of course). Lower amounts of trimethylation at this residue correlated with reduced mechanical stress, and the signaling mechanism for this appears to be the Nesprin-2 protein. That one had already been shown to be involved in mechanotransduction signaling down to the nucleus (it sits on the outside of the nuclear membrane), and here’s a big example of how that matters. Indeed, silencing Nesprin-2 in tumor cells before those implantation experiments mentioned above led to robust tumor growth even in contracting cardiac muscle, so the case looks pretty well proven. All this makes a person wonder if there’s a way to exploit these effects in non-cardiac tissue. Is there some way to stimulate Nesprin-2 signaling to make cells cancer-resistant? What will that do to cells that aren’t constantly pumping away in the heart muscle? Or could there be an outright mechanical stress therapy for localized tumors? We shall see. . .
Show full content
Think about it: have you every heard of a case of heart cancer? It’s very rare indeed, and why that’s the case has been a longstanding puzzle. This new paper (here at Science) seems to have found a big piece of the answer, though.
Cardiomyoctes are a rather special class of cell, but one of the fundamental things that make them special is that they are always in motion and under mechanical stress. If you raise them correctly in a cell culture dish, they will contract rhythmically even in there! And that is indeed the secret behind the lack of cancer in this tissue, apparently. The authors here made sure of the effect by overexpressing mutant K-Ras and deleting p53 in a mouse model, a terrible combination that resulting in multiple cancers in a range of tissues except the heart (even though it did indeed carry these mutations). Meanwhile, cells from various tumor types failed to proliferate as usual when implanted in cardiac tissue, but surgical remodeling of the heart vessels to decrease the mechanical load on the cardiac tissue allowed such tumors to take hold.
Closer study of these tissues implicates chromatin remodeling, specifically methylation of the histone 3 lysine 9 (chromatin changes in general are a big story across oncology, of course). Lower amounts of trimethylation at this residue correlated with reduced mechanical stress, and the signaling mechanism for this appears to be the Nesprin-2 protein. That one had already been shown to be involved in mechanotransduction signaling down to the nucleus (it sits on the outside of the nuclear membrane), and here’s a big example of how that matters. Indeed, silencing Nesprin-2 in tumor cells before those implantation experiments mentioned above led to robust tumor growth even in contracting cardiac muscle, so the case looks pretty well proven.
All this makes a person wonder if there’s a way to exploit these effects in non-cardiac tissue. Is there some way to stimulate Nesprin-2 signaling to make cells cancer-resistant? What will that do to cells that aren’t constantly pumping away in the heart muscle? Or could there be an outright mechanical stress therapy for localized tumors? We shall see. . .
You can put this one in the “reactions I never expected to see” category, because it’s a way to selectively functionalize aryl rings with multiple fluorines on them. And no, I don’t mean “functionalize at the carbon(s) that aren’t fluorinated yet” or even “kick out the most likely SnAr leaving group fluorine”. This is stepping and and replacing fluorines with H, D, alkyls or other aryls. The reason this looks so odd is that most of the time in organic chemistry breaking a C-F bond is going to be an uphill climb. They’re pretty strong as a rule, which is one reason why we medicinal chemists use them as blocking groups on carbons that are likely to undergo oxidative metabolism when a drug candidate hits the CYP enzymes in the liver. Fluorine has a number of other effects that can be attributed to its powerful electronegativity, and fluorination is very likely to change not only the metabolic profile of your compound, but to affect binding to proteins and to change solubility and other physical properties as well. So there are a number of ways to add fluorines to various parts of a molecule, under reaction conditions that range from not-so-bad to dive-behind-a-wall, but taking them back off? Not so many. The one that springs to mind is that I mentioned above, nucleophilic aromatic substitution. You can displace an aryl fluoride if the ring it’s on is suitably activated, and it’s a really useful reaction. para-Fuoro nitrobenzene or para-fluorobenzaldehyde are textbook examples: this reaction goes through an anionic intermediate that then kicks the fluorine back out to restore the aromatic ring, and groups that make that anionic state less painful accelerate things. But if you want to try it on a neutral or electron-rich aryl instead, then good luck to you. The new paper linked above is a different thing entirely, and I don’t think I’ve ever seen anything quite like it. The fluorinated starting material reaction with a pyridine-borane reagent that attacks at a particular C-F bond to generate a radical intermediate, and this is what loses the (solvated) fluoride to generate a borylaryl radical cation with a fluoride as the ion pair. That’s the species that then gets attacked by the coupling partner, forming a new C-C (or CH or CD) bond and kicking out a B-F side product. If you change the nature of the borane-pyridine reagent (by substitutions on the pyridine) you can tune this to take out different fluorines in order, and the authors demonstrate some of these stepwise functionalizations, all the way out to four steps. A variety of groups can be coupled under these conditions (substituted aryls, heteroaryls, substituted alkyls including alpha-amino couplings, alkenes (to give alkyl chains), and more. It’s quite weird to see, and will give retrosynthesis planners an entirely new way to thing about potential routes. And you can of course combine this new chemistry with a step of good ol’ SnAr from a suitable starting material for even more variety. It’s going to take me a bit to start envisioning polyfluoroaryls as versatile starting materials, though!
Show full content
You can put this one in the “reactions I never expected to see” category, because it’s a way to selectively functionalize aryl rings with multiple fluorines on them. And no, I don’t mean “functionalize at the carbon(s) that aren’t fluorinated yet” or even “kick out the most likely SnAr leaving group fluorine”. This is stepping and and replacing fluorines with H, D, alkyls or other aryls.
The reason this looks so odd is that most of the time in organic chemistry breaking a C-F bond is going to be an uphill climb. They’re pretty strong as a rule, which is one reason why we medicinal chemists use them as blocking groups on carbons that are likely to undergo oxidative metabolism when a drug candidate hits the CYP enzymes in the liver. Fluorine has a number of other effects that can be attributed to its powerful electronegativity, and fluorination is very likely to change not only the metabolic profile of your compound, but to affect binding to proteins and to change solubility and other physical properties as well.
So there are a number of ways to add fluorines to various parts of a molecule, under reaction conditions that range from not-so-bad to dive-behind-a-wall, but taking them back off? Not so many. The one that springs to mind is that I mentioned above, nucleophilic aromatic substitution. You can displace an aryl fluoride if the ring it’s on is suitably activated, and it’s a really useful reaction. para-Fuoro nitrobenzene or para-fluorobenzaldehyde are textbook examples: this reaction goes through an anionic intermediate that then kicks the fluorine back out to restore the aromatic ring, and groups that make that anionic state less painful accelerate things. But if you want to try it on a neutral or electron-rich aryl instead, then good luck to you.
The new paper linked above is a different thing entirely, and I don’t think I’ve ever seen anything quite like it. The fluorinated starting material reaction with a pyridine-borane reagent that attacks at a particular C-F bond to generate a radical intermediate, and this is what loses the (solvated) fluoride to generate a borylaryl radical cation with a fluoride as the ion pair. That’s the species that then gets attacked by the coupling partner, forming a new C-C (or CH or CD) bond and kicking out a B-F side product.
If you change the nature of the borane-pyridine reagent (by substitutions on the pyridine) you can tune this to take out different fluorines in order, and the authors demonstrate some of these stepwise functionalizations, all the way out to four steps. A variety of groups can be coupled under these conditions (substituted aryls, heteroaryls, substituted alkyls including alpha-amino couplings, alkenes (to give alkyl chains), and more. It’s quite weird to see, and will give retrosynthesis planners an entirely new way to thing about potential routes. And you can of course combine this new chemistry with a step of good ol’ SnAr from a suitable starting material for even more variety. It’s going to take me a bit to start envisioning polyfluoroaryls as versatile starting materials, though!
We have some really interesting progress in pancreatic cancer to talk about, both on the small-molecule and the mRNA vaccine fronts. Let’s do the small-molecule ones first, because those were honestly more unexpected. So to lead off, Revolution Medicines announced at the AACR meeting in San Diego that their drug daraxonrasib showed strong efficacy in patients with metastatic pancreatic ductal adenocarcinoma (PDAC). That is a very, very difficult to treat population - these people typically have only a few months to live. But the drug seems to have doubled their expected lifespan (13.2 months overall survival versus 6.7 months for chemotherapy standard of care). That is an extraordinary improvement, and I don’t believe that the field has ever seen anything like it in such patients with such advanced disease. The first question is, what’s this molecule? Daraxonrasib is shown at right, and you would be right to assume that its mechanism of action is not simple. It targets the RAS protein, which has been a white whale of oncology for decades. It was long considered undruggable, but there are now approved drugs that target a rare mutant form of it (G12C) that provides a cysteine handle that can be covalently modified. The problem with those is that (1) it’s not a common mutation and (2) it appears relatively easy for tumors to mutate out of this approach. But daraxonrasib’s mechanism is to stabilize a complex between all sorts of RAS isoforms and the widely expressed chaperone protein cyclophilin A. It was designed in a project starting from a natural product (Sanglifehrin A) that binds to CycA, cutting that down to a minimal binding region (that hydrazide part) and then building back with a covalent warhead to target that G12C residue in the cyclophilin-RAS complex. Some of this work goes back to the Warp Drive Bio days, actually. Further work led to RMC-7977 which accomplishes this feat on a much wider variety of RAS proteins, and then to daraxonrasib itself (RMC-6236). You’d have to classify this as a molecular glue, I’d say - cyclophilin A doesn’t really have any affinity for RAS under normal conditions, and neither does the drug. But the induced complex has a binding pocket made from surfaces of both proteins that binds the ligand, and this new complex interferes greatly with RAS’s downstream activities. As you’ll see from that last link, optimizing this structure for that binding mode while giving it good enough pharmacokinetic properties to make it a drug was also not trivial (!) PDAC is known to be one of the most “RAS-addicted” tumor types, so it’s a natural place to try this compound out. That survival benefit is impossible to argue with, but going after RAS does not come without penalties. The company says that the drug was generally well tolerated, but that’s on the oncology scale. Former senator Ben Sasse says that he had been noticing increasing back and abdominal pain, and unfortunately found that that was due to inoperable Stage 4 pancreatic cancer which had already metastatized to liver, lung, and other tissues. If you’re up to the point of overt physical symptoms like that with PDAC, it’s almost certainly bad news; this is one of the cancer types that famously sneaks up on people and is notoriously difficult to catch early. He’s been on daraxonrasib since early this year, and describes it this way: “. . .it’s a nasty drug. It causes crazy stuff like my body can’t grow skin and so I bleed all out of a whole bunch of parts of me that shouldn’t be bleeding” If you go to that link above, be prepared, because he also looks like he’s had aqua regia thrown all over him (and apparently feels a bit like that, too). But his tumor volume has gone down by about 75%, and there’s a very strong chance that he wouldn’t still be alive at all without having gone on the drug. He seems to feel that it’s a worthwhile tradeoff, and I suspect that there will be many others in his situation who agree.
Show full content
We have some really interesting progress in pancreatic cancer to talk about, both on the small-molecule and the mRNA vaccine fronts. Let’s do the small-molecule ones first, because those were honestly more unexpected.
So to lead off, Revolution Medicines announced at the AACR meeting in San Diego that their drug daraxonrasib showed strong efficacy in patients with metastatic pancreatic ductal adenocarcinoma (PDAC). That is a very, very difficult to treat population - these people typically have only a few months to live. But the drug seems to have doubled their expected lifespan (13.2 months overall survival versus 6.7 months for chemotherapy standard of care). That is an extraordinary improvement, and I don’t believe that the field has ever seen anything like it in such patients with such advanced disease.
The first question is, what’s this molecule? Daraxonrasib is shown at right, and you would be right to assume that its mechanism of action is not simple. It targets the RAS protein, which has been a white whale of oncology for decades. It was long considered undruggable, but there are now approved drugs that target a rare mutant form of it (G12C) that provides a cysteine handle that can be covalently modified. The problem with those is that (1) it’s not a common mutation and (2) it appears relatively easy for tumors to mutate out of this approach.
But daraxonrasib’s mechanism is to stabilize a complex between all sorts of RAS isoforms and the widely expressed chaperone protein cyclophilin A. It was designed in a project starting from a natural product (Sanglifehrin A) that binds to CycA, cutting that down to a minimal binding region (that hydrazide part) and then building back with a covalent warhead to target that G12C residue in the cyclophilin-RAS complex. Some of this work goes back to the Warp Drive Bio days, actually. Further work led to RMC-7977 which accomplishes this feat on a much wider variety of RAS proteins, and then to daraxonrasib itself (RMC-6236).
You’d have to classify this as a molecular glue, I’d say - cyclophilin A doesn’t really have any affinity for RAS under normal conditions, and neither does the drug. But the induced complex has a binding pocket made from surfaces of both proteins that binds the ligand, and this new complex interferes greatly with RAS’s downstream activities. As you’ll see from that last link, optimizing this structure for that binding mode while giving it good enough pharmacokinetic properties to make it a drug was also not trivial (!)
PDAC is known to be one of the most “RAS-addicted” tumor types, so it’s a natural place to try this compound out. That survival benefit is impossible to argue with, but going after RAS does not come without penalties. The company says that the drug was generally well tolerated, but that’s on the oncology scale. Former senator Ben Sasse says that he had been noticing increasing back and abdominal pain, and unfortunately found that that was due to inoperable Stage 4 pancreatic cancer which had already metastatized to liver, lung, and other tissues. If you’re up to the point of overt physical symptoms like that with PDAC, it’s almost certainly bad news; this is one of the cancer types that famously sneaks up on people and is notoriously difficult to catch early.
He’s been on daraxonrasib since early this year, and describes it this way: “. . .it’s a nasty drug. It causes crazy stuff like my body can’t grow skin and so I bleed all out of a whole bunch of parts of me that shouldn’t be bleeding” If you go to that link above, be prepared, because he also looks like he’s had aqua regia thrown all over him (and apparently feels a bit like that, too). But his tumor volume has gone down by about 75%, and there’s a very strong chance that he wouldn’t still be alive at all without having gone on the drug. He seems to feel that it’s a worthwhile tradeoff, and I suspect that there will be many others in his situation who agree.
Living in New England, we tend to get out into the yards, parks, and gardens this time of year because we’ve pretty much had it with winter. Of course, winter hasn’t always had it with us - we’ve had frost here the last couple of mornings, so it’s not exactly time to put the tomatos and cucumbers out there yet. But whenever the weather does start to warm up, so (unfortunately) does the threat of tick-borne diseases like Lyme. That one’s named after a town in Connecticut where the disease was first recognized, and the deer ticks that spread the Borrelia burgdorferi infection behind the disease are ubiquitous in this part of the country. If you go walking through the woods or rustling around in your garden bed around here, you would be extremely well advised to look yourself over for ticks afterwards. I have found them many times, occasionally briefly attached, although (as far as I’m aware) I have not had Lyme disease itself yet. But I know quite a few people who have. You do seem to need rather prolonged exposure to an active tick bite to transmit the bacteria, but you also do need to look for them for that to work in your favor! There was at one point a vaccine for the disease (LYMErix from SmithKline, later GlaxoSmithKline). That was approved in 1998 after a trial in over 10,000 people that showed a 76% reduction in disease, with no significant side effects noted during the trial. But it was not without disadvantages. First off, you needed three doses of the vaccine spread out over 12 months - and of course, even with that, protection was obviously not 100%. The duration of the vaccines’s effects (yearly? multiyear?) was also an open question, which presumably would have been answered over the years (that’s a general problem with new vaccines, for obvious reasons). And there was another factor remarked on at the time, which applies to several sorts of medical interventions: the worry that if people felt that they were more “bulletproof” against Lyme that they might not bother as much with other measures to reduce their exposure to tick bites, which (given that 76% efficacy) might lead to more Lyme cases in the unknowingly unprotected, as well as more cases of the other nasty tick-borne diseases like babesiosis, Rocky Mountain spotted fever, and so on). Not what you want! But as this article shows, there were other problems. Reports began to appear of side effects, particularly musculoskeletal ones resembling arthritis. A class-action lawsuit was filed in the year after approval. Looking at the data, though, the VAERS database for adverse events (which is incomplete and rather noisy, but the best we have) showed only 59 such arthritis reports after 1.4 million doses of vaccine. That incidence rate is the same as you would see in people who didn’t get the vaccine, and what’s more, there was no correlation with the second and third doses (as there really should be with any kind of immune-driven effects). The regulatory hearings on all this were. . .contentious. In 2002, GSK withdrew the vaccine from the market in the face of rapidly declining sales, and (according to the article above) settled the class action suits in an agreement that paid the bills of the law firms involved but did not provide any compensation to the “vaccine victims” that were the basis of the suits. So this was not an episode that showed the legal system in its best light. You have to think that GSK was worried about an even bigger legal pile-up if the vaccine remained on the market, and that they did not like their odds of demonstrating to juries across the country that there seemed to be no correlation with arthritis at all. I bring this up because there is another Lyme disease vaccine in the works, from Pfizer and Valneva. This one just completed its Phase III trial, and seems to have shown efficacy between 70 and 75% while targeting the same bacterial protein (OspA) as the earlier vaccine, albeit with much greater coverage of variants. There is room to argue about the efficacy, because there were not enough cases of Lyme were reported during the trial to give statistical significance to the results. Which is extremely annoying. But Pfizer is going to go ahead and submit for regulatory approval, which in the current climate for vaccine regulation is quite the move. And if you think that’s swimming against the current, consider that Moderna has two mRNA Lyme vaccine candidates in trials! Presumably they’re hoping to submit those after a change in administration, and buddy, I for one simply cannot wait. It will be very interesting indeed to see how this goes. The new vaccine also suffers from a multiple-dose schedule - I see from the clinical trial documents that they looked at a three-dose protocol with shots at 2 months and 6 months after the first dose, as well as a two-dose (one six months later). And they also looked at some booster regiments to follow up at months 18, 30, and 42. The data for all these combinations are not available yet, so we’ll have to see if these variations made a difference or not. And should the vaccine reach the market - not a sure thing - we will also have to see how the buzzing clouds of injury lawyers react to the stimulus. . .
Show full content
Living in New England, we tend to get out into the yards, parks, and gardens this time of year because we’ve pretty much had it with winter. Of course, winter hasn’t always had it with us - we’ve had frost here the last couple of mornings, so it’s not exactly time to put the tomatos and cucumbers out there yet. But whenever the weather does start to warm up, so (unfortunately) does the threat of tick-borne diseases like Lyme.
That one’s named after a town in Connecticut where the disease was first recognized, and the deer ticks that spread the Borrelia burgdorferi infection behind the disease are ubiquitous in this part of the country. If you go walking through the woods or rustling around in your garden bed around here, you would be extremely well advised to look yourself over for ticks afterwards. I have found them many times, occasionally briefly attached, although (as far as I’m aware) I have not had Lyme disease itself yet. But I know quite a few people who have. You do seem to need rather prolonged exposure to an active tick bite to transmit the bacteria, but you also do need to look for them for that to work in your favor!
There was at one point a vaccine for the disease (LYMErix from SmithKline, later GlaxoSmithKline). That was approved in 1998 after a trial in over 10,000 people that showed a 76% reduction in disease, with no significant side effects noted during the trial. But it was not without disadvantages. First off, you needed three doses of the vaccine spread out over 12 months - and of course, even with that, protection was obviously not 100%. The duration of the vaccines’s effects (yearly? multiyear?) was also an open question, which presumably would have been answered over the years (that’s a general problem with new vaccines, for obvious reasons). And there was another factor remarked on at the time, which applies to several sorts of medical interventions: the worry that if people felt that they were more “bulletproof” against Lyme that they might not bother as much with other measures to reduce their exposure to tick bites, which (given that 76% efficacy) might lead to more Lyme cases in the unknowingly unprotected, as well as more cases of the other nasty tick-borne diseases like babesiosis, Rocky Mountain spotted fever, and so on). Not what you want!
But as this article shows, there were other problems. Reports began to appear of side effects, particularly musculoskeletal ones resembling arthritis. A class-action lawsuit was filed in the year after approval. Looking at the data, though, the VAERS database for adverse events (which is incomplete and rather noisy, but the best we have) showed only 59 such arthritis reports after 1.4 million doses of vaccine. That incidence rate is the same as you would see in people who didn’t get the vaccine, and what’s more, there was no correlation with the second and third doses (as there really should be with any kind of immune-driven effects). The regulatory hearings on all this were. . .contentious.
In 2002, GSK withdrew the vaccine from the market in the face of rapidly declining sales, and (according to the article above) settled the class action suits in an agreement that paid the bills of the law firms involved but did not provide any compensation to the “vaccine victims” that were the basis of the suits. So this was not an episode that showed the legal system in its best light. You have to think that GSK was worried about an even bigger legal pile-up if the vaccine remained on the market, and that they did not like their odds of demonstrating to juries across the country that there seemed to be no correlation with arthritis at all.
I bring this up because there is another Lyme disease vaccine in the works, from Pfizer and Valneva. This one just completed its Phase III trial, and seems to have shown efficacy between 70 and 75% while targeting the same bacterial protein (OspA) as the earlier vaccine, albeit with much greater coverage of variants. There is room to argue about the efficacy, because there were not enough cases of Lyme were reported during the trial to give statistical significance to the results. Which is extremely annoying. But Pfizer is going to go ahead and submit for regulatory approval, which in the current climate for vaccine regulation is quite the move. And if you think that’s swimming against the current, consider that Moderna has two mRNA Lyme vaccine candidates in trials! Presumably they’re hoping to submit those after a change in administration, and buddy, I for one simply cannot wait.
It will be very interesting indeed to see how this goes. The new vaccine also suffers from a multiple-dose schedule - I see from the clinical trial documents that they looked at a three-dose protocol with shots at 2 months and 6 months after the first dose, as well as a two-dose (one six months later). And they also looked at some booster regiments to follow up at months 18, 30, and 42. The data for all these combinations are not available yet, so we’ll have to see if these variations made a difference or not. And should the vaccine reach the market - not a sure thing - we will also have to see how the buzzing clouds of injury lawyers react to the stimulus. . .
Translation of mRNA into proteins is a nonstop, nonnegotiable process that is essential to the life of a cell, and it has acquired a *lot* of evolutionary tuning over the last few billion years. In critters like us with nuclei and other such organelles (the big happy club of eukaryotes, to which so many of my readers belong as G. K. Chesterton used to say), there’s a very important protein complex called elF4F. That’s short for “eukaryotic initiation factor 4F", and it’s composed of three different subunit proteins. A lot of translation is “cap-dependent”, that is, it requires the presence of some special labels on the 5’ end of the messenger RNAs, and elF4F is what brings those capped RNAs together to the small (40S) ribosomal subunit to get things going. Prokaryotes, that is the bacteria, archaea, and of course the viruses that infect all the forms of life, don’t use elF4F or that mechanism. There are examples of viruses that express proteases that deliberately mess it up, the better to hijack the cell into making their own proteins instead, but that’s as close as you get. Well, until now. This new paper comes from a team studying the (rather weird) “giant DNA viruses”. Those are odd beasts, not least because of their size. They can be visible by regular optical microscopy, and some of them are larger than some types of bacteria (!) They have large double-stranded DNA genomes, and these genomes code for some stuff that you just won’t find in any other viruses. There are enzymes in there that seem to come from glycolysis and tricarboxylic acid pathways, which is rather odd baggage for something that doesn’t actually have any metabolism of its own going on, and some of them also code for metabolic enzymes like CYP P450 subtypes. The belief is that some of these have been jacked from some ancient host cells at an earlier point in evolution and kept around ever since. Not least among their odd features is that these giant viruses can in turn be infected by virophages themselves: virus viruses! “Great fleas have lesser fleas, upon their backs to bite ‘em, and little fleas still lesser fleas, and so ad infinitum”. We can add to that list of unnerving-for-a-virus features the presence of a viral form of elF4F! The authors here find giant DNA viruses deploying it as a weapon - it takes over for the host cell’s initiation factor complex but it only accepts viral RNA. So this allows the viral infection to roll along by co-opting the ribosomal machinery right from the very first stage. The paper shows that this vlF4F complex is also remarkably resistant to cellular stress, continuing to crank out viral proteins under all sorts of conditions. This discovery (and the other odd genes mentioned above) really makes one think about what must be going on. It certainly does seem likely that such machinery was indeed stolen in the distant past from some eukaryotic cell that had already evolved them. But you certainly can’t rule out that infection by these giant DNA viruses, which are ubiquitous, have in turn affected eukaryotic evolution afterwards. Neither side is coming out of this unchanged, and untangling who has done what and to whom (and when!) is going to be quite the project. . .
Show full content
Translation of mRNA into proteins is a nonstop, nonnegotiable process that is essential to the life of a cell, and it has acquired a *lot* of evolutionary tuning over the last few billion years. In critters like us with nuclei and other such organelles (the big happy club of eukaryotes, to which so many of my readers belong as G. K. Chesterton used to say), there’s a very important protein complex called elF4F.
That’s short for “eukaryotic initiation factor 4F", and it’s composed of three different subunit proteins. A lot of translation is “cap-dependent”, that is, it requires the presence of some special labels on the 5’ end of the messenger RNAs, and elF4F is what brings those capped RNAs together to the small (40S) ribosomal subunit to get things going. Prokaryotes, that is the bacteria, archaea, and of course the viruses that infect all the forms of life, don’t use elF4F or that mechanism. There are examples of viruses that express proteases that deliberately mess it up, the better to hijack the cell into making their own proteins instead, but that’s as close as you get.
Well, until now. This new paper comes from a team studying the (rather weird) “giant DNA viruses”. Those are odd beasts, not least because of their size. They can be visible by regular optical microscopy, and some of them are larger than some types of bacteria (!) They have large double-stranded DNA genomes, and these genomes code for some stuff that you just won’t find in any other viruses. There are enzymes in there that seem to come from glycolysis and tricarboxylic acid pathways, which is rather odd baggage for something that doesn’t actually have any metabolism of its own going on, and some of them also code for metabolic enzymes like CYP P450 subtypes. The belief is that some of these have been jacked from some ancient host cells at an earlier point in evolution and kept around ever since. Not least among their odd features is that these giant viruses can in turn be infected by virophages themselves: virus viruses! “Great fleas have lesser fleas, upon their backs to bite ‘em, and little fleas still lesser fleas, and so ad infinitum”.
We can add to that list of unnerving-for-a-virus features the presence of a viral form of elF4F! The authors here find giant DNA viruses deploying it as a weapon - it takes over for the host cell’s initiation factor complex but it only accepts viral RNA. So this allows the viral infection to roll along by co-opting the ribosomal machinery right from the very first stage. The paper shows that this vlF4F complex is also remarkably resistant to cellular stress, continuing to crank out viral proteins under all sorts of conditions.
This discovery (and the other odd genes mentioned above) really makes one think about what must be going on. It certainly does seem likely that such machinery was indeed stolen in the distant past from some eukaryotic cell that had already evolved them. But you certainly can’t rule out that infection by these giant DNA viruses, which are ubiquitous, have in turn affected eukaryotic evolution afterwards. Neither side is coming out of this unchanged, and untangling who has done what and to whom (and when!) is going to be quite the project. . .
This is not going to come as a surprise to readers of this site, because even if you don’t agree with this contention you have certainly at least been exposed to it: the Cochrane Review folks have examined the clinical evidence for anti-amyloid antibodies as therapies for Alzheimer’s and found, well. . .you know what they found: The effect of amyloid‐beta‐targeting monoclonal antibodies on cognitive function and dementia severity at 18 months in people with mild cognitive impairment or mild dementia due to Alzheimer’s disease is trivial, while on functional ability, it is small at best. . .Successful removal of amyloid from the brain does not seem to be associated with clinically meaningful effects in people with mild cognitive impairment or mild dementia due to Alzheimer’s disease. Future research on disease‐modifying treatments for Alzheimer’s disease should focus on other mechanisms of action. They reviewed 17 clinical studies that evaluated seven different monoclonal antibodies. All of these were placebo-controlled, and all were funded by the developers of the different drugs. This covers over 20,000 participants, and the results are clear and consistent across the entire data set, from what I can see: these drugs do not work. They do not improve the condition of people with Alzheimer’s to any noticeable degree: everyone taking them continued and continues to get worse and worse. The evidence for causing even a slight slowdown in the relentless progress of the disease is mostly nonexistent, and any evidence pointing in that direction is - under the most optimistic assumptions - almost certainly too small for anyone to detect in the real world. There have already been objections that this meta-analysis includes earlier antibodies that were abandoned by their developers along with the two most recent ones that were actually approved by the FDA. Personally, I find that to be special pleading (“You should only count the good ones!”), and I also note my own belief that the two that supposedly worked actually failed on efficacy and never should have been approved by the FDA at all. This has turned out not to be merely a crank opinion. The evidence from all these clinical trials did convince the Cochrane reviewers of one thing, though: taking these drugs increases the risk of amyloid-related imaging abnormalities. It is extremely concerning when the most definitive data about an entire class of drugs is about their unwanted side effects. I do not see why these antibodies should be on the market. And needless to say, I agree with the conclusion the authors came to here that it is far past time for something else. Amyloid-directed therapies truly, truly do not appear to be the answer for Alzheimer’s treatment. When I started work in the field back in the early 1990s, I was convinced of the opposite - the evidence looked very strong that defects in amyloid processing were indeed the cause of the disease. But that was thirty-five years ago, thirty-five years in which therapy after therapy after therapy aimed at amyloid mechanisms has failed. I have covered many of these over the years here on the blog, and looking back over the field is a depressing experience. We’re way past persistence, way past focus, way past optimism and multiple shots on goal and old-college-tries. Do something else! For God's sake, do something else.
Show full content
This is not going to come as a surprise to readers of this site, because even if you don’t agree with this contention you have certainly at least been exposed to it: the Cochrane Review folks have examined the clinical evidence for anti-amyloid antibodies as therapies for Alzheimer’s and found, well. . .you know what they found:
The effect of amyloid‐beta‐targeting monoclonal antibodies on cognitive function and dementia severity at 18 months in people with mild cognitive impairment or mild dementia due to Alzheimer’s disease is trivial, while on functional ability, it is small at best. . .Successful removal of amyloid from the brain does not seem to be associated with clinically meaningful effects in people with mild cognitive impairment or mild dementia due to Alzheimer’s disease. Future research on disease‐modifying treatments for Alzheimer’s disease should focus on other mechanisms of action.
They reviewed 17 clinical studies that evaluated seven different monoclonal antibodies. All of these were placebo-controlled, and all were funded by the developers of the different drugs. This covers over 20,000 participants, and the results are clear and consistent across the entire data set, from what I can see: these drugs do not work. They do not improve the condition of people with Alzheimer’s to any noticeable degree: everyone taking them continued and continues to get worse and worse. The evidence for causing even a slight slowdown in the relentless progress of the disease is mostly nonexistent, and any evidence pointing in that direction is - under the most optimistic assumptions - almost certainly too small for anyone to detect in the real world.
There have already been objections that this meta-analysis includes earlier antibodies that were abandoned by their developers along with the two most recent ones that were actually approved by the FDA. Personally, I find that to be special pleading (“You should only count the good ones!”), and I also note my own belief that the two that supposedly worked actually failed on efficacy and never should have been approved by the FDA at all. This has turned out not to be merely a crank opinion.
The evidence from all these clinical trials did convince the Cochrane reviewers of one thing, though: taking these drugs increases the risk of amyloid-related imaging abnormalities. It is extremely concerning when the most definitive data about an entire class of drugs is about their unwanted side effects. I do not see why these antibodies should be on the market.
And needless to say, I agree with the conclusion the authors came to here that it is far past time for something else. Amyloid-directed therapies truly, truly do not appear to be the answer for Alzheimer’s treatment. When I started work in the field back in the early 1990s, I was convinced of the opposite - the evidence looked very strong that defects in amyloid processing were indeed the cause of the disease. But that was thirty-five years ago, thirty-five years in which therapy after therapy after therapy aimed at amyloid mechanisms has failed.
I have covered many of these over the years here on the blog, and looking back over the field is a depressing experience. We’re way past persistence, way past focus, way past optimism and multiple shots on goal and old-college-tries. Do something else! For God's sake, do something else.
Artemisinin-based therapies are the absolute mainstay of malaria treatment the world over, so this new paper deserves attention. The drug is often given in combination with the older aminoquinoline agents like choloroquine, piperaquine, and amodiaquine, but the authors here make a strong case that this is actually counterproductive. As the paper notes, heme is central to the mechanism of action for both kinds of drugs. The aminoquinolines bind to it and affect heme homeostasis, and may well product toxic adducts that inhibit parasite growth. Meanwhile, the famous peroxide group in artemisinin gets cleaved by heme to form its active metabolite which causes a variety of protein alkylation events in situ, damaging the parasites from several directions at once. Naturally enough, people have looked for drug-drug interactions between these two classes, but the paper makes the case that these were not done under realistic conditions to reflect the in vivo state. Pulsing the active dihydroartemisinin (DHA) dose (since it has a short half-life) shows that in chloroquine-resistant parasites the two drugs interfere very strongly. It looks like the quinoline drugs actually block the effects of the active DHA, which is really, really not what you want to be doing. The hypothesis is that the heme complexes formed by the quinoline drugs leave the heme unable to cleave the peroxide bond in artemisinin, and in the chloroquine-resistant ones it appears that transport of it out of the parasite digestive vacuoles in enhanced. The authors show that you can actually rescue all the DHA-induced protein damage in the parasites by giving them chloroquine beforehand! This problem can vary according to the exact combinations used and the background genetics of the parasites themselves, but overall it seems to be quite general across the quinolines and across different peroxide-containing antimalarials. From what I can see, though, chloroquine is definitely the worst for cancelling out artemisinin. These results argue that we need to understand more about the interactions between these antimalarial drugs, and that it’s quite possible that we’ve been impairing malaria therapy out in the field by thinking that we understood enough already (!) We need to at least pick the least antagonistic combinations possible, and with an eye to parasite genetics whenever that’s feasible. Malaria has been an extremely wily enemy, and that hasn’t changed one bit.
Show full content
Artemisinin-based therapies are the absolute mainstay of malaria treatment the world over, so this new paper deserves attention. The drug is often given in combination with the older aminoquinoline agents like choloroquine, piperaquine, and amodiaquine, but the authors here make a strong case that this is actually counterproductive.
As the paper notes, heme is central to the mechanism of action for both kinds of drugs. The aminoquinolines bind to it and affect heme homeostasis, and may well product toxic adducts that inhibit parasite growth. Meanwhile, the famous peroxide group in artemisinin gets cleaved by heme to form its active metabolite which causes a variety of protein alkylation events in situ, damaging the parasites from several directions at once.
Naturally enough, people have looked for drug-drug interactions between these two classes, but the paper makes the case that these were not done under realistic conditions to reflect the in vivo state. Pulsing the active dihydroartemisinin (DHA) dose (since it has a short half-life) shows that in chloroquine-resistant parasites the two drugs interfere very strongly. It looks like the quinoline drugs actually block the effects of the active DHA, which is really, really not what you want to be doing.
The hypothesis is that the heme complexes formed by the quinoline drugs leave the heme unable to cleave the peroxide bond in artemisinin, and in the chloroquine-resistant ones it appears that transport of it out of the parasite digestive vacuoles in enhanced. The authors show that you can actually rescue all the DHA-induced protein damage in the parasites by giving them chloroquine beforehand! This problem can vary according to the exact combinations used and the background genetics of the parasites themselves, but overall it seems to be quite general across the quinolines and across different peroxide-containing antimalarials. From what I can see, though, chloroquine is definitely the worst for cancelling out artemisinin.
These results argue that we need to understand more about the interactions between these antimalarial drugs, and that it’s quite possible that we’ve been impairing malaria therapy out in the field by thinking that we understood enough already (!) We need to at least pick the least antagonistic combinations possible, and with an eye to parasite genetics whenever that’s feasible. Malaria has been an extremely wily enemy, and that hasn’t changed one bit.
I spent a day at Williams College last week, which I enjoyed very much, and I found a part of my lecture there overlapping with a big topic in undergraduate education. I have a section in several of my talks where I speak about AlphaFold-type machine learning and its implications for drug discovery, and that seemed to fit rather closely into concerns that many professors are having about the effect of AI systems on coursework and learning. I’m sure that if that’s your line of work, the topic must come up so relentlessly that people are starting to lose their minds at the prospects of dealing with it again, but out here in blog-readership-land I think it might be worth some discussion. One of the points I make when I talk about AlphaFold gets summed up like this (and it’s something I’ve said here before as well): if the Protein Folding Problem was set by God to force the human race to really understand the mechanisms behind protein structure, then, well. . .we cheated on the exam. Because we don’t understand those factors well enough to calculate such structures de novo, just using what we know about hydrogen bonds, torsional angles, steric hindrance, pi-stacking interactions and all the other things that add up energetically to stable protein conformations. I mean, we know a lot about those things, but we don’t know enough - not enough to take a big sample of protein sequences and derive from first principles the likely protein structures they’ll form. Most definitely we can’t do something like that with anything like the speed and success rate of the pattern-matching provided by AlphaFold-type machine learning. We used the large and well-curated pile of structural data in the PDB to take that shortcut, and it has turned out that proteins use many of the same tricks and patterns and combinations often enough that this approach really has worked out well. Don’t get me wrong - AlphaFold-type structures are far from infallible, and that’s because there are still a great many interesting and important protein structural motifs that are not well enough represented in our structural data sets. The PDB itself is far, far from a random sampling of protein space, of course (for starters, it is extremely biased towards structures that have a greater propensity to form high-quality crystals!) But it still has a lot of great information in it, and the relentless repetition and re-use of structural types in natural protein space gave the human race a big opportunity to bypass all the first-principles stuff. Which we took! And that brings up the question of what all this is for: do you want protein structures because they will tell you more about the complex thermodynamic balancing that goes into protein folding in a general sense, or do you want protein structures because you want to do something else with them? Like drug discovery, industrial enzyme design, all those applications that depend so much more on you just having the answer rather than on how you got to that answer. And here of course is where we split off from education. When you’re learning chemistry and biology, or honestly when you’re learning anything at all, the “just gimme the answer” impulse is toxic behavior that one should avoid. This is why so many writers - and I am very definitely one of them - have such an aversion to the sales pitches for LLM writing assistants offering to compose, revise, summarize the things I’m writing about. Like so many other people, I write to think and I think to write. Putting my thoughts down into some sort of order for a blog post, for example, is one of the ways that I organize my thinking. If some chatbot slurps up the source material, runs it through a blender, and excretes it out again for me in little processed nuggets, that does that thinking process no good whatsoever. But so many chatbot pitches seem to just assume that I want to dodge all that haaaard stuff and just get right to a convenient bullet-pointed answer. So you can see the problem with undergraduate course work, and believe me, professors have been seeing it for quite some time now. You assign your students material to read, digest, and summarize in an assignment because that is supposed to give their minds the experience of taking in this new material, making sense of it, and making enough sense of it to where they can then speak or write coherently about it. It’s work! But that’s one of the few reliable ways, in most cases, to learn anything. Having Chat O Matic give you a handy four-paragraph summary to turn in, though, is a reliable way to learn little or nothing. Box checked, you did the assignment, what’s next? All the situation needs is a professor who’s turned over the first hard steps of grading to chatbot software as well and you can take the darn humans - and their darn brains - right out of the loop. It reminds me of the old Russian joke about “As long as they pretend to pay me, I’ll pretend to work”. This is a tough problem, and the best answers to it are not yet apparent. But everyone seems to be in agreement that “Just let the students fill in the blanks with whatever good answers they can get, however they can get them” has never been a good answer itself, and never will be. Now, those of us doing research in which protein structures can be helpful, we are glad to have to modeled ones that we get (even if we should always remember to take them only for what they’re worth). We have things to do with them, as mentioned above. But I keep thinking that at some point it would do us all good if we understood the material well enough to be able to generate these answers without pattern-matching to structures that we’ve already determined experimentally. It would be valuable to understand hydrogen bonding and pi-stacking and all the rest of it well enough that we could simulate them computationally without generating great big ol’ error bars on the results, and the techniques that we would have to develop to sum all of these things up and balance them out across entire protein structures would actually be quite impressive (they’d have to be!) Are we going to ever do that? I think so. . .but there seems little doubt that AlphaFold and its competitors have taken the pressure off those lines of research. They’re hard questions! And if you would just rather have the answers, well, odds are that we can get you some much more quickly and painlessly. Do you want to know, or do you want to understand? Like the old Jack Benny routine where a robber threatens him with “Your money or your life” and he stalls saying “Ok, ok, I’m thinking about it!”, we have to opportunity to think about this one, too. Lucky us?
Show full content
I spent a day at Williams College last week, which I enjoyed very much, and I found a part of my lecture there overlapping with a big topic in undergraduate education. I have a section in several of my talks where I speak about AlphaFold-type machine learning and its implications for drug discovery, and that seemed to fit rather closely into concerns that many professors are having about the effect of AI systems on coursework and learning. I’m sure that if that’s your line of work, the topic must come up so relentlessly that people are starting to lose their minds at the prospects of dealing with it again, but out here in blog-readership-land I think it might be worth some discussion.
One of the points I make when I talk about AlphaFold gets summed up like this (and it’s something I’ve said here before as well): if the Protein Folding Problem was set by God to force the human race to really understand the mechanisms behind protein structure, then, well. . .we cheated on the exam. Because we don’t understand those factors well enough to calculate such structures de novo, just using what we know about hydrogen bonds, torsional angles, steric hindrance, pi-stacking interactions and all the other things that add up energetically to stable protein conformations. I mean, we know a lot about those things, but we don’t know enough - not enough to take a big sample of protein sequences and derive from first principles the likely protein structures they’ll form. Most definitely we can’t do something like that with anything like the speed and success rate of the pattern-matching provided by AlphaFold-type machine learning.
We used the large and well-curated pile of structural data in the PDB to take that shortcut, and it has turned out that proteins use many of the same tricks and patterns and combinations often enough that this approach really has worked out well. Don’t get me wrong - AlphaFold-type structures are far from infallible, and that’s because there are still a great many interesting and important protein structural motifs that are not well enough represented in our structural data sets. The PDB itself is far, far from a random sampling of protein space, of course (for starters, it is extremely biased towards structures that have a greater propensity to form high-quality crystals!) But it still has a lot of great information in it, and the relentless repetition and re-use of structural types in natural protein space gave the human race a big opportunity to bypass all the first-principles stuff.
Which we took! And that brings up the question of what all this is for: do you want protein structures because they will tell you more about the complex thermodynamic balancing that goes into protein folding in a general sense, or do you want protein structures because you want to do something else with them? Like drug discovery, industrial enzyme design, all those applications that depend so much more on you just having the answer rather than on how you got to that answer.
And here of course is where we split off from education. When you’re learning chemistry and biology, or honestly when you’re learning anything at all, the “just gimme the answer” impulse is toxic behavior that one should avoid. This is why so many writers - and I am very definitely one of them - have such an aversion to the sales pitches for LLM writing assistants offering to compose, revise, summarize the things I’m writing about. Like so many other people, I write to think and I think to write. Putting my thoughts down into some sort of order for a blog post, for example, is one of the ways that I organize my thinking. If some chatbot slurps up the source material, runs it through a blender, and excretes it out again for me in little processed nuggets, that does that thinking process no good whatsoever. But so many chatbot pitches seem to just assume that I want to dodge all that haaaard stuff and just get right to a convenient bullet-pointed answer.
So you can see the problem with undergraduate course work, and believe me, professors have been seeing it for quite some time now. You assign your students material to read, digest, and summarize in an assignment because that is supposed to give their minds the experience of taking in this new material, making sense of it, and making enough sense of it to where they can then speak or write coherently about it. It’s work! But that’s one of the few reliable ways, in most cases, to learn anything. Having Chat O Matic give you a handy four-paragraph summary to turn in, though, is a reliable way to learn little or nothing. Box checked, you did the assignment, what’s next?
All the situation needs is a professor who’s turned over the first hard steps of grading to chatbot software as well and you can take the darn humans - and their darn brains - right out of the loop. It reminds me of the old Russian joke about “As long as they pretend to pay me, I’ll pretend to work”. This is a tough problem, and the best answers to it are not yet apparent. But everyone seems to be in agreement that “Just let the students fill in the blanks with whatever good answers they can get, however they can get them” has never been a good answer itself, and never will be.
Now, those of us doing research in which protein structures can be helpful, we are glad to have to modeled ones that we get (even if we should always remember to take them only for what they’re worth). We have things to do with them, as mentioned above. But I keep thinking that at some point it would do us all good if we understood the material well enough to be able to generate these answers without pattern-matching to structures that we’ve already determined experimentally. It would be valuable to understand hydrogen bonding and pi-stacking and all the rest of it well enough that we could simulate them computationally without generating great big ol’ error bars on the results, and the techniques that we would have to develop to sum all of these things up and balance them out across entire protein structures would actually be quite impressive (they’d have to be!)
Are we going to ever do that? I think so. . .but there seems little doubt that AlphaFold and its competitors have taken the pressure off those lines of research. They’re hard questions! And if you would just rather have the answers, well, odds are that we can get you some much more quickly and painlessly. Do you want to know, or do you want to understand? Like the old Jack Benny routine where a robber threatens him with “Your money or your life” and he stalls saying “Ok, ok, I’m thinking about it!”, we have to opportunity to think about this one, too. Lucky us?
I often get asked what areas of drug discovery look most likely to bear AI-driven advances into the clinic, and my usual answer is “therapeutic antibodies”. Thats because it’s a protein-centric problem in the actual modeling, and we know quite a bit already about antibody structures (at least as compared to the much large wild-type protein structural landscape). And that’s because antibodies themselves are a (relatively!) constrained space within that larger one, although don’t let anyone tell you that it’s a small one. All you can say is that it’s less terrifying than the protein universe at large. And antibodies themselves are rather privileged structures, biologically. Evolution has seen to that, as we all carry a vast array of them swimming around our bloodstream at all time. Their stabilities and plasma half-lives are just off the charts compared to the normal run of proteins, and therapeutically that can allow for dosing every few weeks or months in actual patients. You really do have some advantages in this space. This is not to say that things can’t go wrong. Specificity is a big concern, always. A really good antibody can be exquisitely specific, but getting a really good antibody is not light work. Given the number of possible binding partners out there, surprises can always be lurking. And among those things that bind to your new therapeutic antibody candidate might well be. . .other existing antibodies in some of your patients. Yep, immunogenicity is another big worry, and it’s one that is very hard to get a real-world read on until, well, your antibody drug goes out there into the real world. Everyone’s immune response is different, very much not least because everyone’s suite of ten jillion antibodies is different, and you’ll just have to see what happens in the clinic and in the market. Here’s a preprint on computational approaches to antibody-antigen binding, and just as with small molecules, being able to compute your way to meaningful new possibilities and rankings in that area is an extremely attractive prospect. In this case, the authors took 106 different single-chain nanobodies (a trimmed-down version of the traditional antibody world, with these things being found in animals like llamas and alpacas) and their experimentally determined complexes with specific antigens. Then they produced a random shuffled pile of over 11,000 other possible pairings from all these antigens and antibodies and let three widely used software methods try to distinguish the grain from the chaff. The programs were AlphaFold3, Boltz-2, and Chai-1. Let’s just say right off that none of the three showed any particular ability to pick out the real pairings - that much is clear. Boltz-2 seems to have been more permissive than the other two, ranking more possibilities highly, but that didn’t make a difference in the end. AF3 was apparently the best of the lot, but don’t take that for much: it was still far from being able to reliably tell you a real complex versus a random shuffled one. Another thing to take away from this study was that some of the tools we use to evaluate these things aren’t worth that much, either. The authors used the “clash score” from the TopModel software, which is supposed to look for steric clashes between individual atoms in such complexes, but none of the modeling software programs distinguished themselves. Chai-1 might have done the opposite, actually - its clash scores were several times worse than the other two. But it’s hard to say just how much of a defect that is, because the clash scores for the real complexes were not noticeably different from the randomly faked-up shuffled ones. In fact, there were a number of cases of the shuffled complexes having better scores than anything in the real set, even though these were, in the end, not real antibody-antigen pairs at all. These sorts of effects persisted across a number of different types of evaluation, and there seems to be no way around the conclusion that computational antibody-antigen pair prediction is - at the moment - not much use at all. Comparing the predicted structures of the real pairs with their experimentally determined ones brought this home: the software was happy to assign generic sorts of interactions and plausible regions of contact, but did not pick up on any of the distinctive features that made the real complexes actually work. Here’s the authors summarizing the state of the art: Although de novo generation success rates are increasing into the double digit realm (~10-15% on hard interface tasks), our results indicate that there is a disconnect between de novo design (task: “generate an antibody for a given target without exploiting other information”) and de novo prediction (task: “indicate all the antigens that can be bound by a given antibody, and vice-versa”). Our “real vs shuffled” discrimination benchmark shows that models can generate interfaces that appear plausible, albeit incorrect, across many pairings, resulting in an abundance of false positives that are challenging to triage. And they warn that common workflows in this area “risk conflating structural plausibility with binding specificity”, when they have shown pretty comprehensively that these don’t have much to do with each other. Another warning is that “Overconfident failures, underconfident successes, and weak cross-tool agreement all highlight that current confidence metrics capture internal structural consistency rather than biological correctness”. Words to heed!
Show full content
I often get asked what areas of drug discovery look most likely to bear AI-driven advances into the clinic, and my usual answer is “therapeutic antibodies”. Thats because it’s a protein-centric problem in the actual modeling, and we know quite a bit already about antibody structures (at least as compared to the much large wild-type protein structural landscape). And that’s because antibodies themselves are a (relatively!) constrained space within that larger one, although don’t let anyone tell you that it’s a small one. All you can say is that it’s less terrifying than the protein universe at large.
And antibodies themselves are rather privileged structures, biologically. Evolution has seen to that, as we all carry a vast array of them swimming around our bloodstream at all time. Their stabilities and plasma half-lives are just off the charts compared to the normal run of proteins, and therapeutically that can allow for dosing every few weeks or months in actual patients. You really do have some advantages in this space.
This is not to say that things can’t go wrong. Specificity is a big concern, always. A really good antibody can be exquisitely specific, but getting a really good antibody is not light work. Given the number of possible binding partners out there, surprises can always be lurking. And among those things that bind to your new therapeutic antibody candidate might well be. . .other existing antibodies in some of your patients. Yep, immunogenicity is another big worry, and it’s one that is very hard to get a real-world read on until, well, your antibody drug goes out there into the real world. Everyone’s immune response is different, very much not least because everyone’s suite of ten jillion antibodies is different, and you’ll just have to see what happens in the clinic and in the market.
Here’s a preprint on computational approaches to antibody-antigen binding, and just as with small molecules, being able to compute your way to meaningful new possibilities and rankings in that area is an extremely attractive prospect. In this case, the authors took 106 different single-chain nanobodies (a trimmed-down version of the traditional antibody world, with these things being found in animals like llamas and alpacas) and their experimentally determined complexes with specific antigens. Then they produced a random shuffled pile of over 11,000 other possible pairings from all these antigens and antibodies and let three widely used software methods try to distinguish the grain from the chaff.
The programs were AlphaFold3, Boltz-2, and Chai-1. Let’s just say right off that none of the three showed any particular ability to pick out the real pairings - that much is clear. Boltz-2 seems to have been more permissive than the other two, ranking more possibilities highly, but that didn’t make a difference in the end. AF3 was apparently the best of the lot, but don’t take that for much: it was still far from being able to reliably tell you a real complex versus a random shuffled one.
Another thing to take away from this study was that some of the tools we use to evaluate these things aren’t worth that much, either. The authors used the “clash score” from the TopModel software, which is supposed to look for steric clashes between individual atoms in such complexes, but none of the modeling software programs distinguished themselves. Chai-1 might have done the opposite, actually - its clash scores were several times worse than the other two. But it’s hard to say just how much of a defect that is, because the clash scores for the real complexes were not noticeably different from the randomly faked-up shuffled ones. In fact, there were a number of cases of the shuffled complexes having better scores than anything in the real set, even though these were, in the end, not real antibody-antigen pairs at all.
These sorts of effects persisted across a number of different types of evaluation, and there seems to be no way around the conclusion that computational antibody-antigen pair prediction is - at the moment - not much use at all. Comparing the predicted structures of the real pairs with their experimentally determined ones brought this home: the software was happy to assign generic sorts of interactions and plausible regions of contact, but did not pick up on any of the distinctive features that made the real complexes actually work.
Here’s the authors summarizing the state of the art:
Although de novo generation success rates are increasing into the double digit realm (~10-15% on hard interface tasks), our results indicate that there is a disconnect between de novo design (task: “generate an antibody for a given target without exploiting other information”) and de novo prediction (task: “indicate all the antigens that can be bound by a given antibody, and vice-versa”). Our “real vs shuffled” discrimination benchmark shows that models can generate interfaces that appear plausible, albeit incorrect, across many pairings, resulting in an abundance of false positives that are challenging to triage.
And they warn that common workflows in this area “risk conflating structural plausibility with binding specificity”, when they have shown pretty comprehensively that these don’t have much to do with each other. Another warning is that “Overconfident failures, underconfident successes, and weak cross-tool agreement all highlight that current confidence metrics capture internal structural consistency rather than biological correctness”. Words to heed!
If you look at cells dispassionately, one of the things that strikes is that man, do we ever have a lot of ribosomes. These are of course the protein-synthesis machines that are kept humming nonstop as RNA sequences are translated into protein sequences, and ribosomes themselves are structurally a mixture of proteins and some unusual RNA molecules all their own. The peptidyltransferase center (PTC) where the actual synthesis of new peptide bonds takes place is itself a flat-out ribozyme, an RNA structure that is acting like we expect enzyme proteins to - and doing an extraordinary job of it, too. Mammalian cells are estimated to have millions ribosomes per cell, a figure that many people not not realize, and they take up a substantial part of the the entire cellular volume. That’s also while using up a hefty amount of the cell’s energy budget as well. For fast-growing cells like bacteria, it appears that the final limits to their growth rate are ribosome abundance, the time and energy it takes to make new ones, and the rate at which they themselves can turn out new proteins. In turn, the size limit of bacterial cells is believed to be bounded by how many ribosomes can be fit into a given volume (and how many more would be needed to go larger). And never forget, not least (nowhere near least) on a ribosome’s protein production priority list is. . .making more ribosomal proteins to make still more ribosomes. So this new preprint looks over this evidence and more and asks if perhaps we have been looking at all of biology the wrong way around. Instead of living creatures being cells that have lots of ribosomes, maybe life is all about ribosomes, and cells are the things that have evolved as the most convenient abodes for them. They hypothesis a symbiotic relationship between an early replicase enzyme and an early ribosome-like structure, which as it evolved started to decorate itself with all the other things we associate with living cells. By which I mean little trinkets like, say, metabolism: In this framework, the evolution of the increasingly diverse metabolic pathways catalyzed by the (proto)ribosome-produced enzymes can be viewed as being driven primarily by the need to sustain the activity of the ribosome itself, a principle that continues to govern the functioning of modern cells. In this context, the ribosome is perhaps most pertinently viewed as a selfish entity subjugating the other functional systems of the cell. The authors also note that no virus has ever been discovered that encodes its own functioning ribosome(s), and hypothesize that they went down a different “selfish replicator” pathway. Cells are, in this framework, primarily ribosome-encoding organisms, while viruses can be seen as capsid- or virion-encoding ones. In fact, that might be the only alternative available, given the time, material, and energy costs of building ribosomes from a standing start. As the authors say, “hijacking the host translation apparatus appears to be the only viable strategy for genetic parasites” In the end, it’s impossible to definitely prove a chicken-and-egg question like this one (whether the cell bosses the ribosomes around or vice versa). But I think that the authors are right that it’s worthwhile to consider that the latter might be the case, and to see what that might tell us about cellular evolution from the proposed ancient RNA world. I don’t think I’ll ever look at the little gizmos quite the same way again. . .!
Show full content
If you look at cells dispassionately, one of the things that strikes is that man, do we ever have a lot of ribosomes. These are of course the protein-synthesis machines that are kept humming nonstop as RNA sequences are translated into protein sequences, and ribosomes themselves are structurally a mixture of proteins and some unusual RNA molecules all their own. The peptidyltransferase center (PTC) where the actual synthesis of new peptide bonds takes place is itself a flat-out ribozyme, an RNA structure that is acting like we expect enzyme proteins to - and doing an extraordinary job of it, too.
Mammalian cells are estimated to have millions ribosomes per cell, a figure that many people not not realize, and they take up a substantial part of the the entire cellular volume. That’s also while using up a hefty amount of the cell’s energy budget as well. For fast-growing cells like bacteria, it appears that the final limits to their growth rate are ribosome abundance, the time and energy it takes to make new ones, and the rate at which they themselves can turn out new proteins. In turn, the size limit of bacterial cells is believed to be bounded by how many ribosomes can be fit into a given volume (and how many more would be needed to go larger). And never forget, not least (nowhere near least) on a ribosome’s protein production priority list is. . .making more ribosomal proteins to make still more ribosomes.
So this new preprint looks over this evidence and more and asks if perhaps we have been looking at all of biology the wrong way around. Instead of living creatures being cells that have lots of ribosomes, maybe life is all about ribosomes, and cells are the things that have evolved as the most convenient abodes for them. They hypothesis a symbiotic relationship between an early replicase enzyme and an early ribosome-like structure, which as it evolved started to decorate itself with all the other things we associate with living cells. By which I mean little trinkets like, say, metabolism:
In this framework, the evolution of the increasingly diverse metabolic pathways catalyzed by the (proto)ribosome-produced enzymes can be viewed as being driven primarily by the need to sustain the activity of the ribosome itself, a principle that continues to govern the functioning of modern cells. In this context, the ribosome is perhaps most pertinently viewed as a selfish entity subjugating the other functional systems of the cell.
The authors also note that no virus has ever been discovered that encodes its own functioning ribosome(s), and hypothesize that they went down a different “selfish replicator” pathway. Cells are, in this framework, primarily ribosome-encoding organisms, while viruses can be seen as capsid- or virion-encoding ones. In fact, that might be the only alternative available, given the time, material, and energy costs of building ribosomes from a standing start. As the authors say, “hijacking the host translation apparatus appears to be the only viable strategy for genetic parasites”
In the end, it’s impossible to definitely prove a chicken-and-egg question like this one (whether the cell bosses the ribosomes around or vice versa). But I think that the authors are right that it’s worthwhile to consider that the latter might be the case, and to see what that might tell us about cellular evolution from the proposed ancient RNA world. I don’t think I’ll ever look at the little gizmos quite the same way again. . .!
Longtime readers will recall the fuss I made about the “synthesis machine” work from the Burke group at Illinois. That involved iterative coupling of boronic esters with MIDA boronates, which were then deprotected to plain boronic esters, which could then be coupled with a further MIDA boronate, and you get the idea. It’s a scheme for directed carbon-carbon bond formation, and when you get down to it, a lot of organic synthesis is in fact a scheme for directed carbon-carbon bond formation. What’s more, it held out the promise of being subject to automation, due to the interesting silica gel purification properties of the MIDA boronate intermediates. As you can see from the link above, I went from these results straight to predicting an eventual end to organic synthesis as we know it, and I think it’s safe to say that the bulk of the readership did not follow me to the end of that series of thoughts. It’s also clear here in 2026 that (so far) organic synthesis as we have known it has not come to such an end, but in light of this new paper I’m going to get on everyone’s nerves by revisiting the subject. This is a good point to note (as does the paper itself) that the idea of iterative automatable bond-forming sequences did not stop with that 2016 MIDA work. Burke himself introduced new possibilities with TIDA boronates with carbenoid insertions, and carbenoids also show up here in this context. Then there are other boron-based schemes and silicon-based ones as well, along with single-carbon insertion protocols, as well as nitrogen- and oxygen-atom insertions (see the references in the paper linked above for these). Overall, there are more and more transformations that are pointing in this direction: general (or pretty general) bond-forming reactions that can be chained or combined to generate a very large number of possible structures, and which are often amenable to being done partially or completely by automation. The new work describes a more streamlined iterative coupling protocol, and one with some significant advantages over a lot of earlier work. It uses the long-familiar pinacolatoboronate esters, B(pin), in this case at alkyl carbon centers, and it activates those with t-butyllithium (the use of which is honestly one of the possible weak points in the whole sequence, because I suspect that not every structure will put up with it). That puts a t-butyl on the boron (along with a negative charge and a lithium cation) and such salts, it appears, can often be isolated and stored for later use. They can then be coupled with a variety of other alkyl compounds that have a leaving group on them (iodide or benzenesulfonate, which is converted to the iodide during the course of the mechanism). You use triphenylphosphine, an acetylide ligand, and copper(I) iodide to get a direct sp3-sp3 carbon-carbon coupling reaction which retains stereochemistry at the B(pin) carbon. And that’s something that you don’t often see in this area. A wide variety of B(pin) compounds seem to be tolerated - silyl ethers, acetals, tertiary amines, pyridines, and cyclopropyl groups all survive just fine. But as mentioned above, I’m sure that there’s going to be trouble with intermediates that have protons acidic enough to react with the t-BuLi. As for the electrophile, the paper demonstrated couplings with alkyl chains contining alkenes, alkeynes, aryls, methyl esters, nitriles, and epoxides, which is pretty good. The only amine I see is protected by a phthalimide, though, which is a sign. And you can also come in with electrophiles that have the required leaving group at one end and another B(pin) at the other, which gives you a shot at the iterated-chemistry idea. This gets demonstrated with a synthesis of the natural product spongidepsin, a thirteen-membered macrocycle with both an amide and an ester in the ring and an alkyne-containing side chain. Six of the copper-catalyzed bond-forming reactions take place along the way. That’s a pretty good demonstration of utility, but the authors also note several areas that could be improved (beta-branched electrophiles are bit wonky, and theres room for improvement in the yields in many cases). But there are obviously a lot of things yet to be tried in these systems - look no further than the continued improvements over the years in Pd-catalyzed reactions like the Suzuki coupling and the Buchwald-Hartwig amine couplings to see how long and complex a process that can be. Just seeing stereoselective and stereospecific C-C bond formation between sp3 centers like this being run so blithely is food for thought. These are really important transformations for organic synthesis and (as with the MIDA and TIDA boronate reactions) the appearance of reasonably general solutions in the distance is an exciting possibility. Whether you’re on the automated-synthesis-could-be-coming train or not, these kinds of reactions do, I think, herald different ways of putting molecules together than what most of us learning when we were studying organic chemistry. At the very least these things can give us combinatorial access to a vast number of new structures with less bespoke effort than ever before. Are we going to make organic synthesis less of an intellectual challenge eventually by getting too good at it? Remember, that’s exactly what happened to organic structure determination, from its roots in chemical degradation and total-synthesis-to-confirm-structure. We’re all too young in this field to have experienced it personally, but the 1960s and 70s really destroyed the old way of doing things, and there were people who just retired rather than find themselves replaced by advances in NMR and X-ray diffraction. Don’t rule it out - and remember, there are other things to work on! I'm 64, folks, so it's not going to be a problem for me. But you 23-year-olds, you might want to keep an eye on it. . .
Show full content
Longtime readers will recall the fuss I made about the “synthesis machine” work from the Burke group at Illinois. That involved iterative coupling of boronic esters with MIDA boronates, which were then deprotected to plain boronic esters, which could then be coupled with a further MIDA boronate, and you get the idea. It’s a scheme for directed carbon-carbon bond formation, and when you get down to it, a lot of organic synthesis is in fact a scheme for directed carbon-carbon bond formation. What’s more, it held out the promise of being subject to automation, due to the interesting silica gel purification properties of the MIDA boronate intermediates.
As you can see from the link above, I went from these results straight to predicting an eventual end to organic synthesis as we know it, and I think it’s safe to say that the bulk of the readership did not follow me to the end of that series of thoughts. It’s also clear here in 2026 that (so far) organic synthesis as we have known it has not come to such an end, but in light of this new paper I’m going to get on everyone’s nerves by revisiting the subject.
This is a good point to note (as does the paper itself) that the idea of iterative automatable bond-forming sequences did not stop with that 2016 MIDA work. Burke himself introduced new possibilities with TIDA boronates with carbenoid insertions, and carbenoids also show up here in this context. Then there are other boron-based schemes and silicon-based ones as well, along with single-carbon insertion protocols, as well as nitrogen- and oxygen-atom insertions (see the references in the paper linked above for these). Overall, there are more and more transformations that are pointing in this direction: general (or pretty general) bond-forming reactions that can be chained or combined to generate a very large number of possible structures, and which are often amenable to being done partially or completely by automation.
The new work describes a more streamlined iterative coupling protocol, and one with some significant advantages over a lot of earlier work. It uses the long-familiar pinacolatoboronate esters, B(pin), in this case at alkyl carbon centers, and it activates those with t-butyllithium (the use of which is honestly one of the possible weak points in the whole sequence, because I suspect that not every structure will put up with it). That puts a t-butyl on the boron (along with a negative charge and a lithium cation) and such salts, it appears, can often be isolated and stored for later use.
They can then be coupled with a variety of other alkyl compounds that have a leaving group on them (iodide or benzenesulfonate, which is converted to the iodide during the course of the mechanism). You use triphenylphosphine, an acetylide ligand, and copper(I) iodide to get a direct sp3-sp3 carbon-carbon coupling reaction which retains stereochemistry at the B(pin) carbon. And that’s something that you don’t often see in this area.
A wide variety of B(pin) compounds seem to be tolerated - silyl ethers, acetals, tertiary amines, pyridines, and cyclopropyl groups all survive just fine. But as mentioned above, I’m sure that there’s going to be trouble with intermediates that have protons acidic enough to react with the t-BuLi. As for the electrophile, the paper demonstrated couplings with alkyl chains contining alkenes, alkeynes, aryls, methyl esters, nitriles, and epoxides, which is pretty good. The only amine I see is protected by a phthalimide, though, which is a sign. And you can also come in with electrophiles that have the required leaving group at one end and another B(pin) at the other, which gives you a shot at the iterated-chemistry idea.
This gets demonstrated with a synthesis of the natural product spongidepsin, a thirteen-membered macrocycle with both an amide and an ester in the ring and an alkyne-containing side chain. Six of the copper-catalyzed bond-forming reactions take place along the way. That’s a pretty good demonstration of utility, but the authors also note several areas that could be improved (beta-branched electrophiles are bit wonky, and theres room for improvement in the yields in many cases). But there are obviously a lot of things yet to be tried in these systems - look no further than the continued improvements over the years in Pd-catalyzed reactions like the Suzuki coupling and the Buchwald-Hartwig amine couplings to see how long and complex a process that can be.
Just seeing stereoselective and stereospecific C-C bond formation between sp3 centers like this being run so blithely is food for thought. These are really important transformations for organic synthesis and (as with the MIDA and TIDA boronate reactions) the appearance of reasonably general solutions in the distance is an exciting possibility. Whether you’re on the automated-synthesis-could-be-coming train or not, these kinds of reactions do, I think, herald different ways of putting molecules together than what most of us learning when we were studying organic chemistry. At the very least these things can give us combinatorial access to a vast number of new structures with less bespoke effort than ever before.
Are we going to make organic synthesis less of an intellectual challenge eventually by getting too good at it? Remember, that’s exactly what happened to organic structure determination, from its roots in chemical degradation and total-synthesis-to-confirm-structure. We’re all too young in this field to have experienced it personally, but the 1960s and 70s really destroyed the old way of doing things, and there were people who just retired rather than find themselves replaced by advances in NMR and X-ray diffraction. Don’t rule it out - and remember, there are other things to work on! I'm 64, folks, so it's not going to be a problem for me. But you 23-year-olds, you might want to keep an eye on it. . .
So let’s finally talk about peptides. And I don’t mean peptides as chemists and biologists understand them (short chains of amino acids) I mean “Peptides!”, the hot new wonder drugs that you can order by mail. Oh man. The first barrier to writing about this situation, for someone like me, is that difference in terms. For a chemist, “peptide” has a pretty clear definition: any relatively short chain of amino acids, and when they get longer we go ahead and call them “proteins”, although the dividing line is a matter of personal opinion. So the number of different possible peptides is just ridiculously huge. If you just pick (say) a ten-amino-acid chain with the 20 common amino acids, that’s twenty to the tenth power, which is over ten trillion possibilities. For comparison, it has been about ten trillion seconds since Homo sapiens emerged as a separate species. And since that number is large and contains multitudes, there are lots of physiologically active peptides and an even much more gigantic mass of them that biologically do nothing at all. As you digest any particular piece of protein (being from Arkansas, I recommend a plate of barbecue or fried catfish) the enzymes in your stomach and small intestine are producing huge numbers of progressively shorter peptide chains as they break down that food on the way to stuff that will be absorbed into the bloodstream. And this is while every living cell is making progressively longer peptide chains off the ribosomes, building back up the proteins needed for life. This is why I I had a puzzled look on my face when I first saw people talking excitedly about how they were taking “peptides” as if that were a well-defined category that everyone knew about. Mentioning digestion brings up the question of stability. Your body is also awash with enzymes (proteins themselves!) that do a terrific job of breaking down peptide bonds. So the lifetime of any random protein string in the blood is probably going to be limited, but the subject of today’s post is not random peptides. Nope, it’s amazing wonder peptides ordered from suppliers who mail them directly to your house, stuff that will variously energize your waking hours, cure your diseases, melt your fat deposits, build your muscles, grow your hair, and extend your lifespan. You’re not going to be taking these things orally, because unless a really substantial amount of engineering has gone into it, any given peptide is going get the same treatment from your digestive system as a chicken breast does, i.e. a complete teardown. These mail-order peptides are injectable items. At one point that would have severely limited their use, but the advent of the GLP-1 drugs (proteins as well, which are largely but not exclusively injectables) has made this more acceptable, at least among people who don’t know someone who has to inject insulin (which needless to say is a small protein itself). So the times have come around to make this a real business opportunity, which many suppliers are enthusiastically leaning into. How about the science? It’s the biggest mixed bag you ever saw. There’s no doubt at all that there are some extremely biologically active peptides out there, and more such things are being discovered all the time. In fact, there’s a whole field devoted to looking at peptides that are too short to have been studied by the usual analyses, and those folks are turning up all sorts of activities that we never even appreciated were there. So that’s the first point: there are indeed a whole range of physical and medical effects to be found in these things. Unfortunately, point two is that we barely have any of these effects worked out - at least not to the degree that you would want before you start injecting them into your leg. We’re still finding things out about extremely well known active proteins like insulin, much less more recent discoveries. That lack of knowledge extends - and how - both to their on-target effects (assuming that the target or targets are reasonably well known, which isn’t always the case) and to their off-target toxicities. And there are going to be plenty of cases where yes, Peptide X sure does do that thing you’re interested in, but it turns out that you can’t do That Thing without doing other things that you are surely not interested in. A number of “peptides of abuse” these days, for example, seem to be targeting human growth hormone pathways and associated ones, so let’s use that as an example. The pitch is often something like “Here’s the signal your body uses to build muscle! Take it directly and get going today!”, and with HGH there’s also been a longstanding subculture that treats it as a Fountain of Youth signal of some kind. “Replenish your growth hormone levels”, the idea is, “and dial back the biological clock!” But growth hormone (and I shouldn’t have to say this) is powerful stuff, and it doesn’t just go tell your muscles to swell up. It affects bone tissue and many other tissues as well. I would invite anyone looking to maximize their growth hormone levels to look up a condition called acromegaly, which is what you get when your body keeps on making more growth hormone than you strictly need. Bones in the hands, feet, and head enlarge, and you get all sorts of side effects like joint pain, high blood pressure, type II diabetes, and other things that are probably not mentioned in the peptide supplier’s brochure. Excess growth hormone also increases the risk of some types of cancer, and that is a general problem with any attempt to (re)activate cellular growth pathways. Cancer, when you get right down to it, is a disease of unrestrained cellular growth, and there are a lot of different pathways that can lead to it. Stimulating cell growth out of the blue with systemic injections of synthetic hormone-like peptides is (in my view, and it ain’t just me) an invitation to greater tumor risk. And that’s just for starters. There are two very good pieces over at Stat right now on this topic. One, by physician Vikas Patel, describes a patient who is discontinuing her statin therapy - very inadvisably, given the clinical presentation described - but is enthusiastic about “BPC-157” as an injectable peptide for her knee pain. Says Patel: My patient is refusing a drug studied in 170,000 people because of side effects that a 124,000-person analysis just confirmed do not exist — while injecting a compound studied in 14 humans, from unregulated sources, based on the recommendation of someone who profits from selling it. She’s probably not the only one. And those using it believe they are “doing their own research.” The other one (by Sarah Hood) relates all this to RFJ Jr.’s advocacy. The flip side of “the government shouldn’t be able to force me to vaccinate my kids” is “I should have the right to take whatever medicines I want to without the government getting in my way”. That’s what we’re seeing here. I would bet that many peptide customers see themselves as free agents who have done their own research and are taking their health into their own hands - and dodging past Big Med and the old fossils at the regulatory agencies while they’re at it. But as that article points out (and as I’m doing today), what they’re actually doing is rolling the dice after falling for sales pitches aimed at exactly this sort of customer. That dice rolling doesn’t just end at the unknown mechanisms of action, the lack of human data, and the lack of information about potential side effects. You’re also ordered from suppliers whose manufacturing standards you are in no position whatsoever to check, no matter how much of a free health warrior you might be. You don’t have an LC/MS or an NMR machine in your garage, so you can’t be sure what it is you’re really injecting, how pure it is, or to what extent it’s already deteriorated on standing. There is no one at the other end of the deal who cares very much, either, believe me. Human nature being what it is in this fallen world, you actually need regulatory agencies to force people to care about these issues by threatening them with severe punishments if they don’t. In my own view (and it ain’t just me) you also have regulatory agencies to force people to show that their drugs actually have some benefit before they can sell them, too. But that’s going further and further out of fashion. Can’t get ahold of the New Hotness to inject into your upper thigh if there are a bunch of stick-in-the-mud folks asking for human data, infringing on your freedom and all. What a time to be alive.
Show full content
So let’s finally talk about peptides. And I don’t mean peptides as chemists and biologists understand them (short chains of amino acids) I mean “Peptides!”, the hot new wonder drugs that you can order by mail. Oh man.
The first barrier to writing about this situation, for someone like me, is that difference in terms. For a chemist, “peptide” has a pretty clear definition: any relatively short chain of amino acids, and when they get longer we go ahead and call them “proteins”, although the dividing line is a matter of personal opinion. So the number of different possible peptides is just ridiculously huge. If you just pick (say) a ten-amino-acid chain with the 20 common amino acids, that’s twenty to the tenth power, which is over ten trillion possibilities. For comparison, it has been about ten trillion seconds since Homo sapiens emerged as a separate species.
And since that number is large and contains multitudes, there are lots of physiologically active peptides and an even much more gigantic mass of them that biologically do nothing at all. As you digest any particular piece of protein (being from Arkansas, I recommend a plate of barbecue or fried catfish) the enzymes in your stomach and small intestine are producing huge numbers of progressively shorter peptide chains as they break down that food on the way to stuff that will be absorbed into the bloodstream. And this is while every living cell is making progressively longer peptide chains off the ribosomes, building back up the proteins needed for life. This is why I I had a puzzled look on my face when I first saw people talking excitedly about how they were taking “peptides” as if that were a well-defined category that everyone knew about.
Mentioning digestion brings up the question of stability. Your body is also awash with enzymes (proteins themselves!) that do a terrific job of breaking down peptide bonds. So the lifetime of any random protein string in the blood is probably going to be limited, but the subject of today’s post is not random peptides. Nope, it’s amazing wonder peptides ordered from suppliers who mail them directly to your house, stuff that will variously energize your waking hours, cure your diseases, melt your fat deposits, build your muscles, grow your hair, and extend your lifespan. You’re not going to be taking these things orally, because unless a really substantial amount of engineering has gone into it, any given peptide is going get the same treatment from your digestive system as a chicken breast does, i.e. a complete teardown. These mail-order peptides are injectable items.
At one point that would have severely limited their use, but the advent of the GLP-1 drugs (proteins as well, which are largely but not exclusively injectables) has made this more acceptable, at least among people who don’t know someone who has to inject insulin (which needless to say is a small protein itself). So the times have come around to make this a real business opportunity, which many suppliers are enthusiastically leaning into.
How about the science? It’s the biggest mixed bag you ever saw. There’s no doubt at all that there are some extremely biologically active peptides out there, and more such things are being discovered all the time. In fact, there’s a whole field devoted to looking at peptides that are too short to have been studied by the usual analyses, and those folks are turning up all sorts of activities that we never even appreciated were there. So that’s the first point: there are indeed a whole range of physical and medical effects to be found in these things.
Unfortunately, point two is that we barely have any of these effects worked out - at least not to the degree that you would want before you start injecting them into your leg. We’re still finding things out about extremely well known active proteins like insulin, much less more recent discoveries. That lack of knowledge extends - and how - both to their on-target effects (assuming that the target or targets are reasonably well known, which isn’t always the case) and to their off-target toxicities.
And there are going to be plenty of cases where yes, Peptide X sure does do that thing you’re interested in, but it turns out that you can’t do That Thing without doing other things that you are surely not interested in. A number of “peptides of abuse” these days, for example, seem to be targeting human growth hormone pathways and associated ones, so let’s use that as an example. The pitch is often something like “Here’s the signal your body uses to build muscle! Take it directly and get going today!”, and with HGH there’s also been a longstanding subculture that treats it as a Fountain of Youth signal of some kind. “Replenish your growth hormone levels”, the idea is, “and dial back the biological clock!”
But growth hormone (and I shouldn’t have to say this) is powerful stuff, and it doesn’t just go tell your muscles to swell up. It affects bone tissue and many other tissues as well. I would invite anyone looking to maximize their growth hormone levels to look up a condition called acromegaly, which is what you get when your body keeps on making more growth hormone than you strictly need. Bones in the hands, feet, and head enlarge, and you get all sorts of side effects like joint pain, high blood pressure, type II diabetes, and other things that are probably not mentioned in the peptide supplier’s brochure.
Excess growth hormone also increases the risk of some types of cancer, and that is a general problem with any attempt to (re)activate cellular growth pathways. Cancer, when you get right down to it, is a disease of unrestrained cellular growth, and there are a lot of different pathways that can lead to it. Stimulating cell growth out of the blue with systemic injections of synthetic hormone-like peptides is (in my view, and it ain’t just me) an invitation to greater tumor risk. And that’s just for starters.
There are two very good pieces over at Stat right now on this topic. One, by physician Vikas Patel, describes a patient who is discontinuing her statin therapy - very inadvisably, given the clinical presentation described - but is enthusiastic about “BPC-157” as an injectable peptide for her knee pain. Says Patel:
My patient is refusing a drug studied in 170,000 people because of side effects that a 124,000-person analysis just confirmed do not exist — while injecting a compound studied in 14 humans, from unregulated sources, based on the recommendation of someone who profits from selling it. She’s probably not the only one. And those using it believe they are “doing their own research.”
The other one (by Sarah Hood) relates all this to RFJ Jr.’s advocacy. The flip side of “the government shouldn’t be able to force me to vaccinate my kids” is “I should have the right to take whatever medicines I want to without the government getting in my way”. That’s what we’re seeing here. I would bet that many peptide customers see themselves as free agents who have done their own research and are taking their health into their own hands - and dodging past Big Med and the old fossils at the regulatory agencies while they’re at it. But as that article points out (and as I’m doing today), what they’re actually doing is rolling the dice after falling for sales pitches aimed at exactly this sort of customer.
That dice rolling doesn’t just end at the unknown mechanisms of action, the lack of human data, and the lack of information about potential side effects. You’re also ordered from suppliers whose manufacturing standards you are in no position whatsoever to check, no matter how much of a free health warrior you might be. You don’t have an LC/MS or an NMR machine in your garage, so you can’t be sure what it is you’re really injecting, how pure it is, or to what extent it’s already deteriorated on standing. There is no one at the other end of the deal who cares very much, either, believe me. Human nature being what it is in this fallen world, you actually need regulatory agencies to force people to care about these issues by threatening them with severe punishments if they don’t.
In my own view (and it ain’t just me) you also have regulatory agencies to force people to show that their drugs actually have some benefit before they can sell them, too. But that’s going further and further out of fashion. Can’t get ahold of the New Hotness to inject into your upper thigh if there are a bunch of stick-in-the-mud folks asking for human data, infringing on your freedom and all. What a time to be alive.
Here’s an interesting surprise: an opiod agonist with what seems to be fewer side effects. That’s been a tricky thing to manage - all the morphine-type compounds have the well-known liabilities (addictive potential, respiratory depression, gut motility and more), and the synthetic ligands like fentanyl certainly don’t avoid these either. In recent years there’s been a revival of interest in an older class of molecules called nitazenes, but unfortunately a big part of that interest has come from drug cartels: the nitazenes are extremely potent, and their synthesis gets around the existing barriers to fentanyl precursors and the like. But this group has been doing extensive med-chem work in the area, and they report N-destethylfluoronitazene (DFNZ) as a promising candidate. As they note, the SAR in these compounds is quite complex, but it can give you mu-opiod agonist potency up to 1000x morphine, which is pretty terrifying. That’s the main reason the series was abandoned - there seemed to be no way to work with such compounds as potential drugs while the above-mentioned side effects tracked pretty much along with the analgesia. But here the authors applied the “potency to burn” idea to find better properties (which I have to note was one of the ideas behind the recent “potency above all” paper I blogged about here). Working on the alkoxy chain of these molecules, they found that fluorination in this region led to some interesting compounds. Metabolic studies using liver microsomes showed that their best compound was rapidly transformed into an N-desethyl analog from the starting N,N-diethyl, so they investigated that one as likely to have better half-life. It turned out to keep the selectivity for the mu-opioid receptor and had similar receptor-binding kinetics, although it did have different behavior in G-protein versus beta-arrestin pathways inside the membrane (biased on the whole toward the G-protein side of things). It showed strong analgesic effects in rodent models of both acute and chronic pain, but its behavior was rather different than the classic opioids in other respects: much less effect on body temperature and much less hyperlocomotion. Making an 18F labeled version allowed them to quantify that DFNZ has significantly lower penetration into the CNS than expected and less receptor occupancy in brain tissue. That appears to be through its affinity for efflux transporters, particularly PGP and BCRP. Further work with implantable sensors in the mouse brains showed that brain hypoxia (a real effect of things like fentanyl in the nucleus accumbens region) showed that DFNZ doesn’t have much effect when given at its full analgesic dose. You could get hypoxia at higher doses, but there’s a much greater window. It also has much lower effects on dopaminergic pathways than other opioids. Building on that last result, animal assays for tolerance and withdrawal (major problems, of course, with morphine and fentanyl) showed that DFNZ produced far fewer withdrawal effects and repeated dosing did not seem to cause tolerance. Further behavioral tests all seemed to point to lower addiction potential. And then there’s this rather unusual statement: “Supporting this idea, anecdotal reports from recreational users suggest that certain fluorinated nitazenes lack rewarding effects” So overall this looks like a promising compound to look at for pain relief, and there are probably other nitazene compounds with similar profiles. I hope that this leads to something useful - safe pain relief, especially at morphine-like levels, would be a great advance.
Show full content
Here’s an interesting surprise: an opiod agonist with what seems to be fewer side effects. That’s been a tricky thing to manage - all the morphine-type compounds have the well-known liabilities (addictive potential, respiratory depression, gut motility and more), and the synthetic ligands like fentanyl certainly don’t avoid these either. In recent years there’s been a revival of interest in an older class of molecules called nitazenes, but unfortunately a big part of that interest has come from drug cartels: the nitazenes are extremely potent, and their synthesis gets around the existing barriers to fentanyl precursors and the like.
But this group has been doing extensive med-chem work in the area, and they report N-destethylfluoronitazene (DFNZ) as a promising candidate. As they note, the SAR in these compounds is quite complex, but it can give you mu-opiod agonist potency up to 1000x morphine, which is pretty terrifying. That’s the main reason the series was abandoned - there seemed to be no way to work with such compounds as potential drugs while the above-mentioned side effects tracked pretty much along with the analgesia. But here the authors applied the “potency to burn” idea to find better properties (which I have to note was one of the ideas behind the recent “potency above all” paper I blogged about here).
Working on the alkoxy chain of these molecules, they found that fluorination in this region led to some interesting compounds. Metabolic studies using liver microsomes showed that their best compound was rapidly transformed into an N-desethyl analog from the starting N,N-diethyl, so they investigated that one as likely to have better half-life. It turned out to keep the selectivity for the mu-opioid receptor and had similar receptor-binding kinetics, although it did have different behavior in G-protein versus beta-arrestin pathways inside the membrane (biased on the whole toward the G-protein side of things).
It showed strong analgesic effects in rodent models of both acute and chronic pain, but its behavior was rather different than the classic opioids in other respects: much less effect on body temperature and much less hyperlocomotion. Making an 18F labeled version allowed them to quantify that DFNZ has significantly lower penetration into the CNS than expected and less receptor occupancy in brain tissue. That appears to be through its affinity for efflux transporters, particularly PGP and BCRP. Further work with implantable sensors in the mouse brains showed that brain hypoxia (a real effect of things like fentanyl in the nucleus accumbens region) showed that DFNZ doesn’t have much effect when given at its full analgesic dose. You could get hypoxia at higher doses, but there’s a much greater window. It also has much lower effects on dopaminergic pathways than other opioids.
Building on that last result, animal assays for tolerance and withdrawal (major problems, of course, with morphine and fentanyl) showed that DFNZ produced far fewer withdrawal effects and repeated dosing did not seem to cause tolerance. Further behavioral tests all seemed to point to lower addiction potential. And then there’s this rather unusual statement: “Supporting this idea, anecdotal reports from recreational users suggest that certain fluorinated nitazenes lack rewarding effects”
So overall this looks like a promising compound to look at for pain relief, and there are probably other nitazene compounds with similar profiles. I hope that this leads to something useful - safe pain relief, especially at morphine-like levels, would be a great advance.
By this point a lot of companies have made a lot of PROTAC bifunctionals. Those, as my readers in the business well know, are large species that have two “business ends” tied together by some sort of linking group. One end binds to a protein of interest (POI) and the other to a protein that is involved in the protein degradation machinery (typically an E3 ligase enzyme). Having these in the same molecule lets you bind to your target and then bring the E3 ligase over in proximity. At that point it does what it does for a living, namely slap ubiquitin moeities on available lysine residues, and those go on to signal the cell that this protein has been tagged for removal to the proteasome and destruction for recycled pieces. This lets you stand up there like Zeus while zapping target proteins with destroying lightning bolts, and this often accomplishes much more than just hitting those targets with more traditional small-molecule inhibitors. There are some details to be worked out before you start raining down the divine retribution, though, and a big one is that linking group. How long it should be, how polar, how flexible. . .there are a lot of possibilities, too many to run through them systematically, and they most definitely have effects on the activity of your final molecules. Looking closely, those activity variations are due to several factors. One is whether your bifunctional brings the two proteins together in a productive manner. We don’t always know why some of them work better than the others in this step, but there’s no doubt that some combinations result in a lot more ubquitination than others. Good ol’ pharmacokinetics is (as always!) a consideration, too. These bifunctionals tend to lie pretty far outside most people’s comfort zones in size, molecular weight, number of hydrogen bonding groups, and all those other parameters we used to use to convince ourselves that we know how to make compounds with good properties. Now, many of them do work - more than you would have thought twenty years ago if you’d advanced these structures, I’m sure - but many of them don’t (problems with oral absorption, membrane penetration, metabolic stability and more). There have been a lot of proposals for guidelines and heuristics to help you narrow things down, and this new paper has a look at them in the context of around 1500 PROTAC candidates produced at AstraZeneca over the last few years. One of the properties that has gotten a lot of attention is “chameleonicity”, which is broadly defined as the ability of a molecule to adapt to solvent or membrane environments by changes in its conformation. You can imagine hydrogen bond pairs being buried in a more close conformation, the better to slip through, for example. In molecules as large as PROTACs (and with flexible linkers in them) it’s not hard to think that this could be important. But is it? The authors here found that they could not use any of the proposed chameleonicity parameters to make sense of their experimental oral absorption numbers, unfortunately. Solubility measurements, in either buffer or simulated intestinal fluid, were no help either. The “bifunctional bioavailability index” from Bristol-Myers Squibb was a bit more useful, but still not something you’d want to base your whole program on. The only help the authors found was from the good ol’ Caco-2 assay. That’s one that’s been in use for many years - you grow a layer of those cells (which come from an intestinal epithelial cancer line) and then measure the amount of compound that can move through it, ideally in either direction (because these cell layers do have an “inside” and an “outside”. It’s rather labor-intensive and low-throughput, and there is no guarantee that every membrane you encounter will behave like a lab-grown layer of Caco-2 cells either (quite the opposite!) But it can certainly flag outliers, and that’s what it did with the bifunctionals here. The compounds with a high efflux ratio in the assay (indicating that compounds moved out through the cell layer more readily than they moved in) were notably less likely to have decent oral absorption, which is a sensible result. So you can at least eliminate some candidates that way if you are willing to put in the time and effort. That’s one take-home here, and the other one is that you probably shouldn’t try to shortcut things by doing chameleonicity calculations, either. If there are shortcuts in this area, we haven’t found the good ones yet.
Show full content
By this point a lot of companies have made a lot of PROTAC bifunctionals. Those, as my readers in the business well know, are large species that have two “business ends” tied together by some sort of linking group. One end binds to a protein of interest (POI) and the other to a protein that is involved in the protein degradation machinery (typically an E3 ligase enzyme). Having these in the same molecule lets you bind to your target and then bring the E3 ligase over in proximity. At that point it does what it does for a living, namely slap ubiquitin moeities on available lysine residues, and those go on to signal the cell that this protein has been tagged for removal to the proteasome and destruction for recycled pieces.
This lets you stand up there like Zeus while zapping target proteins with destroying lightning bolts, and this often accomplishes much more than just hitting those targets with more traditional small-molecule inhibitors. There are some details to be worked out before you start raining down the divine retribution, though, and a big one is that linking group. How long it should be, how polar, how flexible. . .there are a lot of possibilities, too many to run through them systematically, and they most definitely have effects on the activity of your final molecules.
Looking closely, those activity variations are due to several factors. One is whether your bifunctional brings the two proteins together in a productive manner. We don’t always know why some of them work better than the others in this step, but there’s no doubt that some combinations result in a lot more ubquitination than others. Good ol’ pharmacokinetics is (as always!) a consideration, too. These bifunctionals tend to lie pretty far outside most people’s comfort zones in size, molecular weight, number of hydrogen bonding groups, and all those other parameters we used to use to convince ourselves that we know how to make compounds with good properties.
Now, many of them do work - more than you would have thought twenty years ago if you’d advanced these structures, I’m sure - but many of them don’t (problems with oral absorption, membrane penetration, metabolic stability and more). There have been a lot of proposals for guidelines and heuristics to help you narrow things down, and this new paper has a look at them in the context of around 1500 PROTAC candidates produced at AstraZeneca over the last few years.
One of the properties that has gotten a lot of attention is “chameleonicity”, which is broadly defined as the ability of a molecule to adapt to solvent or membrane environments by changes in its conformation. You can imagine hydrogen bond pairs being buried in a more close conformation, the better to slip through, for example. In molecules as large as PROTACs (and with flexible linkers in them) it’s not hard to think that this could be important.
But is it? The authors here found that they could not use any of the proposed chameleonicity parameters to make sense of their experimental oral absorption numbers, unfortunately. Solubility measurements, in either buffer or simulated intestinal fluid, were no help either. The “bifunctional bioavailability index” from Bristol-Myers Squibb was a bit more useful, but still not something you’d want to base your whole program on. The only help the authors found was from the good ol’ Caco-2 assay.
That’s one that’s been in use for many years - you grow a layer of those cells (which come from an intestinal epithelial cancer line) and then measure the amount of compound that can move through it, ideally in either direction (because these cell layers do have an “inside” and an “outside”. It’s rather labor-intensive and low-throughput, and there is no guarantee that every membrane you encounter will behave like a lab-grown layer of Caco-2 cells either (quite the opposite!) But it can certainly flag outliers, and that’s what it did with the bifunctionals here. The compounds with a high efflux ratio in the assay (indicating that compounds moved out through the cell layer more readily than they moved in) were notably less likely to have decent oral absorption, which is a sensible result.
So you can at least eliminate some candidates that way if you are willing to put in the time and effort. That’s one take-home here, and the other one is that you probably shouldn’t try to shortcut things by doing chameleonicity calculations, either. If there are shortcuts in this area, we haven’t found the good ones yet.
In today’s “Well, I’ll be darned” category is this paper, which described a way to get drug delivery into the brain that I never would have imagined. The authors are using a “calvarial” mechanism, and I’m certainly in a forgiving mood if you’ve never heard that word before, because I certainly hadn’t. The calvaria, as it turns out, is the top of the skull, and the skull bones have a lot of anatomical detail in them. The inner and outer surfaces are compact bone, and there’s a layer of spongy “cancellous” bone between them. That contains red bone marrow, and there are channels through this tissue (diploic veins) in a rather complex network whose function is still open for debate. The most generally accepted explanation is one big function is as a cooling system for the brain itself (which after all is a very metabolically active organ). But there are routes into the cranial cavity itself, because it turns out that the skull’s bone marrow is in close biological communication with the meningial layers around the brain. There are “skull-meninges channels” for immune cells that develop in that bone marrow to migrate into the brain, and the authors speculated that this might be a Trojan-horse route for therapeutic agents, They injected drug-loaded nanoparticles into that central space of the skull bone layers, and found that indeed, immune cells take these up and then continue on their merry way into the brain loaded with this cargo. They especially go on to target sites of inflammation therein, which give you a real opportunity for targeted delivery. This was demonstrated with the clinically studied oligopeptide agent nerinetide (a neuroprotective), and found that it was effective in a neuroinflammation model at only 20% of the dose needed otherwise. The authors have already tested this in human patients suffering from middle cerebral artery infarction (which sounds like a terrible event, I have to add). This was done in 20 patients randomized to get standard-of-care or that plus through-the-skull (“ICO”) treatment with Y-3, a neuroprotective already approved in China. For what it’s worth, the number of patients showing better clinical scoring was certainly higher in the ICO group, and the main thing you can take from that is that this certainly looks worthy of further work in a larger trial. The tricky part, I think, will be narrowing down for that. There are obviously a number of drugs you could imagine delivering this way for a variety of conditions, and this delivery route is expected to be fairly drug-agnostic. Acute stroke sounds like it will continue to be a good proving ground, but there are plenty of others. And you could also imagine dropping in whole modified cells rather than just having them take up your drug payloads. A lot of work is in store, but the promise of totally skipping the blood-barrier line is certainly worth it.
Show full content
In today’s “Well, I’ll be darned” category is this paper, which described a way to get drug delivery into the brain that I never would have imagined. The authors are using a “calvarial” mechanism, and I’m certainly in a forgiving mood if you’ve never heard that word before, because I certainly hadn’t.
The calvaria, as it turns out, is the top of the skull, and the skull bones have a lot of anatomical detail in them. The inner and outer surfaces are compact bone, and there’s a layer of spongy “cancellous” bone between them. That contains red bone marrow, and there are channels through this tissue (diploic veins) in a rather complex network whose function is still open for debate. The most generally accepted explanation is one big function is as a cooling system for the brain itself (which after all is a very metabolically active organ).
But there are routes into the cranial cavity itself, because it turns out that the skull’s bone marrow is in close biological communication with the meningial layers around the brain. There are “skull-meninges channels” for immune cells that develop in that bone marrow to migrate into the brain, and the authors speculated that this might be a Trojan-horse route for therapeutic agents,
They injected drug-loaded nanoparticles into that central space of the skull bone layers, and found that indeed, immune cells take these up and then continue on their merry way into the brain loaded with this cargo. They especially go on to target sites of inflammation therein, which give you a real opportunity for targeted delivery. This was demonstrated with the clinically studied oligopeptide agent nerinetide (a neuroprotective), and found that it was effective in a neuroinflammation model at only 20% of the dose needed otherwise.
The authors have already tested this in human patients suffering from middle cerebral artery infarction (which sounds like a terrible event, I have to add). This was done in 20 patients randomized to get standard-of-care or that plus through-the-skull (“ICO”) treatment with Y-3, a neuroprotective already approved in China. For what it’s worth, the number of patients showing better clinical scoring was certainly higher in the ICO group, and the main thing you can take from that is that this certainly looks worthy of further work in a larger trial.
The tricky part, I think, will be narrowing down for that. There are obviously a number of drugs you could imagine delivering this way for a variety of conditions, and this delivery route is expected to be fairly drug-agnostic. Acute stroke sounds like it will continue to be a good proving ground, but there are plenty of others. And you could also imagine dropping in whole modified cells rather than just having them take up your drug payloads. A lot of work is in store, but the promise of totally skipping the blood-barrier line is certainly worth it.
Now here’s something that I wouldn’t have guessed. As the world knows, lipid nanoparticles have been the key to getting mRNA vaccines to work, and they are useful for all other attempts to deliver RNA cargoes into cells and probably much besides. A huge amount of effort, time, and money has been put into trying to optimize them for these purposes - all sorts of different lipids and lipid mixes, close attention to size, morphology, and cargo loading, you name it. But as this new paper notes, even with this, mRNA delivery can be rather spotty across different tissue types, for reasons that are still up for debate. The genetic background of the cells and tissues that youre trying to deliver to will clearly have an influence, but even going after what should be the same cell types you can see some real variation (and plenty of changes between in vitro and in vivo behavior). What’s not up for debate, sadly, is that if we’re going to have to just keep trying combinations of lipid carriers to get out of this problem, well, we could be at it a while: a four-component LNP could have up to ten billion potential combinations, given the number of candidates for each of those positions. The authors here note that culturing your target cells in different media leads to very different mRNA LNP outcomes. The standard cell media mixes are formulated to optimize cell growth in culture, and as such contain rather nonphysiological amounts of many nutrients. That was the impetus to develop human plasma-like medium (HPLM) to more closely mimic the in vivo situation. When they used that mix instead of the usual cell media, the authors saw the efficiency of the mRNA delivery drop sharply, which got them to thinking about the effects of all those extra components in the classic brews. It turns out that the HPLM-grown cells show downregulation of several amino acid metabolic pathways (arginine, methione, and proline, for example), and the team hypothesized that this might have something to do with the impaired LNP uptake. Screening supplementation of individual amino acids in HPLM to bring their concentrations up to where they are in other media showed that extra methionine, arginine, and serine had a very significant effect on efficacy of mRNA delivery. Working on the concentrations of these in more detail, a mixture of 30x HPLM background in methionine, 10X arginine, and 30X serine performed well across the board, with five-to tenfold increases in delivery. Cells fed this mixture showed enrichment of endocytosis pathway proteins (but not those associated with endosomal escape, as far as could be seen). The clathrin-independent-carrier (CLIC) pathway seemed to be especially affected, and had already been flagged by other research as important for LNP success. And convincingly, this amino acid supplementation idea worked in animals, showing greatly increased uptake of mRNA cargos both in model systems and with therapeutic cargos. And this worked not only for mRNA delivery, but for in vivo gene editing cargoes with CRISPR-Cas9 components as well. This is good news - it’s hard to see how anyone could have a problem with some extra amino acids being given along with the mRNA doses. So it looks like the metabolic state of the target cells is indeed a key factor in hitting them with lipid nanoparticles, and that this can be easily modified with a cocktail of inexpensive, easily available and easily administered amino acids. I hope this idea gets some intensive scrutiny and that it works as well for others as it did here!
Show full content
Now here’s something that I wouldn’t have guessed. As the world knows, lipid nanoparticles have been the key to getting mRNA vaccines to work, and they are useful for all other attempts to deliver RNA cargoes into cells and probably much besides. A huge amount of effort, time, and money has been put into trying to optimize them for these purposes - all sorts of different lipids and lipid mixes, close attention to size, morphology, and cargo loading, you name it.
But as this new paper notes, even with this, mRNA delivery can be rather spotty across different tissue types, for reasons that are still up for debate. The genetic background of the cells and tissues that youre trying to deliver to will clearly have an influence, but even going after what should be the same cell types you can see some real variation (and plenty of changes between in vitro and in vivo behavior). What’s not up for debate, sadly, is that if we’re going to have to just keep trying combinations of lipid carriers to get out of this problem, well, we could be at it a while: a four-component LNP could have up to ten billion potential combinations, given the number of candidates for each of those positions.
The authors here note that culturing your target cells in different media leads to very different mRNA LNP outcomes. The standard cell media mixes are formulated to optimize cell growth in culture, and as such contain rather nonphysiological amounts of many nutrients. That was the impetus to develop human plasma-like medium (HPLM) to more closely mimic the in vivo situation. When they used that mix instead of the usual cell media, the authors saw the efficiency of the mRNA delivery drop sharply, which got them to thinking about the effects of all those extra components in the classic brews.
It turns out that the HPLM-grown cells show downregulation of several amino acid metabolic pathways (arginine, methione, and proline, for example), and the team hypothesized that this might have something to do with the impaired LNP uptake. Screening supplementation of individual amino acids in HPLM to bring their concentrations up to where they are in other media showed that extra methionine, arginine, and serine had a very significant effect on efficacy of mRNA delivery. Working on the concentrations of these in more detail, a mixture of 30x HPLM background in methionine, 10X arginine, and 30X serine performed well across the board, with five-to tenfold increases in delivery. Cells fed this mixture showed enrichment of endocytosis pathway proteins (but not those associated with endosomal escape, as far as could be seen). The clathrin-independent-carrier (CLIC) pathway seemed to be especially affected, and had already been flagged by other research as important for LNP success.
And convincingly, this amino acid supplementation idea worked in animals, showing greatly increased uptake of mRNA cargos both in model systems and with therapeutic cargos. And this worked not only for mRNA delivery, but for in vivo gene editing cargoes with CRISPR-Cas9 components as well. This is good news - it’s hard to see how anyone could have a problem with some extra amino acids being given along with the mRNA doses.
So it looks like the metabolic state of the target cells is indeed a key factor in hitting them with lipid nanoparticles, and that this can be easily modified with a cocktail of inexpensive, easily available and easily administered amino acids. I hope this idea gets some intensive scrutiny and that it works as well for others as it did here!
The development of chimeric antigen receptor T-cell (CAR-T) therapy continues, and some really interesting new directions are being explored. As it stands, this treatment can be extraordinarily effective in some patients, and these are generally people who have been through every other option for their cancer therapy. But there are some real limitations, even for treating patients with the leukemia/myeloma type cancers that (so far) this mode is best suited for. An obvious one is the sheer amount of time and effort involved. The patient’s own T cells have to be genetically engineered to go after a selection of antigens that is tailored to each person’s case, and the patients themselves have to be “lymphodepleted” with chemotherapy before these are infused back. It’s a difficult, expensive process, and (like anything in this category) is very hard on the patients themselves (although, to be sure, not as hard as dying from the disease some weeks or months later, which by this point is the only remaining alternative). One long-sought goal has been the “off the shelf” CAR-T protocol, where the T cells don’t have to be painstakingly personalized, but can be prepared from the blood cells of other healthy donors, modified to take on the most widespread antigens as is. This has not been easy; let’s just leave it at that rather than review the landscape in detail. But several organizations are continuing to work on this. A team at Vanderbilt, for example, is running a trial that reported some interim results late last year: no dose-limiting tox or severe cytokine release so far, which is good, but it’s too early to get a read on efficacy. And a company called Allogene is running a trial to see if their CAR-T can delay relapse in treated lymphoma patients, which is an interesting idea that hasn’t really been put to the test. So there are a couple of layers of risk in there, which are well explained by Adam Feuerstein here at Stat. First results are supposed to appear next month. Another idea is to produce the CAR-T cells right within the patient’s own body, rather than going through all that ex vivo work. For that, you’re talking gene therapy to introduce the chimeric antigen receptor into the patient’s T cells, and that’s a pretty bold move. A team in Wuhan has just reported some early data on a lentivirus vector they’re using for this purpose (technology that AstraZeneca bought from them last year), and another report also appeared recently. It’s a mixed bag. In this latest paper, five patients with advanced multiple myeloma were studied and followed for months after a single i.v. infusion of the viral vector. There was no lymphodepletion done on the patients beforehand. No dose-limiting toxicity was seen, but all the patients experienced adverse events: four showed cytokine release syndrome, three of them developed infections, and there were also transient liver enzyme elevations. Unfortunately one of the five died outright from “immune effector cell-associated neurotoxicity”, which is of course Not Good. But the others all appear to have responded to the treatment, with three “stringent complete remissions”. Monitoring the blood of these patients showed the uptake of the viral vector and production of the modified T-cells, as well as the gradual restoration of normal B cells along the way as the malignant ones (and presumably their precursors) were destroyed. The authors make the case that this is a distinct clinical course rather than just a faster version of the present CAR-T treatments, and it’s going to need a lot of close attention to understand all the differences. So this is promising and alarming at the same time: it looks like you really can treat refractory multiple myeloma in this fashion, but is there really going to be a 20% fatality rate as you try to do it? You have to hope - and AstraZeneca has to hope - that that’s not the case, but it’s the earliest of early days with this technique, and there’s a lot yet to learn.
Show full content
The development of chimeric antigen receptor T-cell (CAR-T) therapy continues, and some really interesting new directions are being explored. As it stands, this treatment can be extraordinarily effective in some patients, and these are generally people who have been through every other option for their cancer therapy. But there are some real limitations, even for treating patients with the leukemia/myeloma type cancers that (so far) this mode is best suited for.
An obvious one is the sheer amount of time and effort involved. The patient’s own T cells have to be genetically engineered to go after a selection of antigens that is tailored to each person’s case, and the patients themselves have to be “lymphodepleted” with chemotherapy before these are infused back. It’s a difficult, expensive process, and (like anything in this category) is very hard on the patients themselves (although, to be sure, not as hard as dying from the disease some weeks or months later, which by this point is the only remaining alternative).
One long-sought goal has been the “off the shelf” CAR-T protocol, where the T cells don’t have to be painstakingly personalized, but can be prepared from the blood cells of other healthy donors, modified to take on the most widespread antigens as is. This has not been easy; let’s just leave it at that rather than review the landscape in detail. But several organizations are continuing to work on this. A team at Vanderbilt, for example, is running a trial that reported some interim results late last year: no dose-limiting tox or severe cytokine release so far, which is good, but it’s too early to get a read on efficacy. And a company called Allogene is running a trial to see if their CAR-T can delay relapse in treated lymphoma patients, which is an interesting idea that hasn’t really been put to the test. So there are a couple of layers of risk in there, which are well explained by Adam Feuerstein here at Stat. First results are supposed to appear next month.
Another idea is to produce the CAR-T cells right within the patient’s own body, rather than going through all that ex vivo work. For that, you’re talking gene therapy to introduce the chimeric antigen receptor into the patient’s T cells, and that’s a pretty bold move. A team in Wuhan has just reported some early data on a lentivirus vector they’re using for this purpose (technology that AstraZeneca bought from them last year), and another report also appeared recently.
It’s a mixed bag. In this latest paper, five patients with advanced multiple myeloma were studied and followed for months after a single i.v. infusion of the viral vector. There was no lymphodepletion done on the patients beforehand. No dose-limiting toxicity was seen, but all the patients experienced adverse events: four showed cytokine release syndrome, three of them developed infections, and there were also transient liver enzyme elevations. Unfortunately one of the five died outright from “immune effector cell-associated neurotoxicity”, which is of course Not Good. But the others all appear to have responded to the treatment, with three “stringent complete remissions”.
Monitoring the blood of these patients showed the uptake of the viral vector and production of the modified T-cells, as well as the gradual restoration of normal B cells along the way as the malignant ones (and presumably their precursors) were destroyed. The authors make the case that this is a distinct clinical course rather than just a faster version of the present CAR-T treatments, and it’s going to need a lot of close attention to understand all the differences. So this is promising and alarming at the same time: it looks like you really can treat refractory multiple myeloma in this fashion, but is there really going to be a 20% fatality rate as you try to do it? You have to hope - and AstraZeneca has to hope - that that’s not the case, but it’s the earliest of early days with this technique, and there’s a lot yet to learn.
Here’s a look at one of the most famous opioid agonists in the world, fentanyl. You might think that by this point we would have a pretty thorough understanding of the drug’s behavior and the factors that lead to its (notorious) potency, but such GPCR ligands it seems that there’s always more to discover. The authors here are working off recent results from patch-clamp receptor assays in whole cells that suggested that fentanyl (but not the classic opioid ligand morphine) can actually (re)activate the mu-opioid receptor after being washed out of the system. The hypothesis that this led to was that fentanyl can partition into the cell membrane in a way that morphine can’t, and that this protects it from the washout step and acts as a local reservoir of drug that can continue to affect the membrane-bound opioid receptors. The new paper presents some detailed calculations about the membrane behavior of those two drugs along with a third agonist (isotonitazene) along with the antagonist naloxone. You can see the structures of these four drugs and a simulation of the membrane region at right (the small red dots are water molecules; the larger orange balls are the phosphate end of the phospholipid bilayer). What they find, using advanced molecular dynamics calculations, is that there really do seem to be some large differences between fentanyl’s behavior and the others. Kinetically, it seems to penetrate the membrane much faster (100x or more compared the morphine). For one thing, morphine seems to be making more hydrogen-bond interactions with the phospholipid head groups, which slows down its progress. Other differences are how deep in the membrane the deprotonations of the amines in these molecules take place, and how many water molecules they are dragging along with them. (Remember, the general belief is that molecules need to be in neutral form for efficient transport through the membrane bilayer). The simulations recapitulate the “membrane depot” idea for fentanyl (and do not show this effect for the other ligands). Fentanyl molecules permeate the cell membrane and can move rapidly in both directions (into the cytoplasm and back out onto the cell surface). It’s even possible that fentanyl might be hitting some lipid-facing binding spot in the mu-opioid receptor - that is, interacting directly from inside the membrane. The authors also did some real experimental work that verified the reactivation-after-washout results reported earlier, and they found that after such washouts that the cell membranes did continue to “leak” fentanyl molecules out into solution. There are still more details that need to be worked out, but so far the evidence is pointing in this direction.
Show full content
Here’s a look at one of the most famous opioid agonists in the world, fentanyl. You might think that by this point we would have a pretty thorough understanding of the drug’s behavior and the factors that lead to its (notorious) potency, but such GPCR ligands it seems that there’s always more to discover.
The authors here are working off recent results from patch-clamp receptor assays in whole cells that suggested that fentanyl (but not the classic opioid ligand morphine) can actually (re)activate the mu-opioid receptor after being washed out of the system. The hypothesis that this led to was that fentanyl can partition into the cell membrane in a way that morphine can’t, and that this protects it from the washout step and acts as a local reservoir of drug that can continue to affect the membrane-bound opioid receptors. The new paper presents some detailed calculations about the membrane behavior of those two drugs along with a third agonist (isotonitazene) along with the antagonist naloxone.
You can see the structures of these four drugs and a simulation of the membrane region at right (the small red dots are water molecules; the larger orange balls are the phosphate end of the phospholipid bilayer). What they find, using advanced molecular dynamics calculations, is that there really do seem to be some large differences between fentanyl’s behavior and the others. Kinetically, it seems to penetrate the membrane much faster (100x or more compared the morphine). For one thing, morphine seems to be making more hydrogen-bond interactions with the phospholipid head groups, which slows down its progress. Other differences are how deep in the membrane the deprotonations of the amines in these molecules take place, and how many water molecules they are dragging along with them. (Remember, the general belief is that molecules need to be in neutral form for efficient transport through the membrane bilayer).
The simulations recapitulate the “membrane depot” idea for fentanyl (and do not show this effect for the other ligands). Fentanyl molecules permeate the cell membrane and can move rapidly in both directions (into the cytoplasm and back out onto the cell surface). It’s even possible that fentanyl might be hitting some lipid-facing binding spot in the mu-opioid receptor - that is, interacting directly from inside the membrane.
The authors also did some real experimental work that verified the reactivation-after-washout results reported earlier, and they found that after such washouts that the cell membranes did continue to “leak” fentanyl molecules out into solution. There are still more details that need to be worked out, but so far the evidence is pointing in this direction.
Time to think like a biochemist! And that means paying close attention to energy transactions in a cell, because the science of thermodynamics makes it very, very clear there there is always a bill that has to be paid. Most of the time, the currency involved is of course ATP, since breaking that down to ADP gives you an immediate payout. ATP hydrolysis and the other sources of chemical energy (like acetyl-CoA) function as “battery packs” for all sorts of enzymatic processes. If you find an enzyme without one (like acetylcholinesterase), it means that it’s catalyzing a reaction that’s already thermodynamically favorable but just needs a rate increase - indeed, acetylcholine itself is synthesized by use of the energy in the acetyl CoA thioester bond, so breaking it back down is a favorable downhill process. This new paper has found a new source of such stored energy, and it’s a neat one. There’s an evolutionarily conserved protein motif called the “death fold” that’s found in the proteins that effect deliberate cell death (apoptosis), and in several other processes like innate immunity cascades, inflammation signaling through nF-kappaB, and more. These are known to work by self-assembly - the different death-fold domains each interact only with others of their own type and rapidly make large multi-subunit structures made up just of those proteins. That’s all fine, but now look at it from a thermodynamic perspective. With respect to the death-fold proteins involved in immunity and inflammation, this assembly does not appear to be driven by any of those chemical battery sources that we know about - in fact, it doesn’t need one, because it seems to be thermodynamically favorable to start with. But if that’s the case, what’s the barrier to it happening spontaneously? How can you have large local concentrations of death-fold proteins without them suddenly rolling downhill into self-assembly? There have been some rather detailed arguments about the energetics of the whole system, but this new work seems like it’s answered the question. And that answer is supersaturation. This is a phenomenon that you can demonstrate with one of those warming pads that you activate by clicking some sort of metal disk sitting inside them. It just sits around as a bag of goo until you do that, and then it gradually gets thicker as it gives off a pretty respectable amount of heat. These things are filled with a solution of sodium acetate, a very innocuous compound indeed with an interesting property. In general, when you dissolve some substance you will often hit a limit beyond which no more can be dissolved. Think of dumping salt into a glass of water - as you stir it, you will dissolve the salt crystals, but eventually you’ll reach a point where no more of it goes in and you’ll be stirring a layer of salt on the bottom of the glass that will simply not go into solution for you. That is a saturated solution, but it’s important to note that the saturation amount is dependent on temperature: if you heat that glass up, you will be able to dissolve more salt, but if you then let it cool down the “carrying capacity” of the solution drops back down in turn and some of that salt is going to crystallize back out until you get to that saturation point at room temperature again. This is of course how chemists do recrystallizations - making a “supersaturated” solution and letting it drop crystals back out again (preferably slowly and aesthetically!) But there are some odd cases, and sodium acetate is definitely one. You can get pure sodium acetate (anhydrous, no water present), but you can also get a crystalline form that has three water molecules in the crystal per molecule of sodium acetate (the trihydrate form). If you take some of that trihydrate and heat it up, that orderly arrangement breaks up and the water molecules are liberated, whereupon they dissolve the sodium acetate molecules. If you let that solution cool, you’d expect to get those trihydrate crystals forming again in the exact reverse process, but you don’t. The solution gets “stuck” in a higher-energy state because it’s difficult to recreate the trihydrate crystal form from that direction. You end up with a solution that’s (very) supersaturated with respect to sodium acetate trihydrate - the whole thing would turn solid again if it could just find a way to get there. At the bench, chemists often try tricks like scratching the inside of a glass flask under the liquid surface to provide a bit of a kick and form some tiny crystals in such a solution, and that’s what clicking that metal disk does in the heat packs. You form a small number of microscopic sodium acetate trihydrate crystals that way, and that sets things off - now everything starts templating off of those and the whole solution starts coming apart again, with a release of that stored energy, which is given off as heat. (If you toss such a used heat back into boiling water, you can take it right back around again to the supersaturated solution). This, it turns out, seems to be exactly what the death-fold-domain proteins are doing, but only the ones in the innate immunity pathways. The others need cellular energy to assemble, but this particular set can exist as supersaturated solutions under cellular conditions, piled up and ready to go if the right stimulus starts the cascade. The authors refer to this as a “phase-change battery”, and I think that’s exactly right - it’s a stored higher-energy state and it lets the proteins and their signaling network respond quickly without the need for an influx of energy from the rest of the cell. Note the differences, though, between this and the liquid-liquid phase separation found in intracellular condensate droplets. Those are reversible (unless of course something goes wrong, and there are some disease states that seem to be connected to such failures), but the immunity DFD protein assemblies are one-way. That’s in a general sense, too: for many innate immunity pathways, the instructions to the cell are “Kill yourself now before this infectious agent takes hold”. The authors find evidence of evolutionary optimization of this behavior, leading to low saturation concentrations along with barriers to nucleation between individual proteins that give them the chance to pile up into supersaturated states. Their expression levels line up with this interpretation as well, and needless to say, it is very unusual to see proteins that are being expressed at greater than their saturation levels in the cell (!) But very interestingly, the paper shows that the amount of such supersaturation in different cell types seems to be quite well correlated with the lifespan of the cells. The inference is that relying on this mechanism, while useful (and deeply conserved in evolution), also comes with an unavoidable risk of sudden death if something manages to start the self-assembly and release the DFD proteins from their supersaturated state. So these factors have balanced out to a small and optimized group of them, and the cells expressing significant amounts of them are not allowed to hang around forever. The authors sum it up very well: Our findings imply that cells perpetually await death. The theoretical cumulative certainty of stochastic nucleation over time appears to be reflected in the observed relationship of DFD supersaturation to mortality rates across human cell types. We speculate that this underpins a fundamental tradeoff between innate immunity and life expectancy, potentially contributing to age-related inflammation and stem cell exhaustion.
Show full content
Time to think like a biochemist! And that means paying close attention to energy transactions in a cell, because the science of thermodynamics makes it very, very clear there there is always a bill that has to be paid. Most of the time, the currency involved is of course ATP, since breaking that down to ADP gives you an immediate payout. ATP hydrolysis and the other sources of chemical energy (like acetyl-CoA) function as “battery packs” for all sorts of enzymatic processes. If you find an enzyme without one (like acetylcholinesterase), it means that it’s catalyzing a reaction that’s already thermodynamically favorable but just needs a rate increase - indeed, acetylcholine itself is synthesized by use of the energy in the acetyl CoA thioester bond, so breaking it back down is a favorable downhill process.
This new paper has found a new source of such stored energy, and it’s a neat one. There’s an evolutionarily conserved protein motif called the “death fold” that’s found in the proteins that effect deliberate cell death (apoptosis), and in several other processes like innate immunity cascades, inflammation signaling through nF-kappaB, and more. These are known to work by self-assembly - the different death-fold domains each interact only with others of their own type and rapidly make large multi-subunit structures made up just of those proteins.
That’s all fine, but now look at it from a thermodynamic perspective. With respect to the death-fold proteins involved in immunity and inflammation, this assembly does not appear to be driven by any of those chemical battery sources that we know about - in fact, it doesn’t need one, because it seems to be thermodynamically favorable to start with. But if that’s the case, what’s the barrier to it happening spontaneously? How can you have large local concentrations of death-fold proteins without them suddenly rolling downhill into self-assembly? There have been some rather detailed arguments about the energetics of the whole system, but this new work seems like it’s answered the question.
And that answer is supersaturation. This is a phenomenon that you can demonstrate with one of those warming pads that you activate by clicking some sort of metal disk sitting inside them. It just sits around as a bag of goo until you do that, and then it gradually gets thicker as it gives off a pretty respectable amount of heat. These things are filled with a solution of sodium acetate, a very innocuous compound indeed with an interesting property.
In general, when you dissolve some substance you will often hit a limit beyond which no more can be dissolved. Think of dumping salt into a glass of water - as you stir it, you will dissolve the salt crystals, but eventually you’ll reach a point where no more of it goes in and you’ll be stirring a layer of salt on the bottom of the glass that will simply not go into solution for you. That is a saturated solution, but it’s important to note that the saturation amount is dependent on temperature: if you heat that glass up, you will be able to dissolve more salt, but if you then let it cool down the “carrying capacity” of the solution drops back down in turn and some of that salt is going to crystallize back out until you get to that saturation point at room temperature again.
This is of course how chemists do recrystallizations - making a “supersaturated” solution and letting it drop crystals back out again (preferably slowly and aesthetically!) But there are some odd cases, and sodium acetate is definitely one. You can get pure sodium acetate (anhydrous, no water present), but you can also get a crystalline form that has three water molecules in the crystal per molecule of sodium acetate (the trihydrate form). If you take some of that trihydrate and heat it up, that orderly arrangement breaks up and the water molecules are liberated, whereupon they dissolve the sodium acetate molecules. If you let that solution cool, you’d expect to get those trihydrate crystals forming again in the exact reverse process, but you don’t. The solution gets “stuck” in a higher-energy state because it’s difficult to recreate the trihydrate crystal form from that direction. You end up with a solution that’s (very) supersaturated with respect to sodium acetate trihydrate - the whole thing would turn solid again if it could just find a way to get there.
At the bench, chemists often try tricks like scratching the inside of a glass flask under the liquid surface to provide a bit of a kick and form some tiny crystals in such a solution, and that’s what clicking that metal disk does in the heat packs. You form a small number of microscopic sodium acetate trihydrate crystals that way, and that sets things off - now everything starts templating off of those and the whole solution starts coming apart again, with a release of that stored energy, which is given off as heat. (If you toss such a used heat back into boiling water, you can take it right back around again to the supersaturated solution).
This, it turns out, seems to be exactly what the death-fold-domain proteins are doing, but only the ones in the innate immunity pathways. The others need cellular energy to assemble, but this particular set can exist as supersaturated solutions under cellular conditions, piled up and ready to go if the right stimulus starts the cascade. The authors refer to this as a “phase-change battery”, and I think that’s exactly right - it’s a stored higher-energy state and it lets the proteins and their signaling network respond quickly without the need for an influx of energy from the rest of the cell. Note the differences, though, between this and the liquid-liquid phase separation found in intracellular condensate droplets. Those are reversible (unless of course something goes wrong, and there are some disease states that seem to be connected to such failures), but the immunity DFD protein assemblies are one-way. That’s in a general sense, too: for many innate immunity pathways, the instructions to the cell are “Kill yourself now before this infectious agent takes hold”.
The authors find evidence of evolutionary optimization of this behavior, leading to low saturation concentrations along with barriers to nucleation between individual proteins that give them the chance to pile up into supersaturated states. Their expression levels line up with this interpretation as well, and needless to say, it is very unusual to see proteins that are being expressed at greater than their saturation levels in the cell (!) But very interestingly, the paper shows that the amount of such supersaturation in different cell types seems to be quite well correlated with the lifespan of the cells. The inference is that relying on this mechanism, while useful (and deeply conserved in evolution), also comes with an unavoidable risk of sudden death if something manages to start the self-assembly and release the DFD proteins from their supersaturated state. So these factors have balanced out to a small and optimized group of them, and the cells expressing significant amounts of them are not allowed to hang around forever. The authors sum it up very well:
Our findings imply that cells perpetually await death. The theoretical cumulative certainty of stochastic nucleation over time appears to be reflected in the observed relationship of DFD supersaturation to mortality rates across human cell types. We speculate that this underpins a fundamental tradeoff between innate immunity and life expectancy, potentially contributing to age-related inflammation and stem cell exhaustion.
Here’s a paper evaluating a popular AI/ML model for cofolding ligands and proteins, Boltz-2. This is of course a problem of extreme interest to the drug discovery community, as well as to all sorts of people working on cell biology, structural biology, and related fields. It’s been one of the goals for decades to start from scratch with a protein sequence and a small molecule and be able to say “Does this molecule bind to this protein? How well?” And no, we really haven’t been able to do that, not in the way that we’d like (and certainly not on the scale that we’d like). The error bars on those binding predictions have generally been too wide, and that’s both on the underlying structure of the protein and on the energetic implications of how how it interacts with a given small-molecule ligand. And the computational burdens of even getting that far have generally been too great, given the number of conformations you’re likely to need to examine (and the way that you’ll need to evaluate which of those are most plausible relative to the others). Protein structure from scratch was of course a notoriously hard problem for decades as well, but machine learning off the databases of known protein structures (AlphaFold, RosETTAFold and the like) have made terrific progress by identifying the often-reused structural motifs and their effects on overall tertiary protein structure as they’re combined. But protein-with-ligand, that one is still the Holy Grail. If we could get that to work well, and get the speed up and the computational overhead down, then we’d perhaps be able to finally achieve primary-screen nirvana in silico. No need to make and purify protein, no need to have a basement full of hundreds of thousands of small-molecule candidates in vials and plates. No robot arms, no fluorescent plate readers. Just fire up the computational hardware and software and go get some lunch instead. Boltz-2 is one of the open-source alternatives to software like AlphaFold 3, all of which are trying to address this problem. And it is claimed to produce protein structures at AlphaFold levels of accuracy while simultaneously predicting binding affinity energies at a level similar to the most computationally intensive methods (like free energy perturbation) but hundreds of times faster. So as you can imagine, it and the other programs in this space have gotten a lot of attention. As the paper linked above notes, so far it looks like this software is at its best when working with rather locked-down protein structures and known binding structures - that is, when working on Easy Mode. Unfortunately, we don’t spend much time on Easy Mode in the wonder drug factories. We have a lot of other things to worry about: proteins that don’t have much (or any) good experimental structural data, binding sites that depend crucially on the effects of water molecules, small-molecule cofactors, or on the binding of another ligand at a completely different allosteric pocket. And some of the binding events we’re looking at turn out (once we get real-world data) to involve significant shifts in the original protein structure and/or rather odd twists in the conformation of ligand molecules, neither of which are easy to compute your way to. (You end up paying a lot of energetic penalties if you try to advance step-by-step, and the system may well throw in the towel before the big unexpected energetic payoff at the end). In this paper the authors use a set of 943 virtual-screen hits from some very large previous screening efforts (hundreds of millions of candidates), binding to ten different target proteins, and with the associated real-world in vitro binding data already in hand. Those comprise 364 true positives and 579 false positives, as discovered when those assays were run. The paper notes that so far no affinity-prediction systems have been able to really tell those true positives from false positives in this data set, so this is an adversarial challenge for sure, and just the thing to let a hot new piece of software get its virtual teeth into. Most of the targets, as it happens, are G-protein coupled receptors, althought the binding site diversity is still very high. Boltz-2 ran through the 943 candidates at about two minutes per, and it got all of them into the right binding pocket (as it darn well should - all of these targets have ligand-bound structures in the PDB already). And its predictions for “Is this compound likely to be a good binder?” are notably better than any other method tested (with the exception of two targets on which it failed pretty thoroughly). So it really distinguished itself on finding the true positives. This did not seem to be due to similarities between these compounds and the Boltz-2 training set (which is something you always need to be wary of). That said, its actual predictions of affinity were quite poor as compared to the experimentally determined values. And when the actual structures of the ligands in the binding pockets was examined, the Boltz-2 predictions were pretty far off of what is believed to be the actual situation. Even odder, the accuracy of distinguishing true positives did not seem to be affected by the quality of the docking poses, which is rather counterintuitive. At this point the authors were mindful of a report that came out last year about AlphaFold 3 docking predictions. That work noted that AF3 poses and predictions seemed curiously insensitive to amino acid mutations in the binding site(s) that should severely affect such results. These clearly nonphysical results suggest a great deal of overfitting to the training data, or to particular trends in it, and caused those authors to warn people about relying too much on such deep-learning models. So the authors in this latest paper tried the same trick: introducing amino acid changes that would absolutely blow up important polar interactions between the ligands and their binding sites. We’re talking aspartic acid to alanine, that sort of thing, or dropping a proline into the hinge region of a kinase. These are grenades. Unfortunately, Boltz-2 emerged from that challenge with predictions that were for the most part not statistically different from the ones generated from the wild-type structures. What’s more, the poses of the compounds in these messed-up binding sites seemed to have little in common with what it had generated earlier - i.e., it didn’t hang on to the other interactions it had found while just making the best of it with the abrogated ones. Further alanine-scan mutations (up to six per binding site!) made it clear the Boltz-2 just didn’t care much about such petty details. Even reshuffling the target proteins completely and assigning random ones to the ligands (where they would be expected to have no binding whatsoever) only got rid of about half the true-positive recommendations. For the others, predictions of affinity seemed to be almost independent of what target they chose. This is not what you want. In fact, it’s the opposite of what you want. The authors “therefore advise remaining very skeptical with respect to affinity predictions” from the program (and others like it, you’d have to say) and I completely agree. It is very tempting to look at the outputs of such software and to tell yourself that it must have a deep understanding of the physics and energetics of protein folding and compound binding. But that is an illusion. Large computational models do not understand anything, any more than LLM chatbots know what they are “saying”. We have built these systems to provide blurry copies of what were actually useful and pleasing outputs that we generated by our own efforts, and sometimes the results, the extruded simulated products, are worthwhile enough and sometimes they are not. In the case of overfitted models, which these seem to be, we are at great risk of just talking to ourselves and playing our own voices back to no good effect. Understanding is still our human domain. And we need to understand not only the physics of small-molecule interactions, but the workings of our own tools.
Show full content
Here’s a paper evaluating a popular AI/ML model for cofolding ligands and proteins, Boltz-2. This is of course a problem of extreme interest to the drug discovery community, as well as to all sorts of people working on cell biology, structural biology, and related fields. It’s been one of the goals for decades to start from scratch with a protein sequence and a small molecule and be able to say “Does this molecule bind to this protein? How well?”
And no, we really haven’t been able to do that, not in the way that we’d like (and certainly not on the scale that we’d like). The error bars on those binding predictions have generally been too wide, and that’s both on the underlying structure of the protein and on the energetic implications of how how it interacts with a given small-molecule ligand. And the computational burdens of even getting that far have generally been too great, given the number of conformations you’re likely to need to examine (and the way that you’ll need to evaluate which of those are most plausible relative to the others).
Protein structure from scratch was of course a notoriously hard problem for decades as well, but machine learning off the databases of known protein structures (AlphaFold, RosETTAFold and the like) have made terrific progress by identifying the often-reused structural motifs and their effects on overall tertiary protein structure as they’re combined. But protein-with-ligand, that one is still the Holy Grail. If we could get that to work well, and get the speed up and the computational overhead down, then we’d perhaps be able to finally achieve primary-screen nirvana in silico. No need to make and purify protein, no need to have a basement full of hundreds of thousands of small-molecule candidates in vials and plates. No robot arms, no fluorescent plate readers. Just fire up the computational hardware and software and go get some lunch instead.
Boltz-2 is one of the open-source alternatives to software like AlphaFold 3, all of which are trying to address this problem. And it is claimed to produce protein structures at AlphaFold levels of accuracy while simultaneously predicting binding affinity energies at a level similar to the most computationally intensive methods (like free energy perturbation) but hundreds of times faster. So as you can imagine, it and the other programs in this space have gotten a lot of attention.
As the paper linked above notes, so far it looks like this software is at its best when working with rather locked-down protein structures and known binding structures - that is, when working on Easy Mode. Unfortunately, we don’t spend much time on Easy Mode in the wonder drug factories. We have a lot of other things to worry about: proteins that don’t have much (or any) good experimental structural data, binding sites that depend crucially on the effects of water molecules, small-molecule cofactors, or on the binding of another ligand at a completely different allosteric pocket. And some of the binding events we’re looking at turn out (once we get real-world data) to involve significant shifts in the original protein structure and/or rather odd twists in the conformation of ligand molecules, neither of which are easy to compute your way to. (You end up paying a lot of energetic penalties if you try to advance step-by-step, and the system may well throw in the towel before the big unexpected energetic payoff at the end).
In this paper the authors use a set of 943 virtual-screen hits from some very large previous screening efforts (hundreds of millions of candidates), binding to ten different target proteins, and with the associated real-world in vitro binding data already in hand. Those comprise 364 true positives and 579 false positives, as discovered when those assays were run. The paper notes that so far no affinity-prediction systems have been able to really tell those true positives from false positives in this data set, so this is an adversarial challenge for sure, and just the thing to let a hot new piece of software get its virtual teeth into. Most of the targets, as it happens, are G-protein coupled receptors, althought the binding site diversity is still very high.
Boltz-2 ran through the 943 candidates at about two minutes per, and it got all of them into the right binding pocket (as it darn well should - all of these targets have ligand-bound structures in the PDB already). And its predictions for “Is this compound likely to be a good binder?” are notably better than any other method tested (with the exception of two targets on which it failed pretty thoroughly). So it really distinguished itself on finding the true positives. This did not seem to be due to similarities between these compounds and the Boltz-2 training set (which is something you always need to be wary of).
That said, its actual predictions of affinity were quite poor as compared to the experimentally determined values. And when the actual structures of the ligands in the binding pockets was examined, the Boltz-2 predictions were pretty far off of what is believed to be the actual situation. Even odder, the accuracy of distinguishing true positives did not seem to be affected by the quality of the docking poses, which is rather counterintuitive.
At this point the authors were mindful of a report that came out last year about AlphaFold 3 docking predictions. That work noted that AF3 poses and predictions seemed curiously insensitive to amino acid mutations in the binding site(s) that should severely affect such results. These clearly nonphysical results suggest a great deal of overfitting to the training data, or to particular trends in it, and caused those authors to warn people about relying too much on such deep-learning models. So the authors in this latest paper tried the same trick: introducing amino acid changes that would absolutely blow up important polar interactions between the ligands and their binding sites. We’re talking aspartic acid to alanine, that sort of thing, or dropping a proline into the hinge region of a kinase. These are grenades.
Unfortunately, Boltz-2 emerged from that challenge with predictions that were for the most part not statistically different from the ones generated from the wild-type structures. What’s more, the poses of the compounds in these messed-up binding sites seemed to have little in common with what it had generated earlier - i.e., it didn’t hang on to the other interactions it had found while just making the best of it with the abrogated ones. Further alanine-scan mutations (up to six per binding site!) made it clear the Boltz-2 just didn’t care much about such petty details.
Even reshuffling the target proteins completely and assigning random ones to the ligands (where they would be expected to have no binding whatsoever) only got rid of about half the true-positive recommendations. For the others, predictions of affinity seemed to be almost independent of what target they chose. This is not what you want. In fact, it’s the opposite of what you want. The authors “therefore advise remaining very skeptical with respect to affinity predictions” from the program (and others like it, you’d have to say) and I completely agree.
It is very tempting to look at the outputs of such software and to tell yourself that it must have a deep understanding of the physics and energetics of protein folding and compound binding. But that is an illusion. Large computational models do not understand anything, any more than LLM chatbots know what they are “saying”. We have built these systems to provide blurry copies of what were actually useful and pleasing outputs that we generated by our own efforts, and sometimes the results, the extruded simulated products, are worthwhile enough and sometimes they are not. In the case of overfitted models, which these seem to be, we are at great risk of just talking to ourselves and playing our own voices back to no good effect.
Understanding is still our human domain. And we need to understand not only the physics of small-molecule interactions, but the workings of our own tools.
Here’s the other cognition/aging/Alzheimer’s paper that caught my eye. In a similar way to the work I highlighted yesterday on proteins released by the liver affecting the blood-brain barrier and overall brain function, this one is finding another external signal, from from an unexpected direction. The authors studied the intestinal microbiomes of mice as they aged, and found that species that produce medium-chain fatty acids become more and more prevalent. Then a complex series of events start taking place: these metabolites are ligands for the human GPR84 protein and can drive inflammation in myeloid cells through that activation. This in turn weakens the neural traffic through the vagal system, and this loss of “interoceptive” signaling to the brain leads to a decline in hippocampal function. Impaired memory, in other words. Now that’s one that I wouldn’t have seen coming, but as the paper shows in its references, there are a number of other reports pointing in this microbiome/memory direction. Now overall, the signal/noise of microbiome work is not as high as it should be, but papers like this new one are an important step in shoring up such hypotheses, trying to bridge some of the “by some mechanism that we haven’t figured out yet” gaps. It’s not that new and interesting ideas have to eliminate all of those leaps, but if you have to invoke that sort of thing too many times you’re asking for trouble. Here’s how you avoid that (hint: it involves an awful lot of work). One experiment done here was to house very young (two-month old) mice with old (18-month-old) ones, which led to exchange of microbiome species between the two cohorts and an equilibrium that looked quite a bit more like the old ones. This didn’t seem to have any real effects on physical health and energy levels (or even things like exploratory behavior), but the short-term and long-term memory task performance of the young mice declined. To control for social effects, the team tried things like direct faecal microbiome transplants from the old mice into the young ones, and this recapitulated the memory effects all by itself. Meanwhile, co-housing gnotobiotic (germ-free) mice of both age groups did not affect the memories of the younger ones. Similarly, treating regular groups of young and old mice with two weeks of antibiotics also restored memory task performance in both cohorts. And yes, the aged germ-free mice also performed much better on memory tests than the ones with typical microbiomes, so all of these results point in the same direction. There appears to be a gut microbial factor that impairs memory in older mice. Looking at the bacterial species that were present across different ages, Parabacteroides goldsteinii looked like the top candidate. (That one has already been the subject of a great deal of microbiome work in humans, as that link will show) Colonizing either germ-free or post-antibiotic mice with this species alone brought on the memory trouble, but this effect could not be demonstrated with other species that increased with age, nor with some that showed no real change as the mice aged. Looking at the brains of the impaired mice, it appeared that neuronal function was disrupted in the hippocampus and in several areas known to be involved with sensory processing. A weird and interesting result was that many of the neurons involved in the vagus nerve’s gut-to-brain connections express the TRPV1 vanilloid receptor. Chemogenetic silencing of this receptor gave memory behavior similar to the aged mice, while activation of it seemed to restore function in the elderly cohort. That extended even to such low-tech methods as giving the mice capsaicin as a TRPV1 agonist (!) Other gut-responsive signals such as CCK or GLP-1 showed improvements in the presence of added agonists, although their underlying levels were not changed with aging and/or P. goldsteinii infection. Further experiments showed that (as mentioned above) medium-chain fatty acids produced by those bacteria seem to be the actual signal driving these effects. Oral administration of things like decanoic acid and 3-hydroxyoctanoic acid were enough to affect cognition by themselves, and demonstrated effects along the whole causal chain the above work had laid out (vagus nerve activation and the sensory and hippocampal brain regions). These are known to be ligands for GPR84, and the team showed that mice with inactivating mutations in that receptor were immune to the effects of added medium-chain acids and showed delayed onset of memory trouble in general as compared to wild-type mice. The receptor is largely found in myeloid cells (macrophages, monocytes, and neutrophils) and ablating these also restored memory function (demonstrated through a set of bone-marrow experiments). This looks to me like a very solid paper where the authors have tried to shore up every step of their hypothesis. Inflammation-driven defects in interoceptive signaling truly does look like a cause of memory decline in mice: but does it work that way in humans? You can bet that work is going on as we speak to find that out, but this pathway fits in very well with the overall idea that inappropriate inflammation is a driver of age-related brain dysfunction. But I have to say, we weren’t looking for it first in the gut rather than directly in the brain! There’s clearly a lot of work to be done here, and direct pharmacological intervention in these interoceptive pathways could really be beneficial. Starting with more hot sauce, given those capsaicin results? Try it today!
Show full content
Here’s the other cognition/aging/Alzheimer’s paper that caught my eye. In a similar way to the work I highlighted yesterday on proteins released by the liver affecting the blood-brain barrier and overall brain function, this one is finding another external signal, from from an unexpected direction.
The authors studied the intestinal microbiomes of mice as they aged, and found that species that produce medium-chain fatty acids become more and more prevalent. Then a complex series of events start taking place: these metabolites are ligands for the human GPR84 protein and can drive inflammation in myeloid cells through that activation. This in turn weakens the neural traffic through the vagal system, and this loss of “interoceptive” signaling to the brain leads to a decline in hippocampal function. Impaired memory, in other words.
Now that’s one that I wouldn’t have seen coming, but as the paper shows in its references, there are a number of other reports pointing in this microbiome/memory direction. Now overall, the signal/noise of microbiome work is not as high as it should be, but papers like this new one are an important step in shoring up such hypotheses, trying to bridge some of the “by some mechanism that we haven’t figured out yet” gaps. It’s not that new and interesting ideas have to eliminate all of those leaps, but if you have to invoke that sort of thing too many times you’re asking for trouble. Here’s how you avoid that (hint: it involves an awful lot of work).
One experiment done here was to house very young (two-month old) mice with old (18-month-old) ones, which led to exchange of microbiome species between the two cohorts and an equilibrium that looked quite a bit more like the old ones. This didn’t seem to have any real effects on physical health and energy levels (or even things like exploratory behavior), but the short-term and long-term memory task performance of the young mice declined. To control for social effects, the team tried things like direct faecal microbiome transplants from the old mice into the young ones, and this recapitulated the memory effects all by itself. Meanwhile, co-housing gnotobiotic (germ-free) mice of both age groups did not affect the memories of the younger ones. Similarly, treating regular groups of young and old mice with two weeks of antibiotics also restored memory task performance in both cohorts.
And yes, the aged germ-free mice also performed much better on memory tests than the ones with typical microbiomes, so all of these results point in the same direction. There appears to be a gut microbial factor that impairs memory in older mice. Looking at the bacterial species that were present across different ages, Parabacteroides goldsteinii looked like the top candidate. (That one has already been the subject of a great deal of microbiome work in humans, as that link will show) Colonizing either germ-free or post-antibiotic mice with this species alone brought on the memory trouble, but this effect could not be demonstrated with other species that increased with age, nor with some that showed no real change as the mice aged.
Looking at the brains of the impaired mice, it appeared that neuronal function was disrupted in the hippocampus and in several areas known to be involved with sensory processing. A weird and interesting result was that many of the neurons involved in the vagus nerve’s gut-to-brain connections express the TRPV1 vanilloid receptor. Chemogenetic silencing of this receptor gave memory behavior similar to the aged mice, while activation of it seemed to restore function in the elderly cohort. That extended even to such low-tech methods as giving the mice capsaicin as a TRPV1 agonist (!) Other gut-responsive signals such as CCK or GLP-1 showed improvements in the presence of added agonists, although their underlying levels were not changed with aging and/or P. goldsteinii infection.
Further experiments showed that (as mentioned above) medium-chain fatty acids produced by those bacteria seem to be the actual signal driving these effects. Oral administration of things like decanoic acid and 3-hydroxyoctanoic acid were enough to affect cognition by themselves, and demonstrated effects along the whole causal chain the above work had laid out (vagus nerve activation and the sensory and hippocampal brain regions). These are known to be ligands for GPR84, and the team showed that mice with inactivating mutations in that receptor were immune to the effects of added medium-chain acids and showed delayed onset of memory trouble in general as compared to wild-type mice. The receptor is largely found in myeloid cells (macrophages, monocytes, and neutrophils) and ablating these also restored memory function (demonstrated through a set of bone-marrow experiments).
This looks to me like a very solid paper where the authors have tried to shore up every step of their hypothesis. Inflammation-driven defects in interoceptive signaling truly does look like a cause of memory decline in mice: but does it work that way in humans? You can bet that work is going on as we speak to find that out, but this pathway fits in very well with the overall idea that inappropriate inflammation is a driver of age-related brain dysfunction. But I have to say, we weren’t looking for it first in the gut rather than directly in the brain! There’s clearly a lot of work to be done here, and direct pharmacological intervention in these interoceptive pathways could really be beneficial. Starting with more hot sauce, given those capsaicin results? Try it today!
I’d like to highlight a couple of interesting papers with a bearing on Alzheimer’s and aging in general - well, I hope that they will turn out to have one, because right up front I have to note that they’re both in rodent models. But the unusual mechanisms in both cases are the sort of thing that could translate to humans, and could also be fairly readily checked. We’ll do one today and another tomorrow. This paper is building on some recent work showing that transfusing blood plasma from recently-exercised mice to aged sedentary mice can improve their cognitive function. A particular secreted enzyme, glycosylphosphatidylinositol (GPI)-specific phospholipase D1 (GPLD1) seems to be responsible for at least some of this effect, and the authors have named it an “exerkine” (an exercise-induced circulating blood factor). This GPLD1 does not itself seem to be crossing the blood-brain barrier, so in this paper the authors are hunting for peripheral mechanisms of action. That glycophosphatidylinositol is often used as an anchoring group to attach proteins to membranes, and GPLD1 goes around cleaving it. And this paper identifies tissue-nonspecific alkaline phosphatase (TNAP) as a key protein that is cleaved in this manner from cerebrovascular endothelial cells. The expression of that one seems to be significantly increased in aging tissue, although its function(s) are not well worked out yet, and (quite significantly) it was identified in a 2020 paper as a protein associated with impaired blood-brain transport in aging. (In that paper it’s referred to with a different acronym, ALPL). Here the paper shows that increased GPLD1 in the blood leads to higher levels of TNAP protein, since it’s being cleaved off of cell surfaces, and that losing it in this fashion leads to better cognitive performance in aging rodents. As shown in the 2020 paper just linked, you can recapitulate this effect by dosing a small-molecule inhibitor of TNAP, and they also show that increasing GPLD1 levels (by means other than exercise!) will do that, too. Blood-brain barrier function in the hippocampus of the elderly mice improves either way. And the reverse holds true, too: if you give young mice a viral vector to increase expression of TNAP, their BBB function and their cognitive performance goes down as well. These effects track down to the transcriptional level. So this is yet another route by which exercise is good for you! And you’d have to encourage that in general. But these discoveries also open up the possibility that targeting the TNAP enzyme directly could be a useful therapy, exercise aside. But it’s very likely that GPLD1’s beneficial effects aren’t limited to cleaving TNAP either, so that’s another promising area to investigate. The authors note that GPLD1 has been shown to increase after treatment with several lifespan-increasing treatments in mice (caloric restriction, rapamycin and more), so there could certainly be a lot going on. Past that one, there’s a whole list of exercise-induced proteins that have effects all over the body. So while we’re working on duplicating these pharmacologically, getting more of the real thing would surely be a good idea!
Show full content
I’d like to highlight a couple of interesting papers with a bearing on Alzheimer’s and aging in general - well, I hope that they will turn out to have one, because right up front I have to note that they’re both in rodent models. But the unusual mechanisms in both cases are the sort of thing that could translate to humans, and could also be fairly readily checked. We’ll do one today and another tomorrow.
This paper is building on some recent work showing that transfusing blood plasma from recently-exercised mice to aged sedentary mice can improve their cognitive function. A particular secreted enzyme, glycosylphosphatidylinositol (GPI)-specific phospholipase D1 (GPLD1) seems to be responsible for at least some of this effect, and the authors have named it an “exerkine” (an exercise-induced circulating blood factor). This GPLD1 does not itself seem to be crossing the blood-brain barrier, so in this paper the authors are hunting for peripheral mechanisms of action.
That glycophosphatidylinositol is often used as an anchoring group to attach proteins to membranes, and GPLD1 goes around cleaving it. And this paper identifies tissue-nonspecific alkaline phosphatase (TNAP) as a key protein that is cleaved in this manner from cerebrovascular endothelial cells. The expression of that one seems to be significantly increased in aging tissue, although its function(s) are not well worked out yet, and (quite significantly) it was identified in a 2020 paper as a protein associated with impaired blood-brain transport in aging. (In that paper it’s referred to with a different acronym, ALPL).
Here the paper shows that increased GPLD1 in the blood leads to higher levels of TNAP protein, since it’s being cleaved off of cell surfaces, and that losing it in this fashion leads to better cognitive performance in aging rodents. As shown in the 2020 paper just linked, you can recapitulate this effect by dosing a small-molecule inhibitor of TNAP, and they also show that increasing GPLD1 levels (by means other than exercise!) will do that, too. Blood-brain barrier function in the hippocampus of the elderly mice improves either way. And the reverse holds true, too: if you give young mice a viral vector to increase expression of TNAP, their BBB function and their cognitive performance goes down as well. These effects track down to the transcriptional level.
So this is yet another route by which exercise is good for you! And you’d have to encourage that in general. But these discoveries also open up the possibility that targeting the TNAP enzyme directly could be a useful therapy, exercise aside. But it’s very likely that GPLD1’s beneficial effects aren’t limited to cleaving TNAP either, so that’s another promising area to investigate. The authors note that GPLD1 has been shown to increase after treatment with several lifespan-increasing treatments in mice (caloric restriction, rapamycin and more), so there could certainly be a lot going on. Past that one, there’s a whole list of exercise-induced proteins that have effects all over the body. So while we’re working on duplicating these pharmacologically, getting more of the real thing would surely be a good idea!
I really got a kick out of this recent paper, but that’s probably because I’ve always enjoyed the whole “templated ligation” or “kinetic target-guided synthesis” field. There are several ways to describe it in the literature, which doesn’t make searching for papers any easier, but the basic idea is that you get two partner species to react in the presence of a binding site or surface. If you pick your chemistry right, that reaction/ligation/bond formation/condensation is something that normally has a low background rate, but which can be accelerated by the templated binding/induced proximity that’s brought on by the presence of the target species. In the ideal case, then, you get a situation where a large number of reaction combinations (i.e. a large number of products) are available, but the low background rate of the untemplated reaction keeps you from seeing them. The only products that you detect are the ones that had the advantage of being brought together by mutually beneficial binding events - that is, the only compounds that form are the ones that have built-in potent binding. Another way to think of it is arranging things so that a target selects its own ligands for you from the huge pool of combinatorial possibilities. One way people have done this is DCC, dynamic combinatorial chemistry. For that you use reversible bond-forming systems such as thiols/disulfides or perhaps aldehydes and amines/oximes. These are presumably interconverting constantly, mixing and matching in solution, but the addition of a binding template can (in theory) bias the mix towards the ones where the reactants are brought together more frequently, more closely, or for a longer residence time. There are numerous examples of this sort of thing in the literature, but I will admit that personally DCC leaves me a bit cold, because my own biases are for things that can turn into drug leads. The products of DCC-type reactions (the disulfides, oximes, and so on) are often not suitable for that use, and have to be modified to turn into plausible developable compounds. That can be quite a chasm to cross, because when you get right down to it, there’s nothing quite like a disulfide (bond lengths, polarity, torsional angles, electron density), and finding a druglike isostere can be no fun at all. So I’ve tended towards the systems that do irreversible bond formation, things like the Sharpless-style “click” triazole formation from azides and acetylenes. (Note that I don’t mean the copper-catalyzed ones that most people are familiar with - in this case you’re letting the 3+2 cycloaddition take place with no help other than forced proximity). There are other reactions that you can do that with, but each time you have to make sure that you’re not making the background rate too high. I wrote about one of those here a few years ago. The paper linked in the first paragraph is using a cycloaddition reaction (rather like the uncatalyzed triazole formation), the tetrazine-based inverse electron demand Diels-Alder reaction. That one of course has also been used in many chemical biology applications with the reactive trans-cyclooctene as a partner - like the triazole click reaction, the two partners are generally stable in biochemical systems and tend to react only with each other. But the new thing is this paper is that the tetrazine compounds are no mere small-molecule library but a peptide library of roughly one trillion members. And the norbornene coupling partner (a much less reactive one than usually used for "click" reactions) for that IED Diels-Alder is another such library: one trillion peptides each with a norbornene hanging off of them. These are prepared via mRNA display using the RaPID (Random nonstandard Peptides Integrated Discovery) system that uses a reworked genetic code to produce functionalized macrocyclic peptides. Now the question with all such experiments is: how on earth do you find your hits? The larger your libraries of screening and reaction partners, the smaller the amounts of any given hit you can expect to make - and this situation is surely made even worse by product inhibition. That is, if you do make a wonderful new combination, it will have such good binding to the target as to inactivate that binding site and functionally take it out of the experiment. In this case, the cyclic peptides are produced with mRNA barcodes attached to each partner to identify hits at the end, but there’s another key modification: one library of partners has an HA tag built into it (human influenza hemagglutinin), while the other has a biotinylated lysine residue. Chem-bio geeks will immediately see the “tandem purification” scheme set-up: if you run your libraries against a target, you can then take the reaction mixture and (using solid-supported HA antibodies and streptavidin) pull out only those species that have both an HA domain and a biotin - that is, only the species that were formed by the reaction of one library partner with a partner from the other library. Everything else gets washed away - and at the end, you have those mRNA barcodes to amplify and sequence so you can see which two library partners those were. Ah, molecular biology. The authors demonstrate this protocol using a simplified nucleophilic substitution reaction (cysteine on one library and chloroacetamide on the other), and with their workflow in hand, they went on to apply it to the humungous libraries mentioned above using that same cys-chloroacetamide coupling chemistry. The binding partner they chose was a protein of the well-known 14-3-3 type, which certainly has a lot of binding partners in vivo and has been the subject of a lot of drug discovery and chem-bio work over the years. 14-3-3 forms a dimer structure in vivo, and they were going to target that from both sides. They put their thumbs on a scale a bit - who wouldn’t - by starting from random sequences and then enriching both peptide libraries in structures that were found to be monomeric 14-3-3 binders, by running the cyclization step of the library formation(s) in the presence of solid-supported 14-3-3. The enriched libraries were run against each other and the tandem purification described above was applied to only fish out heterodimeric binders (one side from each library set). They also looked for binders from that first selection step (enriching the monomeric species), and both of these yielded hits that converged on a few favored protein sequences. One of these was already well-known to interact with 14-3-3, while the other two hadn’t been reported until now. But unfortunately they didn’t find heterodimeric species after the selections. Undeterred, that’s when they moved on to the inverse-electron-demand Diels-Alder chemistry, which was expected to more of a sure thing when the partners came into close enough view of each other. Learning from their last run, they added a step: heat denaturing at 70C before the tandem purification, to shut down the enrichment of noncovalent ternary complexes, which had been a problem. Experiments with more and less reactive tetrazines also suggested that they really did need something fast and reactive for this system to work out. These experiments converged on exactly the same protein motifs as the previous one (an encouraging sign, and after several more selection control experiments they were convinced that a single hit was the winner of the process. Its formation was accelerated over 500x in the presence of 14-3-3 dimer as opposed to background controls. The two partners were found to have strong binding themselves 430 +/- 70 pM and 700 +/- 300 pM), and the newly formed Diels-Alder product was a 120 +/- 50 pM binder. The team did a number of other very useful experiments looking at linker lengths and reactivity of the tetrazine partner(s). They believe that these will probably have to be fine-tuned each time you do an experiment like this, and suggest various selection protocols to help you narrow that down along the way. The way that 14-3-3 is in equilibrium between two monomers and the dimer form during the experiment surely also complicated things. There’s a lot more work to be done - not least, they note, in making the whole selection process less labor-intensive - but just seeing a trillion-by-trillion matrix of compounds screened successfully by this technique leaves me in awe anyway!
Show full content
I really got a kick out of this recent paper, but that’s probably because I’ve always enjoyed the whole “templated ligation” or “kinetic target-guided synthesis” field. There are several ways to describe it in the literature, which doesn’t make searching for papers any easier, but the basic idea is that you get two partner species to react in the presence of a binding site or surface. If you pick your chemistry right, that reaction/ligation/bond formation/condensation is something that normally has a low background rate, but which can be accelerated by the templated binding/induced proximity that’s brought on by the presence of the target species.
In the ideal case, then, you get a situation where a large number of reaction combinations (i.e. a large number of products) are available, but the low background rate of the untemplated reaction keeps you from seeing them. The only products that you detect are the ones that had the advantage of being brought together by mutually beneficial binding events - that is, the only compounds that form are the ones that have built-in potent binding. Another way to think of it is arranging things so that a target selects its own ligands for you from the huge pool of combinatorial possibilities.
One way people have done this is DCC, dynamic combinatorial chemistry. For that you use reversible bond-forming systems such as thiols/disulfides or perhaps aldehydes and amines/oximes. These are presumably interconverting constantly, mixing and matching in solution, but the addition of a binding template can (in theory) bias the mix towards the ones where the reactants are brought together more frequently, more closely, or for a longer residence time.
There are numerous examples of this sort of thing in the literature, but I will admit that personally DCC leaves me a bit cold, because my own biases are for things that can turn into drug leads. The products of DCC-type reactions (the disulfides, oximes, and so on) are often not suitable for that use, and have to be modified to turn into plausible developable compounds. That can be quite a chasm to cross, because when you get right down to it, there’s nothing quite like a disulfide (bond lengths, polarity, torsional angles, electron density), and finding a druglike isostere can be no fun at all.
So I’ve tended towards the systems that do irreversible bond formation, things like the Sharpless-style “click” triazole formation from azides and acetylenes. (Note that I don’t mean the copper-catalyzed ones that most people are familiar with - in this case you’re letting the 3+2 cycloaddition take place with no help other than forced proximity). There are other reactions that you can do that with, but each time you have to make sure that you’re not making the background rate too high. I wrote about one of those here a few years ago.
The paper linked in the first paragraph is using a cycloaddition reaction (rather like the uncatalyzed triazole formation), the tetrazine-based inverse electron demand Diels-Alder reaction. That one of course has also been used in many chemical biology applications with the reactive trans-cyclooctene as a partner - like the triazole click reaction, the two partners are generally stable in biochemical systems and tend to react only with each other. But the new thing is this paper is that the tetrazine compounds are no mere small-molecule library but a peptide library of roughly one trillion members. And the norbornene coupling partner (a much less reactive one than usually used for "click" reactions) for that IED Diels-Alder is another such library: one trillion peptides each with a norbornene hanging off of them. These are prepared via mRNA display using the RaPID (Random nonstandard Peptides Integrated Discovery) system that uses a reworked genetic code to produce functionalized macrocyclic peptides.
Now the question with all such experiments is: how on earth do you find your hits? The larger your libraries of screening and reaction partners, the smaller the amounts of any given hit you can expect to make - and this situation is surely made even worse by product inhibition. That is, if you do make a wonderful new combination, it will have such good binding to the target as to inactivate that binding site and functionally take it out of the experiment. In this case, the cyclic peptides are produced with mRNA barcodes attached to each partner to identify hits at the end, but there’s another key modification: one library of partners has an HA tag built into it (human influenza hemagglutinin), while the other has a biotinylated lysine residue.
Chem-bio geeks will immediately see the “tandem purification” scheme set-up: if you run your libraries against a target, you can then take the reaction mixture and (using solid-supported HA antibodies and streptavidin) pull out only those species that have both an HA domain and a biotin - that is, only the species that were formed by the reaction of one library partner with a partner from the other library. Everything else gets washed away - and at the end, you have those mRNA barcodes to amplify and sequence so you can see which two library partners those were. Ah, molecular biology.
The authors demonstrate this protocol using a simplified nucleophilic substitution reaction (cysteine on one library and chloroacetamide on the other), and with their workflow in hand, they went on to apply it to the humungous libraries mentioned above using that same cys-chloroacetamide coupling chemistry. The binding partner they chose was a protein of the well-known 14-3-3 type, which certainly has a lot of binding partners in vivo and has been the subject of a lot of drug discovery and chem-bio work over the years. 14-3-3 forms a dimer structure in vivo, and they were going to target that from both sides.
They put their thumbs on a scale a bit - who wouldn’t - by starting from random sequences and then enriching both peptide libraries in structures that were found to be monomeric 14-3-3 binders, by running the cyclization step of the library formation(s) in the presence of solid-supported 14-3-3. The enriched libraries were run against each other and the tandem purification described above was applied to only fish out heterodimeric binders (one side from each library set). They also looked for binders from that first selection step (enriching the monomeric species), and both of these yielded hits that converged on a few favored protein sequences. One of these was already well-known to interact with 14-3-3, while the other two hadn’t been reported until now. But unfortunately they didn’t find heterodimeric species after the selections.
Undeterred, that’s when they moved on to the inverse-electron-demand Diels-Alder chemistry, which was expected to more of a sure thing when the partners came into close enough view of each other. Learning from their last run, they added a step: heat denaturing at 70C before the tandem purification, to shut down the enrichment of noncovalent ternary complexes, which had been a problem. Experiments with more and less reactive tetrazines also suggested that they really did need something fast and reactive for this system to work out.
These experiments converged on exactly the same protein motifs as the previous one (an encouraging sign, and after several more selection control experiments they were convinced that a single hit was the winner of the process. Its formation was accelerated over 500x in the presence of 14-3-3 dimer as opposed to background controls. The two partners were found to have strong binding themselves 430 +/- 70 pM and 700 +/- 300 pM), and the newly formed Diels-Alder product was a 120 +/- 50 pM binder.
The team did a number of other very useful experiments looking at linker lengths and reactivity of the tetrazine partner(s). They believe that these will probably have to be fine-tuned each time you do an experiment like this, and suggest various selection protocols to help you narrow that down along the way. The way that 14-3-3 is in equilibrium between two monomers and the dimer form during the experiment surely also complicated things. There’s a lot more work to be done - not least, they note, in making the whole selection process less labor-intensive - but just seeing a trillion-by-trillion matrix of compounds screened successfully by this technique leaves me in awe anyway!
Let’s start off the week with some good news. There was a small company called Anacor that specialized in boron-containing drug leads, especially for infectious diseases. They had some shots at the clinic in that area and had an antifungal (tavaborole) approved by the FDA, but their biggest success was a topical medication for dermatitis called crisaborole. That’s a PDE4 inhibitor, although it has to be said that the exact mechanism for its efficacy in this disease is (to my knowledge) not completely worked out. I should note that “success” is a relative term: the drug was considered a potential blockbuster in psoriasis treatment when Pfizer bought Anacor for about $99/share in 2016, but its sales never came close to living up to the $5.2 billion cost of that deal. But Anacor’s biggest legacy is going to be a compound called acoziborole which was developed from one of their leads (specifically, its penetration into the central nervous system was improved). It’s an antiprotozoal compound, and Trypanasoma is one of the organisms it targets. That is of course the infectious agent for “sleeping sickness” (trypanosomiasis) in humans, infamously spread in tropical regions of Africa by the bite of the tsetse fly. It has historically been a terrible disease, and before modern treatments were discovered it was almost always fatal. Any survivors tended to have severe neurological damage. A wide variety of efforts targeting insect bite prevention and the life cycle of the tsetse fly itself have reduced the footprint of the disease, fortunately, because it has been a terrible scourge. The last major epidemic seems to have been in 1970, although it still kills several thousand people a year. But that’s down from tens of thousands a year, and in some years far more than that: a 1901 outbreak in what’s now Uganda is estimated to have killed 250,000 people. One of the current best treatments (fexinidazole) was an old Hoechst drug candidate that was revived by the Drugs for Neglected Diseases Initiative. It’s much more effective when given early, and requires a 10-day course of treatment. But the DNDI has been partnering with Sanofi to develop the above-mentioned acoziborole, which cures the disease in a single oral dose of three pills. You can’t ask much more than that! The EMA has just approved it after several trials in Africa and it’s heading for widespread use. This dosing is far easier than any other treatment option and it’s effective against both the early and late stages of the disease. That’s important because lumbar puncture is the only way to distinguish which stage you’re dealing with, and that of course is a significant diagnostic burden. A current trial is seeing if the drug is safe to use in people who just show positive antibody tests to trypanosome exposure, which could even get rid of the longstanding need for disease confirmation via blood testing microscopy. Acoziborole is so effective that its deployment could entirely eliminate transmission of the disease, which would be a great victory for the human race in general. That will take a lot of time and effort, but just seeing it rolled out in the affected regions is absolute undiluted good news. Celebrate it!
Show full content
Let’s start off the week with some good news. There was a small company called Anacor that specialized in boron-containing drug leads, especially for infectious diseases. They had some shots at the clinic in that area and had an antifungal (tavaborole) approved by the FDA, but their biggest success was a topical medication for dermatitis called crisaborole. That’s a PDE4 inhibitor, although it has to be said that the exact mechanism for its efficacy in this disease is (to my knowledge) not completely worked out. I should note that “success” is a relative term: the drug was considered a potential blockbuster in psoriasis treatment when Pfizer bought Anacor for about $99/share in 2016, but its sales never came close to living up to the $5.2 billion cost of that deal.
But Anacor’s biggest legacy is going to be a compound called acoziborole which was developed from one of their leads (specifically, its penetration into the central nervous system was improved). It’s an antiprotozoal compound, and Trypanasoma is one of the organisms it targets. That is of course the infectious agent for “sleeping sickness” (trypanosomiasis) in humans, infamously spread in tropical regions of Africa by the bite of the tsetse fly. It has historically been a terrible disease, and before modern treatments were discovered it was almost always fatal. Any survivors tended to have severe neurological damage. A wide variety of efforts targeting insect bite prevention and the life cycle of the tsetse fly itself have reduced the footprint of the disease, fortunately, because it has been a terrible scourge. The last major epidemic seems to have been in 1970, although it still kills several thousand people a year. But that’s down from tens of thousands a year, and in some years far more than that: a 1901 outbreak in what’s now Uganda is estimated to have killed 250,000 people.
One of the current best treatments (fexinidazole) was an old Hoechst drug candidate that was revived by the Drugs for Neglected Diseases Initiative. It’s much more effective when given early, and requires a 10-day course of treatment. But the DNDI has been partnering with Sanofi to develop the above-mentioned acoziborole, whichcures the disease in a single oral dose of three pills. You can’t ask much more than that! The EMA has just approved it after several trials in Africa and it’s heading for widespread use.
This dosing is far easier than any other treatment option and it’s effective against both the early and late stages of the disease. That’s important because lumbar puncture is the only way to distinguish which stage you’re dealing with, and that of course is a significant diagnostic burden. A current trial is seeing if the drug is safe to use in people who just show positive antibody tests to trypanosome exposure, which could even get rid of the longstanding need for disease confirmation via blood testing microscopy.
Acoziborole is so effective that its deployment could entirely eliminate transmission of the disease, which would be a great victory for the human race in general. That will take a lot of time and effort, but just seeing it rolled out in the affected regions is absolute undiluted good news. Celebrate it!
We find ourselves in a situation where chemistry is intruding on current events, and I’m referring to something that not everyone seems to have thought about: fertilizer, and especially nitrogen fertilizer. What it is, how it’s used, and especially where it comes from. Intro to Fertilizer Now that's a grabber of a subhead, right? But if you’ve bought a bag of the stuff for your garden or your houseplants, you’ve seen the three numbers used to show the strength and balance of any given fertilizer: the first number of that “NPK” rating is nitrogen (elemental nitrogen percentage by weight), the second is phosphorus (phosphate, as percentage by weight of phosphorus pentoxide as a reference compound), and the third is for potassium (percentage by weight of potassium oxide, again as a reference standard). They’re all important, along with other components like calcium, magnesium, sulfur, and some metal micronutrients as well. The discovery that adding things to the soil could improve production goes back to prehistory, and the ancient uses of animal (or indeed human) manures, dead fish, blood meal and the like are all attempts to increase the available nitrogen. (Phosphate is a story of its own, and a wild one - we’ll do that another day). The problem is of course that the carbon (and carbohydrate) content of a harvested crop comes from the carbon dioxide in the air and water from the ground, both of which one hopes will be back for use next growing season. But nitrogen, phosphorus, and other nutrients can and will be depleted as you harvest the crops and haul them off each year. This loss is much faster than weathering, decay and other such processes that might replenish them in the soil, so adding them back is either crucial right now or will become so in the rather near future no matter where you’re farming. And adding still more (up to a point!) will almost always increase your crop yields, which has been a key concern ever since humans have been burying seeds in the dirt. Nitrogen in Particular We are living in a vast sea of nitrogen gas, but it’s so unreactive that it’s useless for biological pathways as it is. You have to somehow reduce to more useful chemical forms like amines, nitrates, and so on. That’s a reaction that soil bacteria had beaten us to by a couple of billion years (the “nitrogen-fixing” ones in the root nodules of legume plants), but doing it on an industrial scale was something else again. By the early 1900s there were a couple of industrial routes to do this: the Birkeland-Eyde process that used electrical discharge arcs and especially the Frank-Caro cyanamide process that used calcium carbide. Anyone familiar with the history of chemistry (or the history of agriculture) will recognize that human society took a significant turn in the early 20th century with the rise of the Haber-Bosch process. This uses metal catalysts at high temperature and pressure to reduce the nitrogen in the air straight down to ammonia in the presence of hydrogen gas. (Thermodynamic side note: the reaction is enthalpically favorable if you add up heats of formation alone, but the entropy term in the free energy equation kills you because you’re taking four equivalents of gas and turning into just two - thus the need for high temperatures and pressure!) The ammonia is largely turned into urea, which makes a very useful solid fertilizer as is, while some of is diverted into nitric acid/nitrate production, with those latter salts being good fertilizers themselves. Some fertilizer applications (corn especially) just use straight ammonia injected into the soil, poisonous though it is (by the way, in the NPK numbering system ammonia gas is an impressive 82-0-0). This chemistry quickly spread through the industrialized world, and Haber chemistry continues to reign supreme. And I do mean supreme. Here comes an extremely important point: somewhere around half the world’s population stays alive because of these industrially-produced fertilizers. So it would behoove one to know where they come from. And here’s where the details of the chemistry as well logistics and economics in general need to be considered. The current situation I slid past it up there, but you may have noticed that hydrogen is needed for this process - ammonia is just hydrogenated nitrogen gas. So where do you get that? By far the major source for hydrogen has been (and remains) “steam reforming” of methane natural gas, which is another catalytic process (using nickel compounds this time). That’s also done at high temperature and pressure, but you have to be sure to remove all traces of sulfur compounds from the methane, because that will absolutely and irreversibly foul those nickel catalysts. You guessed it - that step takes energy too. Chemically, some of that hydrogen production is sent back around for a “hydrodesulfurization” step, yet another high-pressure high-temperature catalytic step (generally a molybdenum-based mixture) that turns the sulfur compounds in the natural gas into hydrogen sulfide, which is then processed into either elemental sulfur or sulfuric acid. This process is done at virtually all petroleum refineries to strip out sulfur for all sorts of later steps, and that produces huge piles of yellow sulfur that can be easily seen from orbit, far more than the world has a present use for. Back in the days before the oil economy people used to mine sulfur, but you can forget about having to do that now in most of the world. So as you count it up, nearly every single step of Haber-Bosch ammonia production needs heat and pressure, making the whole route rather energy-intensive (although it’s still a much better deal than the two earlier routes mentioned above, which is a big reason why it replaced them). And boy, do we ever need nitrogen fertilizer (see the previous section), so that price is paid without blinking. Nitrogen is everywhere, but the siting of big Haber-Bosch installations is far more feasible if you have some place with lots of relatively cheap natural gas (as a fuel for all the heating and as a hydrogen source). The biggest one in the world is in Louisiana, and there are others in Texas and Oklahoma. But even so we still import very significant amounts of urea, and that importation is highest at this exact season. The world nitrogen fertilizer market has really been shaken up by the situation that we have now caused in the Persian Gulf. And after those last couple of paragraphs it’s easy to see why. The Gulf has extraordinary amounts of natural gas, and thus countries in that region have taken advantage of that value-added business opportunity and have become major fertilizer exporters. But not at the moment. Not right when it’s needed in the Northern Hemisphere. All that stuff comes out on huge container ships, down the Gulf and right out the Strait of Hormuz, just like the oil and the liquified natural gas. Prices for all the nitrogen fertilizers were already running high by historic standards before all of this, but now, well. Farm organizations here in the US are calling for financial help from the administration, but after all the tariff nonsense you have to wonder what they’re expecting. Given the reports of mine-laying in the Strait, we might be looking at significant disruptions for some time. I will resist the impulse to rant and rage. But the likely economic and trade consequences for an Iran war should have been apparent to any reasonably well-informed high school student, much less to the planners at the highest levels of the US government. Indeed, I have no doubt that there are surely competent people left in the various cabinet departments who tried to make this case because they felt as if they could do nothing less. But so much for that. Our leaders have opted instead for the feckless, irresponsible, poorly thought-out mess that we see every morning when we check the news. It seems a safe bet that the likes of Pete Hegseth and Donald Trump have never thought about industrial nitrogen fixation in their lives. You’ve probably seen the Trotsky quote about “You may not be interested in war, but war is interested in you”. Well, our current leaders are very, very interested in war, so we have that covered, damn it all. But some of the things they’re not interested in are coming back around to bite us.
Show full content
We find ourselves in a situation where chemistry is intruding on current events, and I’m referring to something that not everyone seems to have thought about: fertilizer, and especially nitrogen fertilizer. What it is, how it’s used, and especially where it comes from.
Intro to Fertilizer
Now that's a grabber of a subhead, right? But if you’ve bought a bag of the stuff for your garden or your houseplants, you’ve seen the three numbers used to show the strength and balance of any given fertilizer: the first number of that “NPK” rating is nitrogen (elemental nitrogen percentage by weight), the second is phosphorus (phosphate, as percentage by weight of phosphorus pentoxide as a reference compound), and the third is for potassium (percentage by weight of potassium oxide, again as a reference standard).
They’re all important, along with other components like calcium, magnesium, sulfur, and some metal micronutrients as well. The discovery that adding things to the soil could improve production goes back to prehistory, and the ancient uses of animal (or indeed human) manures, dead fish, blood meal and the like are all attempts to increase the available nitrogen. (Phosphate is a story of its own, and a wild one - we’ll do that another day). The problem is of course that the carbon (and carbohydrate) content of a harvested crop comes from the carbon dioxide in the air and water from the ground, both of which one hopes will be back for use next growing season. But nitrogen, phosphorus, and other nutrients can and will be depleted as you harvest the crops and haul them off each year. This loss is much faster than weathering, decay and other such processes that might replenish them in the soil, so adding them back is either crucial right now or will become so in the rather near future no matter where you’re farming. And adding still more (up to a point!) will almost always increase your crop yields, which has been a key concern ever since humans have been burying seeds in the dirt.
Nitrogen in Particular
We are living in a vast sea of nitrogen gas, but it’s so unreactive that it’s useless for biological pathways as it is. You have to somehow reduce to more useful chemical forms like amines, nitrates, and so on. That’s a reaction that soil bacteria had beaten us to by a couple of billion years (the “nitrogen-fixing” ones in the root nodules of legume plants), but doing it on an industrial scale was something else again. By the early 1900s there were a couple of industrial routes to do this: the Birkeland-Eyde process that used electrical discharge arcs and especially the Frank-Caro cyanamide process that used calcium carbide.
Anyone familiar with the history of chemistry (or the history of agriculture) will recognize that human society took a significant turn in the early 20th century with the rise of the Haber-Bosch process. This uses metal catalysts at high temperature and pressure to reduce the nitrogen in the air straight down to ammonia in the presence of hydrogen gas. (Thermodynamic side note: the reaction is enthalpically favorable if you add up heats of formation alone, but the entropy term in the free energy equation kills you because you’re taking four equivalents of gas and turning into just two - thus the need for high temperatures and pressure!) The ammonia is largely turned into urea, which makes a very useful solid fertilizer as is, while some of is diverted into nitric acid/nitrate production, with those latter salts being good fertilizers themselves. Some fertilizer applications (corn especially) just use straight ammonia injected into the soil, poisonous though it is (by the way, in the NPK numbering system ammonia gas is an impressive 82-0-0).
This chemistry quickly spread through the industrialized world, and Haber chemistry continues to reign supreme. And I do mean supreme. Here comes an extremely important point: somewhere around half the world’s population stays alive because of these industrially-produced fertilizers. So it would behoove one to know where they come from. And here’s where the details of the chemistry as well logistics and economics in general need to be considered.
The current situation
I slid past it up there, but you may have noticed that hydrogen is needed for this process - ammonia is just hydrogenated nitrogen gas. So where do you get that? By far the major source for hydrogen has been (and remains) “steam reforming” of methane natural gas, which is another catalytic process (using nickel compounds this time). That’s also done at high temperature and pressure, but you have to be sure to remove all traces of sulfur compounds from the methane, because that will absolutely and irreversibly foul those nickel catalysts. You guessed it - that step takes energy too. Chemically, some of that hydrogen production is sent back around for a “hydrodesulfurization” step, yet another high-pressure high-temperature catalytic step (generally a molybdenum-based mixture) that turns the sulfur compounds in the natural gas into hydrogen sulfide, which is then processed into either elemental sulfur or sulfuric acid. This process is done at virtually all petroleum refineries to strip out sulfur for all sorts of later steps, and that produces huge piles of yellow sulfur that can be easily seen from orbit, far more than the world has a present use for. Back in the days before the oil economy people used to mine sulfur, but you can forget about having to do that now in most of the world.
So as you count it up, nearlyevery single step of Haber-Bosch ammonia production needs heat and pressure, making the whole route rather energy-intensive (although it’s still a much better deal than the two earlier routes mentioned above, which is a big reason why it replaced them). And boy, do we ever need nitrogen fertilizer (see the previous section), so that price is paid without blinking. Nitrogen is everywhere, but the siting of big Haber-Bosch installations is far more feasible if you have some place with lots of relatively cheap natural gas (as a fuel for all the heating and as a hydrogen source). The biggest one in the world is in Louisiana, and there are others in Texas and Oklahoma. But even so we still import very significant amounts of urea, and that importation is highest at this exact season.
The world nitrogen fertilizer market has really been shaken up by the situation that we have now caused in the Persian Gulf. And after those last couple of paragraphs it’s easy to see why. The Gulf has extraordinary amounts of natural gas, and thus countries in that region have taken advantage of that value-added business opportunity and have become major fertilizer exporters. But not at the moment. Not right when it’s needed in the Northern Hemisphere. All that stuff comes out on huge container ships, down the Gulf and right out the Strait of Hormuz, just like the oil and the liquified natural gas. Prices for all the nitrogen fertilizers were already running high by historic standards before all of this, but now, well. Farm organizations here in the US are calling for financial help from the administration, but after all the tariff nonsense you have to wonder what they’re expecting. Given the reports of mine-laying in the Strait, we might be looking at significant disruptions for some time.
I will resist the impulse to rant and rage. But the likely economic and trade consequences for an Iran war should have been apparent to any reasonably well-informed high school student, much less to the planners at the highest levels of the US government. Indeed, I have no doubt that there are surely competent people left in the various cabinet departments who tried to make this case because they felt as if they could do nothing less. But so much for that. Our leaders have opted instead for the feckless, irresponsible, poorly thought-out mess that we see every morning when we check the news.
It seems a safe bet that the likes of Pete Hegseth and Donald Trump have never thought about industrial nitrogen fixation in their lives. You’ve probably seen the Trotsky quote about “You may not be interested in war, but war is interested in you”. Well, our current leaders are very, very interested in war, so we have that covered, damn it all. But some of the things they’re not interested in are coming back around to bite us.
I enjoyed this paper, because it plays to my own prejudices. I’ve long believed that assays and experiments should be conducted as close to the real biological systems as is feasible, and that if you are (perforce) starting in more simplified ones, that you should move up the scale as soon as you can. That means if you show up with purified protein assay data, I’ll ask when it’s going into cells, and if you have cell data I’ll ask what it does in whole animals. (You can drop organoids into the middle of that if you really want to). If it works in animals, how’s the two-week tox? If you’ve made it through that, when’s the Phase I trial start? And so on. I’m simplifying here, but you get the idea: keep things moving. Get the closest thing to a real-world readout that you can. The paper linked above provides another example of why you want to do this. The authors are comparing kinase inhibitor selectivity profiles, looking at what you get when you screen a big panel of individual kinases versus what you get in living cells. It shouldn’t come as a surprise that the answers are different, but perhaps it will be one when you see just how different they can be. The authors find examples of compounds that look pretty nonselective in the screening panel but are actually much cleaner in their cellular effects. For example, they show an inhibitor (TPKI-39) that hits at least 15 kinases in the panel, at least half of them with affinities below 100 nanomolar. But in a NanoBRET cell assay format, it only seems to bind to DDR1, DDR2, and FLT1. Similarly, the compound GW296115 shows less thatn 160 nM affinity against at least 27 kinases in individual enzyme assays, but in cell assays it really only hits BRSK1 and BRSK2 (both of which are poorly characterized and could use a decent probe compound). The authors note that both of these compounds (and several of their other examples) are “Type II” kinase inhibitors, which occupy the ATP-binding region of their target enzymes in the inactive conformation (with the “DFG” loop pointed out, as aficionados well appreciate). The Type I inhibitors bind the active form that has the DFG loop in. Years ago there was a general supposition that Type II inhibitors were more rare and more selective, but that hasn’t really held up as more examples have been investigated. But the results here argue that they are perhaps more likely to be selective in actual cells. These results are similar to what is often found in covalent-labeling experiments in living cells, versus cell lysates, versus isolated proteins. As you move along that spectrum, you see more and more things showing up as hits, and this is likely because the non-physiological conditions are exposing protein surfaces (and allowing for protein conformations) that may be inaccessible under live-cell conditions (where partner proteins are present and keeping things more in check). So the take-home here is to look at such compounds in actual cells wherever possible, and to not be too jumpy about selectivity measurements that are produced only under cell-free conditions. There’s a lot going on down there, and you could easily miss out on some useful compounds because you never gave them a chance to show how useful they could be!
Show full content
I enjoyed this paper, because it plays to my own prejudices. I’ve long believed that assays and experiments should be conducted as close to the real biological systems as is feasible, and that if you are (perforce) starting in more simplified ones, that you should move up the scale as soon as you can. That means if you show up with purified protein assay data, I’ll ask when it’s going into cells, and if you have cell data I’ll ask what it does in whole animals. (You can drop organoids into the middle of that if you really want to). If it works in animals, how’s the two-week tox? If you’ve made it through that, when’s the Phase I trial start? And so on. I’m simplifying here, but you get the idea: keep things moving. Get the closest thing to a real-world readout that you can.
The paper linked above provides another example of why you want to do this. The authors are comparing kinase inhibitor selectivity profiles, looking at what you get when you screen a big panel of individual kinases versus what you get in living cells. It shouldn’t come as a surprise that the answers are different, but perhaps it will be one when you see just how different they can be. The authors find examples of compounds that look pretty nonselective in the screening panel but are actually much cleaner in their cellular effects.
For example, they show an inhibitor (TPKI-39) that hits at least 15 kinases in the panel, at least half of them with affinities below 100 nanomolar. But in a NanoBRET cell assay format, it only seems to bind to DDR1, DDR2, and FLT1. Similarly, the compound GW296115 shows less thatn 160 nM affinity against at least 27 kinases in individual enzyme assays, but in cell assays it really only hits BRSK1 and BRSK2 (both of which are poorly characterized and could use a decent probe compound).
The authors note that both of these compounds (and several of their other examples) are “Type II” kinase inhibitors, which occupy the ATP-binding region of their target enzymes in the inactive conformation (with the “DFG” loop pointed out, as aficionados well appreciate). The Type I inhibitors bind the active form that has the DFG loop in. Years ago there was a general supposition that Type II inhibitors were more rare and more selective, but that hasn’t really held up as more examples have been investigated. But the results here argue that they are perhaps more likely to be selective in actual cells.
These results are similar to what is often found in covalent-labeling experiments in living cells, versus cell lysates, versus isolated proteins. As you move along that spectrum, you see more and more things showing up as hits, and this is likely because the non-physiological conditions are exposing protein surfaces (and allowing for protein conformations) that may be inaccessible under live-cell conditions (where partner proteins are present and keeping things more in check).
So the take-home here is to look at such compounds in actual cells wherever possible, and to not be too jumpy about selectivity measurements that are produced only under cell-free conditions. There’s a lot going on down there, and you could easily miss out on some useful compounds because you never gave them a chance to show how useful they could be!
Let’s have a look at the unique situation of UniQure. It’s a story about central nervous system disease therapy, that’s for sure. And it’s a story about clinical trial design, and it’s also a story about new and challenging modes of treatment working their way into such human trails. But unfortunately, it has turned most of all into a story about regulatory affairs as they are run under the Trump administration. This is all about a potential therapy for Huntington’s disease, and God knows that one could use a few more potential therapies. I’ve written about the disease and its biology several times over the years, most recently here and here, and although there are plenty of mysteries about its etiology, there’s also one thing that’s very clear indeed. The exact protein (Huntingtin, and thus the exact gene) that are behind the disease are well-known, and have been known for decades. Anyone who believes that such knowledge leads quickly to treatments or cures is invited to look at this situation to correct their thinking. One thing that’s been equally clear for that length of time is that (like most single-protein diseases) HD looks like a clear target for gene therapy or related approaches that directly affect the production of the mutant protein. But that has not been easy. As that paper illustrates, all sorts of ideas have been researched - RNA interference, antisense oligos, and more. The latter made it all the way to human trials with Roche and IONIS, an antisense oligo that would bind to the messenger RNA for the mutant Huntingtin protein, but the results were very disappointing, for reasons that are still unclear, at least to me. That agent (tominersen) is still in the clinic years later, and is apparently moving forward in a select population (younger patients, less advanced disease) at the highest dose tested - we shall see. UniQure’s approach is broadly similar, but they’re using RNA interference rather than an antisense oligo. Over the years they’ve published on their selection of the appropriate microRNA sequences to achieve this silencing, and they have packaged a group of genes for production of their best candidate into a viral vector (AAV) to deliver that to brain tissue. Now, getting a viral payload into the brain may seem like a difficult task, and by gosh you’re right about that: this therapy (AMT-130) is delivered through the blood-brain barrier by lengthy surgery that involves placement of a microcatheter targeting the caudate and striatum regions of the brain. Nontrivial, for sure. But they went ahead with the trials, and here’s where the trouble starts. The FDA that agreed to the clinical protocol is not the FDA we have today, as Adam Feuerstein explains very well here. He frames that as a disconnect between the Peter Marks era at the agency (which brought us a series of Duchenne muscular dystrophy treatment approvals that in the end appear to do little or nothing for actual Duchenne patientss) and the Vinay Prasad era, which has been far more skeptical of gene therapy. And vaccines. And lots of other things. come to think of it. Like Adam, I would hope for an agency whose polices land somewhere between these two wide extremes, but I hope for a lot of things these days and that’s just one of them. By last Thursday, we were to the point where an “unnamed FDA official” was accusing Uniqure of distorting and manipulating their data in the service of what he called a “failed therapy”. Anyone capable of reading between any lines at all could tell that this was the voice of Vinay Prasad, and indeed Rep. Jake Auchincloss said as much on Friday. But that was the same day that it was announced that Prasad was (once again) leaving the FDA (which news sent Uniqure stock up sharply). It’s been a ride. The dispute, as laid out in the Feuerstein Stat link above, is the interpretation of the AMT-130 results versus controls. The study actually did sham surgery for the control group so no one would know who got the actual treatment, but that control group was unblinded after a year and given the opportunity to get the real therapy as well. From that point on the treatment groups was compared to “wild type” Huntington’s patients rather than to the now-abrogated control group. As it turned out, the one-year results showed a trend, but were not statistically significant. But the two- and three-year analyses looked better, and it’s the three-year data that got all the attention last September. Prasad’s take was that the one-year data indicated failure, and that the attempt to use “natural history of disease” control data after that was inappropriate and an attempt to pull a fast one for regulatory approval. The FDA was trying to request another sham-surgery-controlled trial, this one to be followed up for longer without unblinding. My own take is that the data that we have certainly is worth giving AMT-130 a full regulatory review and quite possibly a provisional approval (to my eyes, it’s a lot stronger than what the FDA was willing to accept with those Duchenne therapies, not that we should make that the standard going forward). The sham surgery for its part is no joke - it doesn’t penetrate the skull like the real one, but the patients are under general anesthesia for 10 to 12 hours, which is a lot to ask for a placebo, because about 10% of the sham group the first time around developed a blood clot during this part. So let’s see what the FDA does now. That’ll depend on who replaces Vinay Prasad, how much of a loon they are, and how much they’re willing to make decisions like this. I mean, the field is broad: you could get a MAHA-talking version of Peter Marks who’s ready to let all sorts of stuff through in case it works, or you could get someone who’s not even sure that RNA exists or that it’s not a secret 5G cell phone technology. Let’s see.
Show full content
Let’s have a look at the unique situation of UniQure. It’s a story about central nervous system disease therapy, that’s for sure. And it’s a story about clinical trial design, and it’s also a story about new and challenging modes of treatment working their way into such human trails. But unfortunately, it has turned most of all into a story about regulatory affairs as they are run under the Trump administration.
This is all about a potential therapy for Huntington’s disease, and God knows that one could use a few more potential therapies. I’ve written about the disease and its biology several times over the years, most recently here and here, and although there are plenty of mysteries about its etiology, there’s also one thing that’s very clear indeed. The exact protein (Huntingtin, and thus the exact gene) that are behind the disease are well-known, and have been known for decades. Anyone who believes that such knowledge leads quickly to treatments or cures is invited to look at this situation to correct their thinking.
One thing that’s been equally clear for that length of time is that (like most single-protein diseases) HD looks like a clear target for gene therapy or related approaches that directly affect the production of the mutant protein. But that has not been easy. As that paper illustrates, all sorts of ideas have been researched - RNA interference, antisense oligos, and more. The latter made it all the way to human trials with Roche and IONIS, an antisense oligo that would bind to the messenger RNA for the mutant Huntingtin protein, but the results were very disappointing, for reasons that are still unclear, at least to me. That agent (tominersen) is still in the clinic years later, and is apparently moving forward in a select population (younger patients, less advanced disease) at the highest dose tested - we shall see.
UniQure’s approach is broadly similar, but they’re using RNA interference rather than an antisense oligo. Over the years they’ve published on their selection of the appropriate microRNA sequences to achieve this silencing, and they have packaged a group of genes for production of their best candidate into a viral vector (AAV) to deliver that to brain tissue. Now, getting a viral payload into the brain may seem like a difficult task, and by gosh you’re right about that: this therapy (AMT-130) is delivered through the blood-brain barrier by lengthy surgery that involves placement of a microcatheter targeting the caudate and striatum regions of the brain. Nontrivial, for sure.
But they went ahead with the trials, and here’s where the trouble starts. The FDA that agreed to the clinical protocol is not the FDA we have today, as Adam Feuerstein explains very well here. He frames that as a disconnect between the Peter Marks era at the agency (which brought us a series of Duchenne muscular dystrophy treatment approvals that in the end appear to do little or nothing for actual Duchenne patientss) and the Vinay Prasad era, which has been far more skeptical of gene therapy. And vaccines. And lots of other things. come to think of it. Like Adam, I would hope for an agency whose polices land somewhere between these two wide extremes, but I hope for a lot of things these days and that’s just one of them.
By last Thursday, we were to the point where an “unnamed FDA official” was accusing Uniqure of distorting and manipulating their data in the service of what he called a “failed therapy”. Anyone capable of reading between any lines at all could tell that this was the voice of Vinay Prasad, and indeed Rep. Jake Auchincloss said as much on Friday. But that was the same day that it was announced that Prasad was (once again) leaving the FDA (which news sent Uniqure stock up sharply). It’s been a ride.
The dispute, as laid out in the Feuerstein Stat link above, is the interpretation of the AMT-130 results versus controls. The study actually did sham surgery for the control group so no one would know who got the actual treatment, but that control group was unblinded after a year and given the opportunity to get the real therapy as well. From that point on the treatment groups was compared to “wild type” Huntington’s patients rather than to the now-abrogated control group. As it turned out, the one-year results showed a trend, but were not statistically significant. But the two- and three-year analyses looked better, and it’s the three-year data that got all the attention last September. Prasad’s take was that the one-year data indicated failure, and that the attempt to use “natural history of disease” control data after that was inappropriate and an attempt to pull a fast one for regulatory approval. The FDA was trying to request another sham-surgery-controlled trial, this one to be followed up for longer without unblinding.
My own take is that the data that we have certainly is worth giving AMT-130 a full regulatory review and quite possibly a provisional approval (to my eyes, it’s a lot stronger than what the FDA was willing to accept with those Duchenne therapies, not that we should make that the standard going forward). The sham surgery for its part is no joke - it doesn’t penetrate the skull like the real one, but the patients are under general anesthesia for 10 to 12 hours, which is a lot to ask for a placebo, because about 10% of the sham group the first time around developed a blood clot during this part.
So let’s see what the FDA does now. That’ll depend on who replaces Vinay Prasad, how much of a loon they are, and how much they’re willing to make decisions like this. I mean, the field is broad: you could get a MAHA-talking version of Peter Marks who’s ready to let all sorts of stuff through in case it works, or you could get someone who’s not even sure that RNA exists or that it’s not a secret 5G cell phone technology. Let’s see.
Here’s a really interesting (but rather unnerving) look at the state of “big data” in biomedical science. The author had been working on some software tools to find copy-pasted blocks of data in large data sets as a method of fraud detection. Such duplications have been the subject of controversy in papers from the Südhof lab at Stanford (and you can get two different sorts of perspective on that case with those links) and also with the Pruitt lab at McMaster. Both of these have been pretty widely covered. But what happens when you turn such software loose on databases that no one is questioning? A part-volunteer effort has now examined 600 such data sets, and from one perspective, the news is pretty good: 97% of them were pretty unremarkable. But that leaves 18 that were serious-looking, and here you can see an examination of three of those. (One of the others had already been retracted!) This Parkinson’s paper from 2016, for example, made a splash with evidence that gut microbiota have a profound influence on Parkinson’s disease. Unfortunately, in the data supporting these conclusions there are duplications in the data that would indeed have affected the results (they make up a significant part of the data in each case). There’s another case on evolved resistance to toxins that has a number of data points that are hard to explain, although it’s also not easy to explain why they would have been altered (as they do appear to be on first reading). A third case seems less problematic - that’s a 2017 study looking for individual behavioral traits in clonal populations of fish. There are four-fold repeating values that seem to have arisen after the merging of two data files, with data being assigned to the wrong rows thenceforth. In this case, the authors of the study readily worked out and admitted to the mistake, and have corrected the published data set. I’m glad to see that the Dryad repository has been supporting this effort by helping to contact journals and authors about the problems that have emerged. The post says that an effort to scan the other Excel-based data sets on the Dryad site (around 24,000 of them!) is up next, and it will be worth finding out if that 3% figure holds. As well as finding out what percentage of those will turn out to be in the “Oops, honest mistake” category versus the “We don’t really see the problem here” one, and especially versus the “How dare you” category. Stay tuned!
Show full content
Here’s a really interesting (but rather unnerving) look at the state of “big data” in biomedical science. The author had been working on some software tools to find copy-pasted blocks of data in large data sets as a method of fraud detection. Such duplications have been the subject of controversy in papers from the Südhof lab at Stanford (and you can get two different sorts of perspective on that case with those links) and also with the Pruitt lab at McMaster. Both of these have been pretty widely covered.
But what happens when you turn such software loose on databases that no one is questioning? A part-volunteer effort has now examined 600 such data sets, and from one perspective, the news is pretty good: 97% of them were pretty unremarkable. But that leaves 18 that were serious-looking, and here you can see an examination of three of those. (One of the others had already been retracted!)
This Parkinson’s paper from 2016, for example, made a splash with evidence that gut microbiota have a profound influence on Parkinson’s disease. Unfortunately, in the data supporting these conclusions there are duplications in the data that would indeed have affected the results (they make up a significant part of the data in each case). There’s another case on evolved resistance to toxins that has a number of data points that are hard to explain, although it’s also not easy to explain why they would have been altered (as they do appear to be on first reading). A third case seems less problematic - that’s a 2017 study looking for individual behavioral traits in clonal populations of fish. There are four-fold repeating values that seem to have arisen after the merging of two data files, with data being assigned to the wrong rows thenceforth. In this case, the authors of the study readily worked out and admitted to the mistake, and have corrected the published data set.
I’m glad to see that the Dryad repository has been supporting this effort by helping to contact journals and authors about the problems that have emerged. The post says that an effort to scan the other Excel-based data sets on the Dryad site (around 24,000 of them!) is up next, and it will be worth finding out if that 3% figure holds. As well as finding out what percentage of those will turn out to be in the “Oops, honest mistake” category versus the “We don’t really see the problem here” one, and especially versus the “How dare you” category. Stay tuned!
I’ve written a number of posts here about weird natural products (and another such roundup is in the works), but I couldn’t resist highlighting this paper. The compounds discussed (which have been found to be produced in bacteria, specifically one human-pathogenic species, Nocardia ninae) are not huge and bewilderingly complex. Natural products have plenty of those kinds of structures, but these are very small molecules indeed. They’re just really unlikely ones. That’s because they’re diazo compounds - yep, like good ol’ diazomethane, the methyl-ester-seeking-chemist’s friend. The diazo group (as you can see at left) is a funny beast. It has one nitrogen with a formal positive charge and another atom with a formal negative charge, so net neutral. You can put the negative charge either on the terminal carbon or the terminal nitrogen, but the truth is kind of in between. It also really looks like something that could find it in its heart to change to plain old dinitrogen gas. And so it does! Unless stabilized by adjacent functional groups, diazo compounds are known to be reactive and touchy, properties that are most on display with the smaller members of the series. You don’t get much smaller than diazomethane, and that one is famous for reacting with carboxylic acids to more-or-less instantly form the corresponding methyl ester (and bubbles of nitrogen gas, if you’re working on a large enough scale, and if you are I would consider your life choices carefully). It does a lot of other interesting reactions, but you will always want to have it in dilute solution when you’re trying those, because it is sensitive to friction, heat, strong light, acid, and all sorts of other things, and it will most definitely explode if you push your luck. It will also poison you without exploding, and this is something that not enough people realize. That’s even more true for the more shelf-stable version (trimethylsilyldiazomethane) because that stability (and the commercial availability of solutions of it) have led some to think that it has correspondingly lower toxicity. This is not the case, and there have been fatalities to illustrate it. Here, the authors demonstrate that this species of bacteria (and likely many others besides) can produce this functional group through the actions of a cluster of enzymes. You start from lysine, and a series of steps goes through a hydroxylamine, disubstituted hydrazine, hydrazone, and then diazo. The authors have recapitulated these steps with isolated enzymes and shown that they occur as a gene cluster as well. The weirdest one (Dob3) is at that final step, and it has an unusual di-iron catalytic center that seems to accept a variety of hydrazone substrates. By trapping with a cyclooctyne, the paper demonstrates that both 4-diazo-3-oxobutanoic acid and diazoacetone are produced by the bacteria. These would be unlikely to be detected directly in most analytical protocols due to the low amounts and the overall sensitivity of the diazos (see above). Now, just why the organisms are producing these things remains an open question. They are certainly unusual reactive species and might be used in further biosynthetic pathways, or they might deter other bacteria from getting too close and messing around. Maybe both! What’s for sure is that considerable metabolic effort goes into making them, so there must be some reasons that have worked out evolutionarily. We shall see. So any of you who had diazoacetone on your list of Likely Natural Products, stop by the customer service desk and collect your winnings. The rest of us will be over here shaking our heads and wondering what’s next. . .
Show full content
I’ve written a number of posts here about weird natural products (and another such roundup is in the works), but I couldn’t resist highlighting this paper. The compounds discussed (which have been found to be produced in bacteria, specifically one human-pathogenic species, Nocardia ninae) are not huge and bewilderingly complex. Natural products have plenty of those kinds of structures, but these are very small molecules indeed. They’re just really unlikely ones.
That’s because they’re diazo compounds - yep, like good ol’ diazomethane, the methyl-ester-seeking-chemist’s friend. The diazo group (as you can see at left) is a funny beast. It has one nitrogen with a formal positive charge and another atom with a formal negative charge, so net neutral. You can put the negative charge either on the terminal carbon or the terminal nitrogen, but the truth is kind of in between. It also really looks like something that could find it in its heart to change to plain old dinitrogen gas. And so it does! Unless stabilized by adjacent functional groups, diazo compounds are known to be reactive and touchy, properties that are most on display with the smaller members of the series.
You don’t get much smaller than diazomethane, and that one is famous for reacting with carboxylic acids to more-or-less instantly form the corresponding methyl ester (and bubbles of nitrogen gas, if you’re working on a large enough scale, and if you are I would consider your life choices carefully). It does a lot of other interesting reactions, but you will always want to have it in dilute solution when you’re trying those, because it is sensitive to friction, heat, strong light, acid, and all sorts of other things, and it will most definitely explode if you push your luck.
It will also poison you without exploding, and this is something that not enough people realize. That’s even more true for the more shelf-stable version (trimethylsilyldiazomethane) because that stability (and the commercial availability of solutions of it) have led some to think that it has correspondingly lower toxicity. This is not the case, and there have been fatalities to illustrate it.
Here, the authors demonstrate that this species of bacteria (and likely many others besides) can produce this functional group through the actions of a cluster of enzymes. You start from lysine, and a series of steps goes through a hydroxylamine, disubstituted hydrazine, hydrazone, and then diazo. The authors have recapitulated these steps with isolated enzymes and shown that they occur as a gene cluster as well. The weirdest one (Dob3) is at that final step, and it has an unusual di-iron catalytic center that seems to accept a variety of hydrazone substrates.
By trapping with a cyclooctyne, the paper demonstrates that both 4-diazo-3-oxobutanoic acid and diazoacetone are produced by the bacteria. These would be unlikely to be detected directly in most analytical protocols due to the low amounts and the overall sensitivity of the diazos (see above). Now, just why the organisms are producing these things remains an open question. They are certainly unusual reactive species and might be used in further biosynthetic pathways, or they might deter other bacteria from getting too close and messing around. Maybe both! What’s for sure is that considerable metabolic effort goes into making them, so there must be some reasons that have worked out evolutionarily. We shall see.
So any of you who had diazoacetone on your list of Likely Natural Products, stop by the customer service desk and collect your winnings. The rest of us will be over here shaking our heads and wondering what’s next. . .
Quick preface: For those who have asked, my in-laws in Iran are (so far) all OK, but we are all of course in an extremely dangerous, stupid, and random situation right now. So yes, it does feel a bit odd to be writing about science and medical topics when there are so many other things going on (and going off) in the world. But this is one way I keep my own equilibrium, and I hope that that word of new discoveries and things that actually seem to be working out for the future can perhaps do the same for you. I’ve written here a number of times about chimeric antigen receptor T (CAR-T) cells a number of times here, going back to the first startling reports of their clinical efficacy. What’s been apparent for years now is the difficulty of getting them to work against solid tumors - that is, ones that are not based on circulating populations of aberrant blood cells. There are a number of reasons for this, and surely more that we haven’t worked out yet, but a big one has been the lack of “clean targets” for the T cells. Ideally you want a surface antigen protein that is expressed primarily by tumor cells and not by normal tissue, and you want a given tumor to express it all the way through. Remember, most solid tumors are heterogeneous mixtures of mutated cancer cells, all fighting it out with each other for resources and room to grow, with new mutations appearing all the time. Single-cell sequencing techniques have made this vividly clear over the years, and this also helps explain why anticancer therapy is so difficult: you may come up with a method to kill off 70, 80, 95% of the cells in a tumor, but you will leave behind a resistant remnant that now is free to expand, its competition having been conveniently removed. And so the tumor recurs some weeks or months later, and this time in an even harder-to-treat form. The surface protein CD70 is apparently not present in the great majority of adult tissues (except for some differentiating immune cell lines), but it does show up in tumors - to a degree. There are cells in many tumor types that are positive for CD70, but they also contain plenty of cells that appear not to be expressing that it at all. This new paper explores the hypothesis that these supposedly negative cells may in fact still be expressing CD70, but at a level too low for the usual methods of detection - and that means at a level too low for conventional CAR-T cells to recognize them, too. The authors find that the CD70 expression is epigenetically silenced by the methylase enzyme EZH2, which trimethylates lysine residues on histone H3 in the chromatin of these cells. That shuts expression down almost completely, but not quite. When compared to actual CD70 knockout cell lines under stringent detection conditions, they still show some low levels. The team then went on to try out one of the higher-sensitivity CAR-T platforms, the “HLA-independent T cell” (HIT) receptor design, aimed at CD70 with the hopes that these could go on to eliminate these low-expression cells. They prepared xenografts in animals taken from extremely resistant human tumors, and treatment with the CD70-HIT T cells showed very strong activity against pancreatic and ovarian tumors, and they propose a chromatin-accessibility assay to determine if a given tumor could be sensitive to this sort of CD70 attack. The authors note that at least 20 solid tumor types have been found to express the protein heterogeneously, and suggest that many (perhaps all) of these are in fact still expressing very low levels of CD70 and could be candidates for such treatment. This could be very good news indeed, and it also opens up a hunt for more such hidden targets that we have been missing. The combination of these and new methods to produce more active T cells is something we should keep an eye out for in clinical trials, and I look forward to following what happens!
Show full content
Quick preface: For those who have asked, my in-laws in Iran are (so far) all OK, but we are all of course in an extremely dangerous, stupid, and random situation right now. So yes, it does feel a bit odd to be writing about science and medical topics when there are so many other things going on (and going off) in the world. But this is one way I keep my own equilibrium, and I hope that that word of new discoveries and things that actually seem to be working out for the future can perhaps do the same for you.
I’ve written here a number of times about chimeric antigen receptor T (CAR-T) cells a number of times here, going back to the first startling reports of their clinical efficacy. What’s been apparent for years now is the difficulty of getting them to work against solid tumors - that is, ones that are not based on circulating populations of aberrant blood cells. There are a number of reasons for this, and surely more that we haven’t worked out yet, but a big one has been the lack of “clean targets” for the T cells. Ideally you want a surface antigen protein that is expressed primarily by tumor cells and not by normal tissue, and you want a given tumor to express it all the way through.
Remember, most solid tumors are heterogeneous mixtures of mutated cancer cells, all fighting it out with each other for resources and room to grow, with new mutations appearing all the time. Single-cell sequencing techniques have made this vividly clear over the years, and this also helps explain why anticancer therapy is so difficult: you may come up with a method to kill off 70, 80, 95% of the cells in a tumor, but you will leave behind a resistant remnant that now is free to expand, its competition having been conveniently removed. And so the tumor recurs some weeks or months later, and this time in an even harder-to-treat form.
The surface protein CD70 is apparently not present in the great majority of adult tissues (except for some differentiating immune cell lines), but it does show up in tumors - to a degree. There are cells in many tumor types that are positive for CD70, but they also contain plenty of cells that appear not to be expressing that it at all. This new paper explores the hypothesis that these supposedly negative cells may in fact still be expressing CD70, but at a level too low for the usual methods of detection - and that means at a level too low for conventional CAR-T cells to recognize them, too. The authors find that the CD70 expression is epigenetically silenced by the methylase enzyme EZH2, which trimethylates lysine residues on histone H3 in the chromatin of these cells. That shuts expression down almost completely, but not quite. When compared to actual CD70 knockout cell lines under stringent detection conditions, they still show some low levels.
The team then went on to try out one of the higher-sensitivity CAR-T platforms, the “HLA-independent T cell” (HIT) receptor design, aimed at CD70 with the hopes that these could go on to eliminate these low-expression cells. They prepared xenografts in animals taken from extremely resistant human tumors, and treatment with the CD70-HIT T cells showed very strong activity against pancreatic and ovarian tumors, and they propose a chromatin-accessibility assay to determine if a given tumor could be sensitive to this sort of CD70 attack. The authors note that at least 20 solid tumor types have been found to express the protein heterogeneously, and suggest that many (perhaps all) of these are in fact still expressing very low levels of CD70 and could be candidates for such treatment.
This could be very good news indeed, and it also opens up a hunt for more such hidden targets that we have been missing. The combination of these and new methods to produce more active T cells is something we should keep an eye out for in clinical trials, and I look forward to following what happens!
Here’s a transcript of a recent podcast over at Nature in their Careers section. They’ve been doing a series on what they feel are taboo topics in the workplace, and previous ones have covered things as disparate as alcohol abuse and religious convictions. This latest one, though hits on a subject that might not sound, to someone who hasn’t been in academia, like that much of a divisive issue. But it is: leaving academia to go into industry. The host (Adam Levy) interviews in some depth two people who’ve had that experience, and they touch on a lot of the features, with a major one the attitude that academia is the only “real” job for someone who came up through academic training. Leaving is then seen as a step down, an admission that you just couldn’t cut it in the work that you’re supposed to be doing. Another problem is how such a move brings up other topics that may not always be aired out: work/life balance, the actual chances for promotion or long-term employment, enjoyment (or lack of it) in dealing with students and grading, and of course salary. Some of these get weaponized in the service of that first big attitude - oh sure, you’re willing to leave academia because you’re being bought off with promises of money and free time, that sort of thing. I didn’t experience this as directly as the people being interviewed, but I did expect (as I finished up undergraduate study and started graduate school) that I was going to end up as a professor somewhere. I came from a small liberal arts school (Hendrix) where the quality of the chemistry instruction was overall quite high (and especially, thanks to Tom Goodwin, in the organic chemistry that I was specializing in). So I envisioned myself as going to a similar sort of situation - a good liberal-arts sort of school where I would teach organic chemistry and do some research on the side. By that point I was already a bit worried about the idea of going to a larger research-intensive university and fighting for tenure there, and what I saw in my own graduate years really firmed those feelings up, let me tell you. I went into my postdoc position sure that there was just no way I was going to do that - it looked like an even worse grind than graduate school, and that by itself had pushed me pretty close to the limits of what I could put up with. As it happened, when time came around to start looking for jobs, there weren’t really any academic jobs of the type I wanted to try for showing up. By that time, though, I had several friends from grad school who’d gone off into industry, and I thought I should follow suit. Thirty-seven years later, here I am! I certainly did see professors who felt that their “favorite” students and post-docs should definitely be aiming at academic jobs (and who felt disappointed or even betrayed if they didn’t). That comes up in the podcast interviews, as well as the feeling that by leaving academia you were somehow letter everyone else down, or that you’d taken up a spot that could have gone to someone who would have stuck with it. These feelings may be natural enough, but I don’t think they’re too accurate or helpful. You could just as easily feel that you’re doing other people (as well as yourself) a favor by creating a space for someone new who might find the situation more to their liking, for example. Overall, my feeling is that the “industry vs. academia” divide is not quite as much of an instant argument-starter as it used to be, but these interviews make it clear that things are a long way from dying down. There seems little doubt that academic careers have just been getting more difficult over the years, and that was true well before the recent attempts at complete destruction of a lot of US academic funding mechanisms and before the direct attacks on academic policies and institutions that have characterized the current administration. Other countries such as the U.K. have been going through unfortunately similar experiences, which perhaps has made it seem less necessary to explain your decision if you decide that it’s time for you yourself to get out. The situation isn’t helped by the processes and customes, some of them of very long standing, that just don’t seem to make sense when you think about them: “I think sometimes it helps when you break outside the academic bubble. Because I remember explaining once to my in-laws (who aren’t in academia) about publishing, and explaining to them that well, you write something, and then you send it off to a journal. And then other people, your friends, will review it for free. And then once you’ve had it accepted, you pay money to get it published. And then you then basically pay money to get it back again. Just trying to explain that to my in-laws, they just said, “I don’t understand. If you’re writing it and you’re doing all the editing, why are you paying to get it out there, and paying to get it back?” And I think there were little things like that that sometimes you get an outside perspective and you go, “Ah, that’s, I guess it’s a point.” It’s not that we don’t have some senseless behaviors in industry; don’t get me wrong. But do we have anything quite as institutionalized and quite as odd as the current publication system? That's just the sort of thing to make you start re-evaluating your life choices. . .
Show full content
Here’s a transcript of a recent podcast over at Nature in their Careers section. They’ve been doing a series on what they feel are taboo topics in the workplace, and previous ones have covered things as disparate as alcohol abuse and religious convictions. This latest one, though hits on a subject that might not sound, to someone who hasn’t been in academia, like that much of a divisive issue. But it is: leaving academia to go into industry.
The host (Adam Levy) interviews in some depth two people who’ve had that experience, and they touch on a lot of the features, with a major one the attitude that academia is the only “real” job for someone who came up through academic training. Leaving is then seen as a step down, an admission that you just couldn’t cut it in the work that you’re supposed to be doing. Another problem is how such a move brings up other topics that may not always be aired out: work/life balance, the actual chances for promotion or long-term employment, enjoyment (or lack of it) in dealing with students and grading, and of course salary. Some of these get weaponized in the service of that first big attitude - oh sure, you’re willing to leave academia because you’re being bought off with promises of money and free time, that sort of thing.
I didn’t experience this as directly as the people being interviewed, but I did expect (as I finished up undergraduate study and started graduate school) that I was going to end up as a professor somewhere. I came from a small liberal arts school (Hendrix) where the quality of the chemistry instruction was overall quite high (and especially, thanks to Tom Goodwin, in the organic chemistry that I was specializing in). So I envisioned myself as going to a similar sort of situation - a good liberal-arts sort of school where I would teach organic chemistry and do some research on the side.
By that point I was already a bit worried about the idea of going to a larger research-intensive university and fighting for tenure there, and what I saw in my own graduate years really firmed those feelings up, let me tell you. I went into my postdoc position sure that there was just no way I was going to do that - it looked like an even worse grind than graduate school, and that by itself had pushed me pretty close to the limits of what I could put up with. As it happened, when time came around to start looking for jobs, there weren’t really any academic jobs of the type I wanted to try for showing up. By that time, though, I had several friends from grad school who’d gone off into industry, and I thought I should follow suit. Thirty-seven years later, here I am!
I certainly did see professors who felt that their “favorite” students and post-docs should definitely be aiming at academic jobs (and who felt disappointed or even betrayed if they didn’t). That comes up in the podcast interviews, as well as the feeling that by leaving academia you were somehow letter everyone else down, or that you’d taken up a spot that could have gone to someone who would have stuck with it. These feelings may be natural enough, but I don’t think they’re too accurate or helpful. You could just as easily feel that you’re doing other people (as well as yourself) a favor by creating a space for someone new who might find the situation more to their liking, for example.
Overall, my feeling is that the “industry vs. academia” divide is not quite as much of an instant argument-starter as it used to be, but these interviews make it clear that things are a long way from dying down. There seems little doubt that academic careers have just been getting more difficult over the years, and that was true well before the recent attempts at complete destruction of a lot of US academic funding mechanisms and before the direct attacks on academic policies and institutions that have characterized the current administration. Other countries such as the U.K. have been going through unfortunately similar experiences, which perhaps has made it seem less necessary to explain your decision if you decide that it’s time for you yourself to get out. The situation isn’t helped by the processes and customes, some of them of very long standing, that just don’t seem to make sense when you think about them:
“I think sometimes it helps when you break outside the academic bubble. Because I remember explaining once to my in-laws (who aren’t in academia) about publishing, and explaining to them that well, you write something, and then you send it off to a journal. And then other people, your friends, will review it for free. And then once you’ve had it accepted, you pay money to get it published. And then you then basically pay money to get it back again.
Just trying to explain that to my in-laws, they just said, “I don’t understand. If you’re writing it and you’re doing all the editing, why are you paying to get it out there, and paying to get it back?”
And I think there were little things like that that sometimes you get an outside perspective and you go, “Ah, that’s, I guess it’s a point.”
It’s not that we don’t have some senseless behaviors in industry; don’t get me wrong. But do we have anything quite as institutionalized and quite as odd as the current publication system? That's just the sort of thing to make you start re-evaluating your life choices. . .
Here’s a real oddity for you in the field of antiviral mechanisms. Over the years people have looked at all sorts of viral infection and propagation mechanisms, trying to find vulnerabilities to exploit. But “cell membrane stiffening” leading to decreased fluidity and impaired viral entry has never, to my knowledge, been one of them. So how to you get that toughened-up membrane state? The authors say that this is downstream of the mechanosensor Piezo1. That’s an interesting system - the receptor is (as the name implies) sensitive to mechanical stress, and it’s found in a whole list of human tissues. It seems to work differently than other known mechanosensors, and roles have been found for it in the cardiovascular system (and thence in the CNS), in lung function, in bone remodeling, cell lineage switching in the marrow, in kidney function, in T-cell migration, and who knows what else. Turns out that there are a lot of uses for being able to sense pressure, liquid flow, viscosity, and movement on a cellular level! And here the authors say that its effects on cell membranes can change the efficiency of viral entry mechanisms. They noticed this via the effects of cell culture density on viral replication - specifically, crowded cells seem to show slower viral spread, which is sort of the opposite of what you might otherwise expect. One big difference in a well-stuffed cell culture is the mechanical stress on the individual cells, so they started looking at that as a possible mechanism. And in fact knocking down expression of Piezo1 seemed to bring back the ability of viruses to infect densely-packed cell cultures, which seems to refute the standard explanations for the effect (“densely packed cells are showing suboptimal growth anyway”, etc.) And more directly, taking regular lower-density cell cultures and mechanically shaking them seemed to bring on enhanced resistance to viral infection. This doesn’t seem to depend so much on Piezo1 expression levels (or changes thereof) as much as it does on activation of the protein. Chemical agonists of Piezo1 had similar effects without the vibrations being necessary. And the cellular entry proteins used by the viruses under study didn’t seem to be affected - rather it was a whole-membrane effect, with changes in lipid-raft behavior, impaired diffusion of probes in the lipid bilayer, and so on. Membrane proteins just seem to have more trouble moving around after the remodeling induced by Piezo1 activation, and this hits the viral entry mechanisms as well - an effect demonstrated across a range of viruses. So this “mechano-antiviral response system” (MARS) may well be a type of innate immunity that we have been completely overlooking, and there may well be ways to leverage it to our advantage. There are those synthetic Piezo1 agonists, for one thing, but more simply, direct mechanical stimulation such as vibration or even massage is worth investigating. There’s a lot to learn about this, and I look forward to seeing other groups replicate and extend these results! And as a corollary, man is there a lot of interesting stuff out there that we haven't realized yet. . .
Show full content
Here’s a real oddity for you in the field of antiviral mechanisms. Over the years people have looked at all sorts of viral infection and propagation mechanisms, trying to find vulnerabilities to exploit. But “cell membrane stiffening” leading to decreased fluidity and impaired viral entry has never, to my knowledge, been one of them.
So how to you get that toughened-up membrane state? The authors say that this is downstream of the mechanosensor Piezo1. That’s an interesting system - the receptor is (as the name implies) sensitive to mechanical stress, and it’s found in a whole list of human tissues. It seems to work differently than other known mechanosensors, and roles have been found for it in the cardiovascular system (and thence in the CNS), in lung function, in bone remodeling, cell lineage switching in the marrow, in kidney function, in T-cell migration, and who knows what else. Turns out that there are a lot of uses for being able to sense pressure, liquid flow, viscosity, and movement on a cellular level!
And here the authors say that its effects on cell membranes can change the efficiency of viral entry mechanisms. They noticed this via the effects of cell culture density on viral replication - specifically, crowded cells seem to show slower viral spread, which is sort of the opposite of what you might otherwise expect. One big difference in a well-stuffed cell culture is the mechanical stress on the individual cells, so they started looking at that as a possible mechanism. And in fact knocking down expression of Piezo1 seemed to bring back the ability of viruses to infect densely-packed cell cultures, which seems to refute the standard explanations for the effect (“densely packed cells are showing suboptimal growth anyway”, etc.)
And more directly, taking regular lower-density cell cultures and mechanically shaking them seemed to bring on enhanced resistance to viral infection. This doesn’t seem to depend so much on Piezo1 expression levels (or changes thereof) as much as it does on activation of the protein. Chemical agonists of Piezo1 had similar effects without the vibrations being necessary. And the cellular entry proteins used by the viruses under study didn’t seem to be affected - rather it was a whole-membrane effect, with changes in lipid-raft behavior, impaired diffusion of probes in the lipid bilayer, and so on. Membrane proteins just seem to have more trouble moving around after the remodeling induced by Piezo1 activation, and this hits the viral entry mechanisms as well - an effect demonstrated across a range of viruses.
So this “mechano-antiviral response system” (MARS) may well be a type of innate immunity that we have been completely overlooking, and there may well be ways to leverage it to our advantage. There are those synthetic Piezo1 agonists, for one thing, but more simply, direct mechanical stimulation such as vibration or even massage is worth investigating. There’s a lot to learn about this, and I look forward to seeing other groups replicate and extend these results! And as a corollary, man is there a lot of interesting stuff out there that we haven't realized yet. . .
One of the things that seems clear (after many years of writing this blog and fielding questions about drug research from all and sundry) is that most people outside the field don’t realize the failure rates for drug trials. And they don’t realize how often it is that seemingly good ideas just don’t pan out. The recurring proposals to de-emphasize confirmatory trials for efficacy in favor of moving ahead with getting medications to patients is an example of this. Advocates often refer to an “invisible graveyard” of patients who died while waiting for more clinical data on something that could have helped them, and I don’t want to make light of that. But there would also be a huge mass of still-pretty-invisible deadweight losses if we waved a lot more therapies through - money and time and effort and hopes squandered on things that turned out to be complete wastes. If you have a too-optimistic view of how often drug trials work and how often new therapeutic ideas come through, then the Invisible Graveyard looms all the larger. But my own experience has been that the wastes in the other direction are pretty fearsome, and that we’ve quietly demonstrated this over and over. It’s just that people tend not to notice it happening, because companies, understandably enough, don’t make a big noise about trials that don’t work out and clinical hypotheses that turned out to be mistaken. There’s a constant background noise of “promising clinical results” stories in the news, but people generally don’t get a chance to see how few of those lead anywhere. I’ve covered a number of such cases here over the years, and believe me, there are so many more where those came from. (That’s not even a comprehensive list from my own site!) All of those examples looked promising in Phase II or in a first Phase III trial, but came to grief on closer examination. This is a good overview of a recent example of this, and it’s not even one drug or even one company. A few years ago, Genentech had promising clinical data on an immuno-oncology idea, an antibody (tirgolumab) targeting TIGIT (which stands for “T-cell immunoreceptor with If and ITIM domains”). This was supposed to enhance to effects of some other existing immunotherapies, and a 2020 report of a trial in 135 patients seemed to show that this was just what was happening. Since that time an estimated 22 companies have taken 30 TIGIT compounds into development and more that 220 clinical trials have been run. Needless to say this has been a huge investment of time, effort and money, so what do we have to show for it here in 2026? Not much. Starting in 2022, trial results began to look significantly less promising, and candidate after candidate came up short. Genentech has actually exited the field entirely, with the failure of a large lung cancer study last year as the final blow. Plenty of other companies have as well, and as this article shows, experts in the field range from deeply pessimistic to well-hey-you-never-know about the prospects for the remaining candidates. One clinician is quoted as saying that “We have had the Phase II hope versus the Phase III reality”, and that really sums it up. And that’s why my argument has always been that we can’t race ahead based on the Phase II hope stage. TIGIT-directed drugs at least don’t seem to have made anyone worse, but some of those earlier links will point to drugs that had promising early trials that no only didn’t work in larger studies, but turned out to have tox liabilities that actually would have ended up injuring and outright killing off more people had they been approved at that stage. As those last two papers linked detail, we have learned an awful lot more about TIGIT during these last few years of failure, and that includes learning the likely reasons for why these compounds didn’t work (too much overlap between its signaling pathways and the existing immuno-oncology drugs is a big part of the explanation). There are still some ideas in the space - AstraZeneca is keeping the faith with an approach of their own, and one can only wish them luck. But at the current state of biomedical knowledge, that luck is still an essential part of success. Hard work is a big part too, for sure. The best expert insights, absolutely. Great big piles of money - goes without saying, doesn’t it? But luck is in there, too, and will be for the foreseeable future. Never let yourself forget that part.
Show full content
One of the things that seems clear (after many years of writing this blog and fielding questions about drug research from all and sundry) is that most people outside the field don’t realize the failure rates for drug trials. And they don’t realize how often it is that seemingly good ideas just don’t pan out.
The recurring proposals to de-emphasize confirmatory trials for efficacy in favor of moving ahead with getting medications to patients is an example of this. Advocates often refer to an “invisible graveyard” of patients who died while waiting for more clinical data on something that could have helped them, and I don’t want to make light of that. But there would also be a huge mass of still-pretty-invisible deadweight losses if we waved a lot more therapies through - money and time and effort and hopes squandered on things that turned out to be complete wastes.
If you have a too-optimistic view of how often drug trials work and how often new therapeutic ideas come through, then the Invisible Graveyard looms all the larger. But my own experience has been that the wastes in the other direction are pretty fearsome, and that we’ve quietly demonstrated this over and over. It’s just that people tend not to notice it happening, because companies, understandably enough, don’t make a big noise about trials that don’t work out and clinical hypotheses that turned out to be mistaken. There’s a constant background noise of “promising clinical results” stories in the news, but people generally don’t get a chance to see how few of those lead anywhere. I’ve covered a number of such cases here over the years, and believe me, there are so many more where those came from. (That’s not even a comprehensive list from my own site!) All of those examples looked promising in Phase II or in a first Phase III trial, but came to grief on closer examination.
This is a good overview of a recent example of this, and it’s not even one drug or even one company. A few years ago, Genentech had promising clinical data on an immuno-oncology idea, an antibody (tirgolumab) targeting TIGIT (which stands for “T-cell immunoreceptor with If and ITIM domains”). This was supposed to enhance to effects of some other existing immunotherapies, and a 2020 report of a trial in 135 patients seemed to show that this was just what was happening.
Since that time an estimated 22 companies have taken 30 TIGIT compounds into development and more that 220 clinical trials have been run. Needless to say this has been a huge investment of time, effort and money, so what do we have to show for it here in 2026? Not much. Starting in 2022, trial results began to look significantly less promising, and candidate after candidate came up short. Genentech has actually exited the field entirely, with the failure of a large lung cancer study last year as the final blow. Plenty of other companies have as well, and as this article shows, experts in the field range from deeply pessimistic to well-hey-you-never-know about the prospects for the remaining candidates. One clinician is quoted as saying that “We have had the Phase II hope versus the Phase III reality”, and that really sums it up.
And that’s why my argument has always been that we can’t race ahead based on the Phase II hope stage. TIGIT-directed drugs at least don’t seem to have made anyone worse, but some of those earlier links will point to drugs that had promising early trials that no only didn’t work in larger studies, but turned out to have tox liabilities that actually would have ended up injuring and outright killing off more people had they been approved at that stage.
As those last two papers linked detail, we have learned an awful lot more about TIGIT during these last few years of failure, and that includes learning the likely reasons for why these compounds didn’t work (too much overlap between its signaling pathways and the existing immuno-oncology drugs is a big part of the explanation). There are still some ideas in the space - AstraZeneca is keeping the faith with an approach of their own, and one can only wish them luck. But at the current state of biomedical knowledge, that luck is still an essential part of success. Hard work is a big part too, for sure. The best expert insights, absolutely. Great big piles of money - goes without saying, doesn’t it? But luck is in there, too, and will be for the foreseeable future. Never let yourself forget that part.
This perspective is well worth reading, even if you find yourself arguing with some parts of it (here's a freely available preprint version). Longtime drug discovery guy Mark Murcko titled it “The Affinity Advantage”, and I think this gets across the point he’s trying to make: “I argue that a greater emphasis on optimizing binding affinity will accelerate drug discovery. Note that “optimizing” is not always synonymous with “maximizing” So let’s talk about that! As the paper notes in the first paragraph, potency is necessary but not sufficient, because there are a lot of other factors that go into making a successful drug. And let’s define “potency” right up front as a functional readout, the phenotypic/pharmacological response in the cell or animal at a given dose. Meanwhile “binding affinity” (which a lot of us loosely use as a synonym) is of course a measure of the interaction between a given compound and a given protein target. In all of our defense, those two are often pretty well correlated, but not always, not over the whole range available, and not in a linear fashion for the most part, either. Murcko’s arguing here that we don’t actually put enough time and effort into optimizing that binding affinity. And I can see where he’s coming from (details below), but I have to say before starting that this triggers some old allergies from back in my own med-chem career. Here are the details: One of the first big programs I worked on was a selective muscarinic antagonist program, and the company’s CNS leadership was constant on the team to make thing more and more potent, more and more selective. Those aren’t irrational demands - muscarinic ligands are notorious (as both antagonists and agonists) for setting off very unwelcome side effects due to lack of selectivity. And for a compound that’s supposed to penetrate the blood-brain barrier, you have a real advantage if your ligand is very potent - that very likely means that less of it has to get into the brain to get your desired effect, and the BBB may well be working diligently to keep very much of your compound getting across already! But in my view, the process was not managed well, and even in my early years in the industry I eventually started wondering what was going on. The modifications that were made to the chemical series to bring on all that potency and selectivity made the lead molecules larger and more complex all the way, and relentlessly degraded their physical and pharmacokinetic properties (absorption, logD, solubility, and so on). In the end, the project delivered a compound that was not exactly what you’d pick for a CNS drug candidate, and indeed it wiped out quickly in development. Now, even a wonderful and perfect antagonist molecule would likely have gone on to do the same (the mechanism of action was, in the end, suspect and no one’s been able to get it to work since), but the team spent several years banging away “improving” compounds that were, I thought, way overengineered for such a long-shot idea. So that’s where I’m starting from! Murcko’s list of the advantages of higher affinity include that pharmacokinetic point I made above, extended to giving a fighting chance to unusual molecular types or difficult mechanisms in general, as well as giving better signal/noise in cell assays, and quicker identification of resistance mechanisms (in infectious disease or oncology). Those make a lot of sense to me. There are a couple of points where I think he may be trying to have it both ways, in that more potent compounds may well be more specific and thus have fewer off-target side effects, but he also mentions potent tool compounds helping to provide early insights into off-target effects as well. He also lists “embolden teams to pursue synthetically challenging compounds” but that (as he notes) would seem to require a better ability to predict gains in affinity than we often have. To its credit, the paper also lists objections to this affinity-above-all approach. They include the danger of ignoring other properties in the drive to increase binding (which is what I think happened in the muscarinic project discussed above), and indeed this might be a constant danger because many of the modifications that are most likely to increase affinity can directly hurt compound properties. I think some of Murcko’s responses reflect (to his own credit) his innately optimistic nature (disclosure: I’ve worked with him!), such as the statement that “If we can design for affinity we can also design away from “antitargets” (i.e. the ones that cause bad side effects). He also confronts the problems with (among others) applying this principle to phenotypic targets (where you start out not even knowing the target you’re hitting, much less the binding affinity of your compounds to it), to target-selection problems, idiosyncratic tox, and to newer complex modalities such as glues and bifunctionals. I’ll send you to read the paper for all the details on these, but in many cases, a major part of the response to these objections is that more potent compounds will made advancing our knowledge in all of these areas more rapid and more secure. And I think that’s probably correct - God knows way too much work has been spent over the years on model system and tool compounds (many of them, grrr, available from commercial catalogs) that are simply not potent enough (and often not selective enough) to draw any firm conclusions from. Working at the other end of the scale, where you know that you’re hitting a defined target very strongly, has to be an improvement. Not least is the ability to have some potency to burn in the service of properties or selectivity, if you have to. But to do that effectively you ideally want to be more potent than you ever might have thought you needed to be to start with! And there’s another advantage in being able to transfer chemical matter and mechanisms to related proteins, using the knowledge that you’ve gained by solid optimization against your original one. It can be a lot better than starting blind again, that’s for sure. What levels of potency are we talking about, then? The paper exhorts med-chem teams not to stop optimizing affinity once you hit nanomolar levels, so that should give you the idea. His general advice is to treat optimizing affinity as an objective that never goes away: “. . .I am suggesting that throughout the duration of a research project, the pursuit of further intrinsic affinity gains is a sound strategy. Improved affinity, if not offset by setbacks in ADME parameters, will generally lead to better outcomes.” This also means avoiding the temptation to narrow down to one lead series too quickly, and to be tolerant of more difficult chemistry in the service of these potency gains. What it doesn’t mean is to ignore selectivity and ADME properties, because that would be a sure route to disaster. No, this is definitely playing the process of drug discovery on Hard Mode, but what other mode is likely to deliver results? It’s a hard game. I do wonder how many organizations will be willing to try it like this, though - I’ve seen plenty of organizations, projects (and project managers) whose patience would not extend far enough to realize all these recommendations. (And so has Mark Murcko!) But it could well be an ideal to shoot for.
Show full content
This perspective is well worth reading, even if you find yourself arguing with some parts of it (here's a freely available preprint version). Longtime drug discovery guy Mark Murcko titled it “The Affinity Advantage”, and I think this gets across the point he’s trying to make: “I argue that a greater emphasis on optimizing binding affinity will accelerate drug discovery. Note that “optimizing” is not always synonymous with “maximizing”
So let’s talk about that! As the paper notes in the first paragraph, potency is necessary but not sufficient, because there are a lot of other factors that go into making a successful drug. And let’s define “potency” right up front as a functional readout, the phenotypic/pharmacological response in the cell or animal at a given dose. Meanwhile “binding affinity” (which a lot of us loosely use as a synonym) is of course a measure of the interaction between a given compound and a given protein target. In all of our defense, those two are often pretty well correlated, but not always, not over the whole range available, and not in a linear fashion for the most part, either.
Murcko’s arguing here that we don’t actually put enough time and effort into optimizing that binding affinity. And I can see where he’s coming from (details below), but I have to say before starting that this triggers some old allergies from back in my own med-chem career. Here are the details:
One of the first big programs I worked on was a selective muscarinic antagonist program, and the company’s CNS leadership was constant on the team to make thing more and more potent, more and more selective. Those aren’t irrational demands - muscarinic ligands are notorious (as both antagonists and agonists) for setting off very unwelcome side effects due to lack of selectivity. And for a compound that’s supposed to penetrate the blood-brain barrier, you have a real advantage if your ligand is very potent - that very likely means that less of it has to get into the brain to get your desired effect, and the BBB may well be working diligently to keep very much of your compound getting across already!
But in my view, the process was not managed well, and even in my early years in the industry I eventually started wondering what was going on. The modifications that were made to the chemical series to bring on all that potency and selectivity made the lead molecules larger and more complex all the way, and relentlessly degraded their physical and pharmacokinetic properties (absorption, logD, solubility, and so on). In the end, the project delivered a compound that was not exactly what you’d pick for a CNS drug candidate, and indeed it wiped out quickly in development. Now, even a wonderful and perfect antagonist molecule would likely have gone on to do the same (the mechanism of action was, in the end, suspect and no one’s been able to get it to work since), but the team spent several years banging away “improving” compounds that were, I thought, way overengineered for such a long-shot idea.
So that’s where I’m starting from! Murcko’s list of the advantages of higher affinity include that pharmacokinetic point I made above, extended to giving a fighting chance to unusual molecular types or difficult mechanisms in general, as well as giving better signal/noise in cell assays, and quicker identification of resistance mechanisms (in infectious disease or oncology). Those make a lot of sense to me. There are a couple of points where I think he may be trying to have it both ways, in that more potent compounds may well be more specific and thus have fewer off-target side effects, but he also mentions potent tool compounds helping to provide early insights into off-target effects as well. He also lists “embolden teams to pursue synthetically challenging compounds” but that (as he notes) would seem to require a better ability to predict gains in affinity than we often have.
To its credit, the paper also lists objections to this affinity-above-all approach. They include the danger of ignoring other properties in the drive to increase binding (which is what I think happened in the muscarinic project discussed above), and indeed this might be a constant danger because many of the modifications that are most likely to increase affinity can directly hurt compound properties. I think some of Murcko’s responses reflect (to his own credit) his innately optimistic nature (disclosure: I’ve worked with him!), such as the statement that “If we can design for affinity we can also design away from “antitargets” (i.e. the ones that cause bad side effects). He also confronts the problems with (among others) applying this principle to phenotypic targets (where you start out not even knowing the target you’re hitting, much less the binding affinity of your compounds to it), to target-selection problems, idiosyncratic tox, and to newer complex modalities such as glues and bifunctionals.
I’ll send you to read the paper for all the details on these, but in many cases, a major part of the response to these objections is that more potent compounds will made advancing our knowledge in all of these areas more rapid and more secure. And I think that’s probably correct - God knows way too much work has been spent over the years on model system and tool compounds (many of them, grrr, available from commercial catalogs) that are simply not potent enough (and often not selective enough) to draw any firm conclusions from. Working at the other end of the scale, where you know that you’re hitting a defined target very strongly, has to be an improvement.
Not least is the ability to have some potency to burn in the service of properties or selectivity, if you have to. But to do that effectively you ideally want to be more potent than you ever might have thought you needed to be to start with! And there’s another advantage in being able to transfer chemical matter and mechanisms to related proteins, using the knowledge that you’ve gained by solid optimization against your original one. It can be a lot better than starting blind again, that’s for sure.
What levels of potency are we talking about, then? The paper exhorts med-chem teams not to stop optimizing affinity once you hit nanomolar levels, so that should give you the idea. His general advice is to treat optimizing affinity as an objective that never goes away: “. . .I am suggesting that throughout the duration of a research project, the pursuit of further intrinsic affinity gains is a sound strategy. Improved affinity, if not offset by setbacks in ADME parameters, will generally lead to better outcomes.” This also means avoiding the temptation to narrow down to one lead series too quickly, and to be tolerant of more difficult chemistry in the service of these potency gains. What it doesn’t mean is to ignore selectivity and ADME properties, because that would be a sure route to disaster.
No, this is definitely playing the process of drug discovery on Hard Mode, but what other mode is likely to deliver results? It’s a hard game. I do wonder how many organizations will be willing to try it like this, though - I’ve seen plenty of organizations, projects (and project managers) whose patience would not extend far enough to realize all these recommendations. (And so has Mark Murcko!) But it could well be an ideal to shoot for.
There’s been a lot of talk in recent years about “liquid biopsies” and especially about the promise of using blood samples to test for a wide variety of cancers at once. In most cases, this involves looking for fragments of tumor-associated DNA or RNA that are circulating in the blood, and on the face of it, that seems like a pretty good plan. But it has some challenges. One is whether or not a given tumor sheds enough signal to be a reliable marker or not, and that depends both on the amount of DNA/RNA and also on its stability in the blood. The second is just deciding what signal you’re looking for, and especially making sure that it’s not something that can lead to false positives when derived from normal tissue. And of course you have the usual analytical concerns when you’re looking for very small amounts of something in a very large and very mixed sample: what’s the signal/noise of your assay technology, and how much does it vary from marker to marker to marker? Then there are larger questions, such as: does this detection - even if it’s reasonably successful - actually benefit the patients? That sounds like an odd thing to ask, but consider what happens if some tumor type only generates a solid readout once it’s too advanced to treat. The real endpoints are improvements in mortality and morbidity, and those can be difficult to prove. A test in this area that’s had a lot of press is Galleri, from a company called Grail. They’ve been selling the test for several years now although it hasn’t yet been FDA-approved, and that means that people are almost entirely paying out of pocket (nearly $1000, and last year the company says that they had 185,000 customers). So how does that work? This business goes on under the provisions of the Clinical Laboratory Improvement Act (CLIA). That is a framework that tries to standardize the complexity and scope of various diagnostic tests, and if you read that link you’ll see that this is just as wooly a prospect as it sounds like. There is a category under the CLIA called “laboratory-developed tests”, which are supposed to be developed and used within a single lab, and the FDA has stated that these can indeed be on the market without prior approval. The 23-and-Me people started by offering their direct-to-consumer sequencing test in just such a fashion, but the FDA (in a rare move) forced them to seek approval as a medical device. I believe that Galleri (and the competing CancerGuard test) are in the “ask your doctor” category rather than being DTC to avoid this sort of problem. People watching this field - physicians, patients, and not least, insurance providers - have been waiting for a large prospective trial in the UK to read out on Galleri, one involving 142,000 patients aged 50 to 77. This is just the sort of thing these tests need, and I have to applaud the Grail folks for being willing to run it rather than just continuing the wink-and-nudge business model (which of course does have its limitations, compared to, say, getting on the Medicare formulary). Over the weekend those results appeared. The readouts were how many people got a cancer diagnosis, compared to a control group that didn’t get the test, and how many of these were in the (far more actionable) stage 1 or stage 2 of the cancer. The hope was to show a 20% reduction in advanced-cancer diagnoses overall. Unfortunately, the total number of people diagnosed at stage 3 or stage 4 did not show any statistically significant reduction in the Galleri patients as compared to controls - a complete miss. Perhaps that endpoint was a bit ambitious, although if you look at the advertising in this field, it’s really the sort of thing that the early-screening companies are implicitly promising. This really is about the worst result that Grail could have had, and it’s not going to help out the other people in this field, either. I would have to think that the pressure is on now for anyone promulgating a multiple-early-detection assay to run a real trial like this one and to show that they don’t get the same disappointing results. And even if they'd met their goals, questions would remain, as pointed out here at Stat: this is still too short a trial to get the sorts of mortality improvement readouts mentioned above. It should be! What we’ve seen here is that what seems to be a sensible idea for early cancer detetion, implemented with the latest technology, can come up completely short. And that should make everyone stop and think for a minute. Lots of reasonable and/or exciting ideas in medicine collapse when you put some weight on them, sadly, and we should never forget it and never pretend that we’ve suddenly rewritten the rulebook and dodged the need for experimental proof.
Show full content
There’s been a lot of talk in recent years about “liquid biopsies” and especially about the promise of using blood samples to test for a wide variety of cancers at once. In most cases, this involves looking for fragments of tumor-associated DNA or RNA that are circulating in the blood, and on the face of it, that seems like a pretty good plan. But it has some challenges.
One is whether or not a given tumor sheds enough signal to be a reliable marker or not, and that depends both on the amount of DNA/RNA and also on its stability in the blood. The second is just deciding what signal you’re looking for, and especially making sure that it’s not something that can lead to false positives when derived from normal tissue. And of course you have the usual analytical concerns when you’re looking for very small amounts of something in a very large and very mixed sample: what’s the signal/noise of your assay technology, and how much does it vary from marker to marker to marker? Then there are larger questions, such as: does this detection - even if it’s reasonably successful - actually benefit the patients? That sounds like an odd thing to ask, but consider what happens if some tumor type only generates a solid readout once it’s too advanced to treat. The real endpoints are improvements in mortality and morbidity, and those can be difficult to prove.
A test in this area that’s had a lot of press is Galleri, from a company called Grail. They’ve been selling the test for several years now although it hasn’t yet been FDA-approved, and that means that people are almost entirely paying out of pocket (nearly $1000, and last year the company says that they had 185,000 customers).
So how does that work? This business goes on under the provisions of the Clinical Laboratory Improvement Act (CLIA). That is a framework that tries to standardize the complexity and scope of various diagnostic tests, and if you read that link you’ll see that this is just as wooly a prospect as it sounds like. There is a category under the CLIA called “laboratory-developed tests”, which are supposed to be developed and used within a single lab, and the FDA has stated that these can indeed be on the market without prior approval. The 23-and-Me people started by offering their direct-to-consumer sequencing test in just such a fashion, but the FDA (in a rare move) forced them to seek approval as a medical device. I believe that Galleri (and the competing CancerGuard test) are in the “ask your doctor” category rather than being DTC to avoid this sort of problem.
People watching this field - physicians, patients, and not least, insurance providers - have been waiting for a large prospective trial in the UK to read out on Galleri, one involving 142,000 patients aged 50 to 77. This is just the sort of thing these tests need, and I have to applaud the Grail folks for being willing to run it rather than just continuing the wink-and-nudge business model (which of course does have its limitations, compared to, say, getting on the Medicare formulary). Over the weekend those results appeared.
The readouts were how many people got a cancer diagnosis, compared to a control group that didn’t get the test, and how many of these were in the (far more actionable) stage 1 or stage 2 of the cancer. The hope was to show a 20% reduction in advanced-cancer diagnoses overall. Unfortunately, the total number of people diagnosed at stage 3 or stage 4 did not show any statistically significant reduction in the Galleri patients as compared to controls - a complete miss.
Perhaps that endpoint was a bit ambitious, although if you look at the advertising in this field, it’s really the sort of thing that the early-screening companies are implicitly promising. This really is about the worst result that Grail could have had, and it’s not going to help out the other people in this field, either. I would have to think that the pressure is on now for anyone promulgating a multiple-early-detection assay to run a real trial like this one and to show that they don’t get the same disappointing results. And even if they'd met their goals, questions would remain, as pointed out here at Stat: this is still too short a trial to get the sorts of mortality improvement readouts mentioned above.
It should be! What we’ve seen here is that what seems to be a sensible idea for early cancer detetion, implemented with the latest technology, can come up completely short. And that should make everyone stop and think for a minute. Lots of reasonable and/or exciting ideas in medicine collapse when you put some weight on them, sadly, and we should never forget it and never pretend that we’ve suddenly rewritten the rulebook and dodged the need for experimental proof.
Here’s a very interesting rundown on an issue that became apparent during the coronavirus vaccine development period. You may remember (if you haven’t expunged that entire period from your mind!) that although the mRNA vaccines were the biggest successes, that adenovirus-vectored vaccines from AstraZeneca and Janssen/J&J were also in the race. But those ran into trouble as they were rolled out into larger populations (I wrote about this here at the time). All drugs (and that includes all vaccines) are susceptible to rare side effects that you often have very little chance of seeing during clinical trials. The Phase III trials for these vaccines involved tens of thousands of patients, too few to pick up such a signal. About 3 out of every 100,000 people who got the AZ vaccine in the UK and about 1 out of every 200,000 who got the J&J vaccine here in the US developed a vaccine-induced immune thrombocytopenia and thrombosis (VITT). By April of 2021, 34 million people in Europe had had the AstraZeneca vaccine, and there had been 222 cases of VITT. And although these effects were obviously uncommon, health authorities often advised people to get different vaccines (especially for younger patients, who appeared to have greater risk for this side effect) since there were vaccine alternatives in many countries. These two coronavirus vaccines are really the only ones I know of that have shown a clear VITT problem - it did take everyone by surprise. There are all sorts of other thrombocytopenia syndromes that can be brought on by administration of heparin (for example) and as a complication of a number of other diseases, so the problems were certainly recognizable when they occurred. In this case, the WHO issued guidelines in July of 2021 on avoiding and managing VITT, and these were later refined. It’s also important to note that the use of both the AZ and J&J vaccines were heavily restricted, since there were clear alternatives available. But as a thought experiment, if there had been no other alternatives, it seems clear that worldwide vaccination with these would have continued, with due attention to the patient populations and close monitoring for VITT should it occur. The benefits still would surely have outweighed the risks, although I’m glad we didn’t have to make that tradeoff. But in the US, the AstraZeneca vaccine was never available and the J&J vaccine’s Emergency Use Authorization expired in May 2023 (by which time actual usage of the vaccine was almost nonexistent). The AZ vaccine is also no longer produced - it was removed from authorization around the world by mid-2024. Since this time there has been a great deal of research to try to figure out the details of what happened. That has implications for adenovirus vector vaccines in general, of course, since it was surely not a coincidence that the two vaccines of that type were the ones that set off the alarms. Natural adenovirus infection has also been suspected to lead to rare thrombocytopenia, and that was very unlikely to be a chance finding, too. J&J, for one, had spent years working up an entire platform for a whole range of vaccines of this type, an effort that was accelerated by the advent of the pandemic. And now we have some answers! This new paper goes into detail. We already knew that the problems were (proximately) caused by antibodies to platelet factor 4. But why were those generated? It turns out that in rare cases, patient reacted to the adenovirus and generated antibodies to adenoviral core protein VII - and these antibodies cross-react with PF4! It looks like people with immunoglobin light-chain allele IGVB-21*02 or 21*03 are the ones susceptible to this - there’s a lysine at position 31 of this sequence that is mutated to a glutamic acid in such patients, and this is what sets off the PF4 binding. What makes this especially tricky is that the germline sequences of all the VITT patients still code for that lysine - it’s a somatic mutation that flips it to glutamate, and it probably occurs in a few B cells in the germinal centers of the lymph nodes. The authors do a thorough job of tracking all this down with a great number of control experiments in the various sequences. This is really good to see. It lays out the problems that future adenovirus vector researchers are going to have to deal with (presumably by altering the sequence of that viral pVII protein!) and it also would seem to help explain some other rare blood-clotting effects after various infections or transfusions. One more note: I mentioned above that all drugs and all vaccines are susceptible to these sorts of rare-but-bad effects. This is not an excuse, but it’s also neither an exaggeration nor an attempt to minimize such problems. This is especially true for anything that touches on the immune system, which absolutely the most variable part of human biochemistry and genetics - it’s evolved to be so! You cannot run large enough clinical trials to catch all such effects beforehand, nor can you calculate and simulate your way around them at that stage, either. About 1% of the population truly has allergic sensitivity to beta-lactam drugs like penicillin, for example, and that is by far the largest and most prominent drug-related immune side effect we know of. That you could probably see in the trials. Everything else is way down there in the tiny populations, and believe me, everyone has their fingers crossed when their new drug rolls out into a really large population for the first time. No one likes to dwell on this, but it’s the truth, and it’s also the truth that there is not much we can do about it. That’s life and that’s the biology of life. That’s why I’m very happy when we can illuminate what’s actually going on, as has happened here. The immune system is large and contains multitudes, but here’s another piece of it we can use.
Show full content
Here’s a very interesting rundown on an issue that became apparent during the coronavirus vaccine development period. You may remember (if you haven’t expunged that entire period from your mind!) that although the mRNA vaccines were the biggest successes, that adenovirus-vectored vaccines from AstraZeneca and Janssen/J&J were also in the race.
But those ran into trouble as they were rolled out into larger populations (I wrote about this here at the time). All drugs (and that includes all vaccines) are susceptible to rare side effects that you often have very little chance of seeing during clinical trials. The Phase III trials for these vaccines involved tens of thousands of patients, too few to pick up such a signal. About 3 out of every 100,000 people who got the AZ vaccine in the UK and about 1 out of every 200,000 who got the J&J vaccine here in the US developed a vaccine-induced immune thrombocytopenia and thrombosis (VITT). By April of 2021, 34 million people in Europe had had the AstraZeneca vaccine, and there had been 222 cases of VITT. And although these effects were obviously uncommon, health authorities often advised people to get different vaccines (especially for younger patients, who appeared to have greater risk for this side effect) since there were vaccine alternatives in many countries.
These two coronavirus vaccines are really the only ones I know of that have shown a clear VITT problem - it did take everyone by surprise. There are all sorts of other thrombocytopenia syndromes that can be brought on by administration of heparin (for example) and as a complication of a number of other diseases, so the problems were certainly recognizable when they occurred. In this case, the WHO issued guidelines in July of 2021 on avoiding and managing VITT, and these were later refined.
It’s also important to note that the use of both the AZ and J&J vaccines were heavily restricted, since there were clear alternatives available. But as a thought experiment, if there had been no other alternatives, it seems clear that worldwide vaccination with these would have continued, with due attention to the patient populations and close monitoring for VITT should it occur. The benefits still would surely have outweighed the risks, although I’m glad we didn’t have to make that tradeoff. But in the US, the AstraZeneca vaccine was never available and the J&J vaccine’s Emergency Use Authorization expired in May 2023 (by which time actual usage of the vaccine was almost nonexistent). The AZ vaccine is also no longer produced - it was removed from authorization around the world by mid-2024.
Since this time there has been a great deal of research to try to figure out the details of what happened. That has implications for adenovirus vector vaccines in general, of course, since it was surely not a coincidence that the two vaccines of that type were the ones that set off the alarms. Natural adenovirus infection has also been suspected to lead to rare thrombocytopenia, and that was very unlikely to be a chance finding, too. J&J, for one, had spent years working up an entire platform for a whole range of vaccines of this type, an effort that was accelerated by the advent of the pandemic.
And now we have some answers! This new paper goes into detail. We already knew that the problems were (proximately) caused by antibodies to platelet factor 4. But why were those generated? It turns out that in rare cases, patient reacted to the adenovirus and generated antibodies to adenoviral core protein VII - and these antibodies cross-react with PF4! It looks like people with immunoglobin light-chain allele IGVB-21*02 or 21*03 are the ones susceptible to this - there’s a lysine at position 31 of this sequence that is mutated to a glutamic acid in such patients, and this is what sets off the PF4 binding. What makes this especially tricky is that the germline sequences of all the VITT patients still code for that lysine - it’s a somatic mutation that flips it to glutamate, and it probably occurs in a few B cells in the germinal centers of the lymph nodes. The authors do a thorough job of tracking all this down with a great number of control experiments in the various sequences.
This is really good to see. It lays out the problems that future adenovirus vector researchers are going to have to deal with (presumably by altering the sequence of that viral pVII protein!) and it also would seem to help explain some other rare blood-clotting effects after various infections or transfusions.
One more note: I mentioned above that all drugs and all vaccines are susceptible to these sorts of rare-but-bad effects. This is not an excuse, but it’s also neither an exaggeration nor an attempt to minimize such problems. This is especially true for anything that touches on the immune system, which absolutely the most variable part of human biochemistry and genetics - it’s evolved to be so!
You cannot run large enough clinical trials to catch all such effects beforehand, nor can you calculate and simulate your way around them at that stage, either. About 1% of the population truly has allergic sensitivity to beta-lactam drugs like penicillin, for example, and that is by far the largest and most prominent drug-related immune side effect we know of. That you could probably see in the trials. Everything else is way down there in the tiny populations, and believe me, everyone has their fingers crossed when their new drug rolls out into a really large population for the first time. No one likes to dwell on this, but it’s the truth, and it’s also the truth that there is not much we can do about it. That’s life and that’s the biology of life. That’s why I’m very happy when we can illuminate what’s actually going on, as has happened here. The immune system is large and contains multitudes, but here’s another piece of it we can use.
After having just written last week about the FDA’s refusal to consider Moderna’s mRNA vaccine application, the agency has apparently reversed course and told the company that they will accept it. According to Stat, the agency will review the vaccine in 50-64-year-olds via the regular pathway and over-65s via accelerated approval with a required post-marketing study if approved, and has set a deadline of August 5. This is good news right on the face of it, and you have to take your good news where you can get it with the current iteration of the federal health agencies. It’s like the other day when Dr. Oz popped up and actually said it might be a good idea for people to get the measles vaccine. I was very glad to see that, too, but my happiness is strongly mitigated by the way that none of this bullshit should be happening at all. In Moderna’s case, Vinay Prasad should never have overruled his advisors in the first place, and the FDA should have accepted the Moderna application as a matter of course. As so many of us familiar with the industry pointed out last week, drug development has long, expensive timelines, and if we’re going to switch to a system where the Big Boss can just randomly decide at the end of the process that he doesn’t like you and he doesn’t like your drug, then things will start breaking down rapidly. No one is going to embark on human trials in order to sound out Vinay Prasad’s ever-changing moods after completion. As Matthew Herper points out here, Prasad seems to just love drama and headlines and sudden surprises, but these things cannot be part of a viable business model. We shouldn’t have an administration that is throwing roadblocks in front of mRNA technology every chance it gets, or one that spends some of its time and effort minimizing the importance of vaccination in general. Believe me, I’m glad that Oz told people to get vaccinated for measles, but look around you: we’re already at the third worst year for the disease in the US since we declared it eliminated in 2000 and we’re only partway through February. Wouldn’t it be better not to have done these things in the first place? Well, we’re way past that now. And I’m not sure that we’re in the clear on the Moderna application either, to be honest. The agency could go through the motions of a full review and turn it down anyway - I mean, from what I’ve seen of the data there would be no good justification for that, but doing some sort of review would at least provide more cover than the “get lost” response from last week. Or they could insist that Moderna run a trial using the agency’s preferred vaccine - or even a placebo, as insane as that sounds medically - because the administration’s public-facing line on vaccines is Careful, Careful, You Just Can’t Be Too Cautious. That’s a load of crap, of course - RFK Jr.’s real line on vaccines is that they’re toxic bioweapons from the evil drug industry - but Careful, Careful will do the job. So as long as the likes of Kennedy, Prasad, Makary et al. are in charge of such things, I wouldn’t regard anything as a done deal. And that gets back to the point that Herper and others are making - that this is no way to run a regulatory agency. We need honesty, we need clarity, and we need a commitment to objective data, and we are simply not going to get any of those things except when someone in the administration feels like play-acting one of them out for the cameras.
Show full content
After having just written last week about the FDA’s refusal to consider Moderna’s mRNA vaccine application, the agency has apparently reversed course and told the company that they will accept it. According to Stat, the agency will review the vaccine in 50-64-year-olds via the regular pathway and over-65s via accelerated approval with a required post-marketing study if approved, and has set a deadline of August 5.
This is good news right on the face of it, and you have to take your good news where you can get it with the current iteration of the federal health agencies. It’s like the other day when Dr. Oz popped up and actually said it might be a good idea for people to get the measles vaccine. I was very glad to see that, too, but my happiness is strongly mitigated by the way that none of this bullshit should be happening at all.
In Moderna’s case, Vinay Prasad should never have overruled his advisors in the first place, and the FDA should have accepted the Moderna application as a matter of course. As so many of us familiar with the industry pointed out last week, drug development has long, expensive timelines, and if we’re going to switch to a system where the Big Boss can just randomly decide at the end of the process that he doesn’t like you and he doesn’t like your drug, then things will start breaking down rapidly. No one is going to embark on human trials in order to sound out Vinay Prasad’s ever-changing moods after completion. As Matthew Herper points out here, Prasad seems to just love drama and headlines and sudden surprises, but these things cannot be part of a viable business model.
We shouldn’t have an administration that is throwing roadblocks in front of mRNA technology every chance it gets, or one that spends some of its time and effort minimizing the importance of vaccination in general. Believe me, I’m glad that Oz told people to get vaccinated for measles, but look around you: we’re already at the third worst year for the disease in the US since we declared it eliminated in 2000 and we’re only partway through February. Wouldn’t it be better not to have done these things in the first place? Well, we’re way past that now.
And I’m not sure that we’re in the clear on the Moderna application either, to be honest. The agency could go through the motions of a full review and turn it down anyway - I mean, from what I’ve seen of the data there would be no good justification for that, but doing some sort of review would at least provide more cover than the “get lost” response from last week. Or they could insist that Moderna run a trial using the agency’s preferred vaccine - or even a placebo, as insane as that sounds medically - because the administration’s public-facing line on vaccines is Careful, Careful, You Just Can’t Be Too Cautious. That’s a load of crap, of course - RFK Jr.’s real line on vaccines is that they’re toxic bioweapons from the evil drug industry - but Careful, Careful will do the job.
So as long as the likes of Kennedy, Prasad, Makary et al. are in charge of such things, I wouldn’t regard anything as a done deal. And that gets back to the point that Herper and others are making - that this is no way to run a regulatory agency. We need honesty, we need clarity, and we need a commitment to objective data, and we are simply not going to get any of those things except when someone in the administration feels like play-acting one of them out for the cameras.
This weekend brought news that the Russian opposition leader Alexei Navalny was poisoned in prison by the compound epibatidine. That is not (to put it delicately) the first thing one would have expected, so I wanted to give a little background on this compound first. It’s a toxin isolated from a frog species found in Ecuador and Peru (and a few of its relatives), and like all poison frogs it is a very festive-looking creature indeed. That is of course a warning to potential predators, as with many brightly colored species around the world, a little evolutionary message to any hungry onlooker that they can afford to be so bright and prominent for a very good reason that you should have had a chance to learn by now. Many such frogs are used by native groups in the New World jungles as arrow-poison sources, although this particular species doesn’t seem to be. It has a simple structure with one rather unusual feature, that 2-chloropyridine group. You do see halogenated natural products, but more often from marine organisms where chlorine and bromine are more easily available. An even weirder-looking related alkaloid with the same group in it (phantasmidine) is also found at lower concentrations in the frogs. Unfortunately, the biosynthesis of these compounds has not yet really been worked out (to my knowledge). It is known, as with most other poison dart frogs, that if you raise them in captivity they do not produce the toxin: there is something in their natural diet or environment that allows for it that is not found under terrarium conditions. Even under jungle conditions, sometimes one population of frogs will have the toxin while another in a different location does not. It is very likely that the frogs do not have the ability to produce the compound on their own, but instead acquire it from their diet of local insects, etc. and then sequester the epibatidine in their skin. This has been documented with both birds and frogs with another such case, batrachotoxin - that one is chemically distinct from epibatidine and is found in a different genus of frogs, but it’s likely a similar underlying story. Not knowing the exact species that produce these compounds has made studying the chemical pathways behind them rather difficult! And as with all such compounds, an immediate question is how the creatures that produce or sequester them manage to avoid poisoning themselves. Edit: here's how they do it, apparently, by co-evolving a mutant form of the receptor. This does not come without a cost, it seems. Another paper reports that this Epibatidine works as a ligand for both the muscarinic and nicotinic receptors - it’s an agonist, substituting for the natural ligand acetylcholine, and in general messing with the cholinergic system is going to lead to some strong effects. If you strongly block such signaling, you have replicated the mode of action of nerve gas, and if you strongly enhance it (as in this case) you can get a range of effects including analgesia and muscle paralysis. That latter one is especially unwelcome in the respiratory and cardiovascular system, clearly, and there is no antidote. The compound’s pharmacologic window between interesting pain relief qualities and seizures-n’-death is unfortunately quite narrow. People have tried to widen it, most notably Abbott (AbbVie) in the 1990s. They did a lot of work in this area looking for a nonopioid pain compound and took a chemical cousin of epibatidine (ABT-594, tebanicline) into human trials. They had definitely gotten rid of the “death” side effect by that point, as the FDA tends to insist on, but the window between analgesia and the remaining side effects was still too small. These included nausea, vomiting, impaired coordination, and apparently rather weird dreams as well. People were dropping out of the treatment group in the Phase II with alarming frequency, and the compound was abandoned. There are still a number of possible opportunities in the selective-nicotinergic-agonist area, but realizing selective cholinergic agonists is a problem that stretches back many decades and no general solutions have been found. OK, back to the present day. The presence of the compound in Navalny’s body seems to be beyond dispute. He died two years ago in a “special regime” prison in Siberia, and his body was returned to his mother. Numerous toxicological examinations have confirmed the epibatidine, which does not undergo much metabolism in the human body. That along with its unusual structure make it very easy to identify. I am in agreement with those who believe that this was a deliberate choice by Vladimir Putin’s regime. After all, they had tried to kill Navalny in 2020 with what was obviously a Russian-manufactured nerve agent, and that was after previous chemical attacks in 2017 and 2019. The use of a tropical frog poison in Siberia is to me a grim joke and a statement that this was obviously an unnatural death that was carried out by people with obvious knowledge of human poisons. You don’t need the frogs: epibatidine itself is not that hard to synthesize in the lab by a variety of published routes. It can be made in quantity by any competent organic chemist who knows enough to take the proper precautions, and Russia as a country has a great many skilled organic chemists. The Russian military and security services have been experts in poisoning people with exotic materials for a long, long time. They know exactly what they are doing from a chemical point of view, even if some of their assassins have not been particularly competent or well-informed themselves. Some have speculated that the authorities wanted to try out the epibatidine route to see how well it worked, but let’s be realistic: they could have done that on all sorts of other Siberian prison inmates without anyone ever hearing about it. I don’t think that the Russian state services have many review-board problems when it comes to running human trials. No, this was murder, obvious murder, and it was set up to be an obvious murder. Vladimir Putin is a corrupt, lawless poisoner, and he has had no qualms about demonstrating this over and over. He’ll order it done again the next time the opportunity presents itself.
Show full content
This weekend brought news that the Russian opposition leader Alexei Navalny was poisoned in prison by the compound epibatidine. That is not (to put it delicately) the first thing one would have expected, so I wanted to give a little background on this compound first.
It’s a toxin isolated from a frog species found in Ecuador and Peru (and a few of its relatives), and like all poison frogs it is a very festive-looking creature indeed. That is of course a warning to potential predators, as with many brightly colored species around the world, a little evolutionary message to any hungry onlooker that they can afford to be so bright and prominent for a very good reason that you should have had a chance to learn by now. Many such frogs are used by native groups in the New World jungles as arrow-poison sources, although this particular species doesn’t seem to be.
It has a simple structure with one rather unusual feature, that 2-chloropyridine group. You do see halogenated natural products, but more often from marine organisms where chlorine and bromine are more easily available. An even weirder-looking related alkaloid with the same group in it (phantasmidine) is also found at lower concentrations in the frogs. Unfortunately, the biosynthesis of these compounds has not yet really been worked out (to my knowledge). It is known, as with most other poison dart frogs, that if you raise them in captivity they do not produce the toxin: there is something in their natural diet or environment that allows for it that is not found under terrarium conditions. Even under jungle conditions, sometimes one population of frogs will have the toxin while another in a different location does not.
It is very likely that the frogs do not have the ability to produce the compound on their own, but instead acquire it from their diet of local insects, etc. and then sequester the epibatidine in their skin. This has been documented with both birds and frogs with another such case, batrachotoxin - that one is chemically distinct from epibatidine and is found in a different genus of frogs, but it’s likely a similar underlying story. Not knowing the exact species that produce these compounds has made studying the chemical pathways behind them rather difficult!
And as with all such compounds, an immediate question is how the creatures that produce or sequester them manage to avoid poisoning themselves. Edit: here's how they do it, apparently, by co-evolving a mutant form of the receptor. This does not come without a cost, it seems. Another paper reports that this Epibatidine works as a ligand for both the muscarinic and nicotinic receptors - it’s an agonist, substituting for the natural ligand acetylcholine, and in general messing with the cholinergic system is going to lead to some strong effects. If you strongly block such signaling, you have replicated the mode of action of nerve gas, and if you strongly enhance it (as in this case) you can get a range of effects including analgesia and muscle paralysis. That latter one is especially unwelcome in the respiratory and cardiovascular system, clearly, and there is no antidote. The compound’s pharmacologic window between interesting pain relief qualities and seizures-n’-death is unfortunately quite narrow.
People have tried to widen it, most notably Abbott (AbbVie) in the 1990s. They did a lot of work in this area looking for a nonopioid pain compound and took a chemical cousin of epibatidine (ABT-594, tebanicline) into human trials. They had definitely gotten rid of the “death” side effect by that point, as the FDA tends to insist on, but the window between analgesia and the remaining side effects was still too small. These included nausea, vomiting, impaired coordination, and apparently rather weird dreams as well. People were dropping out of the treatment group in the Phase II with alarming frequency, and the compound was abandoned. There are still a number of possible opportunities in the selective-nicotinergic-agonist area, but realizing selective cholinergic agonists is a problem that stretches back many decades and no general solutions have been found.
OK, back to the present day. The presence of the compound in Navalny’s body seems to be beyond dispute. He died two years ago in a “special regime” prison in Siberia, and his body was returned to his mother. Numerous toxicological examinations have confirmed the epibatidine, which does not undergo much metabolism in the human body. That along with its unusual structure make it very easy to identify. I am in agreement with those who believe that this was a deliberate choice by Vladimir Putin’s regime. After all, they had tried to kill Navalny in 2020 with what was obviously a Russian-manufactured nerve agent, and that was after previous chemical attacks in 2017 and 2019. The use of a tropical frog poison in Siberia is to me a grim joke and a statement that this was obviously an unnatural death that was carried out by people with obvious knowledge of human poisons. You don’t need the frogs: epibatidine itself is not that hard to synthesize in the lab by a variety of published routes. It can be made in quantity by any competent organic chemist who knows enough to take the proper precautions, and Russia as a country has a great many skilled organic chemists.
The Russian military and security services have been experts in poisoning people with exotic materials for a long, long time. They know exactly what they are doing from a chemical point of view, even if some of their assassins have not been particularly competent or well-informed themselves. Some have speculated that the authorities wanted to try out the epibatidine route to see how well it worked, but let’s be realistic: they could have done that on all sorts of other Siberian prison inmates without anyone ever hearing about it. I don’t think that the Russian state services have many review-board problems when it comes to running human trials.
No, this was murder, obvious murder, and it was set up to be an obvious murder. Vladimir Putin is a corrupt, lawless poisoner, and he has had no qualms about demonstrating this over and over. He’ll order it done again the next time the opportunity presents itself.
Edit: as of February 19, this paper has picked up (rather rapidly) an Editorial Expression of Concern. ". . .concerns have been raised regarding inconsistencies between the registration record of this trial on clinicaltrials.gov and (the) published version (of) the study protocol, as well as with some of the findings in this study" That doesn't sound good so far. The editors say that they are investigating and will report with more details, and I'll add those here as they appear. For now, grain-of-salt treatment (at least) seems appropriate, and I'm not happy to report that. Here’s an oddity that I’m glad was put to the test of a controlled trial. It seems that retrospective studies on cancer immunotherapy patients had suggested that there might be an advantage to giving the infusions earlier in the day, so this team took 210 non-small-cell lung cancer patients and divided them into two groups. One got the infustion early in the day (from 7:30 AM to 2 PM), and the other later (from 3 PM to 8 PM). Some of the patients were taking sintilimab and others pembrolizumab. The results are surprisingly strong: the early-infusion group had a median progression-free survival (PFS) of 11.3 months, while the late-infusion group’s median PFS was 5.7 months. Overall survival was median 28 months and 16.8 months for the two groups, and both results were (as they sound!) highly statistically significant. I’d be willing to bet that even the organizers of this trial weren’t expecting readouts this definitive. Subgroup analysis showed that this held for people taking either of the immunotherapy drugs mentioned. So what’s going on here? Some circadian-rhythm explanation seems inevitable here. The team found that the early-infusion group had increased levels of T cells (and increased levels of activated ones, on top of that), but the connection is between that and diurnal timing is still unclear. That is going to be a very interesting thing to figure out. I am certainly willing to believe nearly anything about the immune system at this point! The first step will be to replicate this effect, of course, and it’s an easy enough intervention that I hope that we see this happening soon.
Show full content
Edit: as of February 19, this paper has picked up (rather rapidly) an Editorial Expression of Concern. ". . .concerns have been raised regarding inconsistencies between the registration record of this trial on clinicaltrials.gov and (the) published version (of) the study protocol, as well as with some of the findings in this study" That doesn't sound good so far. The editors say that they are investigating and will report with more details, and I'll add those here as they appear. For now, grain-of-salt treatment (at least) seems appropriate, and I'm not happy to report that.
Here’s an oddity that I’m glad was put to the test of a controlled trial. It seems that retrospective studies on cancer immunotherapy patients had suggested that there might be an advantage to giving the infusions earlier in the day, so this team took 210 non-small-cell lung cancer patients and divided them into two groups. One got the infustion early in the day (from 7:30 AM to 2 PM), and the other later (from 3 PM to 8 PM). Some of the patients were taking sintilimab and others pembrolizumab.
The results are surprisingly strong: the early-infusion group had a median progression-free survival (PFS) of 11.3 months, while the late-infusion group’s median PFS was 5.7 months. Overall survival was median 28 months and 16.8 months for the two groups, and both results were (as they sound!) highly statistically significant. I’d be willing to bet that even the organizers of this trial weren’t expecting readouts this definitive. Subgroup analysis showed that this held for people taking either of the immunotherapy drugs mentioned.
So what’s going on here? Some circadian-rhythm explanation seems inevitable here. The team found that the early-infusion group had increased levels of T cells (and increased levels of activated ones, on top of that), but the connection is between that and diurnal timing is still unclear. That is going to be a very interesting thing to figure out. I am certainly willing to believe nearly anything about the immune system at this point! The first step will be to replicate this effect, of course, and it’s an easy enough intervention that I hope that we see this happening soon.
I’ve written before a time or two about retroviral DNA, which is something that we’re all carrying around whether we feel like it or not. Over evolutionary time there have been a number of events where human germ cell lines have had these sequences inserted into them by retroviral infection, and we’ve been living with them ever since. Mutations occur, of course, and some of these sequences are rather decayed by now, but it seems that between 5 and 10% of any given person’s genome is retroviral gunk of this sort. And “gunk” is probably how the (relatively few) people who think about these sequences may well think of them. Once in a while there’s a connection found to human disease - I wrote about this in 2015 with a retrovirus called HERV-K and the neurological disease ALS, and that topic is still an active area of research. To give you an idea, the HERV-K group is considered to be one of the more recent retroviral additions to the genome, and it is found in humans, apes, and old-world monkeys in general, but not the new-world species. Which still puts it back quite a ways - it appears to have infected the human line both before and after our phylogenetic split with chimpanzees, for example. Such retroviral sequences, which are normally transcriptionally repressed, can become active in tumor cells (another active area of research) and, most weirdly, during embryogenesis. Those reports have raised the possibility that we humans have adapted over evolutionary time to the presence of these sequences to the point that they are part of embryonic development, and this new paper adds more to that idea. The authors find that transcripts from the retrovirus MLT2A1 (more specifically, chimeric transcripts with some human sequence in them too) seem to be essential towards activation of genomic transcription at the 4-cell and 8-cell stage of embryonic development. That’s mighty early, and obviously mighty important. So the idea that this activation has come to be dependent on a chimeric RNA transcript partially derived from an ancient retroviral infection is really something to think about. This earlier work and this as well had pointed in this direction, and also notes that such transcripts can occur in adult somatic tissues like the pineal gland. MLT2A1 is part of a family (HERV-L) that’s also relatively recent, evolutionarily speaking. . .but not so recent as not to be an underpinning of some of our most fundamental reproductive processes, apparently. It really does make you rethink what it means to be human - or what it means to be any sort of living creature, since you’d have to figure that most complex living creatures share this sort of mixed-up, patched-together character if you just look closely enough. . .
Show full content
I’ve written before a time or two about retroviral DNA, which is something that we’re all carrying around whether we feel like it or not. Over evolutionary time there have been a number of events where human germ cell lines have had these sequences inserted into them by retroviral infection, and we’ve been living with them ever since. Mutations occur, of course, and some of these sequences are rather decayed by now, but it seems that between 5 and 10% of any given person’s genome is retroviral gunk of this sort.
And “gunk” is probably how the (relatively few) people who think about these sequences may well think of them. Once in a while there’s a connection found to human disease - I wrote about this in 2015 with a retrovirus called HERV-K and the neurological disease ALS, and that topic is still an active area of research. To give you an idea, the HERV-K group is considered to be one of the more recent retroviral additions to the genome, and it is found in humans, apes, and old-world monkeys in general, but not the new-world species. Which still puts it back quite a ways - it appears to have infected the human line both before and after our phylogenetic split with chimpanzees, for example.
Such retroviral sequences, which are normally transcriptionally repressed, can become active in tumor cells (another active area of research) and, most weirdly, during embryogenesis. Those reports have raised the possibility that we humans have adapted over evolutionary time to the presence of these sequences to the point that they are part of embryonic development, and this new paper adds more to that idea. The authors find that transcripts from the retrovirus MLT2A1 (more specifically, chimeric transcripts with some human sequence in them too) seem to be essential towards activation of genomic transcription at the 4-cell and 8-cell stage of embryonic development.
That’s mighty early, and obviously mighty important. So the idea that this activation has come to be dependent on a chimeric RNA transcript partially derived from an ancient retroviral infection is really something to think about. This earlier work and this as well had pointed in this direction, and also notes that such transcripts can occur in adult somatic tissues like the pineal gland.
MLT2A1 is part of a family (HERV-L) that’s also relatively recent, evolutionarily speaking. . .but not so recent as not to be an underpinning of some of our most fundamental reproductive processes, apparently. It really does make you rethink what it means to be human - or what it means to be any sort of living creature, since you’d have to figure that most complex living creatures share this sort of mixed-up, patched-together character if you just look closely enough. . .
Last night brought news that the FDA has refused to review Moderna’s application for their new mRNA influenza vaccine, and more details have emerged so far today. All of them are infuriating. Right off, let’s just make clear that an outright refusal-to-review rejection like this is quite unusual, since biopharma companies (large and small) typically work with the FDA during their trials to make sure that things are being run in a way that the agency finds acceptable. Why wouldn’t you? Human clinical trials are famously expensive and resource-intensive, so flying your New Drug Application blind into the regulatory approval process would be extremely foolish. There are a lot of reasons for your drug to get rejected, but getting a “we’re not even going to look at this” is really the most brutal. It is especially unusual for a vaccine. If there is a prior example like this with the FDA, I am unaware of it. Moderna seems baffled by the decision in their public statements, and I can’t blame them. They have been developing this vaccine for years now, running the key trial in over 40,000 patients, and regulatory authorities in every other nation where they have submitted their application have accepted it for review without problems. It appears that the agency - well, Vinay Prasad - told the company that they did not run an adequate trial because they did not compare their vaccine candidate to the best available standard of care. Stat reported today that Prasad overruled others at the agency, who were ready to begin the review, although HHS denies that this was the case. Of course they do. But Prasad signed the rejection letter personally, which is also something I’ve never heard of before, so draw your own conclusions. Moderna tested their mRNA candidate against GSK’s Fluarix quadrivalent vaccine in adults 50 and above. You would not think that this would be that controversial, and certainly not grounds for refusal-to-file, and that’s been Moderna’s position. The company has acknowledged that the FDA suggested testing against a different vaccine, and it would be interesting to know which one, since most of the available influenza vaccines are quadrivalent inactivated-virus vaccines of rather the same class. (My guess is probably Flublok, a recombinant subunit vaccine that does seem to have an advantage in 50+ patients). But at the same time, the agency appears to have signed off on the trial design as proposed, and I can’t see Moderna going ahead with it if the agency had done otherwise. No, this is very much an outlier. And let’s get real here: this application is being denied, personally by Vinay Prasad and against the recommendation of the FDA’s remaining experts, because he and the rest of the Trump administration are hostile to vaccines in general and to mRNA technology in particular. I don’t see how anyone can look at the statements and actions of the political appointees (from RFK Jr. on down) and come away with any other impression. We are deliberately walking away from the most advanced form of one of the most effective public health measures available to the human race, and instead we are investigated older technologies that happen to involve the administration’s friends. Meanwhile, mRNA therapies are under investigation - in more advanced parts of the world - for far more than vaccines, including various types of cancer. But we, on the other hand, seem to be plowing money into ivermectin (of all things) for that purpose. Like so many of the Trump administration’s actions, this is simultaneously weird, dangerous, and profoundly stupid. And we are all going to pay the price for it.
Show full content
Last night brought news that the FDA has refused to review Moderna’s application for their new mRNA influenza vaccine, and more details have emerged so far today. All of them are infuriating.
Right off, let’s just make clear that an outright refusal-to-review rejection like this is quite unusual, since biopharma companies (large and small) typically work with the FDA during their trials to make sure that things are being run in a way that the agency finds acceptable. Why wouldn’t you? Human clinical trials are famously expensive and resource-intensive, so flying your New Drug Application blind into the regulatory approval process would be extremely foolish. There are a lot of reasons for your drug to get rejected, but getting a “we’re not even going to look at this” is really the most brutal.
It is especially unusual for a vaccine. If there is a prior example like this with the FDA, I am unaware of it. Moderna seems baffled by the decision in their public statements, and I can’t blame them. They have been developing this vaccine for years now, running the key trial in over 40,000 patients, and regulatory authorities in every other nation where they have submitted their application have accepted it for review without problems.
It appears that the agency - well, Vinay Prasad - told the company that they did not run an adequate trial because they did not compare their vaccine candidate to the best available standard of care. Stat reported today that Prasad overruled others at the agency, who were ready to begin the review, although HHS denies that this was the case. Of course they do. But Prasad signed the rejection letter personally, which is also something I’ve never heard of before, so draw your own conclusions.
Moderna tested their mRNA candidate against GSK’s Fluarix quadrivalent vaccine in adults 50 and above. You would not think that this would be that controversial, and certainly not grounds for refusal-to-file, and that’s been Moderna’s position. The company has acknowledged that the FDA suggested testing against a different vaccine, and it would be interesting to know which one, since most of the available influenza vaccines are quadrivalent inactivated-virus vaccines of rather the same class. (My guess is probably Flublok, a recombinant subunit vaccine that does seem to have an advantage in 50+ patients). But at the same time, the agency appears to have signed off on the trial design as proposed, and I can’t see Moderna going ahead with it if the agency had done otherwise. No, this is very much an outlier.
And let’s get real here: this application is being denied, personally by Vinay Prasad and against the recommendation of the FDA’s remaining experts, because he and the rest of the Trump administration are hostile to vaccines in general and to mRNA technology in particular. I don’t see how anyone can look at the statements and actions of the political appointees (from RFK Jr. on down) and come away with any other impression. We are deliberately walking away from the most advanced form of one of the most effective public health measures available to the human race, and instead we are investigated older technologies that happen to involve the administration’s friends. Meanwhile, mRNA therapies are under investigation - in more advanced parts of the world - for far more than vaccines, including various types of cancer. But we, on the other hand, seem to be plowing money into ivermectin (of all things) for that purpose.
Like so many of the Trump administration’s actions, this is simultaneously weird, dangerous, and profoundly stupid. And we are all going to pay the price for it.
The list of weird ideas for using bifunctional molecules is nowhere near reaching its end, and this paper is another example of that. There have been scattered reports over the years of small molecules that inactivate particular proteins by causing them to assemble into inappropriate multimeric forms, which are inactive in themselves and/or degraded by cells once formed. The authors here are deliberately aiming at that effect. To do that they target proteins that are known to self-assemble into homodimers (and there are plenty of those out there), and they make compounds that have two identical ligands tethered together with a linker. The idea is that you experiment with these linkers to favor connecting homodimers with each other (rather than turning around and linking on a single homodimer in two places), and that can lead you to a rather large polymeric species (see the illustration). These new assemblies have the same options as mentioned above: either they continue as active species (which is unlikely, considering the rather large change they’ve undergone), or they lose their activity, and/or the new polymeric species gets targeted for degradation by the cellular machinery as some sort of inappropriate aggregate. Proteins have a tendency to clump up, some very much more than others, and there are indeed cellular pathways that monitor for this and try to clear these heaps out through proteasomal degradation, autophagy pathways, or what have you. The authors demonstrate this by making “polymerization-inducing chimera” (PINCH) compounds targeting proteins like Bcl6 and Keap1. As with other bifunctionals, “linkerology” is a real factor here. For the Bcl6 ones, they used a known covalent modifier of the target and found that they needed at least 11 PEG units between the two identical ligands to be effective. But the resulting PINCH species seemed (by imaging studies) to form aggregate clumps of Bcl6 protein, apparently uncleared, and the resulting phenotype was different from what you get with just simple inhibition. For Keap1, they did the same trick with the reversible-covalent compound bardoloxone, and again the linker was extremely important. Two PEG groups worth was fine, but one PEG unit or three instead? No activity. Trying some non-PEG linkers led to a similar mix of “works fine” and “does nothing”, and the most you can say so far is that you’d better be ready to try several! The resulting inactivation of Keap1 also seems to be through aggregate formation, and the effects are notably longer-lasting than plain inhibition. The fate of these aggregates changes depending on what cell line you try this in, interestingly. Some of them just have the Keap1 polymers just sitting around, while others clear them, and the different sensitivities of the different lines to PINCH treatment may well be correlated with this effect. So far, it looks like the polymers themselves are not toxic, and that it’s just their loss of function that is driving the downstream effects. So this is the sort of thing that can keep people busy for a while, what with the variation in linker behavior, cells lines, and so on. Imagine how many different results you’ll see in an actual animal, when you add pharmacokinetics (absorption, distribution, clearance) into the mix! But there could be some really interesting compounds in the space for those with the patience to track them down. . .
Show full content
The list of weird ideas for using bifunctional molecules is nowhere near reaching its end, and this paper is another example of that. There have been scattered reports over the years of small molecules that inactivate particular proteins by causing them to assemble into inappropriate multimeric forms, which are inactive in themselves and/or degraded by cells once formed. The authors here are deliberately aiming at that effect.
To do that they target proteins that are known to self-assemble into homodimers (and there are plenty of those out there), and they make compounds that have two identical ligands tethered together with a linker. The idea is that you experiment with these linkers to favor connecting homodimers with each other (rather than turning around and linking on a single homodimer in two places), and that can lead you to a rather large polymeric species (see the illustration). These new assemblies have the same options as mentioned above: either they continue as active species (which is unlikely, considering the rather large change they’ve undergone), or they lose their activity, and/or the new polymeric species gets targeted for degradation by the cellular machinery as some sort of inappropriate aggregate. Proteins have a tendency to clump up, some very much more than others, and there are indeed cellular pathways that monitor for this and try to clear these heaps out through proteasomal degradation, autophagy pathways, or what have you.
The authors demonstrate this by making “polymerization-inducing chimera” (PINCH) compounds targeting proteins like Bcl6 and Keap1. As with other bifunctionals, “linkerology” is a real factor here. For the Bcl6 ones, they used a known covalent modifier of the target and found that they needed at least 11 PEG units between the two identical ligands to be effective. But the resulting PINCH species seemed (by imaging studies) to form aggregate clumps of Bcl6 protein, apparently uncleared, and the resulting phenotype was different from what you get with just simple inhibition.
For Keap1, they did the same trick with the reversible-covalent compound bardoloxone, and again the linker was extremely important. Two PEG groups worth was fine, but one PEG unit or three instead? No activity. Trying some non-PEG linkers led to a similar mix of “works fine” and “does nothing”, and the most you can say so far is that you’d better be ready to try several! The resulting inactivation of Keap1 also seems to be through aggregate formation, and the effects are notably longer-lasting than plain inhibition.
The fate of these aggregates changes depending on what cell line you try this in, interestingly. Some of them just have the Keap1 polymers just sitting around, while others clear them, and the different sensitivities of the different lines to PINCH treatment may well be correlated with this effect. So far, it looks like the polymers themselves are not toxic, and that it’s just their loss of function that is driving the downstream effects.
So this is the sort of thing that can keep people busy for a while, what with the variation in linker behavior, cells lines, and so on. Imagine how many different results you’ll see in an actual animal, when you add pharmacokinetics (absorption, distribution, clearance) into the mix! But there could be some really interesting compounds in the space for those with the patience to track them down. . .
This is a pretty weird idea, but it seems to work. I’ve written many times about aspect of crystallography, but there’s one great big overarching concern in that field that you have to get past: can you even get crystals of your desired compound at all? Advances in x-ray crystallography and electron diffraction have helped to push the size and availability of useful crystals further and further, and we can now get experimental structures from samples that once would have been considered not even worth looking at. But even very tiny crystals are still crystals: what if you can’t even get that far? That’s what this new paper is addressing, and it fits into a long tradition in the field. That is, “if you can’t get a crystal of your desired molecule, make a derivative of it where you can”. All sorts of things have been used for this trick - you might be able to make a crystalline salt of an amine compound, for example, or you might esterify a carboxylic acid with some chunky, likely-to-pack-well side chain. On the other side of that transformation, you could go after hydroxyl groups on your target molecule and esterify them to see if something will start growing. But as you can tell even from that (very incomplete!) list, this is something of an art form. You can never really be sure which of these changes might do the trick, and you should never forget that each of the new compounds you’ve produced can be examined in all sorts of solvents and conditions when it comes time to crystallize them. This sort of thing can go on for weeks, months. Years. The paper linked above is suggesting what might be a more general method. Over the years, it’s been found that metal complexes consisting of three metal atoms arranged with three substituted pyrazole molecules into a planar triangular ring seem to have a strong affinity for complexation with a variety of organic molecules. Many crystalline species can be produced this way, as you can see here. The species with three silver atoms and three bis(trifluoromethyl)pyrazole molecules seems to be especially useful, as shown here. Even so, there are plenty of species out there that don’t quite come through with this method. So the authors have found that plain ol’ acetyl groups are very good for attracting the metal complex, and show that acetylating recalcitrant molecules turns them into good partners for forming crystalline species. This is demonstrated with a whole range of alcohols, polyols, phenols and polyphenols, and amine species. Some of these are just flat-out liquids at ambient conditions, while others are gummy, gunky, sorta-solids as they stand. A few are actually too volatile for anyone to have successfully crystallized them as well. But they seem to form quite useful crystal species under these conditions, with the acetylations done in situ without isolation, and I’m sure those heavy silver atoms help out quite a bit with the X-ray data. It is rather weird to see orderly crystal structures for things like cyclopentadecanol! Looking through the supplementary data, I see that the R factors for most of these are around 0.06, give or take, which is not sparkling clean but not horrible. Some of them are down in the 0.03 range, while others are more up around 0.10, which is rather looser. Those latter structures would qualify as “good enough for synthetic organic chemists” but I very much suspect they would not please the real crystallographers out there. But for Cro-Magnons like me (and my tribe), a fuzzy structure usually beats no structure at all, so we’ll take what we can get!
Show full content
This is a pretty weird idea, but it seems to work. I’ve written many times about aspect of crystallography, but there’s one great big overarching concern in that field that you have to get past: can you even get crystals of your desired compound at all?
Advances in x-ray crystallography and electron diffraction have helped to push the size and availability of useful crystals further and further, and we can now get experimental structures from samples that once would have been considered not even worth looking at. But even very tiny crystals are still crystals: what if you can’t even get that far? That’s what this new paper is addressing, and it fits into a long tradition in the field.
That is, “if you can’t get a crystal of your desired molecule, make a derivative of it where you can”. All sorts of things have been used for this trick - you might be able to make a crystalline salt of an amine compound, for example, or you might esterify a carboxylic acid with some chunky, likely-to-pack-well side chain. On the other side of that transformation, you could go after hydroxyl groups on your target molecule and esterify them to see if something will start growing. But as you can tell even from that (very incomplete!) list, this is something of an art form. You can never really be sure which of these changes might do the trick, and you should never forget that each of the new compounds you’ve produced can be examined in all sorts of solvents and conditions when it comes time to crystallize them. This sort of thing can go on for weeks, months. Years.
The paper linked above is suggesting what might be a more general method. Over the years, it’s been found that metal complexes consisting of three metal atoms arranged with three substituted pyrazole molecules into a planar triangular ring seem to have a strong affinity for complexation with a variety of organic molecules. Many crystalline species can be produced this way, as you can see here. The species with three silver atoms and three bis(trifluoromethyl)pyrazole molecules seems to be especially useful, as shown here. Even so, there are plenty of species out there that don’t quite come through with this method.
So the authors have found that plain ol’ acetyl groups are very good for attracting the metal complex, and show that acetylating recalcitrant molecules turns them into good partners for forming crystalline species. This is demonstrated with a whole range of alcohols, polyols, phenols and polyphenols, and amine species. Some of these are just flat-out liquids at ambient conditions, while others are gummy, gunky, sorta-solids as they stand. A few are actually too volatile for anyone to have successfully crystallized them as well. But they seem to form quite useful crystal species under these conditions, with the acetylations done in situ without isolation, and I’m sure those heavy silver atoms help out quite a bit with the X-ray data. It is rather weird to see orderly crystal structures for things like cyclopentadecanol!
Looking through the supplementary data, I see that the R factors for most of these are around 0.06, give or take, which is not sparkling clean but not horrible. Some of them are down in the 0.03 range, while others are more up around 0.10, which is rather looser. Those latter structures would qualify as “good enough for synthetic organic chemists” but I very much suspect they would not please the real crystallographers out there. But for Cro-Magnons like me (and my tribe), a fuzzy structure usually beats no structure at all, so we’ll take what we can get!
Here, thanks to Milkshake over at Org Prep Daily, is an example of what the scientific literature is slowly turning into under the onslaught of chatbots. The Royal Society of Chemistry journal Sustainable Energy and Fuels has a paper in it that's been up for a while (published in August of 2024) with the unremarkable title of "Critical insights into eutectic molten hydroxide electrolysis for sustainable green hydrogen production". It's not a topic that I pay that much attention to, and it's not a journal that I think I've ever read a paper from anyway. So why am I noting it here? Well, try to read the thing. Just try. It starts off sounding pedestrian but sane, but that doesn't last: Embarking on a journey at the intersection of innovation and sustainability, this research review delves into the realm of hydrogen gas production through a lens of unprecedented possibilities. Driven by concerns over environmental impact and the ever-increasing demand for clean energy, the focus shifts towards the electrochemical process of splitting steam for hydrogen production via eutectic molten hydroxide electrolysis. This exploration is not merely a scientific pursuit; it is a quest to redefine our energy landscape. Imagine a novel reference electrode, a stable companion crafted from the fusion of Ni/Ni(OH)2 and an ionic membrane. . .Through meticulous exploration and theoretical contemplation, this review sets out to redefine the boundaries of hydrogen gas production, laying the groundwork for a sustainable energy future. This review transcends the ordinary, unlocking the secrets that propel us toward a cleaner, brighter tomorrow. Helloooo, chatbot. The whole damn thing is written like this, and the adjective-laden prose really starts to grate in the context of a supposed scientific article. But who am I kidding? This crap would grate in the context of an ad for a used-car lot. Everything is novel and exciting and unique and transformative and unusual and important, and the preferred chatbot verbs (like "delve") get a real workout along the way, too. The whole thing ends up sounding like the grandiose fantasies of someone who has either taken a not-very-entertaining recreational drug or has had a (hopefully temporary) injury to their centers of speech. Maybe both. Some sections flutter back down into reality, but then you get things like this: In the captivating domain of electrochemical exploration, the platinum electrode assumes the spotlight. A meticulous cyclic voltammetry analysis at 550 °C, immersed in molten NaOH, unveils the nuanced interplay of redox peaks, symbolic of the reduction of a delicate oxide film enveloping the platinum wire's surface.135Fig. 3(D) presents the cyclic voltammograms from Ge et al.'s study135 employing platinum as the working electrode. Each peak narrates a unique story: the cathodic current peak C1 signifies the poetic reduction of the oxide film; the captivating surge in cathodic current at C2 (−0.4 V) unfolds a ballet of hydrogen gas evolution; the anodic current peak O1 depicts the stoic oxidation of the oxide film, and the enchanting O2 serves as a crescendo harmonizing with the birth of oxygen gas. The saga continues beyond platinum, venturing into the realm of noble nickel. Its cyclic voltammetry narrative in molten NaOH reveals a tapestry of redox peaks akin to its platinum counterpart. The cathodic sonnet at C3 serenades the reduction of a wispy oxide film caressing the nickel surface. The authors, if that's the word we're looking for, are from eight different locations ranging from Penn State through Nottingham, Morocco, Moscow, Sarajevo, Islamabad and more. But there is nowhere on earth where people talk (or write) like this, or at least there shouldn't be. The authors never should have let this manuscript go out in this form, and needless to say, neither should the editors. If that's the word we're looking for. Come on, people: if you're going to use the automatic word-rearrangers, at least try to cover your tracks a tiny bit. This is what we have to look forward to? Journals filling up with this bilge, this useless wordy debris? Gosh, that's going to work out great with all those machine-learning algorithms for collating human knowledge - just wait until we pour these buckets of sludge into them. That'll produce all kinds of captivating insights for us to delve into. Oh, God.
Show full content
Here, thanks to Milkshake over at Org Prep Daily, is an example of what the scientific literature is slowly turning into under the onslaught of chatbots. The Royal Society of Chemistry journal Sustainable Energy and Fuelshas a paper in it that's been up for a while (published in August of 2024) with the unremarkable title of "Critical insights into eutectic molten hydroxide electrolysis for sustainable green hydrogen production". It's not a topic that I pay that much attention to, and it's not a journal that I think I've ever read a paper from anyway. So why am I noting it here?
Well, try to read the thing. Just try. It starts off sounding pedestrian but sane, but that doesn't last:
Embarking on a journey at the intersection of innovation and sustainability, this research review delves into the realm of hydrogen gas production through a lens of unprecedented possibilities. Driven by concerns over environmental impact and the ever-increasing demand for clean energy, the focus shifts towards the electrochemical process of splitting steam for hydrogen production via eutectic molten hydroxide electrolysis. This exploration is not merely a scientific pursuit; it is a quest to redefine our energy landscape. Imagine a novel reference electrode, a stable companion crafted from the fusion of Ni/Ni(OH)2 and an ionic membrane. . .Through meticulous exploration and theoretical contemplation, this review sets out to redefine the boundaries of hydrogen gas production, laying the groundwork for a sustainable energy future. This review transcends the ordinary, unlocking the secrets that propel us toward a cleaner, brighter tomorrow.
Helloooo, chatbot. The whole damn thing is written like this, and the adjective-laden prose really starts to grate in the context of a supposed scientific article. But who am I kidding? This crap would grate in the context of an ad for a used-car lot. Everything is novel and exciting and unique and transformative and unusual and important, and the preferred chatbot verbs (like "delve") get a real workout along the way, too. The whole thing ends up sounding like the grandiose fantasies of someone who has either taken a not-very-entertaining recreational drug or has had a (hopefully temporary) injury to their centers of speech. Maybe both. Some sections flutter back down into reality, but then you get things like this:
In the captivating domain of electrochemical exploration, the platinum electrode assumes the spotlight. A meticulous cyclic voltammetry analysis at 550 °C, immersed in molten NaOH, unveils the nuanced interplay of redox peaks, symbolic of the reduction of a delicate oxide film enveloping the platinum wire's surface.135Fig. 3(D) presents the cyclic voltammograms from Ge et al.'s study135 employing platinum as the working electrode. Each peak narrates a unique story: the cathodic current peak C1 signifies the poetic reduction of the oxide film; the captivating surge in cathodic current at C2 (−0.4 V) unfolds a ballet of hydrogen gas evolution; the anodic current peak O1 depicts the stoic oxidation of the oxide film, and the enchanting O2 serves as a crescendo harmonizing with the birth of oxygen gas. The saga continues beyond platinum, venturing into the realm of noble nickel. Its cyclic voltammetry narrative in molten NaOH reveals a tapestry of redox peaks akin to its platinum counterpart. The cathodic sonnet at C3 serenades the reduction of a wispy oxide film caressing the nickel surface.
The authors, if that's the word we're looking for, are from eight different locations ranging from Penn State through Nottingham, Morocco, Moscow, Sarajevo, Islamabad and more. But there is nowhere on earth where people talk (or write) like this, or at least there shouldn't be. The authors never should have let this manuscript go out in this form, and needless to say, neither should the editors. If that's the word we're looking for. Come on, people: if you're going to use the automatic word-rearrangers, at least try to cover your tracks a tiny bit.
This is what we have to look forward to? Journals filling up with this bilge, this useless wordy debris? Gosh, that's going to work out great with all those machine-learning algorithms for collating human knowledge - just wait until we pour these buckets of sludge into them. That'll produce all kinds of captivating insights for us to delve into. Oh, God.
Here’s one that I hadn’t heard of: there is epidemiological evidence that a medical history of cancer seems to decrease the risk of an Alzheimer’s diagnosis. From the looks of it, this observation has survived several attempts to make it go away. That’s exactly how you should treat such an interesting hypothesis, because it gets much more interesting if you can’t easily dismiss it. Very roughly, a cancer diagnosis appears to lower the risk of later Alzheimer’s disease by around 10% in large cohorts, and it’s certainly possible that there are factors (and particular cohorts) that could make this figure go up or down. But why is there such an effect at all? That’s the big question, and here’s a new paper in Cell that takes a crack at answering it. It has to be said up front that this is another mouse-Alzheimer’s-model study, so it can’t be taken as definitive, but it definitely points to some things that need to be further investigated in human subjects. In such rodent models (which are dependent on engineered overabundance of amyloid proteins), the authors find that a particular protein secreted by peripheral tumor cells, cystatin-C, appears to decrease the amyloid burden. The proposed mechanism is activation of the TREM2 protein by cyc-C, and the connection there is that mutations leading the loss of function in TREM2 have already been shown to be associated with increased Alzheimer’s risk and severity. It appears to be involved in the robustness of the response of microglial cells to the amyloid plaques found in Alzheimer’s tissue, so it’s equally plausible that dysfunction could lead to trouble and that activation could be beneficial. In fact, there is already a whole literature connecting cystatin-C with Alzheimer’s. It was shown in 2007 to decrease amyloid deposition in just such mouse Alzheimer’s models, although those authors did not make the TREM2 connection at the time (their explanation involved direct binding to soluble amyloid, and that keeping it from forming plaques). Another paper published just a week or two before also came to similar conclusions - the authors overexpressed cys-C in the brains of Alzheimer’s-model mice and showed that it modified the risk of the amyloidosis that these animal develop. From what I can see, the latter paper is referenced in this new article, while the former one does not appear to be, which to me is rather a large omission. Lower circulating levels of cys-C have been noted as a possible Alzheimer’s risk factor, but I don’t see that paper referenced here either. To confuse things more, there is at least one study associating higher circulating levels of cys-C with impaired cognition, although this varied greatly by patient group (this one is cited). It also has to be noted that there was recently an attempt to activate TREM2 directly with an antibody as a means of Alzheimer’s therapy, but the trial did not show any benefit. There is definitely room to argue about how much microglial activation took place as a result of treatment, though, so this result can’t be taken as a definitive statement about TREM2 therapy. This antibody and this clinical trial do not appear to be mentioned in the paper under discussion. Biologically, this is a pretty complicated story. TREM2 is indeed expressed by many types of tumor cells, and high levels of it are considered a bad prognostic sign for many cancers. That’s probably because it’s a sign of a high tumor burden to start with, but I’m not completely sure that the causality all runs that way. Meanwhile, on the cys-C side of things, its levels are considered a marker of renal function, higher ones being worse. That is likely because (as a small protein) it’s a good marker of glomerular filtration in general, so it’s just reporting on what’s already going on in the kidney. But there have been proposals that its association with higher mortality rates are greater than this explanation can account for, so there may be other things at work. There was also a paper just last month on the signaling functions of oligomeric cystatin C that suggests that it binds directly to receptors on myeloid cells and actually decreases their activity - in fact, deleting the gene for cystatin C in their mouse models impaired tumor growth while overexpressing it actually sped up cancer progression. No, there’s a lot going on here, and not all of it is pointing in the same direction. So this new paper, while certainly of great interest, lands in a field that has already had a good deal of attention from the Alzheimer’s community, and it also lands in a very complex signaling landscape. There’s going to have to be a lot more work done before we understand the cystatin C/TREM2/amyloid connection that this work has uncovered, but I hope it eventually does lead to some useful therapeutic ideas.
Show full content
Here’s one that I hadn’t heard of: there is epidemiological evidence that a medical history of cancer seems to decrease the risk of an Alzheimer’s diagnosis. From the looks of it, this observation has survived several attempts to make it go away. That’s exactly how you should treat such an interesting hypothesis, because it gets much more interesting if you can’t easily dismiss it. Very roughly, a cancer diagnosis appears to lower the risk of later Alzheimer’s disease by around 10% in large cohorts, and it’s certainly possible that there are factors (and particular cohorts) that could make this figure go up or down.
But why is there such an effect at all? That’s the big question, and here’s a new paper in Cell that takes a crack at answering it. It has to be said up front that this is another mouse-Alzheimer’s-model study, so it can’t be taken as definitive, but it definitely points to some things that need to be further investigated in human subjects.
In such rodent models (which are dependent on engineered overabundance of amyloid proteins), the authors find that a particular protein secreted by peripheral tumor cells, cystatin-C, appears to decrease the amyloid burden. The proposed mechanism is activation of the TREM2 protein by cyc-C, and the connection there is that mutations leading the loss of function in TREM2 have already been shown to be associated with increased Alzheimer’s risk and severity. It appears to be involved in the robustness of the response of microglial cells to the amyloid plaques found in Alzheimer’s tissue, so it’s equally plausible that dysfunction could lead to trouble and that activation could be beneficial.
In fact, there is already a whole literature connecting cystatin-C with Alzheimer’s. It was shown in 2007 to decrease amyloid deposition in just such mouse Alzheimer’s models, although those authors did not make the TREM2 connection at the time (their explanation involved direct binding to soluble amyloid, and that keeping it from forming plaques). Another paper published just a week or two before also came to similar conclusions - the authors overexpressed cys-C in the brains of Alzheimer’s-model mice and showed that it modified the risk of the amyloidosis that these animal develop. From what I can see, the latter paper is referenced in this new article, while the former one does not appear to be, which to me is rather a large omission. Lower circulating levels of cys-C have been noted as a possible Alzheimer’s risk factor, but I don’t see that paper referenced here either. To confuse things more, there is at least one study associating higher circulating levels of cys-C with impaired cognition, although this varied greatly by patient group (this one is cited).
It also has to be noted that there was recently an attempt to activate TREM2 directly with an antibody as a means of Alzheimer’s therapy, but the trial did not show any benefit. There is definitely room to argue about how much microglial activation took place as a result of treatment, though, so this result can’t be taken as a definitive statement about TREM2 therapy. This antibody and this clinical trial do not appear to be mentioned in the paper under discussion.
Biologically, this is a pretty complicated story. TREM2 is indeed expressed by many types of tumor cells, and high levels of it are considered a bad prognostic sign for many cancers. That’s probably because it’s a sign of a high tumor burden to start with, but I’m not completely sure that the causality all runs that way. Meanwhile, on the cys-C side of things, its levels are considered a marker of renal function, higher ones being worse. That is likely because (as a small protein) it’s a good marker of glomerular filtration in general, so it’s just reporting on what’s already going on in the kidney. But there have been proposals that its association with higher mortality rates are greater than this explanation can account for, so there may be other things at work. There was also a paper just last month on the signaling functions of oligomeric cystatin C that suggests that it binds directly to receptors on myeloid cells and actually decreases their activity - in fact, deleting the gene for cystatin C in their mouse models impaired tumor growth while overexpressing it actually sped up cancer progression. No, there’s a lot going on here, and not all of it is pointing in the same direction.
So this new paper, while certainly of great interest, lands in a field that has already had a good deal of attention from the Alzheimer’s community, and it also lands in a very complex signaling landscape. There’s going to have to be a lot more work done before we understand the cystatin C/TREM2/amyloid connection that this work has uncovered, but I hope it eventually does lead to some useful therapeutic ideas.
This is a brief but informative article at Nature Reviews Drug Discovery from folks at Boston Consulting Group looking at the 2025 drug approvals at the FDA. There were 54 such (excluding diagnostic imaging agents), which is consistent with the landscape since 2014 (the average since then has been exactly that!) 2005-2013 average, by contrast, was 30 new drug approvals. In fact, in that post 2014-period, there’s only been one year (2016, with 28 approvals) that wasn’t higher than any year in the 2005-2013 era. We are clearly approving a lot more drugs in this era. But are we making more money from them? Total peak sales (inflation-adjusted) are indeed higher in the post 2014-era, as they certainly should be. But it’s quite possible that we’re looking at third year in a row of declining peak sales, which has not happened over the last 20 years at all. 2022 peaks sales are listed as 114 billion, and 2023 as 94. Then 2024 is 81 billion (which is an upwards revision from the initial estimate of about 60 billion), and 2025 is estimated at 62 billion. Now that one will likely be revised upwards over the next year or two - that’s typically what happens, because we’re (a) not all that great at forecasting sales and (b) motivated not to overestimate them beforehand, preferring to surprise people with better-than-expected figures. But we’ll have to see if 2025 manages to not be that third year of decline. Where is this trend coming from? Almost certainly due to a relative shortage of great big huge blockbuster drugs. As the article notes, only 4 of the approved drugs from 2025 are forecast to hit peak sales of over three billion per year. It’s not that the industry hasn’t had some underperforming years in the modern era (like 2108, with 62 approvals and $59 billion in peak sales, or 2020, with 56 approvals and $49 billion peak sales). But 2022, by contrast, had a relatively low 43 approvals but hit $114 billion in peak sales, and no one’s sure when and if we’ll get back to a figure like that. I’d say that some of this is likely due to the more “orphan-y” small disease focus of drug development at many smaller companies. And you can’t ignore the more-targeted nature of modern drugs, particularly in oncology. This trend has been coming on for a long time, and as our knowledge increases of the different drivers behind what appear to be similar disease in a patient population, we’re likely to see even more of it. Oncology indeed was the most-represented drug category in this year’s list (16 of the 54, but it has to be said that the estimates for those drugs’ share of peak sales does hold its own (30% of the approvals, 36% of estimated peak sales). We’ll revisit these figures in another year or two and see how they hold up. It will be interesting - likely not cheering, but definitely interesting - to see how four years of the current FDA leadership will affect these numbers as well. . .
Show full content
This is a brief but informative article at Nature Reviews Drug Discovery from folks at Boston Consulting Group looking at the 2025 drug approvals at the FDA. There were 54 such (excluding diagnostic imaging agents), which is consistent with the landscape since 2014 (the average since then has been exactly that!) 2005-2013 average, by contrast, was 30 new drug approvals. In fact, in that post 2014-period, there’s only been one year (2016, with 28 approvals) that wasn’t higher than any year in the 2005-2013 era. We are clearly approving a lot more drugs in this era.
But are we making more money from them? Total peak sales (inflation-adjusted) are indeed higher in the post 2014-era, as they certainly should be. But it’s quite possible that we’re looking at third year in a row of declining peak sales, which has not happened over the last 20 years at all.
2022 peaks sales are listed as 114 billion, and 2023 as 94. Then 2024 is 81 billion (which is an upwards revision from the initial estimate of about 60 billion), and 2025 is estimated at 62 billion. Now that one will likely be revised upwards over the next year or two - that’s typically what happens, because we’re (a) not all that great at forecasting sales and (b) motivated not to overestimate them beforehand, preferring to surprise people with better-than-expected figures. But we’ll have to see if 2025 manages to not be that third year of decline.
Where is this trend coming from? Almost certainly due to a relative shortage of great big huge blockbuster drugs. As the article notes, only 4 of the approved drugs from 2025 are forecast to hit peak sales of over three billion per year. It’s not that the industry hasn’t had some underperforming years in the modern era (like 2108, with 62 approvals and $59 billion in peak sales, or 2020, with 56 approvals and $49 billion peak sales). But 2022, by contrast, had a relatively low 43 approvals but hit $114 billion in peak sales, and no one’s sure when and if we’ll get back to a figure like that.
I’d say that some of this is likely due to the more “orphan-y” small disease focus of drug development at many smaller companies. And you can’t ignore the more-targeted nature of modern drugs, particularly in oncology. This trend has been coming on for a long time, and as our knowledge increases of the different drivers behind what appear to be similar disease in a patient population, we’re likely to see even more of it. Oncology indeed was the most-represented drug category in this year’s list (16 of the 54, but it has to be said that the estimates for those drugs’ share of peak sales does hold its own (30% of the approvals, 36% of estimated peak sales).
We’ll revisit these figures in another year or two and see how they hold up. It will be interesting - likely not cheering, but definitely interesting - to see how four years of the current FDA leadership will affect these numbers as well. . .
This new paper is worth examining as the probable state of the art in LLM-based chemical reaction handling and prediction. The authors report a system (MOSAIC, Multiple Optimized Systems for AI-assisted Chemical prediction) that takes a graphical representation of a proposed new reaction and attempts to produce a written synthetic procedure to realize it in a lab. This is done by creating a fingerprint profile of the proposed reaction using RDKit and Morgan representations of the starting materials and the desired product, and the calculated transformation is then binned into one of many reaction classes, which are represented as cells in a Voronoi diagram. One of those cells/centroids, for example, could represent Buchwald-Hartwig couplings onto aryl bromides, another onto aryl triflates, with all sorts of other cells assigned to all sorts of other chemical transformations from the literature, patents and journals, down to levels like “nitro reduction to amine using tin chloride”. Each of these have had their experimental procedures read and retained as fodder for the LLM phase of things, and the Llama-3.1-8B LLM architecture is used to generate 2,498 separate mini-expert-systems corresponding to reaction types. In the end, you would enter a drawing of your proposed reaction, and the system would spit out a text describing an experimental procedure to get this reaction to work, complete with solvents, temperatures, times, stoichiometries, etc., along with a predicted yield. These are of course reassembled from existing human-produced text procedures, in the same way that any LLM blends and remixes the textual data sets it’s been trained on. The key tricks here are the step that takes the drawn reaction and bins it into the correct Voronoi region (those 2,948 different reactions) and then the LLM step that takes the procedures it has for that sort of reaction and attempts to whip up one that might work for you. So let’s get down to what really matters to most of us: how well does it work? The authors tried feeding known reactions into the system and found that in single-shot predictions it gets the correct solvent about 30% of the time and the correct reagents about 22% of the time. That doesn’t sound so good, but to be fair, many times the answers come out as close-and-chemically-plausible. Allowing for such partial matches, you get 52% hits for solvents and 45% for reagents. If you let several of the many expert systems (the top three of them) pick and also count partial matches, which to me is the most generous interpretation I’m willing to lend credence to, you hit 76% for reagents and 55% for solvents. I will say that my reading of the paper doesn’t leave me certain how the top three expert systems are selected each time. At any rate, the system almost always gets something right, which one of those point-of-view results: for a computational system that’s an encouraging sign that you may well be on the right track, but I would not hire a lab assistant if that phrase was in their letter of recommendation. Applying the software to classes of catalytic reactions (Heck, Suzuki, Buchwald-Hartwig, Sonogashira, etc.) seems to have gone fairly well (these would be some where there are extensive experimental procedures available). The model’s predictions are not as good as others that have been specifically trained on these reaction types, but it’s quite good for a generalist approach. The team also put in 52 new molecules that looked plausible but had not been described yet in the literature, and 37 of these turned out to be makeable with the program’s recommendations (35 using the top recommendation, and the two others by going down to a lower-scoring alternative). Unfortunately, the full paper is not yet available with all its supplementary data, and I look forward to examining this list more closely. Articles about the paper have made many comments about how these molecules could represent new directions in pharmaceutical structures, materials, polymers, and so on, but honestly to me that’s just noise. Making new small molecules that aren’t in the literature is not a challenge in itself - it’s the predicting of usable ways of doing it that could mean more. I make previously unknown molecules all the time, via my own predictions of reactions and reaction conditions, and my success rate is reasonably high. What I would very much like to know is how much better (or worse) MOSAIC is at it, and whether it can save me some time along the way to think about other things (see below). That is to say, I would like to see how its predictions compare to what I (or any other experienced chemist) might have predicted based on a quick pass through literature databases. I take the point that the MOSAIC system has to some extent already had those literature passes done for it while building its various LLM modules, so it could in theory save time compared to bespoke searching. But those time savings will disappear quickly if it suggests more unproductive reactions than I can suggest myself! And that brings up the usual thoughts about the purpose of such software (and indeed, hardware) assistance. I’ve referred to this as “redefining grunt work”, by which I mean taking things (in this case) that once were considered at the center of a synthetic chemist’s job and gradually moving them into the category of “necessary work that this machine over here can speed up for you” or even “necessary work that this machine will just do for you while you do something else”. And that means, as I’ve said before, that we chemists have to be alert not only to the encroachment of software onto our sacred turf, but (since that’s likely going to happen anyway) to also be alert on how best to turn that situation to our advantage. We have to be ready to spend our energies on higher-level problems: if we’re not thinking all the time about How To Make These Compounds, we should be use that time to think harder about What Compounds Need to be Made. And on top of that, Why We Should Be Making Them in the First Place. Those are going to rather more difficult for any LLM to help out with!
Show full content
This new paper is worth examining as the probable state of the art in LLM-based chemical reaction handling and prediction. The authors report a system (MOSAIC, Multiple Optimized Systems for AI-assisted Chemical prediction) that takes a graphical representation of a proposed new reaction and attempts to produce a written synthetic procedure to realize it in a lab.
This is done by creating a fingerprint profile of the proposed reaction using RDKit and Morgan representations of the starting materials and the desired product, and the calculated transformation is then binned into one of many reaction classes, which are represented as cells in a Voronoi diagram. One of those cells/centroids, for example, could represent Buchwald-Hartwig couplings onto aryl bromides, another onto aryl triflates, with all sorts of other cells assigned to all sorts of other chemical transformations from the literature, patents and journals, down to levels like “nitro reduction to amine using tin chloride”. Each of these have had their experimental procedures read and retained as fodder for the LLM phase of things, and the Llama-3.1-8B LLM architecture is used to generate 2,498 separate mini-expert-systems corresponding to reaction types.
In the end, you would enter a drawing of your proposed reaction, and the system would spit out a text describing an experimental procedure to get this reaction to work, complete with solvents, temperatures, times, stoichiometries, etc., along with a predicted yield. These are of course reassembled from existing human-produced text procedures, in the same way that any LLM blends and remixes the textual data sets it’s been trained on. The key tricks here are the step that takes the drawn reaction and bins it into the correct Voronoi region (those 2,948 different reactions) and then the LLM step that takes the procedures it has for that sort of reaction and attempts to whip up one that might work for you.
So let’s get down to what really matters to most of us: how well does it work? The authors tried feeding known reactions into the system and found that in single-shot predictions it gets the correct solvent about 30% of the time and the correct reagents about 22% of the time. That doesn’t sound so good, but to be fair, many times the answers come out as close-and-chemically-plausible. Allowing for such partial matches, you get 52% hits for solvents and 45% for reagents. If you let several of the many expert systems (the top three of them) pick and also count partial matches, which to me is the most generous interpretation I’m willing to lend credence to, you hit 76% for reagents and 55% for solvents. I will say that my reading of the paper doesn’t leave me certain how the top three expert systems are selected each time.
At any rate, the system almost always gets something right, which one of those point-of-view results: for a computational system that’s an encouraging sign that you may well be on the right track, but I would not hire a lab assistant if that phrase was in their letter of recommendation.
Applying the software to classes of catalytic reactions (Heck, Suzuki, Buchwald-Hartwig, Sonogashira, etc.) seems to have gone fairly well (these would be some where there are extensive experimental procedures available). The model’s predictions are not as good as others that have been specifically trained on these reaction types, but it’s quite good for a generalist approach. The team also put in 52 new molecules that looked plausible but had not been described yet in the literature, and 37 of these turned out to be makeable with the program’s recommendations (35 using the top recommendation, and the two others by going down to a lower-scoring alternative). Unfortunately, the full paper is not yet available with all its supplementary data, and I look forward to examining this list more closely.
Articles about the paper have made many comments about how these molecules could represent new directions in pharmaceutical structures, materials, polymers, and so on, but honestly to me that’s just noise. Making new small molecules that aren’t in the literature is not a challenge in itself - it’s the predicting of usable ways of doing it that could mean more. I make previously unknown molecules all the time, via my own predictions of reactions and reaction conditions, and my success rate is reasonably high. What I would very much like to know is how much better (or worse) MOSAIC is at it, and whether it can save me some time along the way to think about other things (see below).
That is to say, I would like to see how its predictions compare to what I (or any other experienced chemist) might have predicted based on a quick pass through literature databases. I take the point that the MOSAIC system has to some extent already had those literature passes done for it while building its various LLM modules, so it could in theory save time compared to bespoke searching. But those time savings will disappear quickly if it suggests more unproductive reactions than I can suggest myself!
And that brings up the usual thoughts about the purpose of such software (and indeed, hardware) assistance. I’ve referred to this as “redefining grunt work”, by which I mean taking things (in this case) that once were considered at the center of a synthetic chemist’s job and gradually moving them into the category of “necessary work that this machine over here can speed up for you” or even “necessary work that this machine will just do for you while you do something else”. And that means, as I’ve said before, that we chemists have to be alert not only to the encroachment of software onto our sacred turf, but (since that’s likely going to happen anyway) to also be alert on how best to turn that situation to our advantage. We have to be ready to spend our energies on higher-level problems: if we’re not thinking all the time about How To Make These Compounds, we should be use that time to think harder about What Compounds Need to be Made. And on top of that, Why We Should Be Making Them in the First Place. Those are going to rather more difficult for any LLM to help out with!
Not many people outside of infectious disease specialists may realize it, but the order Mononegavirales is really bad news for human health. Inside that one you can find measles (the fashionable infection of 2026, damn it all), RSV (always with us), mumps, rabies, and even Ebola, which I very much hope does not become a hot item in any year. There are plenty of differences between all these (there are eleven families in this order), but something that the Mononegavirales species have in common is the existence of “viral factories” (VFs). These are concentrated blobs of viral proteins that form in infected cells and serve to crank out the pieces of new viral particles for further infection. They are, in fact, phase-separated condensates (which shows again how useful that physical behavior is across different systems - I wrote about these most recently here). But there’s been a mystery about them, as this paper explains well. It’s generally believed, with good reason, that such condensates can only form when the concentration of the proteins that make it up get over a certain threshold. But when an infection is just starting out, there doesn’t seem to be any way for that to be possible. You’d need viral factory condensates to make that much protein, and you can’t condense to get such VFs unless the protein is already there - or can you? The authors show the way out of this paradox. For RSV, viral factories are formed by the viral nucleoprotein (N), the viral phosphoprotein (P) and also contain the “large” protein (L) and its cofactors, the viral RNA polymerase, and various RNA transcripts. But there are “pre-replication centers” (PRCs) that form before these VFs are able to completely assemble, and these are imaged here for the first time. They are the seeds of the VF condensate formation, what is basically a feed-forward process: protein replication starts at a lower and less efficient level, but these viral proteins are strongly recruited to the PRCs in turn, which makes them even more productive, which makes more protein, and. . .you get the idea! Before long you have the full-fledged viral factories that have been known for some time as a hallmark of RSV-infected cells. This is how the condensates get bootstrapped from low-concentration beginnings. An unexpected result was that when you look at individual RSV particles (virions) themselves, some of them are much more “PRC-competent” than others. Indeed some of the virions are actually pretty terrible at replication, because they don’t have pre-formed PRCs ready to go in them when they infect a cell. It looks very much like an RSV infection in a whole animal is driven by the virions that do have the PRCs assembled for delivery; the others turn more or less into bystanders (although what viral proteins they do produce probably get recruited over to those other strongly-binding PRCs from other virions that have hit the same cell). But there’s a lot of cell-to-cell heterogeneity in an RSV infection, and these results suggest why: some of these cells have been hit by far more PRC-active virions and some of them haven’t. This raises a lot of interesting questions, for sure. What exactly are the factors that make PRCs assemble more in some virions than others? Do the PRCs themselves vary in their ability to nucleate viral factories in turn, and if so, what factors drive those differences? A larger question is evolutionary: you’d think that there would be a selection advantage in having efficient PRC formation and that over time you just wouldn’t see those less efficient virions at all. This makes you wonder if there really is an effective selection mechanism at the genetic level or if there’s some random process that’s mixing things up at a slightly later stage. And moving beyond the Mononegavirales order, there are plenty of other viruses that have to deal with the starting-from-scratch problem when they first infect a cell. Indeed, there are many other kinds that seem to form condensates during their attacks on cells. Do they also do some kind of condensate-seeding trick to get things going? Or will that possibly turn out to be a trick that just the crazily-infectious ones have hit on? And as the authors note, there are certainly also implications for condensate formation in general, as we work out the sequences and interactions that make this feed-forward process work so well. Onward. . .
Show full content
Not many people outside of infectious disease specialists may realize it, but the order Mononegavirales is really bad news for human health. Inside that one you can find measles (the fashionable infection of 2026, damn it all), RSV (always with us), mumps, rabies, and even Ebola, which I very much hope does not become a hot item in any year.
There are plenty of differences between all these (there are eleven families in this order), but something that the Mononegavirales species have in common is the existence of “viral factories” (VFs). These are concentrated blobs of viral proteins that form in infected cells and serve to crank out the pieces of new viral particles for further infection. They are, in fact, phase-separated condensates (which shows again how useful that physical behavior is across different systems - I wrote about these most recently here). But there’s been a mystery about them, as this paper explains well. It’s generally believed, with good reason, that such condensates can only form when the concentration of the proteins that make it up get over a certain threshold. But when an infection is just starting out, there doesn’t seem to be any way for that to be possible. You’d need viral factory condensates to make that much protein, and you can’t condense to get such VFs unless the protein is already there - or can you?
The authors show the way out of this paradox. For RSV, viral factories are formed by the viral nucleoprotein (N), the viral phosphoprotein (P) and also contain the “large” protein (L) and its cofactors, the viral RNA polymerase, and various RNA transcripts. But there are “pre-replication centers” (PRCs) that form before these VFs are able to completely assemble, and these are imaged here for the first time. They are the seeds of the VF condensate formation, what is basically a feed-forward process: protein replication starts at a lower and less efficient level, but these viral proteins are strongly recruited to the PRCs in turn, which makes them even more productive, which makes more protein, and. . .you get the idea! Before long you have the full-fledged viral factories that have been known for some time as a hallmark of RSV-infected cells. This is how the condensates get bootstrapped from low-concentration beginnings.
An unexpected result was that when you look at individual RSV particles (virions) themselves, some of them are much more “PRC-competent” than others. Indeed some of the virions are actually pretty terrible at replication, because they don’t have pre-formed PRCs ready to go in them when they infect a cell. It looks very much like an RSV infection in a whole animal is driven by the virions that do have the PRCs assembled for delivery; the others turn more or less into bystanders (although what viral proteins they do produce probably get recruited over to those other strongly-binding PRCs from other virions that have hit the same cell).
But there’s a lot of cell-to-cell heterogeneity in an RSV infection, and these results suggest why: some of these cells have been hit by far more PRC-active virions and some of them haven’t. This raises a lot of interesting questions, for sure. What exactly are the factors that make PRCs assemble more in some virions than others? Do the PRCs themselves vary in their ability to nucleate viral factories in turn, and if so, what factors drive those differences? A larger question is evolutionary: you’d think that there would be a selection advantage in having efficient PRC formation and that over time you just wouldn’t see those less efficient virions at all. This makes you wonder if there really is an effective selection mechanism at the genetic level or if there’s some random process that’s mixing things up at a slightly later stage.
And moving beyond the Mononegavirales order, there are plenty of other viruses that have to deal with the starting-from-scratch problem when they first infect a cell. Indeed, there are many other kinds that seem to form condensates during their attacks on cells. Do they also do some kind of condensate-seeding trick to get things going? Or will that possibly turn out to be a trick that just the crazily-infectious ones have hit on? And as the authors note, there are certainly also implications for condensate formation in general, as we work out the sequences and interactions that make this feed-forward process work so well. Onward. . .
Here’s a paper that illustrates an important topic in med-chem, one that an awful lot of ink and pixels have been spilled on over the years. When we talk about affinity of a drug to a target, the binding constants that we measure have a lot of thermodynamics packed inside them. Like every other chemical reaction and interaction, the favorable ones show a decrease in overall Gibbs free energy for the system (delta-G), but one should never forget that the equation for that energy change has two terms. You have enthalpy (delta-H), which consists of a lot of the things that we typically think of driving binding interactions (acid-base pairs, hydrogen bonding, pi-electron interactions, and so on), but there’s also that temperature-and-entropy term (T delta-S). Entropy is a bit more slippery concept, but one way to start thinking about it (although not the whole story) is order and disorder. Compare the starting state and the end state of your process: in which one of them are the components more orderly (fixed in their conformations and positions, for example) or disorderly (able to move around more freely)? As the reaction proceeds, how does the total amount of that order and disorder change? “More disorderly” is by itself energetically favored, as a look around your surroundings will generally demonstrate. That tends to hold whether you’re looking at your your chemical reactions, your bookshelf, your laundry, inside your refrigerator, or at the state of your nation’s political system. But totaling up that entropy in a binding event is no small matter. You have to look at the ligand that’s binding, of course, and you’d think that much of the time it’s going to lose entropy as it binds (since it’s snuggling into position in the binding site itself, as opposed to floating around out there in solution). But that “floating around in solution” brings you to consider the water molecules that it’s surrounded by out there. If they’re forming a fairly orderly solvation shell around your ligand, that’s going to be broken up as it moves into the binding site, and you might pick up some favorable increased entropy that way. But then there’s that binding site! What’s the entropic state of the protein target before and after binding - more ordered overall, or not? Remember that distant domains might be changing position, not just the areas around the binding site, and they all have water molecules around them, too. The binding site itself may have some key water molecules involved in its structure, and the changes there can run the whole range of positive or negative entropic effects depending on the situation. There are a lot of different single-water-molecule situations with proteins! It is indeed a pain in the rear, to use a thermodynamic term of the art. In many situations, enthalpic effects and entropic effects seem to be working at cross purposes to each other. This “entropy-enthalpy compensation” is what people have been arguing about for at least the last thirty years, because it sometimes seems like some perverse but inescapable law of nature and sometimes like just an artifact of how we’re viewing the problem. And it does have to be said that the two don’t cancel each other out all the time, or we’d have no way to optimize the binding of our drug candidates at all! The paper linked above is looking at an old tricyclic drug, doxepin, and its (rather strong) binding to the histamine-1 receptor. Like a lot of other simple tricyclics of that general class, it binds to all sorts of other stuff as well, as do its metabolites, making it a messy proposition in vivo. You can see the list at that link. But it has had many years of use as an antihistamine, antipsychotic, anxiolytic, sleeping aid, and so on, although it's largely fading into the past in most of these areas. My first thought when I saw the structure was "I'll bet that stuff can put you on the floor", and I believe that's an accurate statement. You’ll note because of that double bond that there are two isomers, Z and E doxepin (from the good ol’ German “zusammen” and “entgegen” - if you keep digging in organic chemistry you’ll eventually hit a German layer). The Z reproducibly binds better than the E (two- to five-fold better depending on your assay) but they’re both down in the lower nanomolar range. What the present paper finds, on close examination by isothermal calorimetry, is that the Z isomer’s binding is almost entirely enthalpy-driven with only a very small change in the entropy term. The E isomer, though, is notably less enthalpically favorable, but makes up a lot of that with an improved entropy term. And there’s why we keep talking about entropy-enthalpy compensation! Put simply, maybe too simply, the Z isomer has better interactions with the protein itself, but those remove a lot of its conformational flexibility. Meanwhile, the E isomer doesn’t have as strong an enthalpy hand to play, but since it doesn’t lose as much flexibility while binding it doesn’t take the loss-of-entropy hit along the way like the Z isomer had to. So the two of them end up much closer than you otherwise might have guessed. Studies on mutant receptors showed that a particular tyrosine hydroxyl group in the receptor is a big player in these differences. If you mutate that one to a valine, the two isomers bind almost identically, and with almost identical values for their entropy and enthalpy terms, to boot. It’s pointed toward the tricyclic ring of the structure (but isn’t making a hydrogen bond with the oxygen up there, if that’s what you were thinking). Your first guess might also have been something to do with the basic nitrogen down at the other end of the molecule, but that would also have come up short; things don’t seem to differ much down there for the two isomers. Subtle details all the way down! But that’s medicinal chemistry, and that’s just one of the many reasons why it ain’t easy. . .
Show full content
Here’s a paper that illustrates an important topic in med-chem, one that an awful lot of ink and pixels have been spilled on over the years. When we talk about affinity of a drug to a target, the binding constants that we measure have a lot of thermodynamics packed inside them. Like every other chemical reaction and interaction, the favorable ones show a decrease in overall Gibbs free energy for the system (delta-G), but one should never forget that the equation for that energy change has two terms.
You have enthalpy (delta-H), which consists of a lot of the things that we typically think of driving binding interactions (acid-base pairs, hydrogen bonding, pi-electron interactions, and so on), but there’s also that temperature-and-entropy term (T delta-S). Entropy is a bit more slippery concept, but one way to start thinking about it (although not the whole story) is order and disorder. Compare the starting state and the end state of your process: in which one of them are the components more orderly (fixed in their conformations and positions, for example) or disorderly (able to move around more freely)? As the reaction proceeds, how does the total amount of that order and disorder change? “More disorderly” is by itself energetically favored, as a look around your surroundings will generally demonstrate. That tends to hold whether you’re looking at your your chemical reactions, your bookshelf, your laundry, inside your refrigerator, or at the state of your nation’s political system.
But totaling up that entropy in a binding event is no small matter. You have to look at the ligand that’s binding, of course, and you’d think that much of the time it’s going to lose entropy as it binds (since it’s snuggling into position in the binding site itself, as opposed to floating around out there in solution). But that “floating around in solution” brings you to consider the water molecules that it’s surrounded by out there. If they’re forming a fairly orderly solvation shell around your ligand, that’s going to be broken up as it moves into the binding site, and you might pick up some favorable increased entropy that way. But then there’s that binding site! What’s the entropic state of the protein target before and after binding - more ordered overall, or not? Remember that distant domains might be changing position, not just the areas around the binding site, and they all have water molecules around them, too. The binding site itself may have some key water molecules involved in its structure, and the changes there can run the whole range of positive or negative entropic effects depending on the situation. There are a lot of different single-water-molecule situations with proteins! It is indeed a pain in the rear, to use a thermodynamic term of the art.
In many situations, enthalpic effects and entropic effects seem to be working at cross purposes to each other. This “entropy-enthalpy compensation” is what people have been arguing about for at least the last thirty years, because it sometimes seems like some perverse but inescapable law of nature and sometimes like just an artifact of how we’re viewing the problem. And it does have to be said that the two don’t cancel each other out all the time, or we’d have no way to optimize the binding of our drug candidates at all!
The paper linked above is looking at an old tricyclic drug, doxepin, and its (rather strong) binding to the histamine-1 receptor. Like a lot of other simple tricyclics of that general class, it binds to all sorts of other stuff as well, as do its metabolites, making it a messy proposition in vivo. You can see the list at that link. But it has had many years of use as an antihistamine, antipsychotic, anxiolytic, sleeping aid, and so on, although it's largely fading into the past in most of these areas. My first thought when I saw the structure was "I'll bet that stuff can put you on the floor", and I believe that's an accurate statement.
You’ll note because of that double bond that there are two isomers, Z and E doxepin (from the good ol’ German “zusammen” and “entgegen” - if you keep digging in organic chemistry you’ll eventually hit a German layer). The Z reproducibly binds better than the E (two- to five-fold better depending on your assay) but they’re both down in the lower nanomolar range. What the present paper finds, on close examination by isothermal calorimetry, is that the Z isomer’s binding is almost entirely enthalpy-driven with only a very small change in the entropy term. The E isomer, though, is notably less enthalpically favorable, but makes up a lot of that with an improved entropy term. And there’s why we keep talking about entropy-enthalpy compensation!
Put simply, maybe too simply, the Z isomer has better interactions with the protein itself, but those remove a lot of its conformational flexibility. Meanwhile, the E isomer doesn’t have as strong an enthalpy hand to play, but since it doesn’t lose as much flexibility while binding it doesn’t take the loss-of-entropy hit along the way like the Z isomer had to. So the two of them end up much closer than you otherwise might have guessed.
Studies on mutant receptors showed that a particular tyrosine hydroxyl group in the receptor is a big player in these differences. If you mutate that one to a valine, the two isomers bind almost identically, and with almost identical values for their entropy and enthalpy terms, to boot. It’s pointed toward the tricyclic ring of the structure (but isn’t making a hydrogen bond with the oxygen up there, if that’s what you were thinking). Your first guess might also have been something to do with the basic nitrogen down at the other end of the molecule, but that would also have come up short; things don’t seem to differ much down there for the two isomers.
Subtle details all the way down! But that’s medicinal chemistry, and that’s just one of the many reasons why it ain’t easy. . .
Let’s take a look at the vaccine situation here in the US, because it’s (1) important and (2) not very pretty. A lot of it can be summed up in the recent statements by the head of the Advisory Committee on Immunization Practices (ACIP) at the CDC, Dr. Kirk Milhoan. This is of course the committee that last year saw its entire roster fired by Robert F. Kennedy Jr. and replaced by what I will politely refer to as a group of his stooges. All of them are “vaccine skeptics”, by which you may read “against vaccines in general. Milhoan himself was elevated to his position at ACIP in December, and does not always seem to have been happy with the promotion (there are also reports that he has been fired from his cardiology position over his appointment due to protests from patients). It was in December that ACIP removed mandatory hepatitis B vaccination from the infant schedule, replacing it with “shared decision making” in the name of individual choice. This vaccine in particular has been a focus of Trump administration’s public health agencies, because they have been pushing for a placebo-controlled trial of it. This may sound odd for a vaccine that has been given (in one form or another) for many years now, and has been shown in very large populations to both sharply decrease the levels of hepatitis B itself but also (years later) the incidence of liver cancer (for which long-term hepatitis infection is a major cause). Under these conditions, the idea of a placebo-controlled trial seems outright inhuman, and you would wonder who might consent to such a thing. The idea was to run this with a Danish team in the impoverished west African nation of Guinea-Bissau, but a November coup in that country seems to have disrupted this. As that article will show you, there’s been a great deal of confusion about who proposed the trial, who approved it, whether relevant medical ethics committees were even informed about it, and what its benefits were purported to be for Guinea-Bissau and for medical science in general. That confusion extended to whether or not the trial had been cancelled, but it seems that the current answer, thank God, is “yes, it has been”. The overall plan was for 7,000 infants to be vaccinated at birth, with another 7,000 to wait six weeks - during which time, especially under conditions that unfortunately prevail in much of the country, you could expect a good number of them to be exposed to the hepatitis B virus. How anyone could condone such an experiment on these children is a question for criminal psychologists to answer. It’s especially infuriating because the whole “They’re experimenting on our babies!” line is a mainstay of the anti-vaccine movement in the US and other countries. These people regard mandatory vaccine schedules as illegal and immoral coercion and never tire of waving the banner of individual freedom to refuse vaccines entirely. That “shared-decision-making phrase” came up, and how, as ACIP recently proceeded to put half a dozen other childhood vaccines in the same category. And Kirk Milhoan himself did a thorough job of revealing this worldview on a recent podcast. He feels that mandatory vaccination for diseases like polio and measles is (his phrase) “medical battery” and that the right to refuse the shots outweighs and medical or public health objections. His worldview is completely aligned with his boss (RFK Jr.): summarized, that “these diseases generally weren’t so bad anyway, and they’ve pretty much disappeared because of Modern Times in general (sanitation, etc.) and not because of vaccination per se, and anyway the vaccines themselves have such terrible risks and side effects that not giving them is obviously the right choice.” I cannot put into words how much I disagree with all of these contentions. We are going to pay a terrible price for sowing mistrust of these vaccines - which is what these decisions do - and for telling parents that they should just do their own thing because none of these shots are required, anyway. Response from actual physicians and medical experts has been (as you’d expect) extremely unfavorable, and the American Academy of Pediatrics has come out completely against the new loosened recommendations. I certainly hope that most doctors will hold the line here, but what this means, no matter what, is that it will be easier than ever for children not to receive these shots. Which means it will be easier than ever to get the diseases that they prevent. All of this is taking place against a backdrop of increasing measles infections all around the country (and in other industrialized countries that have recently been letting their guard down for similar political reasons). We used to have that disease eliminated in this country, for God’s sake. This is a crime, a crime of such stubborn stupidity that it’s hard to even get your head around it, and here we are doing it to ourselves. And this is of course not just being fought out in the infant vaccination schedules. This administration is hostile to vaccination from every direction, which is why it appears that the CDC is now doing a deliberately worse job of even tracking the underlying diseases (as well as vaccination rates). If the data can’t be relied on, then everyone can just throw up their hands, right? So we have the spectacle of an administration who is supposedly so dedicated to the pursuit of medical truth that they want to run placebo-controlled trials for vaccines that have been shown effective for decades, and on the other hand they’re destroying the public data record so that everything turns into “Gosh, who can say?” Downstream of all this chaos are the people doing vaccine research and manufacturing. All these changes in the government recommendations have the potential to throw insurance coverage into turmoil, and you can absolutely see that a goal of the administration is to remove liability shields for vaccine makers so they can spend every day of every year fighting in court. Lastly, Moderna has just announced that they’re cutting back late-stage vaccine trials due to the uncertainty in the US market. This is not good news, and while some of it may well be due to Moderna’s own problems, the underlying point stands. Unfortunately. This is all just horrible, and we're apparently all just going to have to live through it while the fools and grifters have their way. And while we do everything we can to replace them at the first chance.
Show full content
Let’s take a look at the vaccine situation here in the US, because it’s (1) important and (2) not very pretty. A lot of it can be summed up in the recent statements by the head of the Advisory Committee on Immunization Practices (ACIP) at the CDC, Dr. Kirk Milhoan. This is of course the committee that last year saw its entire roster fired by Robert F. Kennedy Jr. and replaced by what I will politely refer to as a group of his stooges. All of them are “vaccine skeptics”, by which you may read “against vaccines in general. Milhoan himself was elevated to his position at ACIP in December, and does not always seem to have been happy with the promotion (there are also reports that he has been fired from his cardiology position over his appointment due to protests from patients).
It was in December that ACIP removed mandatory hepatitis B vaccination from the infant schedule, replacing it with “shared decision making” in the name of individual choice. This vaccine in particular has been a focus of Trump administration’s public health agencies, because they have been pushing for a placebo-controlled trial of it. This may sound odd for a vaccine that has been given (in one form or another) for many years now, and has been shown in very large populations to both sharply decrease the levels of hepatitis B itself but also (years later) the incidence of liver cancer (for which long-term hepatitis infection is a major cause).
Under these conditions, the idea of a placebo-controlled trial seems outright inhuman, and you would wonder who might consent to such a thing. The idea was to run this with a Danish team in the impoverished west African nation of Guinea-Bissau, but a November coup in that country seems to have disrupted this. As that article will show you, there’s been a great deal of confusion about who proposed the trial, who approved it, whether relevant medical ethics committees were even informed about it, and what its benefits were purported to be for Guinea-Bissau and for medical science in general. That confusion extended to whether or not the trial had been cancelled, but it seems that the current answer, thank God, is “yes, it has been”. The overall plan was for 7,000 infants to be vaccinated at birth, with another 7,000 to wait six weeks - during which time, especially under conditions that unfortunately prevail in much of the country, you could expect a good number of them to be exposed to the hepatitis B virus. How anyone could condone such an experiment on these children is a question for criminal psychologists to answer.
It’s especially infuriating because the whole “They’re experimenting on our babies!” line is a mainstay of the anti-vaccine movement in the US and other countries. These people regard mandatory vaccine schedules as illegal and immoral coercion and never tire of waving the banner of individual freedom to refuse vaccines entirely.
That “shared-decision-making phrase” came up, and how, as ACIP recently proceeded to put half a dozen other childhood vaccines in the same category. And Kirk Milhoan himself did a thorough job of revealing this worldview on a recent podcast. He feels that mandatory vaccination for diseases like polio and measles is (his phrase) “medical battery” and that the right to refuse the shots outweighs and medical or public health objections. His worldview is completely aligned with his boss (RFK Jr.): summarized, that “these diseases generally weren’t so bad anyway, and they’ve pretty much disappeared because of Modern Times in general (sanitation, etc.) and not because of vaccination per se, and anyway the vaccines themselves have such terrible risks and side effects that not giving them is obviously the right choice.”
I cannot put into words how much I disagree with all of these contentions. We are going to pay a terrible price for sowing mistrust of these vaccines - which is what these decisions do - and for telling parents that they should just do their own thing because none of these shots are required, anyway. Response from actual physicians and medical experts has been (as you’d expect) extremely unfavorable, and the American Academy of Pediatrics has come out completely against the new loosened recommendations. I certainly hope that most doctors will hold the line here, but what this means, no matter what, is that it will be easier than ever for children not to receive these shots.
Which means it will be easier than ever to get the diseases that they prevent. All of this is taking place against a backdrop of increasing measles infections all around the country (and in other industrialized countries that have recently been letting their guard down for similar political reasons). We used to have that disease eliminated in this country, for God’s sake. This is a crime, a crime of such stubborn stupidity that it’s hard to even get your head around it, and here we are doing it to ourselves.
And this is of course not just being fought out in the infant vaccination schedules. This administration is hostile to vaccination from every direction, which is why it appears that the CDC is now doing a deliberately worse job of even tracking the underlying diseases (as well as vaccination rates). If the data can’t be relied on, then everyone can just throw up their hands, right? So we have the spectacle of an administration who is supposedly so dedicated to the pursuit of medical truth that they want to run placebo-controlled trials for vaccines that have been shown effective for decades, and on the other hand they’re destroying the public data record so that everything turns into “Gosh, who can say?”
Downstream of all this chaos are the people doing vaccine research and manufacturing. All these changes in the government recommendations have the potential to throw insurance coverage into turmoil, and you can absolutely see that a goal of the administration is to remove liability shields for vaccine makers so they can spend every day of every year fighting in court. Lastly, Moderna has just announced that they’re cutting back late-stage vaccine trials due to the uncertainty in the US market. This is not good news, and while some of it may well be due to Moderna’s own problems, the underlying point stands. Unfortunately. This is all just horrible, and we're apparently all just going to have to live through it while the fools and grifters have their way. And while we do everything we can to replace them at the first chance.
This is one of those papers that really makes you wonder what you’re missing when you look around you. The authors are looking at what might sound like a rather uninteresting question - what happens when you dissolve alkylamines in water? I would have said (and I’ll bet that many of you would have too) that well, you get a solution of alkylamine molecules in water. I mean, there will be solvation shells and there will be hydrogen bonding around the amine ends and some local rough order of the water molecules around each amine molecule, but all of it will be dynamic at room temperature, too. Right? Well, not for the first time, and not for the last, There’s More To It Than That. X-ray studies over the years have indicated what the authors describe as “a very peculiar microheterogeneity” being produced when (for example) hexylamine is dissolved in water. This is probably the most detailed look at the X-ray scattering data yet, across the whole series from butylamine (pee-yew) up to nonylamine (not smelling all that much better, honestly) in different proportions with purified water. These were checked across a range of concentrations and (in some cases) across a range of temperatures as well. What comes out is a rather surprising picture. The amine headgroups (NH2) do indeed interact strongly with the water molecules, with water being the H-bond donor to the acceptor nitrogen atoms. This leads to amine saturation around water domains in solution. At lower amounts of alkylamines and/or higher temperatures, though, what you actually have is more like two liquids in the same container. One of these is pretty much straight water, and the other of these is an amine/water mixture, a demixing that leads to actual phase separation because there aren’t enough amine molecules to stabilize the “patches of water” situation any more. But at the other end of the scale, high amine/low water, you have pockets of water in the bulk amine, each of which is surrounded by amines with their nitrogen atoms pointed at the interfaces. But at lower temperatures (and even more so at lower amine concentrations) you get a lamellar structure, with stacked layers of water and amines partaking of that amine-head-group interaction at the boundaries. You can think of this situation as flat patches of more or less pure water surrounded by double-layered sheets of more or less pure amine. It’s a surprisingly orderly structure for what’s supposed to be a well-mixed solution, and as the authors note, it’s the sort of thing you’d expect to see more in soft-matter gelatinous situations. This lamellar structure demixes immediately once temperature or amine concentration thresholds are crossed. You’d hardly notice any of this stuff as you mix these liquids together! Only the higher-temperature phase separation would be visible to the naked eye, I would think. It makes you wonder how many more things we miss. The physical world is full of shapes, textures, and structures that exist without us even realizing them. . .cue up William Blake, who himself is not generally miscible with discussions of hydrogen bonding and phase diagrams.
Show full content
This is one of those papers that really makes you wonder what you’re missing when you look around you. The authors are looking at what might sound like a rather uninteresting question - what happens when you dissolve alkylamines in water? I would have said (and I’ll bet that many of you would have too) that well, you get a solution of alkylamine molecules in water. I mean, there will be solvation shells and there will be hydrogen bonding around the amine ends and some local rough order of the water molecules around each amine molecule, but all of it will be dynamic at room temperature, too. Right?
Well, not for the first time, and not for the last, There’s More To It Than That. X-ray studies over the years have indicated what the authors describe as “a very peculiar microheterogeneity” being produced when (for example) hexylamine is dissolved in water. This is probably the most detailed look at the X-ray scattering data yet, across the whole series from butylamine (pee-yew) up to nonylamine (not smelling all that much better, honestly) in different proportions with purified water. These were checked across a range of concentrations and (in some cases) across a range of temperatures as well.
What comes out is a rather surprising picture. The amine headgroups (NH2) do indeed interact strongly with the water molecules, with water being the H-bond donor to the acceptor nitrogen atoms. This leads to amine saturation around water domains in solution. At lower amounts of alkylamines and/or higher temperatures, though, what you actually have is more like two liquids in the same container. One of these is pretty much straight water, and the other of these is an amine/water mixture, a demixing that leads to actual phase separation because there aren’t enough amine molecules to stabilize the “patches of water” situation any more. But at the other end of the scale, high amine/low water, you have pockets of water in the bulk amine, each of which is surrounded by amines with their nitrogen atoms pointed at the interfaces.
But at lower temperatures (and even more so at lower amine concentrations) you get a lamellar structure, with stacked layers of water and amines partaking of that amine-head-group interaction at the boundaries. You can think of this situation as flat patches of more or less pure water surrounded by double-layered sheets of more or less pure amine. It’s a surprisingly orderly structure for what’s supposed to be a well-mixed solution, and as the authors note, it’s the sort of thing you’d expect to see more in soft-matter gelatinous situations. This lamellar structure demixes immediately once temperature or amine concentration thresholds are crossed.
You’d hardly notice any of this stuff as you mix these liquids together! Only the higher-temperature phase separation would be visible to the naked eye, I would think. It makes you wonder how many more things we miss. The physical world is full of shapes, textures, and structures that exist without us even realizing them. . .cue up William Blake, who himself is not generally miscible with discussions of hydrogen bonding and phase diagrams.
I’ve written a number of times over the years here about the placebo effect, and needless to note there’s still a lot more to be said. This new paper is stark evidence of that! It notes that there is evidence that (at least in mice) that the dopaminergic mesolimbic pathway is known to help modulate immune function, but is also involved in expectations of positive outcomes. This raises the possibility of a direct neural/biochemical linkage that might be behind some placebo effects, and the authors have had a chance to put that idea to the test. The experimental design is quite something. They’re looking for increased activity in the ventral tegmenal area (VTA) as well as the bilaterial nucleus accumbens region of the brain, and this is determined by functional MRI imaging. But the volunteers in this trial got to watch the imaging, and got to practice strategies in increase that activity. This feedback could be any combination of “perceptual, affective, cognitive or meta-cognitive mental contents”, basically whatever did the job in the imaginative-positive-thoughts area, and each participant worked out whatever strategy was best to increase this dopaminergic mesolimbic activity. There were two control groups: one of them used the fMRI feedback to learn to increase activity in some other brain region that had nothing to do with the mesolimbic pathways, and another group didn’t do the fMRI feedback practice at all. After the last training session/evaluation, all participants received a hepatitis B vaccine, and all were evaluated after 14 and 28 days to see what the antibody response was. A subset of patients were also evaluated after three months as well. One potential complication was that even the people who weren’t specifically training up on the dopaminergic pathways might be getting some benefit there, because the entire training process is an anticipation/reward paradigm. Another variable is that some people simply don’t respond to the Hep B vaccine (there turned out to be seven of these, evenly distributed among the groups, and they were excluded from later analysis). The results are quite interesting: the patients who learned to upregulate their VTA activity really did show enhanced response to the vaccine. They could be distinguished from the ones who did mostly upregulation of nucleus accumbens activity, and also from those who learned to upregulate activity in other brain regions not having to do with the dopaminergic/mesolimbic system. The people in the latter group who did show increased VTA activity anyway also showed better vaccine response, interestingly. The authors also tried to find other differences between the people who upregulated VTA, such as testing them for behavioral measures of response to incentives and approach/avoidance tendencies, but nothing really showed up by these measures. The authors: “Altogether, our study demonstrates that upregulating the VTA with repeated fMRI-NF training is associated with a stronger post-vaccination immune response in humans. Considering the lack of evidence for alternative interpretations, our findings suggest a top-down brain-immune regulation mechanism, similar to that previously described in rodents” The authors suggest that a follow-up study could concentrate more on specific VTA activation rather than other brain regions (in light of their results here), but also note that there are other physiological and psychological factors that they could have missed that could have affected the results. (One of these is the involvement of the VTA in wakefulness and the effects of sleep duration and habits on immune response). The effect of other immune challenges would also be very useful to check, and of course all of these would be better run on a larger scale to get even better statistical power. But on the face of it, this does look like confirmation of an effect that already seems well-documented in rodents and now appears to extend to humans. It appears that our immune systems are to some degree cross-wired with our moods and specifically our expectations of rewards and positive outcomes. That at least is one physical, neurochemical connection between the world of thoughts and emotions and the (seemingly unrelated) world of B cells and antigen display. What others are out there?
Show full content
I’ve written a number of times over the years here about the placebo effect, and needless to note there’s still a lot more to be said. This new paper is stark evidence of that! It notes that there is evidence that (at least in mice) that the dopaminergic mesolimbic pathway is known to help modulate immune function, but is also involved in expectations of positive outcomes. This raises the possibility of a direct neural/biochemical linkage that might be behind some placebo effects, and the authors have had a chance to put that idea to the test.
The experimental design is quite something. They’re looking for increased activity in the ventral tegmenal area (VTA) as well as the bilaterial nucleus accumbens region of the brain, and this is determined by functional MRI imaging. But the volunteers in this trial got to watch the imaging, and got to practice strategies in increase that activity. This feedback could be any combination of “perceptual, affective, cognitive or meta-cognitive mental contents”, basically whatever did the job in the imaginative-positive-thoughts area, and each participant worked out whatever strategy was best to increase this dopaminergic mesolimbic activity.
There were two control groups: one of them used the fMRI feedback to learn to increase activity in some other brain region that had nothing to do with the mesolimbic pathways, and another group didn’t do the fMRI feedback practice at all. After the last training session/evaluation, all participants received a hepatitis B vaccine, and all were evaluated after 14 and 28 days to see what the antibody response was. A subset of patients were also evaluated after three months as well.
One potential complication was that even the people who weren’t specifically training up on the dopaminergic pathways might be getting some benefit there, because the entire training process is an anticipation/reward paradigm. Another variable is that some people simply don’t respond to the Hep B vaccine (there turned out to be seven of these, evenly distributed among the groups, and they were excluded from later analysis).
The results are quite interesting: the patients who learned to upregulate their VTA activity really did show enhanced response to the vaccine. They could be distinguished from the ones who did mostly upregulation of nucleus accumbens activity, and also from those who learned to upregulate activity in other brain regions not having to do with the dopaminergic/mesolimbic system. The people in the latter group who did show increased VTA activity anyway also showed better vaccine response, interestingly. The authors also tried to find other differences between the people who upregulated VTA, such as testing them for behavioral measures of response to incentives and approach/avoidance tendencies, but nothing really showed up by these measures. The authors:
“Altogether, our study demonstrates that upregulating the VTA with repeated fMRI-NF training is associated with a stronger post-vaccination immune response in humans. Considering the lack of evidence for alternative interpretations, our findings suggest a top-down brain-immune regulation mechanism, similar to that previously described in rodents”
The authors suggest that a follow-up study could concentrate more on specific VTA activation rather than other brain regions (in light of their results here), but also note that there are other physiological and psychological factors that they could have missed that could have affected the results. (One of these is the involvement of the VTA in wakefulness and the effects of sleep duration and habits on immune response). The effect of other immune challenges would also be very useful to check, and of course all of these would be better run on a larger scale to get even better statistical power.
But on the face of it, this does look like confirmation of an effect that already seems well-documented in rodents and now appears to extend to humans. It appears that our immune systems are to some degree cross-wired with our moods and specifically our expectations of rewards and positive outcomes. That at least is one physical, neurochemical connection between the world of thoughts and emotions and the (seemingly unrelated) world of B cells and antigen display. What others are out there?
I think that this article is a useful one for medicinal chemists to read, because it gets at a lot of the “received wisdom” about what sorts of compounds are good development candidates. On the most basic level, there’s not much disagreement: you would like an orally delivered agent with excellent potency and selectivity, given at a relatively low dose with good bioavailability and once a day, that has normal-looking pharmacokinetics to give you blood levels that cover the target of interest, that is metabolized into a short list of inactive compounds that are excreted through the usual pathways. No one’s going to complain about a profile like that. But what if you don’t have a profile like that? And let’s face it, you may well not. Getting all those factors above to line up so well is not so easy, and you will often find yourself wondering which of these desirable qualities you’re going to to have to throw overboard and to what extent. The paper referenced above looks at orally active drugs approved from 2020 to 2024 (104 of them) and analyzes them for how close they are to that ideal profile. And there are plenty of deviations! For one thing, only two-thirds of these drugs are dosed once a day, and the authors note that first-in-class or orphan drugs are more likely to land in the BID or TID dosing domains (which makes sense). Drugs in those latter categories also tend to have higher daily doses, and both of these reflect choices made along the way to get such programs over the finish lines. There relatively few (32) of these drugs with bioavailability numbers in their documentation, but it’s noteworthy that there are several with rather low numbers (below 25%). The most common range is 60 to 70%, but the numbers are all up and down the scale. Interestingly, there are more drugs on the list with daily doses of 300mg and up than there are with daily doses of 50mg or below. Metabolically, nearly half of the drugs on the list are main metabolized by CYP3A4, which is about what you’d expect. That shows up in the analysis of drug-drug interactions as well, where by far the largest categories are due to CYP3A4 induction, inhibition, or just being a competing substrate for the enzyme. 14 of the 104 drugs are known to have active metabolites. And the figure that really surprised me was that 22% of these recently approved drugs come out of the gate with a black-box warning, which is the most serious level that the FDA mandates in labeling. Most of these are cardiovascular, but there are all sorts of things that show up (risk of infection, suicidal ideation). The authors summarize this by warning drug discovery and development teams that too-strict development criteria could well be hurting one’s chances for a successful program. There is a natural tendency to avoid sending things into clinical trials that have too many strikes against them - after all, clinical trials are pretty damned expensive - but we might want to recalibrate our “too many” settings a bit. We should also consider loosening up a bit on chemotypes, because the paper notes that there are several drugs that have made it through in the last few years with features that most of us put in the “would rather not” category, such as acetylenes and various N-O bonds (nitro, oxime, N-oxide, etc.) It’s not easy, as the paper notes, to be a champion for a compound that others might consider “undruglike”. But in the end, the real question is not what people think about a drug’s structure or its properties, but how it actually performs in human patients. Particularly if the disease you’re working on is underserved or the target you’re working on is something that hasn’t been tried before, you should be ready to broaden your point of view.
Show full content
I think that this article is a useful one for medicinal chemists to read, because it gets at a lot of the “received wisdom” about what sorts of compounds are good development candidates. On the most basic level, there’s not much disagreement: you would like an orally delivered agent with excellent potency and selectivity, given at a relatively low dose with good bioavailability and once a day, that has normal-looking pharmacokinetics to give you blood levels that cover the target of interest, that is metabolized into a short list of inactive compounds that are excreted through the usual pathways. No one’s going to complain about a profile like that.
But what if you don’t have a profile like that? And let’s face it, you may well not. Getting all those factors above to line up so well is not so easy, and you will often find yourself wondering which of these desirable qualities you’re going to to have to throw overboard and to what extent. The paper referenced above looks at orally active drugs approved from 2020 to 2024 (104 of them) and analyzes them for how close they are to that ideal profile. And there are plenty of deviations!
For one thing, only two-thirds of these drugs are dosed once a day, and the authors note that first-in-class or orphan drugs are more likely to land in the BID or TID dosing domains (which makes sense). Drugs in those latter categories also tend to have higher daily doses, and both of these reflect choices made along the way to get such programs over the finish lines. There relatively few (32) of these drugs with bioavailability numbers in their documentation, but it’s noteworthy that there are several with rather low numbers (below 25%). The most common range is 60 to 70%, but the numbers are all up and down the scale. Interestingly, there are more drugs on the list with daily doses of 300mg and up than there are with daily doses of 50mg or below.
Metabolically, nearly half of the drugs on the list are main metabolized by CYP3A4, which is about what you’d expect. That shows up in the analysis of drug-drug interactions as well, where by far the largest categories are due to CYP3A4 induction, inhibition, or just being a competing substrate for the enzyme. 14 of the 104 drugs are known to have active metabolites. And the figure that really surprised me was that 22% of these recently approved drugs come out of the gate with a black-box warning, which is the most serious level that the FDA mandates in labeling. Most of these are cardiovascular, but there are all sorts of things that show up (risk of infection, suicidal ideation).
The authors summarize this by warning drug discovery and development teams that too-strict development criteria could well be hurting one’s chances for a successful program. There is a natural tendency to avoid sending things into clinical trials that have too many strikes against them - after all, clinical trials are pretty damned expensive - but we might want to recalibrate our “too many” settings a bit. We should also consider loosening up a bit on chemotypes, because the paper notes that there are several drugs that have made it through in the last few years with features that most of us put in the “would rather not” category, such as acetylenes and various N-O bonds (nitro, oxime, N-oxide, etc.)
It’s not easy, as the paper notes, to be a champion for a compound that others might consider “undruglike”. But in the end, the real question is not what people think about a drug’s structure or its properties, but how it actually performs in human patients. Particularly if the disease you’re working on is underserved or the target you’re working on is something that hasn’t been tried before, you should be ready to broaden your point of view.
Ah, G-protein coupled receptors. They’re an absolute mainstay of drug discovery and have been for many decades, and one of the first projects I worked in when I joined the industry was an effort to produce selective muscarinic M2 antagonists. A few years back I had the chance to meet Bob Lefkowitz (Nobel for GPCRs, along with Brian Kobilka), and I mentioned to him that back in the early 1990s when I was working on them I thought I understood them reasonably well. He grinned when I told him that as time went on I decided that I actually didn’t know jack #$! about them, and that this became more apparent every passing year. This new paper emphasizes that situation! To recap for folks who don’t do this for a living, GPCRs are ubiquitous proteins found on the cell surface, which extend all the way to the inside cytoplasm. There are whole families of these things (with the classic opioid, dopamine, and serotonin ones being famous examples), and they are a way for cells to receive chemical signals from outside that get turned into intracellular changes as these receptors get activated or inactivated. Those “G proteins” in the name are associated with the inner poking-into-the-cell region of the GPCR. The receptor itself changes shape as small molecule ligands (like dopamine!) bind to it or leave on the outside surface, and the activities of the G proteins on the inside change in response to those movements in turn. And those activities include formation or cleavage of a number of prominent “second messenger” molecules like cyclic AMP and others, which set off all kinds of activity as their concentrations move up and down. It’s really a sort of physical toggle switch to get signals through the cell membrane, and on close inspection they truly resemble an old Rube Goldberg machine where the cat jumps for the toy mouse, moving a lever that makes the ball go down and hit the bell that wakes up the bird, which flies over to the. . . Complications ensue. As mentioned, there are often a whole list of GPCRs that respond to the same ligand but are hooked up to different second messengers and are found on different types of cells. That’s one level. Another is that the ligands that activate these receptors (“agonists” in the nomenclature) don’t always activate them to the same degree - there are “partial agonists” that can have different effects than the “full agonists”. Then you have molecules that bind to the receptor but don’t activate it (and in fact can keep agonists from binding while they’re around) - those are “antagonists”, and there are several ways that they can bind to make this happen. Yet another variation is the way that some receptors are set in the “always on” position, activating their second messenger proteins all the time, until a ligand comes along and binds to them to change their shape that then actually shuts them off instead. These are “inverse agonists”. Then you have the way that some particular GPCR subtypes seem to cluster together in defined groups on the cell surface and affect each other’s signaling and behavior as opposed to how they act when they’re studied in isolation. I’m not even going to start on the subtleties on the inner loops of these proteins; suffice it to say that there are whole other signaling families that can also bind down there in addition to the various G-proteins. Oh, it’s a mess. This latest paper is actually trying to bring some of that mess into better focus, and they’re using my old friend and sparring partner the M2 muscarinic receptor as an example. It’s a good choice because it’s been shown to have some large structural changes on ligand binding. Through genetic manipulation and fluorescence microscopy, they’re able to set up a number of structural reporters for the physical state of the receptor protein (the fluorescent groups respond to changes in their environment, such as being shifted into a more or less polar region). And what they find is that this garden-variety GPCR has a whole range of behaviors when exposed to various agonist molecules, shifting around on millisecond time scales between several conformational states with equilibrium constants between each of them. Different agonist molecules hit in different ways when examined at this level of resolution - there are several distinct active states for this GPCR and they interact with the G-proteins in different ways. The authors dryly note that “Overall, our study reveals that the activation of a GPCR in intact cells may be far more complex than previous biophysical studies with isolated receptors have suggested”. And as I mentioned above, it’s not like the previous studies have all led to a simple or orderly model themselves! But this is probably a first look at what’s really happening with these receptors, and we’re going to have to get more details as we try to figure out whether we can use this knowledge to our advantage with new ligands. One way or another, we all now officially know more about GPCRs than we did, and we all now officially understand them less. Welcome to the club!
Show full content
Ah, G-protein coupled receptors. They’re an absolute mainstay of drug discovery and have been for many decades, and one of the first projects I worked in when I joined the industry was an effort to produce selective muscarinic M2 antagonists. A few years back I had the chance to meet Bob Lefkowitz (Nobel for GPCRs, along with Brian Kobilka), and I mentioned to him that back in the early 1990s when I was working on them I thought I understood them reasonably well. He grinned when I told him that as time went on I decided that I actually didn’t know jack #$! about them, and that this became more apparent every passing year. This new paper emphasizes that situation!
To recap for folks who don’t do this for a living, GPCRs are ubiquitous proteins found on the cell surface, which extend all the way to the inside cytoplasm. There are whole families of these things (with the classic opioid, dopamine, and serotonin ones being famous examples), and they are a way for cells to receive chemical signals from outside that get turned into intracellular changes as these receptors get activated or inactivated.
Those “G proteins” in the name are associated with the inner poking-into-the-cell region of the GPCR. The receptor itself changes shape as small molecule ligands (like dopamine!) bind to it or leave on the outside surface, and the activities of the G proteins on the inside change in response to those movements in turn. And those activities include formation or cleavage of a number of prominent “second messenger” molecules like cyclic AMP and others, which set off all kinds of activity as their concentrations move up and down. It’s really a sort of physical toggle switch to get signals through the cell membrane, and on close inspection they truly resemble an old Rube Goldberg machine where the cat jumps for the toy mouse, moving a lever that makes the ball go down and hit the bell that wakes up the bird, which flies over to the. . .
Complications ensue. As mentioned, there are often a whole list of GPCRs that respond to the same ligand but are hooked up to different second messengers and are found on different types of cells. That’s one level. Another is that the ligands that activate these receptors (“agonists” in the nomenclature) don’t always activate them to the same degree - there are “partial agonists” that can have different effects than the “full agonists”. Then you have molecules that bind to the receptor but don’t activate it (and in fact can keep agonists from binding while they’re around) - those are “antagonists”, and there are several ways that they can bind to make this happen. Yet another variation is the way that some receptors are set in the “always on” position, activating their second messenger proteins all the time, until a ligand comes along and binds to them to change their shape that then actually shuts them off instead. These are “inverse agonists”. Then you have the way that some particular GPCR subtypes seem to cluster together in defined groups on the cell surface and affect each other’s signaling and behavior as opposed to how they act when they’re studied in isolation. I’m not even going to start on the subtleties on the inner loops of these proteins; suffice it to say that there are whole other signaling families that can also bind down there in addition to the various G-proteins. Oh, it’s a mess.
This latest paper is actually trying to bring some of that mess into better focus, and they’re using my old friend and sparring partner the M2 muscarinic receptor as an example. It’s a good choice because it’s been shown to have some large structural changes on ligand binding. Through genetic manipulation and fluorescence microscopy, they’re able to set up a number of structural reporters for the physical state of the receptor protein (the fluorescent groups respond to changes in their environment, such as being shifted into a more or less polar region). And what they find is that this garden-variety GPCR has a whole range of behaviors when exposed to various agonist molecules, shifting around on millisecond time scales between several conformational states with equilibrium constants between each of them. Different agonist molecules hit in different ways when examined at this level of resolution - there are several distinct active states for this GPCR and they interact with the G-proteins in different ways.
The authors dryly note that “Overall, our study reveals that the activation of a GPCR in intact cells may be far more complex than previous biophysical studies with isolated receptors have suggested”. And as I mentioned above, it’s not like the previous studies have all led to a simple or orderly model themselves! But this is probably a first look at what’s really happening with these receptors, and we’re going to have to get more details as we try to figure out whether we can use this knowledge to our advantage with new ligands. One way or another, we all now officially know more about GPCRs than we did, and we all now officially understand them less. Welcome to the club!
I haven’t blogged on the microplastics-in-human-tissue reports, but they have certainly been disturbing. Over the last few years, there have been studies suggesting that such species have been accumulating in human brain tissue, the cardiovascular system, testicular tissue and more. There are obviously a lot of microplastic particles out there, considering the environmental wear on so many years of plastic packinging, etc., and it seems unlikely that they’re improving anything. But I will admit to being surprised at the idea of them accumulating in human tissues to this extent. Well, it looks like these results are becoming the site of an analytical-techniques dispute, at least according to the Guardian. Here, for example, is a “Matters Arising” communication about the brain microplastics paper, and its authors say that the original paper does not have enough controls for its methods (pyrolysis GC/MS). They note that the sample preparation techniques used are especially tricky for brain tissue, with its very high lipid content, and that long-chain fatty acids (found naturally in such tissue) can produce polyethylene-like fragments in the GC/MS analysis. They refer to “broader, ongoing gaps in analytical rigor” in this area, and call for researchers to use standardized methods with plenty of internal controls, blank experiments, background corrections, and so on. Similarly, the cardiovascular microplastics paper has come under similar criticism. Those authors point out that the risk of contamination of surgical tissue samples with microplastics during their collection is high, and the paper makes no mention of safeguards to deal with that problem. There were also no blank samples tested, as far as can be seen. Furthermore, the size of the particles noted was much smaller than those seen in other literature reports, with no explanation of how these differences might have come about, and the authors believe that these and other factors could make the paper’s data and conclusions unreliable. Other such criticisms accompany other prominent papers in the field. There seems to be a general problem of groups publishing in this area who have not been sufficiently aware of all the ways that such analyses (which are getting close to the limits of detection) might go wrong. Or perhaps they haven’t been burned enough in the past! This is a tricky area, because you don’t want to see legitimate scientific criticisms used by various yahoos to proclaim that the whole idea of microplastic contamination is bogus. But if we’re going to get a handle on how much of a problem it is in biological systems - and we certainly should - we need numbers that we can trust. Discussing analytical techniques and standards - disagreeing about them very much included - is an essential part of doing good analytical chemistry. That’s how science is supposed to work. Your methods, results, and ideas need to be strong enough to stand up under informed criticism, and if they aren’t, you go back and fix them or you withdraw your claims. Let’s see how this one shakes out!
Show full content
I haven’t blogged on the microplastics-in-human-tissue reports, but they have certainly been disturbing. Over the last few years, there have been studies suggesting that such species have been accumulating in human brain tissue, the cardiovascular system, testicular tissue and more. There are obviously a lot of microplastic particles out there, considering the environmental wear on so many years of plastic packinging, etc., and it seems unlikely that they’re improving anything. But I will admit to being surprised at the idea of them accumulating in human tissues to this extent.
Well, it looks like these results are becoming the site of an analytical-techniques dispute, at least according to the Guardian. Here, for example, is a “Matters Arising” communication about the brain microplastics paper, and its authors say that the original paper does not have enough controls for its methods (pyrolysis GC/MS). They note that the sample preparation techniques used are especially tricky for brain tissue, with its very high lipid content, and that long-chain fatty acids (found naturally in such tissue) can produce polyethylene-like fragments in the GC/MS analysis. They refer to “broader, ongoing gaps in analytical rigor” in this area, and call for researchers to use standardized methods with plenty of internal controls, blank experiments, background corrections, and so on.
Similarly, the cardiovascular microplastics paper has come under similar criticism. Those authors point out that the risk of contamination of surgical tissue samples with microplastics during their collection is high, and the paper makes no mention of safeguards to deal with that problem. There were also no blank samples tested, as far as can be seen. Furthermore, the size of the particles noted was much smaller than those seen in other literature reports, with no explanation of how these differences might have come about, and the authors believe that these and other factors could make the paper’s data and conclusions unreliable. Other such criticisms accompany other prominent papers in the field.
There seems to be a general problem of groups publishing in this area who have not been sufficiently aware of all the ways that such analyses (which are getting close to the limits of detection) might go wrong. Or perhaps they haven’t been burned enough in the past! This is a tricky area, because you don’t want to see legitimate scientific criticisms used by various yahoos to proclaim that the whole idea of microplastic contamination is bogus. But if we’re going to get a handle on how much of a problem it is in biological systems - and we certainly should - we need numbers that we can trust.
Discussing analytical techniques and standards - disagreeing about them very much included - is an essential part of doing good analytical chemistry. That’s how science is supposed to work. Your methods, results, and ideas need to be strong enough to stand up under informed criticism, and if they aren’t, you go back and fix them or you withdraw your claims. Let’s see how this one shakes out!
Working with elemental metals as reagents really takes you back a century or two. That’s partly because iron, zinc, magnesium etc. (as powders, shot, or turnings) are indeed things that have been used in chemistry labs for that long, and partly because we don’t understand a lot of what’s happening at their surfaces much better than we did in 1826. When you’re making organometallic reagents, there’s a long list of activation methods to get things going at the metal surface. Some of these are physical (stirring, crushing, sonication) and some are chemical (iodine, 1,2-dibromoethane, TMS chloride) but what all of them are supposed to do is initiate the reaction and speed it up by providing more surface area in general and less of it coated with metal oxides. Their efficacy is (or can be) real, but not all of them work every time, to put it lightly. The Grignards and organozinc reagents can be particularly fraught, with all sorts of eye-of-newt techniques that have been passed down through the literature and in lab lore. This new paper might be taking one off the list, though. The authors specifically investigate 1,2-dibromoethane’s use in forming organozinc reagents, really getting down in the weeds with kinetic measurements via NMR. And they find that no matter how they try, they can’t get the reagent to help the rate of organozinc formation (!) What does seem to help is the stirring associated with the pretreatment activation methods, but if you leave the dibromoethane out of them you get the same enhancements. Trimethylsilyl chloride, on the other hand, really does seem to help, but the reports of using it along with dibromoethane turn out to not need the latter reagent at all. At the risk of reopening the Stir Wars, I'd be interested in seeing yields, etc. with the nonstirred version, activation time aside. The authors recommend ditching dibromoethane entirely and just going to pre-stirring as an activation method, with TMS chloride for recalcitrant cases. Based on their evidence, I see little room to argue!
Show full content
Working with elemental metals as reagents really takes you back a century or two. That’s partly because iron, zinc, magnesium etc. (as powders, shot, or turnings) are indeed things that have been used in chemistry labs for that long, and partly because we don’t understand a lot of what’s happening at their surfaces much better than we did in 1826.
When you’re making organometallic reagents, there’s a long list of activation methods to get things going at the metal surface. Some of these are physical (stirring, crushing, sonication) and some are chemical (iodine, 1,2-dibromoethane, TMS chloride) but what all of them are supposed to do is initiate the reaction and speed it up by providing more surface area in general and less of it coated with metal oxides. Their efficacy is (or can be) real, but not all of them work every time, to put it lightly. The Grignards and organozinc reagents can be particularly fraught, with all sorts of eye-of-newt techniques that have been passed down through the literature and in lab lore.
This new paper might be taking one off the list, though. The authors specifically investigate 1,2-dibromoethane’s use in forming organozinc reagents, really getting down in the weeds with kinetic measurements via NMR. And they find that no matter how they try, they can’t get the reagent to help the rate of organozinc formation (!) What does seem to help is the stirring associated with the pretreatment activation methods, but if you leave the dibromoethane out of them you get the same enhancements. Trimethylsilyl chloride, on the other hand, really does seem to help, but the reports of using it along with dibromoethane turn out to not need the latter reagent at all. At the risk of reopening the Stir Wars, I'd be interested in seeing yields, etc. with the nonstirred version, activation time aside.
The authors recommend ditching dibromoethane entirely and just going to pre-stirring as an activation method, with TMS chloride for recalcitrant cases. Based on their evidence, I see little room to argue!
I believe that mRNA vaccines are a real advance in the field, but there’s no doubt that we’re in the early days of their use. And given the complexities of the immune system, it’s not surprising that we haven’t worked out all the details - if you really want to get down to it, we don’t have all the details on any therapy involving the immune system at all. Which doesn’t mean that we shouldn’t use them! Let’s get even more real, and say that the same statement applies to small molecule drugs as well, whether they have an immune component or not. We can see the outcomes, we can measure risk/reward and safety/efficacy, and make informed decisions about when to use them and for what. I’m beating on this point so hard, as I’m sure people will have realized, because of the violent hostility of the current US federal health authorities towards vaccination, which is generally given an extremely annoying won’t-someone-think-of-the-children spin about how gosh, we haven’t worked out all the details yet so why are we running cruel experiments on helpless pediatric patients, etc. This is how you can easily weaponize the “precautionary principle” into making sure that you take no actions at all because you can’t understand their full consequences down to the last detail (so you can’t be sure that nothing bad is happening and if you can’t be sure than how can you in good conscience and so on). This new paper, for example, has more details about how mRNA formulations induce such useful immune responses. It’s already been noted that the lipid nanoparticles involves in packing the mRNA payloads have some inherent adjuvant activity, which is convenient. Non-native mRNA certainly has such activity, too, but in the vaccines the use of modified nucleosides turns down a lot of the innate immune system activity that would otherwise kick in. Almost all conventional vaccines have some sort of adjuvant to stimulate the immune system and make the antibody-eliciting effects more prominent (indeed, some of them would be almost useless without it at the dosages administered). I’ve written about adjuvants before, but there’s obviously a lot more to say about them because again, there’s a lot that we haven’t uncovered about their modes of action. The LNP and mRNA components are both playing key roles in these vaccine effects, and I’m going to reproduce the graphical abstract of the paper to give you some idea of what’s going on. Well, perhaps. One thing it’s sure to do is make you glad that you’re not an immunologist, because I assure you that this is a very simplified picture as well. What you’re seeing are effects on dendritic cells (the “DC”) in the middle, and it turns out that both components of the vaccine are acting on these but through different pathways. To add to the fun, both involve CD4-bearing T cells and the associated T-follicular helper (Tfh) cells, but again through different pathways. This research group has evidence that the nucleoside-modified mRNA used in the vaccines does not completely make them invisible to the innate immune system’s pattern-recognition-receptors that are always watching for foreign mRNA. Those are also why the mRNA constructs used in the vaccines need to be well purified to remove double-stranded species, which will really trip those receptors for you if present. It turns out that the purified, nucleoside-modified mRNAs do set off a Type I interferon response that had not been worked out before, and the authors propose that there must be a less-characterized pattern recognition receptor or perhaps a yet-undescribed mechanism of action for existing ones that causes this. They haven’t found either one yet, but their evidence strongly suggests that something like one of these has to be out there. But the responses to both the lipid nanoparticles and the mRNA species are important here, and when things go correctly they actually reinforce each other. There are other experiments in the paper that show that the LNP response is a local one, almost entirely occurring in the nearly draining lymph node to the vaccine injection site, instead of a system-wide effect. It doesn’t even look as if you have to inject them at the same time or in any physically coupled formulation (as we do) - the effect works either way. But since we need the LNPs to protect the mRNA and to get good uptake into cells, it’s certainly good that they’re such mechanistic partners. This could well help to explain the failures of many other plausible mRNA delivery systems that were tried in the earlier years of such research. The hope is that as we study these systems more closely we can work out how to hit this balance by design rather than by trying years of things that don’t work as well (or at all!) We’ve got a ways to go before we get to that point, but learning all the tiny switches and dials of the vaccine immune response is going to be a very worthwhile endeavor.
Show full content
I believe that mRNA vaccines are a real advance in the field, but there’s no doubt that we’re in the early days of their use. And given the complexities of the immune system, it’s not surprising that we haven’t worked out all the details - if you really want to get down to it, we don’t have all the details on any therapy involving the immune system at all. Which doesn’t mean that we shouldn’t use them! Let’s get even more real, and say that the same statement applies to small molecule drugs as well, whether they have an immune component or not. We can see the outcomes, we can measure risk/reward and safety/efficacy, and make informed decisions about when to use them and for what.
I’m beating on this point so hard, as I’m sure people will have realized, because of the violent hostility of the current US federal health authorities towards vaccination, which is generally given an extremely annoying won’t-someone-think-of-the-children spin about how gosh, we haven’t worked out all the details yet so why are we running cruel experiments on helpless pediatric patients, etc. This is how you can easily weaponize the “precautionary principle” into making sure that you take no actions at all because you can’t understand their full consequences down to the last detail (so you can’t be sure that nothing bad is happening and if you can’t be sure than how can you in good conscience and so on).
This new paper, for example, has more details about how mRNA formulations induce such useful immune responses. It’s already been noted that the lipid nanoparticles involves in packing the mRNA payloads have some inherent adjuvant activity, which is convenient. Non-native mRNA certainly has such activity, too, but in the vaccines the use of modified nucleosides turns down a lot of the innate immune system activity that would otherwise kick in. Almost all conventional vaccines have some sort of adjuvant to stimulate the immune system and make the antibody-eliciting effects more prominent (indeed, some of them would be almost useless without it at the dosages administered). I’ve written about adjuvants before, but there’s obviously a lot more to say about them because again, there’s a lot that we haven’t uncovered about their modes of action.
The LNP and mRNA components are both playing key roles in these vaccine effects, and I’m going to reproduce the graphical abstract of the paper to give you some idea of what’s going on. Well, perhaps. One thing it’s sure to do is make you glad that you’re not an immunologist, because I assure you that this is a very simplified picture as well. What you’re seeing are effects on dendritic cells (the “DC”) in the middle, and it turns out that both components of the vaccine are acting on these but through different pathways. To add to the fun, both involve CD4-bearing T cells and the associated T-follicular helper (Tfh) cells, but again through different pathways.
This research group has evidence that the nucleoside-modified mRNA used in the vaccines does not completely make them invisible to the innate immune system’s pattern-recognition-receptors that are always watching for foreign mRNA. Those are also why the mRNA constructs used in the vaccines need to be well purified to remove double-stranded species, which will really trip those receptors for you if present. It turns out that the purified, nucleoside-modified mRNAs do set off a Type I interferon response that had not been worked out before, and the authors propose that there must be a less-characterized pattern recognition receptor or perhaps a yet-undescribed mechanism of action for existing ones that causes this. They haven’t found either one yet, but their evidence strongly suggests that something like one of these has to be out there.
But the responses to both the lipid nanoparticles and the mRNA species are important here, and when things go correctly they actually reinforce each other. There are other experiments in the paper that show that the LNP response is a local one, almost entirely occurring in the nearly draining lymph node to the vaccine injection site, instead of a system-wide effect. It doesn’t even look as if you have to inject them at the same time or in any physically coupled formulation (as we do) - the effect works either way. But since we need the LNPs to protect the mRNA and to get good uptake into cells, it’s certainly good that they’re such mechanistic partners. This could well help to explain the failures of many other plausible mRNA delivery systems that were tried in the earlier years of such research.
The hope is that as we study these systems more closely we can work out how to hit this balance by design rather than by trying years of things that don’t work as well (or at all!) We’ve got a ways to go before we get to that point, but learning all the tiny switches and dials of the vaccine immune response is going to be a very worthwhile endeavor.
What's the worst thing that can happen when you take your new drug candidate into the clinic? One's first thought might be "That it turns out not to do anything", but believe me, that is not the worst outcome, bad news though it is. No, at the bottom of the list is finding out that your compound actually harms patients or even kills them - a rare outcome, to be sure, but not an impossible one by any means and not at all unknown. We'd surely see such disasters more often without the two-species-tox requirements in preclinical testing, but even after clearing (say) rat and dog you can uncover something really interesting about humans that you didn't know before. Save us all from such discoveries. Short of that, I'd say that the situation described here at Science is the next worse. A small company took an unusual-mechanism drug for Alzheimer’s into the clinic, which sounds like one of those "Stop me if you've heard this one" stories. But that setup has (to date) relentlessly ended with ". . .and guess what? It didn't do anything". But that's not what happened here. The initial readouts for T3D's drug candidate, T3D-959, actually looked pretty encouraging, and I don't doubt that they set off a round of surprised celebration when they came in. As you may have guessed from the name, this is a company focused on the glucose-handling hypothesis of Alzheimer’s ("type III diabetes"), and the drug is a PPAR gamma-delta ligand that was being repurposed for this trial. (I should say here that taking a CNS-acting nuclear receptor compound into the clinic for a neurodegenerative disease is a pretty bold move - the PPAR boom of the early 2000s proved that we really don't understand that biology very well, and we have proven in extravagant detail that we don't understand Alzheimer’s very well, either. These two huge steaming heaps of uncertainty and hidden details are not likely to cancel each other out when piled together). But mechanism aside, what T3D found when they started looking into the data was horribly unexpected: But before it trumpeted the good news, the small company took a closer look at detailed data for each participant. It found “a nightmare scenario,” according to a July 2025 legal complaint filed by T3D: The results were “medically impossible.” Some Alzheimer’s patients in the placebo group were reported as improving—even though the disease inexorably erodes cognitive abilities. Many trial participants did not even have the memory-robbing condition, the company claimed, and there was no sign of T3D-959 in blood samples from others who purportedly received it. That's a $35 million bonfire right there, folks, and I don't think that the lawsuit, even if successful, is going to make the situation whole considering the huge opportunity costs involved. If these accusations are correct, this is about as throughly bungled as a clinical trial can get. Now, I know what you're thinking: that this trial was run by some newly minted CRO in the hinterlands of China or India, and that would give all of us a chance to roll our eyes and make world-weary comments about how you get what you pay for. But that's not what we're looking at there. The CRO involved was founded 25 years ago in New York by a professor from Mt. Sinai hospital, and the trial was conducted at several centers here in the US. Well, South Florida anyway, which one must admit has had a rather fast-and-loose reputation in medical developments from time to time. But not this loose, not did-we-dose-those-folks-or-not loose. The error bars on human clinical data (especially CNS!) are quite large enough as it stands, thanks, and you don't have to add in too much incompetence to get an absolutely useless stew of numbers. That's one of the other things that surprises me about this situation, that the CRO was willing to send back the data to the company in that condition. Shouldn't a CNS-focused clinical research team have noted the problems before things got to that point? The question is not only what kind of multithumbed minions dosed the patients and collected the data - you also have to wonder about the hapless managers who greenlighted the resulting mishmosh. It seems to be a recurring problem, though. As you can see from the link above, Science's team found that several of these clinical centers have previously fouled up trial data for other companies and other drugs, and that alone should make any responsible CRO avoid them like radioactive waste zones. But the problems are even deeper. It appears that many of these Miami-area centers are cheerfully enrolling "professional patients" who are signing up for as many trials as possible to collect the payments and benefits, and who are likely as not throwing the pills themselves away. By this point in the article, my head was in my hands, and I'm sure that's a common reaction. This sort of idiotic fraud is the exact opposite of medical research. It is of course not in the interest of any drug development company to have this kind of nonsense happening - all it does is kill any chances your drug might have had to demonstrate efficacy, and on the off-chance it shows any, it kills any chance of any reputable regulatory authority ever believing it. This business is hard enough already. So don't get your drugs tested in South Florida, folks! And don't assume that South Florida is the only location of such bullshit factories, either. . .
Show full content
What's the worst thing that can happen when you take your new drug candidate into the clinic? One's first thought might be "That it turns out not to do anything", but believe me, that is not the worst outcome, bad news though it is. No, at the bottom of the list is finding out that your compound actually harms patients or even kills them - a rare outcome, to be sure, but not an impossible one by any means and not at all unknown. We'd surely see such disasters more often without the two-species-tox requirements in preclinical testing, but even after clearing (say) rat and dog you can uncover something really interesting about humans that you didn't know before. Save us all from such discoveries.
Short of that, I'd say that the situation described here at Science is the next worse. A small company took an unusual-mechanism drug for Alzheimer’s into the clinic, which sounds like one of those "Stop me if you've heard this one" stories. But that setup has (to date) relentlessly ended with ". . .and guess what? It didn't do anything". But that's not what happened here. The initial readouts for T3D's drug candidate, T3D-959, actually looked pretty encouraging, and I don't doubt that they set off a round of surprised celebration when they came in. As you may have guessed from the name, this is a company focused on the glucose-handling hypothesis of Alzheimer’s ("type III diabetes"), and the drug is a PPAR gamma-delta ligand that was being repurposed for this trial. (I should say here that taking a CNS-acting nuclear receptor compound into the clinic for a neurodegenerative disease is a pretty bold move - the PPAR boom of the early 2000s proved that we really don't understand that biology very well, and we have proven in extravagant detail that we don't understand Alzheimer’s very well, either. These two huge steaming heaps of uncertainty and hidden details are not likely to cancel each other out when piled together).
But mechanism aside, what T3D found when they started looking into the data was horribly unexpected:
But before it trumpeted the good news, the small company took a closer look at detailed data for each participant. It found “a nightmare scenario,” according to a July 2025 legal complaint filed by T3D: The results were “medically impossible.” Some Alzheimer’s patients in the placebo group were reported as improving—even though the disease inexorably erodes cognitive abilities. Many trial participants did not even have the memory-robbing condition, the company claimed, and there was no sign of T3D-959 in blood samples from others who purportedly received it.
That's a $35 million bonfire right there, folks, and I don't think that the lawsuit, even if successful, is going to make the situation whole considering the huge opportunity costs involved. If these accusations are correct, this is about as throughly bungled as a clinical trial can get. Now, I know what you're thinking: that this trial was run by some newly minted CRO in the hinterlands of China or India, and that would give all of us a chance to roll our eyes and make world-weary comments about how you get what you pay for. But that's not what we're looking at there. The CRO involved was founded 25 years ago in New York by a professor from Mt. Sinai hospital, and the trial was conducted at several centers here in the US. Well, South Florida anyway, which one must admit has had a rather fast-and-loose reputation in medical developments from time to time.
But not this loose, not did-we-dose-those-folks-or-not loose. The error bars on human clinical data (especially CNS!) are quite large enough as it stands, thanks, and you don't have to add in too much incompetence to get an absolutely useless stew of numbers. That's one of the other things that surprises me about this situation, that the CRO was willing to send back the data to the company in that condition. Shouldn't a CNS-focused clinical research team have noted the problems before things got to that point? The question is not only what kind of multithumbed minions dosed the patients and collected the data - you also have to wonder about the hapless managers who greenlighted the resulting mishmosh.
It seems to be a recurring problem, though. As you can see from the link above, Science's team found that several of these clinical centers have previously fouled up trial data for other companies and other drugs, and that alone should make any responsible CRO avoid them like radioactive waste zones. But the problems are even deeper. It appears that many of these Miami-area centers are cheerfully enrolling "professional patients" who are signing up for as many trials as possible to collect the payments and benefits, and who are likely as not throwing the pills themselves away. By this point in the article, my head was in my hands, and I'm sure that's a common reaction. This sort of idiotic fraud is the exact opposite of medical research.
It is of course not in the interest of any drug development company to have this kind of nonsense happening - all it does is kill any chances your drug might have had to demonstrate efficacy, and on the off-chance it shows any, it kills any chance of any reputable regulatory authority ever believing it. This business is hard enough already. So don't get your drugs tested in South Florida, folks! And don't assume that South Florida is the only location of such bullshit factories, either. . .
Sorry to have been out of touch for the last couple of days, but I'm currently fighting off what appears to be some annoying virus. Mild upper respiratory symptoms, recurrent fever, body aches - the whole package. Interestingly, antibody tests show negative for both Covid and influenza A and B. So I don't know quite what I'm dealing with there, but the default flulike treatment regimen is in effect: as much sleep as feasible, lots of liquids, light on the food, occasional ibuprofen. All I can say is that I'll be glad to shake this one off! A new blog post here will be a sure sign. See everyone soon, I hope. . .
Show full content
Sorry to have been out of touch for the last couple of days, but I'm currently fighting off what appears to be some annoying virus. Mild upper respiratory symptoms, recurrent fever, body aches - the whole package. Interestingly, antibody tests show negative for both Covid and influenza A and B. So I don't know quite what I'm dealing with there, but the default flulike treatment regimen is in effect: as much sleep as feasible, lots of liquids, light on the food, occasional ibuprofen. All I can say is that I'll be glad to shake this one off! A new blog post here will be a sure sign. See everyone soon, I hope. . .
Many people have asked me about my opinion on this recent Alzheimer’s paper, and my opinion lands in the large zone of “interesting work that I hope is followed up on”. That may come as a disappointment, because some of the headlines about this paper have been breathless Cure-For-Alzheimer’s stuff, which is always a danger in this area, and for all I know there may be some folks out there who’d like to see me dismiss these results as yet another hopeless Alzheimer’s quest. But that’s the hard part in working in this area (as I did once) or in writing about it as I do now: there are very few Alzheimer’s quests that are completely hopeless. Unfortunately, that phrase better describes our current ability to reverse the disease’s damage. I myself don’t think that the current antibody therapies even do a useful job of slowing it down, and no one is claiming that they can cause people to regain function that they have lost. But the title of this new paper starts off with “Pharmacologic reversal of advance Alzheimer’s disease in mice”, which is an attention-getter for sure. The compound under study is called P7C3-A20, and here’s a 2014 paper on the neuroprotective effects of compounds in this class, along with a paper on their effects after traumatic brain injury. So this is not a sudden new development, and that’s not even the beginning of the story. The compound itself was described in a 2010 paper by the same group as the best hit out of a phenotypic screen for beneficial CNS compounds, and you’ll see references in those 2014 papers to effects on other neurological injuries or neurodegenerative disorders. These effects seemed consistent not with stimulating new neuronal growth per se, but with protecting the survival of new neurons as they are produced. The application of this compound to Alzheimer’s is not a surprise, and to be honest I’m surprised that it’s taken as long as it has given the earlier publications. The 2014 work proposed a target for P7C3-A20, namely the nicotine adenine dinucleotide (NAD) salvage pathway. The compound appears to bind to (and enhance the activity of) the enzyme nicotinamide ribosyltransferase (NAMPT), which is the key enzyme in NAD re-synthesis. NAD itself is of course known as a very important small molecule in biochemistry, participating in a whole list of redox activities in the cell (many of which take place in mitochondria) and acting as a cofactor for a range of enzymes. Many of these reactions produce nicotinamide itself as a consequence of their mechanisms, and NAMPT is the enzyme for the rate-limiting step that produces fresh NAD from nicotinamide. (NAD can also be made de novo, but the “salvage pathway” is extremely important because not enough NAD can be produced otherwise). Given its importance, the role of NAD levels and handling in diseases like Alzheimer’s has been investigated for a long time now. But just outright taking NAD supplements is not a slam-dunk idea, because many types of cancer cells have even higher NAD requirements than normal tissue. The worry has been that you will just be doing any incipient cancers a favor by raising your NAD levels across the board even though you might also be helping out beneficial cell lines like tumor-infiltrating T cells at the same time. Some studies have shown increased risks of some types of cancer with NAD supplementation, and overall, the recommendation has been to hold off on doing that until we understand the risk/benefit landscape better. (The paper under discussion has many links to the literature on this topic). In this case, though, P7C3-A20 is claimed to restore NAD homeostasis without producing excess NAD per se, which would seem to be a better outcome. The new paper shows evidence for the robustness of NAD homeostasis and severity of Alzheimer’s, and this goes all the way up to humans. That’s through testing of brain sample of people who died showing signs of Alzheimer’s pathology in their tissue, while not exhibiting notable cognitive deficits. That possible human connection is a big deal here, because the rest of the study is done in mouse models. These show some very interesting results where it does look like you can make the case that some deficits are being reversed, and that’s a rare thing to see in any neuronal-level CNS work. But I m not a good customer for rodent models of Alzheimer’s. My big problem is that rodent’s don’t actually get Alzheimer’s. Nor, to the best of my knowledge, do any other animals except humans. And that’s not just because we know human behaviors so much better or because we have all these higher brain functions to lose. No other animal shows the cellular-level brain pathology associated with Azlheimer’s disease. If you want to see something like that, you have to engineer it in. How though, you might wonder, do we engineer Alzheimer’s mice when we’re still not certain of the fundamental causes of Alzheimer’s? Well should you ask. As the world knows, the main hypothesis for Alzheimer’s etiology over the decades has to do with the beta-amyloid protein,its overall levels and its handling in various regions of the brain. There are a lot of good reasons for that, as detailed in this post. And since I mention that, let me reiterate that no, the whole amyloid/Alzheimer’s field has not been based just on some work that has now been shown to be fraudulent (a statement that I still see being made by people not familiar with the history). That work did damage, but it was mostly by providing more confidence to many investigators to believe what they believed already. You can see where this is going: the mouse models for Alzheimer’s are generally animals that have had mutations placed in pathways for amyloid production and handling. If you are more confident that amyloid handling really is a fundamental part of Alzheimer’s pathology, then this probably won’t bother you much. If (like me) you have come to doubt that connection more and more over the years, you may be less enthusiastic. But the glass-half-full position is that whether or not such animals are experiencing “real Alzheimer’s” or even something close to it, they most definitely are experiencing neuronal stress. And a compound that seems to alleviate it would be very much worth pursuing further. So that’s where I land here. I think this compound (and/or this mechanism of action) seems worth pursuing in human trials after the usual preclinical checks, and my main question now is why it hasn’t been, after all these years and all these publications. I fear that part of the answer could well be the dominance that amyloid mechanisms (and to a lesser extent, tau protein mechanisms) have had over the field. This concentration looked for many years like a welcome level of targeted effort on the most promising hypothesis in the field, but as the decades have worn on, less so. Amyloidocentric ideas have advanced to the clinic over and over, and the absolute best of them have been - in my view - underwhelming and practically useless. If amyloid were really as central a player as we all used to think, this just should not have been the case. The resulting starvation of alternate approaches might be well illustrated by the long-running story of P7C3-A20. I do wonder, though, if there are other factors at work. After all, other long-shot ideas in Alzheimer’s and neurodegeneration in general have made it into at least small trials based on what may well be less promising results than these. The potential for such drugs is so huge that you can often get people to put a little backing behind them, but I haven’t even seen that much here (from what I can see, P7C3-A20 has never made it into a human trial at all). Is there a Rest of the Story here? And where does the story we know about go at this point? I’ll be watching with interest.
Show full content
Many people have asked me about my opinion on this recent Alzheimer’s paper, and my opinion lands in the large zone of “interesting work that I hope is followed up on”. That may come as a disappointment, because some of the headlines about this paper have been breathless Cure-For-Alzheimer’s stuff, which is always a danger in this area, and for all I know there may be some folks out there who’d like to see me dismiss these results as yet another hopeless Alzheimer’s quest.
But that’s the hard part in working in this area (as I did once) or in writing about it as I do now: there are very few Alzheimer’s quests that are completely hopeless. Unfortunately, that phrase better describes our current ability to reverse the disease’s damage. I myself don’t think that the current antibody therapies even do a useful job of slowing it down, and no one is claiming that they can cause people to regain function that they have lost. But the title of this new paper starts off with “Pharmacologic reversal of advance Alzheimer’s disease in mice”, which is an attention-getter for sure.
The compound under study is called P7C3-A20, and here’s a 2014 paper on the neuroprotective effects of compounds in this class, along with a paper on their effects after traumatic brain injury. So this is not a sudden new development, and that’s not even the beginning of the story. The compound itself was described in a 2010 paper by the same group as the best hit out of a phenotypic screen for beneficial CNS compounds, and you’ll see references in those 2014 papers to effects on other neurological injuries or neurodegenerative disorders. These effects seemed consistent not with stimulating new neuronal growth per se, but with protecting the survival of new neurons as they are produced. The application of this compound to Alzheimer’s is not a surprise, and to be honest I’m surprised that it’s taken as long as it has given the earlier publications.
The 2014 work proposed a target for P7C3-A20, namely the nicotine adenine dinucleotide (NAD) salvage pathway. The compound appears to bind to (and enhance the activity of) the enzyme nicotinamide ribosyltransferase (NAMPT), which is the key enzyme in NAD re-synthesis. NAD itself is of course known as a very important small molecule in biochemistry, participating in a whole list of redox activities in the cell (many of which take place in mitochondria) and acting as a cofactor for a range of enzymes. Many of these reactions produce nicotinamide itself as a consequence of their mechanisms, and NAMPT is the enzyme for the rate-limiting step that produces fresh NAD from nicotinamide. (NAD can also be made de novo, but the “salvage pathway” is extremely important because not enough NAD can be produced otherwise).
Given its importance, the role of NAD levels and handling in diseases like Alzheimer’s has been investigated for a long time now. But just outright taking NAD supplements is not a slam-dunk idea, because many types of cancer cells have even higher NAD requirements than normal tissue. The worry has been that you will just be doing any incipient cancers a favor by raising your NAD levels across the board even though you might also be helping out beneficial cell lines like tumor-infiltrating T cells at the same time. Some studies have shown increased risks of some types of cancer with NAD supplementation, and overall, the recommendation has been to hold off on doing that until we understand the risk/benefit landscape better. (The paper under discussion has many links to the literature on this topic).
In this case, though, P7C3-A20 is claimed to restore NAD homeostasis without producing excess NAD per se, which would seem to be a better outcome. The new paper shows evidence for the robustness of NAD homeostasis and severity of Alzheimer’s, and this goes all the way up to humans. That’s through testing of brain sample of people who died showing signs of Alzheimer’s pathology in their tissue, while not exhibiting notable cognitive deficits.
That possible human connection is a big deal here, because the rest of the study is done in mouse models. These show some very interesting results where it does look like you can make the case that some deficits are being reversed, and that’s a rare thing to see in any neuronal-level CNS work. But I m not a good customer for rodent models of Alzheimer’s. My big problem is that rodent’s don’t actually get Alzheimer’s. Nor, to the best of my knowledge, do any other animals except humans. And that’s not just because we know human behaviors so much better or because we have all these higher brain functions to lose. No other animal shows the cellular-level brain pathology associated with Azlheimer’s disease. If you want to see something like that, you have to engineer it in.
How though, you might wonder, do we engineer Alzheimer’s mice when we’re still not certain of the fundamental causes of Alzheimer’s? Well should you ask. As the world knows, the main hypothesis for Alzheimer’s etiology over the decades has to do with the beta-amyloid protein,its overall levels and its handling in various regions of the brain. There are a lot of good reasons for that, as detailed in this post. And since I mention that, let me reiterate that no, the whole amyloid/Alzheimer’s field has not been based just on some work that has now been shown to be fraudulent (a statement that I still see being made by people not familiar with the history). That work did damage, but it was mostly by providing more confidence to many investigators to believe what they believed already.
You can see where this is going: the mouse models for Alzheimer’s are generally animals that have had mutations placed in pathways for amyloid production and handling. If you are more confident that amyloid handling really is a fundamental part of Alzheimer’s pathology, then this probably won’t bother you much. If (like me) you have come to doubt that connection more and more over the years, you may be less enthusiastic. But the glass-half-full position is that whether or not such animals are experiencing “real Alzheimer’s” or even something close to it, they most definitely are experiencing neuronal stress. And a compound that seems to alleviate it would be very much worth pursuing further.
So that’s where I land here. I think this compound (and/or this mechanism of action) seems worth pursuing in human trials after the usual preclinical checks, and my main question now is why it hasn’t been, after all these years and all these publications. I fear that part of the answer could well be the dominance that amyloid mechanisms (and to a lesser extent, tau protein mechanisms) have had over the field. This concentration looked for many years like a welcome level of targeted effort on the most promising hypothesis in the field, but as the decades have worn on, less so. Amyloidocentric ideas have advanced to the clinic over and over, and the absolute best of them have been - in my view - underwhelming and practically useless. If amyloid were really as central a player as we all used to think, this just should not have been the case.
The resulting starvation of alternate approaches might be well illustrated by the long-running story of P7C3-A20. I do wonder, though, if there are other factors at work. After all, other long-shot ideas in Alzheimer’s and neurodegeneration in general have made it into at least small trials based on what may well be less promising results than these. The potential for such drugs is so huge that you can often get people to put a little backing behind them, but I haven’t even seen that much here (from what I can see, P7C3-A20 has never made it into a human trial at all). Is there a Rest of the Story here? And where does the story we know about go at this point? I’ll be watching with interest.
Happy New Year! Science blogging resumes on Monday. But I have to get this off my chest first: Let me start off this first post of 2026 by acknowledging that 2025 was, in many ways, a horrible year. I say that particularly by the subject matters of this site - biomedical research and science in general. The Trump administration moved quickly last year to attack federal research funding from every single angle they could think of, and went on to do major damage to every science-based agency. The NIH, CDC, FDA and others will need years to recover what’s been done to them. Likewise, US public health will suffer. It already is. I note that we now have reached a level of measles cases not seen in this country for decades, and one can only expect the numbers for every other vaccine-preventable disease to rise as well. This is of course infuriating, because it is so senseless and was so avoidable, but here we are. Children will die who did not have to, and it was a choice. Just as gutting USAID was one of the early choices of the administration, an act of petty vandalism which has led (and will continue to lead) to even more deaths and suffering in the countries that used to get the aid. Never forget: whenever Elon Musk walks up to a podium to bullshit you about colonizing Mars or whatever, he is walking on the bodies of dead children. He yanked their food and medicine away while telling lies and cracking stupid jokes. There’s plenty of blood to go around, though. I have often stated my belief that Robert F. Kennedy, Jr. has it on his hands after his actions during a measles epidemic in Samoa, and now he will get a chance to replicate that disaster on a grander scale, as befits his self-image. Kennedy, our HHS secretary, believes that the 1918 flu pandemic and the 1980s rise of HIB somehow happened because of vaccine research. Marty Makary, our head of the FDA, believes that Lyme disease is an escape biowarfare agent. Jay Bhattacharya, among other things, believes that he is fit to lead the NIH. We are being governed by grinning idiots and sociopaths, and any assessment of our situation in 2026 has to take that as a precondition. This of course goes all the way to the top. Donald Trump is (by a wide margin) the worst president this country has ever had, and his administration is a catastrophe. I will not be discussing my reasons here for those assertions - suffice it to say that I could do a solid 50,000 words on this topic with copious links, citations, and video clips. But to what point? What I think is more important is that any assessment of our situation also has to come to terms with the fact that he was nonetheless re-elected. I am very much cheered by his cratering approval ratings and by the increasing numbers of people speaking out against his policies, but (as with the vaccine situation), all this was avoidable. Dwelling on that, though, can be a mistake. Just saying “This didn’t have to happen” avoids thinking about why it did, and risks not realizing what bad shape our institutions and our civil society had to be in for things to get to this point. Those institutions very much include the House of Representatives, the Senate, and the Supreme Court. There are multiple points of failure here, and the irony is not lost on me that we are heading into the 250th anniversary of this country’s founding while surrounded by horrible evidence that its system of government is in terrible need of repair. But that’s what I’m dedicating this year to. How can I do otherwise? Despair is a sin. And I’m not going to let this convicted felon and adjudicated sexual predator, this incompetent would-be-authoritarian disgrace of a human being run me out of my own country. I am going to continue to speak out against the harmful acts of this administration and I am going to continue to work to replace it with something better. This year, that means the mid-term elections. I always get some fashionably cynical replies to this sort of statement, stuff like “Oh you poor little lamb, you still think there will be mid-term elections”. Keep that one to yourself, because do you know why I think we’re having those elections and that they’ll go terribly for the Trump administration? By the increasingly frantic actions being taken to try to mess around with them. It's the same as Trump’s flailing, incoherent attempts to deal with the Jeffrey Epstein material: these are not the actions of someone who is unconcerned. So keep up the pressure. Speak up and give others the courage to do the same. Flip those seats. Drive more pro-Trump legislators into retirement (or anti-Trump ones who are too old or timid for the job!) Push those approval ratings even lower to give the cowards and toadies reasons to doubt their choices and their futures. Sow discord among the various Republican factions who are already beginning to fight it out for their positions in a post-Trump era. We have to make that post-Trump era happen, and the sooner we get started on it, the better. And let’s start talking about all the things that will have to be done to make this more of a functioning country again when we get there. In years to come, I will never be able to look back on 2025 as anything other than a disaster. But I want to be able to look back on 2026 as when we started to turn things around.
Show full content
Happy New Year! Science blogging resumes on Monday. But I have to get this off my chest first:
Let me start off this first post of 2026 by acknowledging that 2025 was, in many ways, a horrible year. I say that particularly by the subject matters of this site - biomedical research and science in general. The Trump administration moved quickly last year to attack federal research funding from every single angle they could think of, and went on to do major damage to every science-based agency. The NIH, CDC, FDA and others will need years to recover what’s been done to them.
Likewise, US public health will suffer. It already is. I note that we now have reached a level of measles cases not seen in this country for decades, and one can only expect the numbers for every other vaccine-preventable disease to rise as well. This is of course infuriating, because it is so senseless and was so avoidable, but here we are. Children will die who did not have to, and it was a choice. Just as gutting USAID was one of the early choices of the administration, an act of petty vandalism which has led (and will continue to lead) to even more deaths and suffering in the countries that used to get the aid. Never forget: whenever Elon Musk walks up to a podium to bullshit you about colonizing Mars or whatever, he is walking on the bodies of dead children. He yanked their food and medicine away while telling lies and cracking stupid jokes.
There’s plenty of blood to go around, though. I have often stated my belief that Robert F. Kennedy, Jr. has it on his hands after his actions during a measles epidemic in Samoa, and now he will get a chance to replicate that disaster on a grander scale, as befits his self-image. Kennedy, our HHS secretary, believes that the 1918 flu pandemic and the 1980s rise of HIB somehow happened because of vaccine research. Marty Makary, our head of the FDA, believes that Lyme disease is an escape biowarfare agent. Jay Bhattacharya, among other things, believes that he is fit to lead the NIH. We are being governed by grinning idiots and sociopaths, and any assessment of our situation in 2026 has to take that as a precondition.
This of course goes all the way to the top. Donald Trump is (by a wide margin) the worst president this country has ever had, and his administration is a catastrophe. I will not be discussing my reasons here for those assertions - suffice it to say that I could do a solid 50,000 words on this topic with copious links, citations, and video clips. But to what point? What I think is more important is that any assessment of our situation also has to come to terms with the fact that he was nonetheless re-elected. I am very much cheered by his cratering approval ratings and by the increasing numbers of people speaking out against his policies, but (as with the vaccine situation), all this was avoidable. Dwelling on that, though, can be a mistake.
Just saying “This didn’t have to happen” avoids thinking about why it did, and risks not realizing what bad shape our institutions and our civil society had to be in for things to get to this point. Those institutions very much include the House of Representatives, the Senate, and the Supreme Court. There are multiple points of failure here, and the irony is not lost on me that we are heading into the 250th anniversary of this country’s founding while surrounded by horrible evidence that its system of government is in terrible need of repair. But that’s what I’m dedicating this year to.
How can I do otherwise? Despair is a sin. And I’m not going to let this convicted felon and adjudicated sexual predator, this incompetent would-be-authoritarian disgrace of a human being run me out of my own country. I am going to continue to speak out against the harmful acts of this administration and I am going to continue to work to replace it with something better. This year, that means the mid-term elections. I always get some fashionably cynical replies to this sort of statement, stuff like “Oh you poor little lamb, you still think there will be mid-term elections”. Keep that one to yourself, because do you know why I think we’re having those elections and that they’ll go terribly for the Trump administration? By the increasingly frantic actions being taken to try to mess around with them. It's the same as Trump’s flailing, incoherent attempts to deal with the Jeffrey Epstein material: these are not the actions of someone who is unconcerned.
So keep up the pressure. Speak up and give others the courage to do the same. Flip those seats. Drive more pro-Trump legislators into retirement (or anti-Trump ones who are too old or timid for the job!) Push those approval ratings even lower to give the cowards and toadies reasons to doubt their choices and their futures. Sow discord among the various Republican factions who are already beginning to fight it out for their positions in a post-Trump era. We have to make that post-Trump era happen, and the sooner we get started on it, the better. And let’s start talking about all the things that will have to be done to make this more of a functioning country again when we get there.
In years to come, I will never be able to look back on 2025 as anything other than a disaster. But I want to be able to look back on 2026 as when we started to turn things around.
Actual scientific work is definitely slowing down around here as the holidays approach. As I often do, I wanted to (re)post some recipes of foods that go well with the season (or perhaps with cold weather in general). I've rounded up a few new photos of the results as well! All of these are synthesized regularly here at Pipeline HQ, and I can personally vouch for all of them after long experience. If you'd like to see the whole range of recipes posted here over the years, this page (and the ones following it) should bring them up! First off, some desserts. I think the first recipe I posted here (some years ago!) was for chocolate pecan pie, and I still get emails about it. Here's a canonical version, and the result of this year's repeat is shown at left. This one got a bit dark on the top, but it was (like all the others) consumed vigorously. In recent years I've added the recipe for cranberry-lime pie, also shown at right, which has a pH-driven color shift midway through the preparation. The chocolate pecan pie accomplishes many things, but it does not change color. I have also made a batch of these gingersnaps, which get takers both in the earlier soft-cookie part of their lifespan and in the "snappier" phase later on. That's a shot of today's batch, actually. This next recipe may not sound so much like cold-weather food, because it depends on summer blackberries. But the great thing about cobblers is that they work identically with fresh or with frozen berries, and I have a weighed one-cobbler-quantity bag of blackberries that my wife and I picked last September waiting for the right moment. A commercial bag of them will do the trick, too, and is a lot better than going without blackberry cobbler at all. A batch with biscuit-style topping is shown at right. If you're not put off by the idea of a more summery dessert in December (apologies to the Southern Hemisphere folks; you're ready to roll by now!) then you could also try the key lime pie, or even the lime sorbet. I'm not quite ready for that second one myself this time of the year, but the only lime sorbet I've had that competes with it is from Berthillon in Paris (no worries, they have me beaten on everything else!) If it's your sort of thing you have to give it a try at some point, because it's unstoppable. Now to some good stuff to eat before you get to the desserts! A classic dish for cold temperatures is French onion soup, and that recipe is my take on it. A recent effort is shown at left. As many of you no doubt already realize, a lot of online recipes grievously misstate the amount of time that it takes the caramelize the onions for this one. Now, this is not an all-day lashed-to-the-stove job, but neither will it be done in the breezy ten or fifteen minutes that some people insist is enough. But if you have some beef stock and chicken stock (or can buy some!) that is really the only labor-intensive part of the whole thing. Another good cold-weather dish is chicken paprikash, and that recipe makes appearances around here in the wintertime. With noodles and some sour cream, it's pretty hearty stuff. Similarly robust is this chicken pot pie, shown at right, which differs from its commercial counterparts in many ways, the first being that it contains no potatoes whatsoever. Staying with the chicken theme, we made a batch of this chicken-noodle-and-corn soup here just the other night, and I've been making it for over thirty years and can testify to its healing powers. The original recipe, being Pennsylvania Dutch, has saffron in it, but in our household saffron is reserved for the Iranian side of the menu (!) So if you try it that way, let me know. Here's a beef dish that takes a bit of preparation, but odds are you've never had it before, since it's mostly seen in Germany and adjacent areas: rindsrouladen. This one I haven't made in a while, but it always reminds me of my father's cooking, since he made once in a while while my brother and I were growing up. During my post-doc in Germany I ordered it and found that the local version was identical in all ways with his preparation, which I was glad to report back to him! Here's a side dish that (since I'm from Arkansas) I consider to go with most anything: cornbread (show at left, in a prep from the extensive Pipeline kitchens). You can (as the recipe indicates) add corn kernals, chopped onion, diced green chiles, cheese, or all sorts of other stuff to take it in a savory direction. Or you can make it plain and have it with whatever jam or preserves you wish! This is a Southern-style cornbread, that is to say Not Sweet And Cakey, so if you want sweetness you'll have to bring it along yourself at serving time. OK, that should be enough to get us all through a few days, at least. I'll be pretty quiet here this week, but blogging intermittently until the beginning of 2026. I'd like to wish everyone a happy and healthy holiday season, whichever ones you might celebrate. And if your celebration just includes whipping up some good food and having some people sit down and eat it with you, then consider yourself fortunate and enjoy. That's what I do, and I try never to forget the "fortunate" part.
Show full content
Actual scientific work is definitely slowing down around here as the holidays approach. As I often do, I wanted to (re)post some recipes of foods that go well with the season (or perhaps with cold weather in general). I've rounded up a few new photos of the results as well! All of these are synthesized regularly here at Pipeline HQ, and I can personally vouch for all of them after long experience. If you'd like to see the whole range of recipes posted here over the years, this page (and the ones following it) should bring them up!
First off, some desserts. I think the first recipe I posted here (some years ago!) was for chocolate pecan pie, and I still get emails about it. Here's a canonical version, and the result of this year's repeat is shown at left. This one got a bit dark on the top, but it was (like all the others) consumed vigorously. In recent years I've added the recipe for cranberry-lime pie, also shown at right, which has a pH-driven color shift midway through the preparation. The chocolate pecan pie accomplishes many things, but it does not change color.
I have also made a batch of these gingersnaps, which get takers both in the earlier soft-cookie part of their lifespan and in the "snappier" phase later on. That's a shot of today's batch, actually.
This next recipe may not sound so much like cold-weather food, because it depends on summer blackberries. But the great thing about cobblers is that they work identically with fresh or with frozen berries, and I have a weighed one-cobbler-quantity bag of blackberries that my wife and I picked last September waiting for the right moment. A commercial bag of them will do the trick, too, and is a lot better than going without blackberry cobbler at all. A batch with biscuit-style topping is shown at right.
If you're not put off by the idea of a more summery dessert in December (apologies to the Southern Hemisphere folks; you're ready to roll by now!) then you could also try the key lime pie, or even the lime sorbet. I'm not quite ready for that second one myself this time of the year, but the only lime sorbet I've had that competes with it is from Berthillon in Paris (no worries, they have me beaten on everything else!) If it's your sort of thing you have to give it a try at some point, because it's unstoppable.
Now to some good stuff to eat before you get to the desserts! A classic dish for cold temperatures is French onion soup, and that recipe is my take on it. A recent effort is shown at left. As many of you no doubt already realize, a lot of online recipes grievously misstate the amount of time that it takes the caramelize the onions for this one. Now, this is not an all-day lashed-to-the-stove job, but neither will it be done in the breezy ten or fifteen minutes that some people insist is enough. But if you have some beef stock and chicken stock (or can buy some!) that is really the only labor-intensive part of the whole thing.
Another good cold-weather dish is chicken paprikash, and that recipe makes appearances around here in the wintertime. With noodles and some sour cream, it's pretty hearty stuff. Similarly robust is this chicken pot pie, shown at right, which differs from its commercial counterparts in many ways, the first being that it contains no potatoes whatsoever.
Staying with the chicken theme, we made a batch of this chicken-noodle-and-corn soup here just the other night, and I've been making it for over thirty years and can testify to its healing powers. The original recipe, being Pennsylvania Dutch, has saffron in it, but in our household saffron is reserved for the Iranian side of the menu (!) So if you try it that way, let me know.
Here's a beef dish that takes a bit of preparation, but odds are you've never had it before, since it's mostly seen in Germany and adjacent areas: rindsrouladen. This one I haven't made in a while, but it always reminds me of my father's cooking, since he made once in a while while my brother and I were growing up. During my post-doc in Germany I ordered it and found that the local version was identical in all ways with his preparation, which I was glad to report back to him!
Here's a side dish that (since I'm from Arkansas) I consider to go with most anything: cornbread (show at left, in a prep from the extensive Pipeline kitchens). You can (as the recipe indicates) add corn kernals, chopped onion, diced green chiles, cheese, or all sorts of other stuff to take it in a savory direction. Or you can make it plain and have it with whatever jam or preserves you wish! This is a Southern-style cornbread, that is to say Not Sweet And Cakey, so if you want sweetness you'll have to bring it along yourself at serving time.
OK, that should be enough to get us all through a few days, at least. I'll be pretty quiet here this week, but blogging intermittently until the beginning of 2026. I'd like to wish everyone a happy and healthy holiday season, whichever ones you might celebrate. And if your celebration just includes whipping up some good food and having some people sit down and eat it with you, then consider yourself fortunate and enjoy. That's what I do, and I try never to forget the "fortunate" part.
One of the constant themes of cell biology, chemical biology, and drug discovery is trying to find out where things are in cells. It isn’t easy! You can’t just pick out a small molecule, a protein, a lipid, a polysaccharide or what have you and chase it around a living cell with some sort of microscope, because there’s generally either no good way to detect such things specifically or certainly no way to do it on the small scale required. It’s also difficult to just reach into cells and pluck out specific structures or regions and comprehensively inventory them, although over the years we’ve become better and better at it - for some specific cellular fractions and some specific classes of compounds, to be sure. But the most widely uses methods continue to rely on imaging of cells in the presence of some sort of fluorescent system that causes things to light up so we can finally see them. As the practitioners of these arts well know, this is a tricky business, because it’s all too easy to change the properties of what you’re trying to study by hanging a nice bright fluorophore group off of it. Function, transport, eventual localization, half-life and stability: all of these can be messed up in unpredictable ways, so every time you get fluorescent labeling data it’s good scientific discipline to ask yourself how it might be misleading, and what you can do to cross-check the numbers. There are a lot of ingenious ways to deal with these problems, and this recent paper illustrates one of them. The authors are trying to answer a difficult but important question: what’s inside mitochondria, anyway? Those little guys are of course crucial for life, and we know of a great many disorders involving them that we would like to understand better (and to fix). But taking a roll call inside them is very much nontrivial. Part of the problem is that there are many proteins (and other substances) that have fractions both inside and outside the microchondria, with no firm guarantees that they’re doing the same things in both compartments. Another headache is that the mitochondria themselves have some pretty specific zones - the mitochondrial outer membrane, the inner membrane, the distinct space between those two, and the matrix on the inside once you’re past all those other zones. There are all sorts of transport proteins policing these layers, moving stuff in both directions under different cellular conditions, so it’s a complex and dynamic environment. In some cases, you can look at protein sequences and find “targeting sequences”, short stretches of amino acids that interact with such transport proteins to help sort things out. Previous work by these authors, though, had shown that there are plenty of proteins that are localized to the mitochondria that don’t have any of the known targeting sequences, along with plenty of proteins that do seem to have them that don’t seem to be imported into mitochondria! So you can only get so far by looking for these localization sequences. This most recent paper uses “split-reporter” technology, which takes advantage of a property that some proteins have where they can be basically broken into two pieces that are capable of reassembly into a functional whole under cellular conditions. Here they do this with a Green Fluorescent Protein, and GFPs are of course absolutely workhorses of chemical and cellular biology research for their detectability. This team breaks the GFP protein into a large fragment and a small one, and takes advantage of the way that mitochondria have their own localized DNA (a remnant of the long-ago time when they were free-living organisms, before becoming symbiotic with other cells). DNA coding for the large GFP piece is inserted into the mitochondrial sequences, while the proteins under investigation all get tagged with the small piece. By this method, the only green fluorescence will be when one of those proteins shows up inside the mitochondria, where it will recombine into a functional GFP and shine a green beacon at its location. Since it’s the days of modern chemical biology, the authors did this on the entire yeast proteome, tagging everything with the small-GFP sequence. Meanwhile they generated a yeast strain that had the “big GFP” incorporated into its mitochondria. Taking the all-labeled-protein cells and depleting them of their mitochondrial DNA, then immediately allowing them to mate with the big-GFP strain produced a generation that had both sides of the split-GFP system in the same yeast cells. They then imaged these under a wide variety of growth conditions (different food sources and different sort of cellular stress). Interestingly, those food-and-stress variation didn’t seem to yield much - the mitochondrial results all looked pretty similar without signs of lots of different proteins being imported under different conditions. And a lot of the proteins that showed up were already known to be mitochondrially localized (which is a good thing - if those hadn’t shown up it would have set off alarm bells that something was wrong with the method in general). But they were able to pick out 56 new proteins that had never been imaged in mitochondria before, 39 of which also show up in other regions of the cell (that problem mentioned earlier). Eight of them were never known to be associated with mitochondria at all. As the authors note, their C-terminal tagging is certainly capable of missing some proteins, although it’s hard to say how many. The method seems to have good sensitivity, so they’re more worried about misses due to problems with the labeling than about misses due to low abundance, at least for now. One weird thing is that five of the eight new annotations turn out to be ribosomal proteins, and you wouldn’t expect those to have any real functions inside mitochondria (!) Their properties, to be sure, do seem to be more like other mitochondrial proteins than your average nonmitochondrial ones, so there could well be something new going on here, but who knows what! More experiment will be needed to figure that out. But one of the other results does inspire confidence: a protein called Gpp1 was found in the mitochondrial set that no one had ever annotated that way before, and the authors confirmed that there area actually two different transcriptional isoforms of it (a fact not appreciated before) and that one goes to the cytosol and one to the mitochondria. Why, and what it’s doing there, will be the subject of future work! But overall, this paper illustrates the amount of work that had to go into this localization research, and the amount of cross-checking needed to make sure that you can believe your own results. People are chipping away at this sort of thing all over the cell and adding to our knowledge all the time - because, as I like to point out, you certainly can’t go ask an AI model about these things and get any meaningful answers. . .
Show full content
One of the constant themes of cell biology, chemical biology, and drug discovery is trying to find out where things are in cells. It isn’t easy! You can’t just pick out a small molecule, a protein, a lipid, a polysaccharide or what have you and chase it around a living cell with some sort of microscope, because there’s generally either no good way to detect such things specifically or certainly no way to do it on the small scale required. It’s also difficult to just reach into cells and pluck out specific structures or regions and comprehensively inventory them, although over the years we’ve become better and better at it - for some specific cellular fractions and some specific classes of compounds, to be sure.
But the most widely uses methods continue to rely on imaging of cells in the presence of some sort of fluorescent system that causes things to light up so we can finally see them. As the practitioners of these arts well know, this is a tricky business, because it’s all too easy to change the properties of what you’re trying to study by hanging a nice bright fluorophore group off of it. Function, transport, eventual localization, half-life and stability: all of these can be messed up in unpredictable ways, so every time you get fluorescent labeling data it’s good scientific discipline to ask yourself how it might be misleading, and what you can do to cross-check the numbers.
There are a lot of ingenious ways to deal with these problems, and this recent paper illustrates one of them. The authors are trying to answer a difficult but important question: what’s inside mitochondria, anyway? Those little guys are of course crucial for life, and we know of a great many disorders involving them that we would like to understand better (and to fix). But taking a roll call inside them is very much nontrivial. Part of the problem is that there are many proteins (and other substances) that have fractions both inside and outside the microchondria, with no firm guarantees that they’re doing the same things in both compartments. Another headache is that the mitochondria themselves have some pretty specific zones - the mitochondrial outer membrane, the inner membrane, the distinct space between those two, and the matrix on the inside once you’re past all those other zones. There are all sorts of transport proteins policing these layers, moving stuff in both directions under different cellular conditions, so it’s a complex and dynamic environment.
In some cases, you can look at protein sequences and find “targeting sequences”, short stretches of amino acids that interact with such transport proteins to help sort things out. Previous work by these authors, though, had shown that there are plenty of proteins that are localized to the mitochondria that don’t have any of the known targeting sequences, along with plenty of proteins that do seem to have them that don’t seem to be imported into mitochondria! So you can only get so far by looking for these localization sequences.
This most recent paper uses “split-reporter” technology, which takes advantage of a property that some proteins have where they can be basically broken into two pieces that are capable of reassembly into a functional whole under cellular conditions. Here they do this with a Green Fluorescent Protein, and GFPs are of course absolutely workhorses of chemical and cellular biology research for their detectability. This team breaks the GFP protein into a large fragment and a small one, and takes advantage of the way that mitochondria have their own localized DNA (a remnant of the long-ago time when they were free-living organisms, before becoming symbiotic with other cells). DNA coding for the large GFP piece is inserted into the mitochondrial sequences, while the proteins under investigation all get tagged with the small piece. By this method, the only green fluorescence will be when one of those proteins shows up inside the mitochondria, where it will recombine into a functional GFP and shine a green beacon at its location.
Since it’s the days of modern chemical biology, the authors did this on the entire yeast proteome, tagging everything with the small-GFP sequence. Meanwhile they generated a yeast strain that had the “big GFP” incorporated into its mitochondria. Taking the all-labeled-protein cells and depleting them of their mitochondrial DNA, then immediately allowing them to mate with the big-GFP strain produced a generation that had both sides of the split-GFP system in the same yeast cells. They then imaged these under a wide variety of growth conditions (different food sources and different sort of cellular stress).
Interestingly, those food-and-stress variation didn’t seem to yield much - the mitochondrial results all looked pretty similar without signs of lots of different proteins being imported under different conditions. And a lot of the proteins that showed up were already known to be mitochondrially localized (which is a good thing - if those hadn’t shown up it would have set off alarm bells that something was wrong with the method in general). But they were able to pick out 56 new proteins that had never been imaged in mitochondria before, 39 of which also show up in other regions of the cell (that problem mentioned earlier). Eight of them were never known to be associated with mitochondria at all.
As the authors note, their C-terminal tagging is certainly capable of missing some proteins, although it’s hard to say how many. The method seems to have good sensitivity, so they’re more worried about misses due to problems with the labeling than about misses due to low abundance, at least for now. One weird thing is that five of the eight new annotations turn out to be ribosomal proteins, and you wouldn’t expect those to have any real functions inside mitochondria (!) Their properties, to be sure, do seem to be more like other mitochondrial proteins than your average nonmitochondrial ones, so there could well be something new going on here, but who knows what! More experiment will be needed to figure that out. But one of the other results does inspire confidence: a protein called Gpp1 was found in the mitochondrial set that no one had ever annotated that way before, and the authors confirmed that there area actually two different transcriptional isoforms of it (a fact not appreciated before) and that one goes to the cytosol and one to the mitochondria. Why, and what it’s doing there, will be the subject of future work!
But overall, this paper illustrates the amount of work that had to go into this localization research, and the amount of cross-checking needed to make sure that you can believe your own results. People are chipping away at this sort of thing all over the cell and adding to our knowledge all the time - because, as I like to point out, you certainly can’t go ask an AI model about these things and get any meaningful answers. . .
I’ve kicked around the idea of doing some “Where Are They Now” posts here, and I think that this is as good an occasion as any. Back in 2018, I wrote about a new company called Verge Genomics that was using AI/ML methods to go after neuroscience targets like ALS, and my eye was caught by statements like “Instead of tediously screening millions of drugs, the algorithm will computationally predict drugs that work” and “We've discovered a way to map out the hundreds of genes that cause a disease, and then find drugs that target all the genes at once” As you’ll see from that post, I was indeed skeptical, not least because connecting genomics with small-molecule drug discovery is hard and connecting genomics with neurodegenerative disease is really hard, too. So multiplying those two together did not, to me, seem to give much room for optimism. But optimism was definitely what Verge was selling back then, and I did wish them luck, with the proviso that they were certainly going to need it. That luck has run out. Endpoints reports that the company’s only clinical candidate in ALS (VRG50635) has failed in a Phase Ib trial. That was a PIKfyve inhibitor, targeting a phosphoinositide kinase with a great many cellular functions. It’s been of interest in oncology, infectious disease, diabetes, and several other conditions besides, and Verge helped identify it as an ALS target as well. Work has been going on for some years around the industry on inhibitors of that enzyme (which include apilimod, referenced here during the pandemic). Another small company, Acurastem, is continuing their own work on PIVfyve and ALS, and I can only wish them better fortune. Verge, though, is now shutting down all their own drug discovery efforts and switching (they hope) to a partnership mode with their technology. It appears from the article that a lot of the company’s management has left in the last few months, and I’m not sure how many scientists are still at the company at this point. It looks like they’ve auctioned off all of the lab hardware, right down to the water purification equipment. And that news (from October) is particularly grim when you contrast it with this publicity piece from the Chinese news site 36Kr. That one came out about a week before the equipment auction started, but you’d never know that anything like that was in the works - it’s all dynamic leadership, paradigm-busting science, lists of accomplishments and milestones and partnerships and funding. In that PR-drenched world, VRG50635 is “bringing new hope to millions of patients around the world”. The Chinese-intensive focus of the 36Kr piece goes out of its way to emphasize that Verge is not slowing itself down by trying to do only partnerships - no, they’re going it alone and making “world-class original breakthroughs”. You can’t believe everything you read, folks. Especially not gauzy publicity writeups.
Show full content
I’ve kicked around the idea of doing some “Where Are They Now” posts here, and I think that this is as good an occasion as any. Back in 2018, I wrote about a new company called Verge Genomics that was using AI/ML methods to go after neuroscience targets like ALS, and my eye was caught by statements like “Instead of tediously screening millions of drugs, the algorithm will computationally predict drugs that work” and “We've discovered a way to map out the hundreds of genes that cause a disease, and then find drugs that target all the genes at once”
As you’ll see from that post, I was indeed skeptical, not least because connecting genomics with small-molecule drug discovery is hard and connecting genomics with neurodegenerative disease is really hard, too. So multiplying those two together did not, to me, seem to give much room for optimism. But optimism was definitely what Verge was selling back then, and I did wish them luck, with the proviso that they were certainly going to need it.
That luck has run out. Endpointsreports that the company’s only clinical candidate in ALS (VRG50635) has failed in a Phase Ib trial. That was a PIKfyve inhibitor, targeting a phosphoinositide kinase with a great many cellular functions. It’s been of interest in oncology, infectious disease, diabetes, and several other conditions besides, and Verge helped identify it as an ALS target as well. Work has been going on for some years around the industry on inhibitors of that enzyme (which include apilimod, referenced here during the pandemic). Another small company, Acurastem, is continuing their own work on PIVfyve and ALS, and I can only wish them better fortune.
Verge, though, is now shutting down all their own drug discovery efforts and switching (they hope) to a partnership mode with their technology. It appears from the article that a lot of the company’s management has left in the last few months, and I’m not sure how many scientists are still at the company at this point. It looks like they’ve auctioned off all of the lab hardware, right down to the water purification equipment.
And that news (from October) is particularly grim when you contrast it with this publicity piece from the Chinese news site 36Kr. That one came out about a week before the equipment auction started, but you’d never know that anything like that was in the works - it’s all dynamic leadership, paradigm-busting science, lists of accomplishments and milestones and partnerships and funding. In that PR-drenched world, VRG50635 is “bringing new hope to millions of patients around the world”. The Chinese-intensive focus of the 36Kr piece goes out of its way to emphasize that Verge is not slowing itself down by trying to do only partnerships - no, they’re going it alone and making “world-class original breakthroughs”. You can’t believe everything you read, folks. Especially not gauzy publicity writeups.
Here’s a neat look at microbial natural products from a chemical diversity standpoint. Of course, natural products have a fearsome (and well-earned) reputation for displaying structures that we humans would never have gotten around to making - or even thinking of - but once you get that internalized, there are some interesting patterns and lessons. The author (Roger Linington at Simon Fraser Univ.) is also looking at two trends that at first seem to be at odds with each other: the bulk of newly described natural products (from microorganisms here, but I believe that this applies to other sources as well) are variations on scaffolds that have already been described. But at the same time, genomics studies suggest that there must be many more as-yet-unseen structures waiting for us out there. Perhaps these are produced at very low levels, or only under certain conditions (the various explanations for “cryptic natural products”). But there are a great many more apparent biosynthetic gene clusters out there than there are actual natural products, no matter what. If you look through the whole Natural Product Atlas database in terms of chemical similarity, you find under this paper’s (reasonable) criteria that there the 36,454 compounds therein generate 4,148 clusters of two or more compounds, and those account for about 30,000 of the total (meaning that about 6,000 of them are actually singletons with no close structural relatives, or not yet). About 1200 of the clusters have five or more members; the median is 3. Inside a given cluster, even the ones with a large number of members, it’s very much “variations on a theme”, with comparatively minor changes. But those known variations are almost always just a tiny fraction of what would be possible, and the natural products whose biosynthetic pathways are known illustrate that. The polyketides, for example, are a well-represented group in the Atlas, with 312 16-membered macrolactones (for example). But that’s nothing compared to the hundreds of thousands of possible and easily plausible structures that could exist. As the paper notes, though, the number of natural products that we actually see is constrained not just by what is possible, but what is retained in nature (presumably because it’s been found to be useful). And they are of course also constrained by how we isolate and identify them, including where we look, the extraction and purification procedures we use, and our ability to decipher their structures. There are some interesting trends over time. That percentage of singleton compounds has remained roughly the same over the last fifty years, although the percentage of molecules that were singletons at the time of their discovery has slowly gone down. That last effect suggests that by now we have discovered a good percentage of the microbial natural product scaffolds that we are equipped to find. But we also have to remember that many natural products are discovered as part of a small group right from the start - and the number that have remained singletons for decades stands as a reminder of what natural product diversity is capable of. So there’s still a lot out there, but by this time we might well be generating even more natural-product-ish diversity on our own by synthesis - that is, exploring that huge chemical space that seems to be unused in nature. We will probably find that some of is unused for good reason, because it doesn’t seem to do much, but the flip side of that argument is that we have uses for things that nature has never needed. I’ve lost track of the number of times I’ve seen a declaration of a revival in interest in natural products (and this paper adds another such to the pile!) What’s for sure, though, is that they have never gone away, and they’re a long way from ever doing so. They aren’t as central to drug discovery as they were decades (centuries!) ago, but irrelevance is nowhere in sight.
Show full content
Here’s a neat look at microbial natural products from a chemical diversity standpoint. Of course, natural products have a fearsome (and well-earned) reputation for displaying structures that we humans would never have gotten around to making - or even thinking of - but once you get that internalized, there are some interesting patterns and lessons.
The author (Roger Linington at Simon Fraser Univ.) is also looking at two trends that at first seem to be at odds with each other: the bulk of newly described natural products (from microorganisms here, but I believe that this applies to other sources as well) are variations on scaffolds that have already been described. But at the same time, genomics studies suggest that there must be many more as-yet-unseen structures waiting for us out there. Perhaps these are produced at very low levels, or only under certain conditions (the various explanations for “cryptic natural products”). But there are a great many more apparent biosynthetic gene clusters out there than there are actual natural products, no matter what.
If you look through the whole Natural Product Atlas database in terms of chemical similarity, you find under this paper’s (reasonable) criteria that there the 36,454 compounds therein generate 4,148 clusters of two or more compounds, and those account for about 30,000 of the total (meaning that about 6,000 of them are actually singletons with no close structural relatives, or not yet). About 1200 of the clusters have five or more members; the median is 3. Inside a given cluster, even the ones with a large number of members, it’s very much “variations on a theme”, with comparatively minor changes.
But those known variations are almost always just a tiny fraction of what would be possible, and the natural products whose biosynthetic pathways are known illustrate that. The polyketides, for example, are a well-represented group in the Atlas, with 312 16-membered macrolactones (for example). But that’s nothing compared to the hundreds of thousands of possible and easily plausible structures that could exist. As the paper notes, though, the number of natural products that we actually see is constrained not just by what is possible, but what is retained in nature (presumably because it’s been found to be useful). And they are of course also constrained by how we isolate and identify them, including where we look, the extraction and purification procedures we use, and our ability to decipher their structures.
There are some interesting trends over time. That percentage of singleton compounds has remained roughly the same over the last fifty years, although the percentage of molecules that were singletons at the time of their discovery has slowly gone down. That last effect suggests that by now we have discovered a good percentage of the microbial natural product scaffolds that we are equipped to find. But we also have to remember that many natural products are discovered as part of a small group right from the start - and the number that have remained singletons for decades stands as a reminder of what natural product diversity is capable of.
So there’s still a lot out there, but by this time we might well be generating even more natural-product-ish diversity on our own by synthesis - that is, exploring that huge chemical space that seems to be unused in nature. We will probably find that some of is unused for good reason, because it doesn’t seem to do much, but the flip side of that argument is that we have uses for things that nature has never needed.
I’ve lost track of the number of times I’ve seen a declaration of a revival in interest in natural products (and this paper adds another such to the pile!) What’s for sure, though, is that they have never gone away, and they’re a long way from ever doing so. They aren’t as central to drug discovery as they were decades (centuries!) ago, but irrelevance is nowhere in sight.
We have been making small-molecule inhibitors of kinase enzymes for quite a while now in medicinal chemistry, and I would not even want to guess how many such compounds have been described in the literature. As an aside, this is usually the point where someone who’s been around as long as I have recalls that there was a time when people thought - not without some reason - that making such selective kinase inhibitors might not be possible at all. After all, you’re just hitting an ATP binding site in all these enzymes, and y’know, how different can they be from each other? It didn’t help that some of the early kinase inhibitors were indeed blunderbuss compounds that inhibited a wide range of the enzymes, but in the end, getting selectivity (while not trivial) was generally possible, at least to useful degrees. And in more recent years we have been spending a lot of time and effort in this business making targeted protein degrader molecules, hijacking the normal cell pathways that lead to protein turnover to send specific proteins to the shredder well before their time for our own reasons. I’ve written many times about this here, and will doubtless write many more, because this is still very much a work in progress with a lot of unknowns. But the overall idea of being able to make proteins basically disappear rather than just clogging up one of their binding sites is very interesting indeed. Those two things might be more related than we thought, though! This new paper builds on some scattered reports over the years that some kinase enzymes seem to have shorter half-lives in cells once they’ve been exposed to small-molecule inhibitors. The authors studied this pretty systematically and found that this is indeed the case: looking at 98 different kinase enzymes with 1,570 inhibitors turned up 160 example of “inhibitor-induced destabilization”. It’s important to make a mental distinction here: many proteins (kinase enzymes included) are thermally stabilized by binding to ligands. That’s the basis for a whole suite of assays that measure such stability changes with temperature, and often enough (although not invariably!) tighter-binding ligands make such proteins more and more resistant to denaturation/unfolding as the temperature goes up. But that sort of thermodynamic stability is not what we’re talking about in this new case - this is stability with reference to the cellular environment. That makes it important to figure out what the basis for this effect is. In some (perhaps many) cases it may be that inhibitor binding changes the behavior of these kinases in their binding to chaperone proteins. Those are pretty ubiquitous in living systems, with some of the best-known being the “HSP” (heat shock protein) family that are upregulated under thermal stress to help keep things together. But even without stress there’s a lot of proteinaceous hand-holding going on to keep things properly folded and in the right conformations. Disturbing that could well be expected to lead to more vulnerability to the ubiquitination/hauling off to the proteasome pathways. But the authors show that that’s not always the explanation. Sometimes it appears that inhibitor binding just puts kinases into conformations that are better targets for ubiquitination and degradation, and you’d have to think that there are evolutionary reasons for such behavior. Kinases (at least so far) seem to be more susceptible to this sort of behavior than other proteins, and their central role in maintaining or altering the phosphorylation state of other proteins makes them good candidates for layer upon layer of overlapping regulatory pathways. So this might be another one of them! The paper also shows an example with the BLK enzyme where binding to an inhibitor sets off vulnerability to the gamma-secretase enzyme, which cleaves off a myristoyl group that is crucial for BLK’s membrane localization. Without this anchor it floats off into the cytoplasm and it’s apparently intrinsically unstable out there and gets cleared. So there are several pathways at work. All of these modes of action - and who’s to say that there aren’t more - lead to ideas for pharmaceutical intervention. If we’ve inadvertently been changing the cellular stability of some enzymes by inhibiting them, then we should probably get a grip on that process so we can exploit it on purpose. Chaperone dissociation, exposing new protein surfaces to degradation tags, changing cellular localization - these are all pathways that we could potentially use to our advantage once we get a better idea of how they work and which proteins are most vulnerable. There’s always more to do!
Show full content
We have been making small-molecule inhibitors of kinase enzymes for quite a while now in medicinal chemistry, and I would not even want to guess how many such compounds have been described in the literature. As an aside, this is usually the point where someone who’s been around as long as I have recalls that there was a time when people thought - not without some reason - that making such selective kinase inhibitors might not be possible at all. After all, you’re just hitting an ATP binding site in all these enzymes, and y’know, how different can they be from each other? It didn’t help that some of the early kinase inhibitors were indeed blunderbuss compounds that inhibited a wide range of the enzymes, but in the end, getting selectivity (while not trivial) was generally possible, at least to useful degrees.
And in more recent years we have been spending a lot of time and effort in this business making targeted protein degrader molecules, hijacking the normal cell pathways that lead to protein turnover to send specific proteins to the shredder well before their time for our own reasons. I’ve written many times about this here, and will doubtless write many more, because this is still very much a work in progress with a lot of unknowns. But the overall idea of being able to make proteins basically disappear rather than just clogging up one of their binding sites is very interesting indeed.
Those two things might be more related than we thought, though! This new paper builds on some scattered reports over the years that some kinase enzymes seem to have shorter half-lives in cells once they’ve been exposed to small-molecule inhibitors. The authors studied this pretty systematically and found that this is indeed the case: looking at 98 different kinase enzymes with 1,570 inhibitors turned up 160 example of “inhibitor-induced destabilization”.
It’s important to make a mental distinction here: many proteins (kinase enzymes included) are thermally stabilized by binding to ligands. That’s the basis for a whole suite of assays that measure such stability changes with temperature, and often enough (although not invariably!) tighter-binding ligands make such proteins more and more resistant to denaturation/unfolding as the temperature goes up. But that sort of thermodynamic stability is not what we’re talking about in this new case - this is stability with reference to the cellular environment.
That makes it important to figure out what the basis for this effect is. In some (perhaps many) cases it may be that inhibitor binding changes the behavior of these kinases in their binding to chaperone proteins. Those are pretty ubiquitous in living systems, with some of the best-known being the “HSP” (heat shock protein) family that are upregulated under thermal stress to help keep things together. But even without stress there’s a lot of proteinaceous hand-holding going on to keep things properly folded and in the right conformations. Disturbing that could well be expected to lead to more vulnerability to the ubiquitination/hauling off to the proteasome pathways.
But the authors show that that’s not always the explanation. Sometimes it appears that inhibitor binding just puts kinases into conformations that are better targets for ubiquitination and degradation, and you’d have to think that there are evolutionary reasons for such behavior. Kinases (at least so far) seem to be more susceptible to this sort of behavior than other proteins, and their central role in maintaining or altering the phosphorylation state of other proteins makes them good candidates for layer upon layer of overlapping regulatory pathways. So this might be another one of them!
The paper also shows an example with the BLK enzyme where binding to an inhibitor sets off vulnerability to the gamma-secretase enzyme, which cleaves off a myristoyl group that is crucial for BLK’s membrane localization. Without this anchor it floats off into the cytoplasm and it’s apparently intrinsically unstable out there and gets cleared. So there are several pathways at work.
All of these modes of action - and who’s to say that there aren’t more - lead to ideas for pharmaceutical intervention. If we’ve inadvertently been changing the cellular stability of some enzymes by inhibiting them, then we should probably get a grip on that process so we can exploit it on purpose. Chaperone dissociation, exposing new protein surfaces to degradation tags, changing cellular localization - these are all pathways that we could potentially use to our advantage once we get a better idea of how they work and which proteins are most vulnerable. There’s always more to do!
It’s been obvious for many years now that growing antibiotic resistance is a problem, and that it could turn into a very bad one. There has been a great deal of work put into trying to understand the nature of these resistance pathways, but if you’re studying bacterial pathogens in the modern world, you’re showing up at the crime scene long after the break-in. You might be surprised to learn (I was!) that there is actually a resource of pathogenic bacteria from the pre-antibiotic era. The “Murray collection” has several hundred bacterial varieties in it, mostly from the Enterobacteriaceae and specifically a number of Escherichia, Shigella, Klebsiella, Salmonella, and Enterobacter species that were collected from 1917 to 1954 by Everitt George Dunne Murray during his long career. These were stored as cultures on agar slopes, and curation of this collection was continued by his son Robert Everitt George Murray during the mid-20th century. In the early 1980s, subcultures of all of these were transferred to the National Collection of Type Cultures in the UK, where they are still available today. This is obviously a uniquely valuable resource, and the world of infectious disease biology is indebted to the Murrays (father and son) for what must have seemed at times like a very odd use of time and effort. Over the years there have been many studies of the Murray cultures, and a number of very interesting things have been discovered - for example, it was found a few years ago that the majority of the Klebsiella strains in the collection were resistant to penicillin before penicillin even came into any widespread use. They were already prepared with beta-lactamases, presumably due to the natural occurence of such antibiotics in soils and other locations. Here’s a new paper studying the Murray strains, specifically looking at the DNA plasmids that these bacteria carry. Those are the unfortunately all-too-swappable elements that bacteria trade around, and are a primary method by which resistance spreads through a population. They find that the great majority of the Murray-ra plasmids aren’t carrying many resistance genes per se. About 23% of the old plasmids have actually never been seen again in bacterial sequence databases in the modern era, but there are some from all the way back to 1917 that are still around (in modified form) in 2020. What this team found was that modern bacterial pathogens are dominated by large plasmids that have incorporated the older ones in their sequences, with several lines of evidence suggesting that they’re the product of multiple fusion events over the years. As mentioned, the early plasmids have low levels of resistance genes scattered among them, mostly efflux pumps and a few for dealing with cationic peptides. The modern plasmids. . .are not like that. 38% of them are carrying resistance genes, often multiple copies, conferring resistance to a whole range of agents, many of them the broad-spectrum or “last resort” antibiotics, oh joy. The peak are plasmids that carry up to 40 different resistance genes, spanning a dozen different antibiotic types. The authors were able to see several broad types of plasmid in the modern samples, and it appears that those behemoth polyresistance modern plasmids are probably short-lived, with a lot of the nasty diversity lurking in smaller, more stable and persistent sequences that are mixes of old-fashioned Murray sequences (and their descendants) with modern resistance genes. But as they authors note, it’s not just the presence of resistance genes that determines that fate of all these bacterial plasmids - there are clearly evolutionary forces at work beyond just those from antibiotics, and those need to be better understood. The bacteria are nowhere near giving up all their secrets.
Show full content
It’s been obvious for many years now that growing antibiotic resistance is a problem, and that it could turn into a very bad one. There has been a great deal of work put into trying to understand the nature of these resistance pathways, but if you’re studying bacterial pathogens in the modern world, you’re showing up at the crime scene long after the break-in.
You might be surprised to learn (I was!) that there is actually a resource of pathogenic bacteria from the pre-antibiotic era. The “Murray collection” has several hundred bacterial varieties in it, mostly from the Enterobacteriaceae and specifically a number of Escherichia, Shigella, Klebsiella, Salmonella, and Enterobacter species that were collected from 1917 to 1954 by Everitt George Dunne Murray during his long career. These were stored as cultures on agar slopes, and curation of this collection was continued by his son Robert Everitt George Murray during the mid-20th century. In the early 1980s, subcultures of all of these were transferred to the National Collection of Type Cultures in the UK, where they are still available today.
This is obviously a uniquely valuable resource, and the world of infectious disease biology is indebted to the Murrays (father and son) for what must have seemed at times like a very odd use of time and effort. Over the years there have been many studies of the Murray cultures, and a number of very interesting things have been discovered - for example, it was found a few years ago that the majority of the Klebsiella strains in the collection were resistant to penicillin before penicillin even came into any widespread use. They were already prepared with beta-lactamases, presumably due to the natural occurence of such antibiotics in soils and other locations.
Here’s a new paper studying the Murray strains, specifically looking at the DNA plasmids that these bacteria carry. Those are the unfortunately all-too-swappable elements that bacteria trade around, and are a primary method by which resistance spreads through a population. They find that the great majority of the Murray-ra plasmids aren’t carrying many resistance genes per se. About 23% of the old plasmids have actually never been seen again in bacterial sequence databases in the modern era, but there are some from all the way back to 1917 that are still around (in modified form) in 2020. What this team found was that modern bacterial pathogens are dominated by large plasmids that have incorporated the older ones in their sequences, with several lines of evidence suggesting that they’re the product of multiple fusion events over the years.
As mentioned, the early plasmids have low levels of resistance genes scattered among them, mostly efflux pumps and a few for dealing with cationic peptides. The modern plasmids. . .are not like that. 38% of them are carrying resistance genes, often multiple copies, conferring resistance to a whole range of agents, many of them the broad-spectrum or “last resort” antibiotics, oh joy. The peak are plasmids that carry up to 40 different resistance genes, spanning a dozen different antibiotic types.
The authors were able to see several broad types of plasmid in the modern samples, and it appears that those behemoth polyresistance modern plasmids are probably short-lived, with a lot of the nasty diversity lurking in smaller, more stable and persistent sequences that are mixes of old-fashioned Murray sequences (and their descendants) with modern resistance genes. But as they authors note, it’s not just the presence of resistance genes that determines that fate of all these bacterial plasmids - there are clearly evolutionary forces at work beyond just those from antibiotics, and those need to be better understood. The bacteria are nowhere near giving up all their secrets.
I found this to be an interesting paper, and it uses an idea that’s not always easy to realize. There are a lot of time when we’d like to be able to use small proteins and peptides as drugs, but they often have poor pharmacokinetics (absorption, membrane penetration, and most especially metabolic liability). In addition, some of these small-protein ideas can end up being immunogenic, since your body can react to them like the foreign substances they are, with your immune system taken them as signs of some sort of viral or bacterial attack. As an aside, the current vogue for “peptides” (very loosely defined) among wellness types, bodybuilders, and other such groups is grimly amusing from a medicinal chemistry point of view. Most of these people have no actual idea of what they’re talking about, and “peptide” gets used as a synonym for “cool dietary supplement known to guru insiders” with no further thought. So asking someone if they take peptides just gets a blank stare from anyone who knows biochemistry, since your body is swimming with tens/hundreds of thousands of different short protein sequences that fit inside that name. And the number of different possible peptides, even reasonably short ones, is just beyond human comprehension. It’s not a very useful term when thrown around like this! Anyway, what the new paper above is discussing is more in the “miniprotein” class - that is, long enough to fold itself into a defined three-dimensional structure. In this case, they’re about forty amino acids long. And the authors are looking to compare hit rates against a difficult target (PD-1/PD-L1) versus shorter sequences I blogged just the other day about a quick computational approach that came up empty against this one, and it’s really no wonder - that interface features two rather flat beta-sheet surfaces that don’t give small structures very much to grab on to, as far as anyone’s been able to see. And they’re also interested in making these miniprotein candidates out of D-amino acids. Those are of course the mirror-image forms of the ones used normally in living systems, and the idea there is that a peptide or protein made out of such enantiomers will probably not be a substrate for hydrolase enzymes (and may be less immunogenic as well). But you’re not going to be able to take a sequence that you know binds to you target, synthesize it with all the amino acids flipped into the D stereochemistry, and expect it to bind. You’ve made a different compound entirely! So the D-proteins you get out of such a screening/synthesis exercise would not be expected to look much like sequences you’re seen before - you’re just using this non-natural stereochemistry to put the side chains and functional groups into the right position to hit the target, one way or another. But how do you do that? There are, as mentioned, ridiculously huge numbers of potential protein candidates available, and the best ways to produce and evaluate huge mixed libraries of them rely on living-system-derived techniques like phage display (subject of the 2018 Chemistry Nobel, I might add). But the phages and bacteria you’d use for such library preparation are of course all using the good ol’ L amino acids that we’ve stuck with for a few billion years now. Well, here’s a way out: if you can go to the trouble of making a supply of your target protein (the one that you’re screening for binders to) as an all-D protein, then you can screen against whatever normal phage libraries you want. If you identify a binder, then all you have to do is synthesize that hit out of D amino acids so it can bind the natural L-protein target! This mirror-image screening was pioneered back in the 1990s and has been used in a number of applications since for people interested in D-protein screening hits. Now, making any substantial sized protein out of all D amino acids is not enjoyable, but it’s gotten more feasible over the years with advances in organic-chemistry routes to protein synthesis. And that’s what this paper did, making a 146-amino acid construct of the PD-1 extracellular domain as the screening target. (It was assembled from seven shorter sequences by various chemical ligation reactions - not the work of a moment, but certainly doable if you have the time, skill, and patience. Oh, and the money. That too. Most of the commercial screening libraries of various-length peptides didn’t yield much of interest. But their miniprotein library, which had been targeted towards a different purpose entirely (SUMO binding) from another project, actually yielded a micromolar binder. They used that sequence to prepare a new phage library with several hundred million proteins randomized around this one. The twenty or so best survivors from four successive rounds of screening and enrichment turned out to be pretty good when tested as individual compounds - many of them were sub-micromolar and one of them came in with an IC50 of 60 nanomolar (but was about a hundredfold less potent in the cell assay, which is certainly not unheard of either, unfortunately). NMR evidence seems to confirm direct binding to the PD-1 protein, albeit without many structural details. This comes under the heading of “an interesting start”, and it certainly seems to demonstrate that miniproteins can be ligands for such difficult targets. Turning D proteins of any sort into drugs is something that hasn’t quite crossed the finish line yet, although there are two such candidates in clinical trials against other targets. It seems that the advantages of greater stability and lower immunogenicity can be realized with these things, although you certainly don’t want to take either one of those as given without proving it for yourself. There are though, an insane number of possible miniprotein scaffolds and designs, each of which can partake in the equally insane variety of protein sequences, so you’d have to think that there are drugs in there somewhere!
Show full content
I found this to be an interesting paper, and it uses an idea that’s not always easy to realize. There are a lot of time when we’d like to be able to use small proteins and peptides as drugs, but they often have poor pharmacokinetics (absorption, membrane penetration, and most especially metabolic liability). In addition, some of these small-protein ideas can end up being immunogenic, since your body can react to them like the foreign substances they are, with your immune system taken them as signs of some sort of viral or bacterial attack.
As an aside, the current vogue for “peptides” (very loosely defined) among wellness types, bodybuilders, and other such groups is grimly amusing from a medicinal chemistry point of view. Most of these people have no actual idea of what they’re talking about, and “peptide” gets used as a synonym for “cool dietary supplement known to guru insiders” with no further thought. So asking someone if they take peptides just gets a blank stare from anyone who knows biochemistry, since your body is swimming with tens/hundreds of thousands of different short protein sequences that fit inside that name. And the number of different possible peptides, even reasonably short ones, is just beyond human comprehension. It’s not a very useful term when thrown around like this!
Anyway, what the new paper above is discussing is more in the “miniprotein” class - that is, long enough to fold itself into a defined three-dimensional structure. In this case, they’re about forty amino acids long. And the authors are looking to compare hit rates against a difficult target (PD-1/PD-L1) versus shorter sequences I blogged just the other day about a quick computational approach that came up empty against this one, and it’s really no wonder - that interface features two rather flat beta-sheet surfaces that don’t give small structures very much to grab on to, as far as anyone’s been able to see.
And they’re also interested in making these miniprotein candidates out of D-amino acids. Those are of course the mirror-image forms of the ones used normally in living systems, and the idea there is that a peptide or protein made out of such enantiomers will probably not be a substrate for hydrolase enzymes (and may be less immunogenic as well). But you’re not going to be able to take a sequence that you know binds to you target, synthesize it with all the amino acids flipped into the D stereochemistry, and expect it to bind. You’ve made a different compound entirely! So the D-proteins you get out of such a screening/synthesis exercise would not be expected to look much like sequences you’re seen before - you’re just using this non-natural stereochemistry to put the side chains and functional groups into the right position to hit the target, one way or another.
But how do you do that? There are, as mentioned, ridiculously huge numbers of potential protein candidates available, and the best ways to produce and evaluate huge mixed libraries of them rely on living-system-derived techniques like phage display (subject of the 2018 Chemistry Nobel, I might add). But the phages and bacteria you’d use for such library preparation are of course all using the good ol’ L amino acids that we’ve stuck with for a few billion years now. Well, here’s a way out: if you can go to the trouble of making a supply of your target protein (the one that you’re screening for binders to) as an all-D protein, then you can screen against whatever normal phage libraries you want. If you identify a binder, then all you have to do is synthesize that hit out of D amino acids so it can bind the natural L-protein target! This mirror-image screening was pioneered back in the 1990s and has been used in a number of applications since for people interested in D-protein screening hits.
Now, making any substantial sized protein out of all D amino acids is not enjoyable, but it’s gotten more feasible over the years with advances in organic-chemistry routes to protein synthesis. And that’s what this paper did, making a 146-amino acid construct of the PD-1 extracellular domain as the screening target. (It was assembled from seven shorter sequences by various chemical ligation reactions - not the work of a moment, but certainly doable if you have the time, skill, and patience. Oh, and the money. That too.
Most of the commercial screening libraries of various-length peptides didn’t yield much of interest. But their miniprotein library, which had been targeted towards a different purpose entirely (SUMO binding) from another project, actually yielded a micromolar binder. They used that sequence to prepare a new phage library with several hundred million proteins randomized around this one. The twenty or so best survivors from four successive rounds of screening and enrichment turned out to be pretty good when tested as individual compounds - many of them were sub-micromolar and one of them came in with an IC50 of 60 nanomolar (but was about a hundredfold less potent in the cell assay, which is certainly not unheard of either, unfortunately). NMR evidence seems to confirm direct binding to the PD-1 protein, albeit without many structural details.
This comes under the heading of “an interesting start”, and it certainly seems to demonstrate that miniproteins can be ligands for such difficult targets. Turning D proteins of any sort into drugs is something that hasn’t quite crossed the finish line yet, although there are two such candidates in clinical trials against other targets. It seems that the advantages of greater stability and lower immunogenicity can be realized with these things, although you certainly don’t want to take either one of those as given without proving it for yourself. There are though, an insane number of possible miniprotein scaffolds and designs, each of which can partake in the equally insane variety of protein sequences, so you’d have to think that there are drugs in there somewhere!
I enjoyed reading this new synthetic paper, because I can still remember when I learned about the good ol’ Sandmeyer reaction in sophomore organic chemistry class and these authors are among the many people trying to replace it. There’s a reason for that, because while the Sandmeyer is definitely old, it ain’t always good. It’s a reaction that lets you take an aromatic amine (of which there are a great many) and convert that amino group into a wide variety of others, which is in theory extremely useful. It goes back to 1884 (!), and a lot of people don’t realize that its discovery was more or less accidental. Traugott Sandmeyer (you don’t run into many Traugotts these days) was trying to do an arylamine coupling, for sure, by first converting the amine into a reactive benzenediazonium chloride salt. But when he brought in his coupling partner (copper acetylide) he didn’t get a new bond between the aryl ring and the acetylene - instead, he got chlorobenzene, because the chloride counterion hopped in there instead. That’s because those aryldiazoniums are pretty lively creatures. A constant theme of the “Things I Won’t Work With” posts here is the yearning that polynitrogen compounds have to turn into nitrogen gas, and that’s just what happens here. A diazonium salt just needs the lightest push to have its two nitrogen atoms bubble away as N2 gas, leaving an extremely reactive aryl radical behind. In practice, copper salts are the classic way to run the reaction, because the copper decomposes the diazonium readily (and forms a reactive arylcopper species as an intermediate). I was taken with the reaction because of the way it was drawn in our textbook - an arylamine at the center of the diagram, with arrows radiating out from it as it transformed into an aryl chloride, aryl bromide, a phenol, an aryl nitrile, etc. It looked kind of magical, because up until then I had thought of these things are sort of separate orders of things - it was like watching some sort of amazing farm animal that could give birth to fish, frogs, or birds depending on what you fed it. I wanted to go run one right then! Variations of the reaction will bring in a fluorine or a trifluoromethyl group and allow for some other types of coupling reactions as well. So you might imagine that Sandmeyers just get used all the time, but ’tis not the case. Those aryldiazonium salts are just too touchy and hazardous, especially on scale, and especially if you have fantasies about making a big batch and storing it in a flask to turn it into whatever you want to later on. Don’t try that! The reaction can be run industrially, but only with extremely careful attention to detail, and there have been many injuries (and even fatalities) over the years, some of which are detailed here. In general, the less you handle or concentrate the diazonium salt the better off you'll be, but there are no guarantees. There have naturally been many attempts to improve on this situation, and the new paper linked above summarizes several of them. But their new method seems like it might be more general than most, assuming that your starting material can tolerate nitration conditions. That’s because you take the arylamine and make an N-nitro derivative of it, and this can be made to kick out an OH after tautomerization and make a very labile NNO derivative that leaves as gaseous nitrous oxide in a pretty direct analogy to the Sandmeyer being driven by loss of nitrogen. A big difference, though, is that all of this happens in the same pot in situ - you start with the arylamine and isolate the coupling product, with no isolation or handling of any of the intermediates. Mechanistically, this one also looks much more like an aryl cation intermediate than an aryl radical. The authors demonstrate this across a very wide range of substrates, including a lot of heterocyclic amines where the classic Sandmeyer doesn’t perform well, and they bring in halides, thiocyanates, OH, O-tosyl, O-triflate and a number of other coupling partners, including some intramolecular C-C bond forming reactions. They also demonstrate one-pot procedures to take some of these on to now-traditional metal-catalyzed couplings of many types, which gives you a lot of versatility in a synthetic sequence. If the calorimetric profiles look good, this reaction could be quite useful on a larger scale - I’m sure that there will be some process chemists taking a look at that to make sure that the nitration step and the subsequent loss of nitrous oxide don’t present possible thermodynamic hazards. On the bench scale, though, this looks a lot more doable than the Sandmeyer, which as a reaction may be headed into the History of Organic Chemistry department and leaving the Useful Laboratory Transformations one. And to be honest, it was always an uneasy member of that second group anyway!
Show full content
I enjoyed reading this new synthetic paper, because I can still remember when I learned about the good ol’ Sandmeyer reaction in sophomore organic chemistry class and these authors are among the many people trying to replace it.
There’s a reason for that, because while the Sandmeyer is definitely old, it ain’t always good. It’s a reaction that lets you take an aromatic amine (of which there are a great many) and convert that amino group into a wide variety of others, which is in theory extremely useful. It goes back to 1884 (!), and a lot of people don’t realize that its discovery was more or less accidental. Traugott Sandmeyer (you don’t run into many Traugotts these days) was trying to do an arylamine coupling, for sure, by first converting the amine into a reactive benzenediazonium chloride salt. But when he brought in his coupling partner (copper acetylide) he didn’t get a new bond between the aryl ring and the acetylene - instead, he got chlorobenzene, because the chloride counterion hopped in there instead.
That’s because those aryldiazoniums are pretty lively creatures. A constant theme of the “Things I Won’t Work With” posts here is the yearning that polynitrogen compounds have to turn into nitrogen gas, and that’s just what happens here. A diazonium salt just needs the lightest push to have its two nitrogen atoms bubble away as N2 gas, leaving an extremely reactive aryl radical behind. In practice, copper salts are the classic way to run the reaction, because the copper decomposes the diazonium readily (and forms a reactive arylcopper species as an intermediate). I was taken with the reaction because of the way it was drawn in our textbook - an arylamine at the center of the diagram, with arrows radiating out from it as it transformed into an aryl chloride, aryl bromide, a phenol, an aryl nitrile, etc. It looked kind of magical, because up until then I had thought of these things are sort of separate orders of things - it was like watching some sort of amazing farm animal that could give birth to fish, frogs, or birds depending on what you fed it. I wanted to go run one right then!
Variations of the reaction will bring in a fluorine or a trifluoromethyl group and allow for some other types of coupling reactions as well. So you might imagine that Sandmeyers just get used all the time, but ’tis not the case. Those aryldiazonium salts are just too touchy and hazardous, especially on scale, and especially if you have fantasies about making a big batch and storing it in a flask to turn it into whatever you want to later on. Don’t try that! The reaction can be run industrially, but only with extremely careful attention to detail, and there have been many injuries (and even fatalities) over the years, some of which are detailed here. In general, the less you handle or concentrate the diazonium salt the better off you'll be, but there are no guarantees.
There have naturally been many attempts to improve on this situation, and the new paper linked above summarizes several of them. But their new method seems like it might be more general than most, assuming that your starting material can tolerate nitration conditions. That’s because you take the arylamine and make an N-nitro derivative of it, and this can be made to kick out an OH after tautomerization and make a very labile NNO derivative that leaves as gaseous nitrous oxide in a pretty direct analogy to the Sandmeyer being driven by loss of nitrogen. A big difference, though, is that all of this happens in the same pot in situ - you start with the arylamine and isolate the coupling product, with no isolation or handling of any of the intermediates. Mechanistically, this one also looks much more like an aryl cation intermediate than an aryl radical.
The authors demonstrate this across a very wide range of substrates, including a lot of heterocyclic amines where the classic Sandmeyer doesn’t perform well, and they bring in halides, thiocyanates, OH, O-tosyl, O-triflate and a number of other coupling partners, including some intramolecular C-C bond forming reactions. They also demonstrate one-pot procedures to take some of these on to now-traditional metal-catalyzed couplings of many types, which gives you a lot of versatility in a synthetic sequence.
If the calorimetric profiles look good, this reaction could be quite useful on a larger scale - I’m sure that there will be some process chemists taking a look at that to make sure that the nitration step and the subsequent loss of nitrous oxide don’t present possible thermodynamic hazards. On the bench scale, though, this looks a lot more doable than the Sandmeyer, which as a reaction may be headed into the History of Organic Chemistry department and leaving the Useful Laboratory Transformations one. And to be honest, it was always an uneasy member of that second group anyway!
The whole “Are there weird electric fields at the surface of water droplets” question is not getting any easier to understand. I last wrote about this topic here). There are several subsidiary questions that go off in different directions: does unusual chemistry actually happen at air-water interfaces (and is this more prominent, as you might expect if true, in small droplets that are mostly surface area?) Or are these reports (mostly) experimental artifacts? If it does happen, what’s the mechanism? Does it have to do with unusually high electric fields (and field gradients) as some have proposed, or is it something else? Is it specific to air-water interfaces, or does it happen in the vacuum as well? And as for the later, how about under mass spec conditions (with electric charges all over the place?) This latest paper will not douse the flames. They’re using a technique called “vibrational sum frequency generation” spectroscopy which seems well suited to studying surface phenomena. The technique probes changes in the OH stretching band in water’s IR spectrum, and the wavenumber values you get are well known to be correlated with local electric field strength. The authors take pains to go into detail about what is meant by “electric field strength”, to their credit - is it from a roughly static arrangement of water molecule dipoles at the surface, or is it the sum of a bunch of dynamic fluctuations (in the hydrogen-bonding network,e etc.)? But no matter how you get there, the arguing is most vigorous over the idea that such fields can be large enough to change chemical reactivity, and there are very passionate advocates on both sides of that question. The authors are studying extended thin films of water by SFG spectroscopy, and they make predictions of what they should see for the OH stretches under different conditions. Specifically, there should be a red-shifted zone under any sort of relatively high and relatively persistent electric field at the surface (as opposed to the spectra of bulk water), and they go into detail about what the different sorts of fields (and mechanisms for generating them) might mean to the spectra. Now, people have of course tried to estimate these field strengths before, through other techniques, and some measurements come in at 10 to 20 MV/cm, which is pretty strong stuff. Values of this size, the authors note here, should be very apparent in these SFG measurements (and you can indeed see things shift around dramatically in SFG with small amounts of ionic solutes in water or with changes in pH). But they don’t see it! These findings indicate that exceptionally large interfacial fields, as sometimes invoked to explain microdroplet reactivity, are not supported by our SFG results. Within the framework of vibrational SFG, we conclude that there is no evidence for such fields at the neat air−water interface. The interpretation of interfacial electric field-driven catalysis in microdroplet experiments must therefore be reconsidered in light of the absence of spectroscopic field signatures at water interfaces. Yeah, it seems as if it should be. But as the authors note, you don’t have to reach for electric field explanations to deal with unusual interface chemistry - there could be altered solvation chemistry, redox pathways, evaporative concentration and other factors at work, and these aren’t mutually exclusive, either. So this certainly doesn’t mean that all the reports of interesting surface chemistry are wrong - but people may have been reaching for the wrong way to explain them. Let’s see how this goes over!
Show full content
The whole “Are there weird electric fields at the surface of water droplets” question is not getting any easier to understand. I last wrote about this topic here). There are several subsidiary questions that go off in different directions: does unusual chemistry actually happen at air-water interfaces (and is this more prominent, as you might expect if true, in small droplets that are mostly surface area?) Or are these reports (mostly) experimental artifacts? If it does happen, what’s the mechanism? Does it have to do with unusually high electric fields (and field gradients) as some have proposed, or is it something else? Is it specific to air-water interfaces, or does it happen in the vacuum as well? And as for the later, how about under mass spec conditions (with electric charges all over the place?)
This latest paper will not douse the flames. They’re using a technique called “vibrational sum frequency generation” spectroscopy which seems well suited to studying surface phenomena. The technique probes changes in the OH stretching band in water’s IR spectrum, and the wavenumber values you get are well known to be correlated with local electric field strength. The authors take pains to go into detail about what is meant by “electric field strength”, to their credit - is it from a roughly static arrangement of water molecule dipoles at the surface, or is it the sum of a bunch of dynamic fluctuations (in the hydrogen-bonding network,e etc.)? But no matter how you get there, the arguing is most vigorous over the idea that such fields can be large enough to change chemical reactivity, and there are very passionate advocates on both sides of that question.
The authors are studying extended thin films of water by SFG spectroscopy, and they make predictions of what they should see for the OH stretches under different conditions. Specifically, there should be a red-shifted zone under any sort of relatively high and relatively persistent electric field at the surface (as opposed to the spectra of bulk water), and they go into detail about what the different sorts of fields (and mechanisms for generating them) might mean to the spectra.
Now, people have of course tried to estimate these field strengths before, through other techniques, and some measurements come in at 10 to 20 MV/cm, which is pretty strong stuff. Values of this size, the authors note here, should be very apparent in these SFG measurements (and you can indeed see things shift around dramatically in SFG with small amounts of ionic solutes in water or with changes in pH). But they don’t see it!
These findings indicate that exceptionally large interfacial fields, as sometimes invoked to explain microdroplet reactivity, are not supported by our SFG results. Within the framework of vibrational SFG, we conclude that there is no evidence for such fields at the neat air−water interface. The interpretation of interfacial electric field-driven catalysis in microdroplet experiments must therefore be reconsidered in light of the absence of spectroscopic field signatures at water interfaces.
Yeah, it seems as if it should be. But as the authors note, you don’t have to reach for electric field explanations to deal with unusual interface chemistry - there could be altered solvation chemistry, redox pathways, evaporative concentration and other factors at work, and these aren’t mutually exclusive, either. So this certainly doesn’t mean that all the reports of interesting surface chemistry are wrong - but people may have been reaching for the wrong way to explain them. Let’s see how this goes over!
Some chemistry today, drawn from real life (mine, anyway). I was setting up a short series of palladium-catalyzed couplings the other day (Buchwald-Hartwig type, C-N bond formation), and since there were very close precedents to my structures in the chemical literature, I naturally just borrowed the known conditions. There was nothing out of the ordinary about them; it seemed as if they’d work about as well on my starting aryl bromides as it did on the ones already described. (Edit: added some more references to this post after its first publication). Well, they didn’t, of course. Which is the way of such metal-catalyzed couplings, which is why there are fifty gazillion ways of running them in the literature. They work until they don't! You can vary the catalyst ligands, first off, and boy are there are lot of them out there. You can change up the solvent, and the base needed for the reactions to go. There are other additives to try, and you can even vary the source of the palladium. (These days, if you know the system well enough and have some money to spend, you can order “pre-cat” materials where the ligand/Pd complex is already formed for you). In fact, here’s a recent Organic Process Research and Development paper that investigates that last variable in great detail: some catalyst systems don’t seem to care where their palladium comes from, while others care very much indeed, in case you were wondering. But I had no desire to wander off and try a whole list of reaction conditions. In the manner of discovery biopharma chemists everywhere, I didn’t want to perfect my reaction - I just wanted it to make a reasonable amount of product so I could get on to the important stuff! I was staring at my compounds and trying to think about what made them different from the known examples, and the main thing was that I had an extra functional group at the other end of the molecule. I hadn’t thought it would be a problem, but I wondered if it was perhaps sensitive to the base I was using (which was good ol’ cesium carbonate). So I was very interested indeed when I saw this new JACS paper from the Hartwig group themselves. It goes into great detail about the use of a base that I’d heard of but never actually tried, potassium 2-ethylhexanoate (K-2-EH). That might be an obscure-sounding reagent (along with the starting 2-ethylhexanoic acid) unless you’re a Real Industrial Chemist. Those compounds show up in a lot of polymer, coating, formulation, and materials science applications, and the acid is one of the largest-scale compounds of its kind produced industrially. So you can buy big ol’ bottles of the sodium and potassium salts relatively cheaply, and the potassium one is especially notable for dissolving in all kinds of organic solvents (where a lot of other potassium salts and carboxylates may not). The Hartwig group found that it’s an excellent choice in the C-N couplings that bear the name, partly because of that solubility and partly because it’s a much milder base than many that people reach for. I read up on that, checked our inventory, and found a bottle of the stuff one floor below me. A milder base was about the only idea that I had to fix my problem, so it seemed like a good opportunity to try it out. I should note that the Buchwald group at MIT has also investigated some bases with properties of this sort, such as NaOTMS, which may itself be more effective for secondary amines. It's fascinating to me that these transformations have been investigated for so many years now (with these two research groups providing many major advances!) and there are still improvements like this coming along. Who knows what the final forms will be, if there are indeed final forms to be had? Well by golly, I checked this morning and the reaction is making beautifully clean product, as opposed to the mixture of dark gunk I got with the cesium carbonate conditions. It is relatively rare that we get to actually figure out what’s going wrong with our reactions (unless you’re a process chemist, in which case that is your entire job!) But it’s also rare to fix things cleanly on the first shot - I can count the number of times I’ve been able to turn things around like this with one change on the fingers of my hands. Maybe just one hand, and that’s after forty years at the bench. That’s not as grim as it sounds, because remember, over most of that span I’ve been in the world where (as I like to say) there are two yields for reactions: Enough and Not Enough. Most of the time, even a relatively crappy conversion, the sort of thing a process chemist would not put up with for ten seconds, has been Enough, and I move on. But when all your starting material turns to gorp, you don’t have that option. Honestly, I would have settled just for a better product/gorp ratio, but what I got was the cleanest coupling reaction I’ve run in a long time. So thanks to Hartwig and collaborators, and those of you troubleshooting Pd reactions, try a K-2-EH run and see if it helps! Now I can move on (after another step or two) to the real reason I'm making these compounds, which is to do something very odd to an unsuspecting protein, and sadly I can't talk about that. But without making the needed compounds, you can't test out those weirdo ideas, can you? I'm glad these are now unsnarled.
Show full content
Some chemistry today, drawn from real life (mine, anyway). I was setting up a short series of palladium-catalyzed couplings the other day (Buchwald-Hartwig type, C-N bond formation), and since there were very close precedents to my structures in the chemical literature, I naturally just borrowed the known conditions. There was nothing out of the ordinary about them; it seemed as if they’d work about as well on my starting aryl bromides as it did on the ones already described. (Edit: added some more references to this post after its first publication).
Well, they didn’t, of course. Which is the way of such metal-catalyzed couplings, which is why there are fifty gazillion ways of running them in the literature. They work until they don't! You can vary the catalyst ligands, first off, and boy are there are lot of them out there. You can change up the solvent, and the base needed for the reactions to go. There are other additives to try, and you can even vary the source of the palladium. (These days, if you know the system well enough and have some money to spend, you can order “pre-cat” materials where the ligand/Pd complex is already formed for you). In fact, here’s a recent Organic Process Research and Development paper that investigates that last variable in great detail: some catalyst systems don’t seem to care where their palladium comes from, while others care very much indeed, in case you were wondering.
But I had no desire to wander off and try a whole list of reaction conditions. In the manner of discovery biopharma chemists everywhere, I didn’t want to perfect my reaction - I just wanted it to make a reasonable amount of product so I could get on to the important stuff! I was staring at my compounds and trying to think about what made them different from the known examples, and the main thing was that I had an extra functional group at the other end of the molecule. I hadn’t thought it would be a problem, but I wondered if it was perhaps sensitive to the base I was using (which was good ol’ cesium carbonate). So I was very interested indeed when I saw this new JACS paper from the Hartwig group themselves.
It goes into great detail about the use of a base that I’d heard of but never actually tried, potassium 2-ethylhexanoate (K-2-EH). That might be an obscure-sounding reagent (along with the starting 2-ethylhexanoic acid) unless you’re a Real Industrial Chemist. Those compounds show up in a lot of polymer, coating, formulation, and materials science applications, and the acid is one of the largest-scale compounds of its kind produced industrially. So you can buy big ol’ bottles of the sodium and potassium salts relatively cheaply, and the potassium one is especially notable for dissolving in all kinds of organic solvents (where a lot of other potassium salts and carboxylates may not).
The Hartwig group found that it’s an excellent choice in the C-N couplings that bear the name, partly because of that solubility and partly because it’s a much milder base than many that people reach for. I read up on that, checked our inventory, and found a bottle of the stuff one floor below me. A milder base was about the only idea that I had to fix my problem, so it seemed like a good opportunity to try it out. I should note that the Buchwald group at MIT has also investigated some bases with properties of this sort, such as NaOTMS, which may itself be more effective for secondary amines. It's fascinating to me that these transformations have been investigated for so many years now (with these two research groups providing many major advances!) and there are still improvements like this coming along. Who knows what the final forms will be, if there are indeed final forms to be had?
Well by golly, I checked this morning and the reaction is making beautifully clean product, as opposed to the mixture of dark gunk I got with the cesium carbonate conditions. It is relatively rare that we get to actually figure out what’s going wrong with our reactions (unless you’re a process chemist, in which case that is your entire job!) But it’s also rare to fix things cleanly on the first shot - I can count the number of times I’ve been able to turn things around like this with one change on the fingers of my hands. Maybe just one hand, and that’s after forty years at the bench.
That’s not as grim as it sounds, because remember, over most of that span I’ve been in the world where (as I like to say) there are two yields for reactions: Enough and Not Enough. Most of the time, even a relatively crappy conversion, the sort of thing a process chemist would not put up with for ten seconds, has been Enough, and I move on. But when all your starting material turns to gorp, you don’t have that option. Honestly, I would have settled just for a better product/gorp ratio, but what I got was the cleanest coupling reaction I’ve run in a long time. So thanks to Hartwig and collaborators, and those of you troubleshooting Pd reactions, try a K-2-EH run and see if it helps!
Now I can move on (after another step or two) to the real reason I'm making these compounds, which is to do something very odd to an unsuspecting protein, and sadly I can't talk about that. But without making the needed compounds, you can't test out those weirdo ideas, can you? I'm glad these are now unsnarled.
Well, here’s another report of a GLP-1 agent being tried in Alzheimer’s patients (after this recent post was written). The last one didn’t show much, so let’s have a look. In this trial, 204 patients with mild-to-moderate Alzheimer’s were treated with liraglutide (daily injection) versus placebo for one year. The doses started at 0.6mg and worked up to 1.8mg, which is the typical starting dose for diabetes therapy working up to the maximum approved one. The primary outcome was looking for changes in cerebral glucose rate, with secondary endpoints of safety and cognitive assessment changes. That primary endpoint is there because of a good deal of research over the years suggesting that glucose metabolism in Alzheimer’s brain tissue is abnormally low. This may well be linked to insulin resistance in tissues overall, which has led to some characterization of Alzheimer’s as “Type III diabetes” (given the complexities of both conditions, that formulation makes me a bit nervous to tell you the truth). But there are numerous studies in human patients and animal models alike that point towards both those effects, and thus the interest in GLP-1 agonists as potential therapies. The results are. . .mixed. The primary endpoint first: there was no change in cerebral glucose metabolism between the treated patients and the placebo group. It has already been shown that liraglutide does reach pharmacologically active doses in the brain, so that doesn’t seem to have been the problem here. Given these results, one might want to treat the secondary endpoints with some caution, because the underlying hypothesis for the whole trial seems to have been undermined. At any rate, there does seem to have been an improvement in one of the three cognitive tests systems used (ADAS-Exec) but not in the other two (ADCS-ADL or CDR-SoB). I always wonder what to make of results like this, which are distressingly common in Alzheimer’s trials, and I cannot shake the belief that a robust treatment would not be so dependent on questions being asked and the scoring methods used in the evaluations. Looking at the individual scores (Figure 3 in the paper), the latter two evaluations do indeed look almost identical to placebo. The ADAS-Exec differences seem driven by a few patients that did indeed score better in the treatment group, versus a slightly larger cohort that scored notably worse in the placebo group: otherwise, the bulk of the individual patient scores overlap in what to me look like similarly-shaped clouds. That’s not to say that these results aren’t real - for example, it could be possible that the disappearance of the longer tail of poor performance as seen the in the placebo group is due to the treatment - but at the very least they do not appear very strong. The team also looked at volumes of various brain regions via MRI. Volume measures on the hippocampus, entrorhinal cortex, and ventricles did not show any differences between the two groups (all lower), but the temporal lobe showed lower reduction in volume in the treatment group. This was significant (mean and standard error) although the 95% confidence intervals appear to overlap in the individual-patient scores. What this means is unclear; I’m not sure if anyone would have picked the temporal lobe beforehand as the most likely place to see a significant effect. So there are some effects, although none of them bowl a person over, and (as mentioned) evidence for the main hypothesis that a GLP-1 drug would improve cerebral glucose metabolism did not turn up here at all. I think that the “Type III Diabetes” pitch itself (at least in its strongest and simplest form) is taking some hits here with the results of these GLP-1/Alzheimer’s trials. But so far, Alzheimer’s has not yielded to any simple explanations from any direction.
Show full content
Well, here’s another report of a GLP-1 agent being tried in Alzheimer’s patients (after this recent post was written). The last one didn’t show much, so let’s have a look.
In this trial, 204 patients with mild-to-moderate Alzheimer’s were treated with liraglutide (daily injection) versus placebo for one year. The doses started at 0.6mg and worked up to 1.8mg, which is the typical starting dose for diabetes therapy working up to the maximum approved one. The primary outcome was looking for changes in cerebral glucose rate, with secondary endpoints of safety and cognitive assessment changes. That primary endpoint is there because of a good deal of research over the years suggesting that glucose metabolism in Alzheimer’s brain tissue is abnormally low. This may well be linked to insulin resistance in tissues overall, which has led to some characterization of Alzheimer’s as “Type III diabetes” (given the complexities of both conditions, that formulation makes me a bit nervous to tell you the truth). But there are numerous studies in human patients and animal models alike that point towards both those effects, and thus the interest in GLP-1 agonists as potential therapies.
The results are. . .mixed. The primary endpoint first: there was no change in cerebral glucose metabolism between the treated patients and the placebo group. It has already been shown that liraglutide does reach pharmacologically active doses in the brain, so that doesn’t seem to have been the problem here. Given these results, one might want to treat the secondary endpoints with some caution, because the underlying hypothesis for the whole trial seems to have been undermined. At any rate, there does seem to have been an improvement in one of the three cognitive tests systems used (ADAS-Exec) but not in the other two (ADCS-ADL or CDR-SoB). I always wonder what to make of results like this, which are distressingly common in Alzheimer’s trials, and I cannot shake the belief that a robust treatment would not be so dependent on questions being asked and the scoring methods used in the evaluations.
Looking at the individual scores (Figure 3 in the paper), the latter two evaluations do indeed look almost identical to placebo. The ADAS-Exec differences seem driven by a few patients that did indeed score better in the treatment group, versus a slightly larger cohort that scored notably worse in the placebo group: otherwise, the bulk of the individual patient scores overlap in what to me look like similarly-shaped clouds. That’s not to say that these results aren’t real - for example, it could be possible that the disappearance of the longer tail of poor performance as seen the in the placebo group is due to the treatment - but at the very least they do not appear very strong.
The team also looked at volumes of various brain regions via MRI. Volume measures on the hippocampus, entrorhinal cortex, and ventricles did not show any differences between the two groups (all lower), but the temporal lobe showed lower reduction in volume in the treatment group. This was significant (mean and standard error) although the 95% confidence intervals appear to overlap in the individual-patient scores. What this means is unclear; I’m not sure if anyone would have picked the temporal lobe beforehand as the most likely place to see a significant effect.
So there are some effects, although none of them bowl a person over, and (as mentioned) evidence for the main hypothesis that a GLP-1 drug would improve cerebral glucose metabolism did not turn up here at all. I think that the “Type III Diabetes” pitch itself (at least in its strongest and simplest form) is taking some hits here with the results of these GLP-1/Alzheimer’s trials. But so far, Alzheimer’s has not yielded to any simple explanations from any direction.
It is next to impossible to keep up with all the chaos in the Trump administration’s staffing of the public health agencies (NIH, FDA, and all the ones under them). I haven’t tried to cover it all blow-by-blow, because it would exhaust me and exhaust you the readers, and to what end? You can make out the main points of the Trump/RFK Jr. approach pretty easily. One of those is clearly, obviously, a multipronged attack on the practice of vaccination. Every new appointment at the CDC seems like yet another “longtime skeptic” or “maverick thinker”, but what they all have in common is a long public record of anti-vaccine activism. It comes from all sorts of directions: claims of covered-up vaccine damage or deaths, assignment of every possible harm to vaccines while ignoring their benefits, just-asking-questions approaches to every phase of the development and approval process, attempts to repeal mandatory childhood vaccination laws and to mess with the childhood vaccine schedule in any fashion that can make it more likely that fewer children get vaccinated, attempts to rework liability shields for vaccine manufacturers, generalized spreading of fear and uncertainty. . .the lot. Everything points in the same direction, constantly. We can wonder about where this relentless hostility to the very idea of vaccinating children comes from (I have a lot of ideas myself). Perhaps some of it, in some cases, is a sincere (but extremely misguided) belief that vaccination really does harm children and the entire population, instead of the actual situation (out here in reality) where it saves lives and mitigates suffering. But some of it is just a grab for fame, power, and (let’s not pretend otherwise) money. RFK Jr. himself has made millions and millions of dollars being a loud public anti-vaccine advocate, and there’s a lot more cash out there to be scooped up. My own instinct is to impute no reasonable or honorable motives to the man at all - his conduct over the years doesn’t argue for either. This isn’t just my complaint, obviously. Reliable observers all over the biopharma world are aghast at what’s happening. Here’s Steve Usdin at BioCentury on the latest vaccine news, here’s Stat on RFK Jr. and on vaccine policy, here are editorials and expressions of concern at the New York Times, at CIDRAP, at the Guardian, at the Wall Street Journal. . .and when those last two are lined up on the same side, you have to realize that there’s a problem. Meanwhile, at the FDA, confusion reigns. Which is exactly what you don’t want in such a regulatory agency. The FDA needs clarity and consistency, firmness of purpose and to ability to let everyone know where everyone stands. Clinical trial designs, manufacturing oversight, fast-track designations, drug approvals (and conditional approvals and rescinded approvals). . .there is no room for winging it. Drug development is a long and hideously expensive task with plenty of twists and turns, and without clear well-thought-out regulatory processes it can rapidly descend into a nightmarish free-for-all that wastes time, wastes money, and endangers the public. A big part of my loud objections to some of the FDA decisions and approvals over the years has been when they don’t seem to be following their own rules, because that’s such a dangerous way to work. Welcome to 2025, then. I can’t tell what the hell is going on, and I don’t think anyone else can either. Vinay Prasad was apparently pushed out in late July, then came back two weeks later under circumstances that have still not been explained. George Tidmarsh was appointed to lead CDER (the Center for Drug Evaluation and Research) in July, but was forced out in early November amid reports that no one was interested in taking over the job at all given the chaos. Back in mid-November Richard Pazdur was announced as the choice to fill that role, a move that many people found surprisingly sane, given his long track record at the agency. But no fears about that: Stat just broke the news this morning that Pazdur is apparently planning to leave the FDA entirely (according to two sources who seem to have heard his announcement earlier today). This is all perfectly in line with the rest of the Trump administration: we are being lead by a swarm of incompetent ideologues and ignorant grifters who are relentlessly ruining everything they touch. Every part of the federal government is being degraded right in front of us, and it’s happening with the smiling approval of the most useless Congress in the country’s history. I write about the biopharma industry here, so I’ve focused on that, but the rot is everywhere. Speed the day when we can start cleaning it up.
Show full content
It is next to impossible to keep up with all the chaos in the Trump administration’s staffing of the public health agencies (NIH, FDA, and all the ones under them). I haven’t tried to cover it all blow-by-blow, because it would exhaust me and exhaust you the readers, and to what end? You can make out the main points of the Trump/RFK Jr. approach pretty easily.
One of those is clearly, obviously, a multipronged attack on the practice of vaccination. Every new appointment at the CDC seems like yet another “longtime skeptic” or “maverick thinker”, but what they all have in common is a long public record of anti-vaccine activism. It comes from all sorts of directions: claims of covered-up vaccine damage or deaths, assignment of every possible harm to vaccines while ignoring their benefits, just-asking-questions approaches to every phase of the development and approval process, attempts to repeal mandatory childhood vaccination laws and to mess with the childhood vaccine schedule in any fashion that can make it more likely that fewer children get vaccinated, attempts to rework liability shields for vaccine manufacturers, generalized spreading of fear and uncertainty. . .the lot.
Everything points in the same direction, constantly. We can wonder about where this relentless hostility to the very idea of vaccinating children comes from (I have a lot of ideas myself). Perhaps some of it, in some cases, is a sincere (but extremely misguided) belief that vaccination really does harm children and the entire population, instead of the actual situation (out here in reality) where it saves lives and mitigates suffering. But some of it is just a grab for fame, power, and (let’s not pretend otherwise) money. RFK Jr. himself has made millions and millions of dollars being a loud public anti-vaccine advocate, and there’s a lot more cash out there to be scooped up. My own instinct is to impute no reasonable or honorable motives to the man at all - his conduct over the years doesn’t argue for either.
This isn’t just my complaint, obviously. Reliable observers all over the biopharma world are aghast at what’s happening. Here’s Steve Usdin at BioCenturyon the latest vaccine news, here’s Staton RFK Jr. and on vaccine policy, here are editorials and expressions of concern at the New York Times, at CIDRAP, at the Guardian, at the Wall Street Journal. . .and when those last two are lined up on the same side, you have to realize that there’s a problem.
Meanwhile, at the FDA, confusion reigns. Which is exactly what you don’t want in such a regulatory agency. The FDA needs clarity and consistency, firmness of purpose and to ability to let everyone know where everyone stands. Clinical trial designs, manufacturing oversight, fast-track designations, drug approvals (and conditional approvals and rescinded approvals). . .there is no room for winging it. Drug development is a long and hideously expensive task with plenty of twists and turns, and without clear well-thought-out regulatory processes it can rapidly descend into a nightmarish free-for-all that wastes time, wastes money, and endangers the public. A big part of my loud objections to some of the FDA decisions and approvals over the years has been when they don’t seem to be following their own rules, because that’s such a dangerous way to work.
Welcome to 2025, then. I can’t tell what the hell is going on, and I don’t think anyone else can either. Vinay Prasad was apparently pushed out in late July, then came back two weeks later under circumstances that have still not been explained. George Tidmarsh was appointed to lead CDER (the Center for Drug Evaluation and Research) in July, but was forced out in early November amid reports that no one was interested in taking over the job at all given the chaos. Back in mid-November Richard Pazdur was announced as the choice to fill that role, a move that many people found surprisingly sane, given his long track record at the agency. But no fears about that: Stat just broke the news this morning that Pazdur is apparently planning to leave the FDA entirely (according to two sources who seem to have heard his announcement earlier today).
This is all perfectly in line with the rest of the Trump administration: we are being lead by a swarm of incompetent ideologues and ignorant grifters who are relentlessly ruining everything they touch. Every part of the federal government is being degraded right in front of us, and it’s happening with the smiling approval of the most useless Congress in the country’s history. I write about the biopharma industry here, so I’ve focused on that, but the rot is everywhere. Speed the day when we can start cleaning it up.
This is a good look at the current state of the art in creating peptidic binding ligands to protein targets out of thin air - well, “one-shot computation” is probably the more preferred term, but you get the idea. One expects this to start to be more feasible with such peptide ligands because of several factors: the modular nature of the peptides, for starters, combined with a limited number of amino acid building blocks both make the problem more bounded and discrete. And the remarkable successes of protein folding prediction software show that we have a reasonably good handle on peptide-structure-from-peptide-sequence in general, thanks to the huge pile of validated protein structural data. So let’s see how things go. The authors are using BindCraft, a freely available program which was originally developed for design of miniproteins. As the authors say, though, miniproteins are still rather different beasts than shorter peptides, not least because they’re (by definition) showing distinct tertiary structures, as opposed to the far more mobile peptides. It wasn’t clear if the BindCraft software would be able to produce useful small peptide binders at all, but the team set the parameters of the program to generate 10-20-mer peptides and had at it. They started off with MDM2, a protein with high-quality structure data available, and they specified the known MDM2-p53 interface area as the part to design against. Note that this already gives you a couple of important steps to solving the problem (the partner has a stable structure, which you know in detail, and you also know just the region to target). They took 20 peptides suggestions and put them to the experimental test, finding that seven of them showed clear binding behavior, while the other 13 were weak at best (or just not binding at all, for the most part). The hits showed 65-165 nM binding, and six of them recapitulated a known hotspot triad of amino acids that have already been shown to be important for hitting this site. They then tried WDR5, another protein that’s had a substantial amount of work done on it, both for protein binders and for small molecules. It’s known to have two binding sites of interest, the WIN site and the MYC-binding one, and again you see that you’re starting with some advantages. The group generated 100 BindCraft-suggested peptides for each of those, with the top ten of each synthesized for some real experimental validation. Interestingly, though, none of the WIN-targeting candidates seemed to bind at all! On the other hand, six of the MYC-site targeting ones showed real binding, with moderate Kd values of 219 to 650 nM. Some of these do seem to be different than other binding motifs reported in the literature, which is a good sign. The team noted that many of the candidates were good ol’ alpha-helices, and selected one of these Myc-binding candidates for “peptide stapling” to lock that structure in place. Binding of the stapled candidate improved about six-fold to 39 nM, which seems to be good evidence. The last test was generating candidates against another binding interface that is well-described in structural biology, the PD-1/PD-L1 pair. But in this case, despite the structural knowledge and ability to point right at the key regions to target, none of the peptide candidates for either protein’s binding site seemed to show any activity at all. I think this is a pretty honest assessment, and I very much appreciate the authors providing it instead of dwelling on the most successful parts. Overall, the software gets some things right and totally whiffs on others, and one of the things to note is that it’s not obvious which of those situations you’re going to find yourself in! But that’s generally the problem with purely computational approaches; you have to try things out and get your bearings. But when BindCraft can get its digital teeth into a problem, it does pretty well, and that’s creditable performance for a program that wasn’t really designed to be doing quite what it’s asked to be doing here. But that has to be balanced against the fact that (as mentioned) all three of these test cases were on proteins (and binding sites) that have been thoroughly characterized experimentally - your success rate would surely be lower if you try something like this on less-well-understood cases. So this glass is half full. There was a time when results like these would have still knocked people over, and it’s a testimony to how far computational protein design has come that now they fall into the “interesting” category instead. But given that BindCraft is freely available, does not need any serious computational resources (these runs were done on a system with a single GPU) and that the entry barriers to preparing proteins this size are low, I’d say this easily falls into the “Why not try it?” category if small protein ligands are your goal.
Show full content
This is a good look at the current state of the art in creating peptidic binding ligands to protein targets out of thin air - well, “one-shot computation” is probably the more preferred term, but you get the idea. One expects this to start to be more feasible with such peptide ligands because of several factors: the modular nature of the peptides, for starters, combined with a limited number of amino acid building blocks both make the problem more bounded and discrete. And the remarkable successes of protein folding prediction software show that we have a reasonably good handle on peptide-structure-from-peptide-sequence in general, thanks to the huge pile of validated protein structural data.
So let’s see how things go. The authors are using BindCraft, a freely available program which was originally developed for design of miniproteins. As the authors say, though, miniproteins are still rather different beasts than shorter peptides, not least because they’re (by definition) showing distinct tertiary structures, as opposed to the far more mobile peptides. It wasn’t clear if the BindCraft software would be able to produce useful small peptide binders at all, but the team set the parameters of the program to generate 10-20-mer peptides and had at it.
They started off with MDM2, a protein with high-quality structure data available, and they specified the known MDM2-p53 interface area as the part to design against. Note that this already gives you a couple of important steps to solving the problem (the partner has a stable structure, which you know in detail, and you also know just the region to target). They took 20 peptides suggestions and put them to the experimental test, finding that seven of them showed clear binding behavior, while the other 13 were weak at best (or just not binding at all, for the most part). The hits showed 65-165 nM binding, and six of them recapitulated a known hotspot triad of amino acids that have already been shown to be important for hitting this site.
They then tried WDR5, another protein that’s had a substantial amount of work done on it, both for protein binders and for small molecules. It’s known to have two binding sites of interest, the WIN site and the MYC-binding one, and again you see that you’re starting with some advantages. The group generated 100 BindCraft-suggested peptides for each of those, with the top ten of each synthesized for some real experimental validation. Interestingly, though, none of the WIN-targeting candidates seemed to bind at all! On the other hand, six of the MYC-site targeting ones showed real binding, with moderate Kd values of 219 to 650 nM. Some of these do seem to be different than other binding motifs reported in the literature, which is a good sign. The team noted that many of the candidates were good ol’ alpha-helices, and selected one of these Myc-binding candidates for “peptide stapling” to lock that structure in place. Binding of the stapled candidate improved about six-fold to 39 nM, which seems to be good evidence.
The last test was generating candidates against another binding interface that is well-described in structural biology, the PD-1/PD-L1 pair. But in this case, despite the structural knowledge and ability to point right at the key regions to target, none of the peptide candidates for either protein’s binding site seemed to show any activity at all.
I think this is a pretty honest assessment, and I very much appreciate the authors providing it instead of dwelling on the most successful parts. Overall, the software gets some things right and totally whiffs on others, and one of the things to note is that it’s not obvious which of those situations you’re going to find yourself in! But that’s generally the problem with purely computational approaches; you have to try things out and get your bearings. But when BindCraft can get its digital teeth into a problem, it does pretty well, and that’s creditable performance for a program that wasn’t really designed to be doing quite what it’s asked to be doing here. But that has to be balanced against the fact that (as mentioned) all three of these test cases were on proteins (and binding sites) that have been thoroughly characterized experimentally - your success rate would surely be lower if you try something like this on less-well-understood cases.
So this glass is half full. There was a time when results like these would have still knocked people over, and it’s a testimony to how far computational protein design has come that now they fall into the “interesting” category instead. But given that BindCraft is freely available, does not need any serious computational resources (these runs were done on a system with a single GPU) and that the entry barriers to preparing proteins this size are low, I’d say this easily falls into the “Why not try it?” category if small protein ligands are your goal.
There have been some interesting failures recently in Alzheimer’s trials. As long-time readers will know, I consider basically all Alzheimer’s drug trials to have failed to one degree or another, and particularly when it comes to clearing the “will improve patient’s lives in the real world without putting them at too much risk” hurdle. But these two are notable because they’re aimed outside the usual amyloid zone. First off, Novo Nordisk reported that semaglutide (the company’s GLP-1 agonist drug, of course) failed in two Alzheimer’s trials. This was going to be a long shot, but long shots are worth taking in this area if you can afford to try them. Studies of thousands of patients with early cognitive impairment who took an oral form of semaglutide (Rybelsus, currently approved as a diabetes therapy) did not show improvements in mental function as compared to placebo. The company says that the treatment group showed “improvement of Alzheimer’s disease-related biomarkers” in both trials, although it does not (as far as I can see) say what those biomarkers were. And I would wonder how good they are as indicators given that you can show improvements in them and still not beat placebo, personally. The company’s stock took a hit on the news, which is kind of strange. Surely people weren’t betting on this succeeding? But Novo investors have been a jumpy bunch for a while now as Eli Lilly’s star continues to ascend in this area, so the sight of another possible life preserver disappearing might have been enough by itself. At any rate, it does appear as if there’s a disease where GLP-1 drugs are not actually beneficial. Novo had some better news today, though, with a once-weekly shot/once-daily pill combination for amycretin, a dual GLP-1/amylin agonist. I see that people are not quite giving up on the GLP-1/Alzheimer’s idea, but it has to be considered an even longer shot than before. There’s also news in the anti-tau protein area. That’s long been considered a possible Alzheimer’s target, and by “long” I mean decades. But it’s been hard to put that idea to the test in the clinic. Unfortunately, in the last couple of years it has been possible, and the results have not been good so far. Early last year a Lilly candidate (LY3372689, ceperognastat) failed its own trial. Earlier this year Asceneuron halted work on its own oral anti-tau drug candidate (ASN51), and Biogen stopped BIIB113, another similar effort. Now all of these are (were) O-GlcNAcase inhibitors, so you could easily make the case that the problem is that might not be a good mechanism to target tau, even if tau itself is a valid idea. But last year Roche bailed on a collaboration for an anti-tau antibody, which went on to fail its trials shortly afterwards. And the latest news is that J&J’s shot at an anti-tau antibody (posdinemab) has also failed its pivotal trial, with no efficacy seen in slowing the disease at the two-year mark. There are other tau programs that are now in the clinic, but they’re clearly going to have to bring something unusual to make you think that they will show interesting levels of efficacy at this point. Good luck, folks. . .
Show full content
There have been some interesting failures recently in Alzheimer’s trials. As long-time readers will know, I consider basically all Alzheimer’s drug trials to have failed to one degree or another, and particularly when it comes to clearing the “will improve patient’s lives in the real world without putting them at too much risk” hurdle. But these two are notable because they’re aimed outside the usual amyloid zone.
First off, Novo Nordisk reported that semaglutide (the company’s GLP-1 agonist drug, of course) failed in two Alzheimer’s trials. This was going to be a long shot, but long shots are worth taking in this area if you can afford to try them. Studies of thousands of patients with early cognitive impairment who took an oral form of semaglutide (Rybelsus, currently approved as a diabetes therapy) did not show improvements in mental function as compared to placebo. The company says that the treatment group showed “improvement of Alzheimer’s disease-related biomarkers” in both trials, although it does not (as far as I can see) say what those biomarkers were. And I would wonder how good they are as indicators given that you can show improvements in them and still not beat placebo, personally.
The company’s stock took a hit on the news, which is kind of strange. Surely people weren’t betting on this succeeding? But Novo investors have been a jumpy bunch for a while now as Eli Lilly’s star continues to ascend in this area, so the sight of another possible life preserver disappearing might have been enough by itself. At any rate, it does appear as if there’s a disease where GLP-1 drugs are not actually beneficial. Novo had some better news today, though, with a once-weekly shot/once-daily pill combination for amycretin, a dual GLP-1/amylin agonist. I see that people are not quite giving up on the GLP-1/Alzheimer’s idea, but it has to be considered an even longer shot than before.
There’s also news in the anti-tau protein area. That’s long been considered a possible Alzheimer’s target, and by “long” I mean decades. But it’s been hard to put that idea to the test in the clinic. Unfortunately, in the last couple of years it has been possible, and the results have not been good so far. Early last year a Lilly candidate (LY3372689, ceperognastat) failed its own trial. Earlier this year Asceneuron halted work on its own oral anti-tau drug candidate (ASN51), and Biogen stopped BIIB113, another similar effort.
Now all of these are (were) O-GlcNAcase inhibitors, so you could easily make the case that the problem is that might not be a good mechanism to target tau, even if tau itself is a valid idea. But last year Roche bailed on a collaboration for an anti-tau antibody, which went on to fail its trials shortly afterwards. And the latest news is that J&J’s shot at an anti-tau antibody (posdinemab) has also failed its pivotal trial, with no efficacy seen in slowing the disease at the two-year mark. There are other tau programs that are now in the clinic, but they’re clearly going to have to bring something unusual to make you think that they will show interesting levels of efficacy at this point. Good luck, folks. . .
This paper advances a metabolic hypothesis that I certainly didn’t see coming. One of the great mysteries about alcoholism/alcohol use disorder is its underlying biochemical drivers. There’s obviously a behavioral and psychological component, but there are physical and metabolic ones too, and trying to untangle those has been an effort of many decades. The authors here note a long-running series of observations about alcohol and sugar consumption. For one thing, laboratory rats that are given access to alcohol will in turn consume more sugar, and those that are more dependent on sugar will in turn consume more alcohol. This may not just be a laboratory curiosity: there have been several reports over the years that humans with alcohol use disorder have a greater preference for sucrose, although this finding has been complicated by differences in rating “pleasure” versus “intensity” of the sucrose effects, and a more recent study says that the reported preferences did not reproduce in their hands (although the idea does not seem to be going away). But you can produce entire lineages of high-alcohol-drinking rodent models, and these seem to show quite substantial preferences for sweetness compared to controls. The reasons behind all this have been unclear, and as you’d expect there have been ideas about some sort of hedonic phenotype with a disrupted risk/reward system underlying these behaviors. The paper linked in the first line of this post is an attempt to pull known metabolic pathways into the discussion, as well as the idea that there may be some pathways in common. They’ve uncovered evidence that alcohol consumption may (at least in part) be driven by fructose metabolism. Ketohexokinase is the enzyme that processes fructose to fructose-1-phosphate (the first step in fructose metabolism), and animals that have had this enzyme knocked out systemically (or who have been treated with a known inhibitor of it) consume less alcohol under free-choice conditions. And they also consume significantly less sugar under such conditions as well. One reason for that could be that fructose metabolism in the liver stimulates the production of the alcohol dehydrogenase enzyme, and if this enzyme is not upregulated, drinking becomes a much less pleasant experience. The experience of some Asian populations where ALDH levels are low serve as an example (consumption of alcohol causes skin flushing and headaches, among other symptoms, due to increased amounts of acetaldehyde being formed). But it’s not a simple story: as the authors show (and have shown in earlier work), tissue-specific knockouts of ketohexokinase can produce very different effects. KO in hepatocytes only lowers alcohol consumption to a degree, and these animal retain a strong preference for sugar. But KHK knockout in the gut tissue decreases alcohol use much more robustly, quite possibly through a mechanism involving GLP-1 (which is of course a hot topic of its own in gut/CNS mechanisms!) Those gut-knockout KHK animals also consume markedly less sugar when given a choice. There also appears to be a connection with alcohol-associated liver damage. Remarkably, the hepatocyte KHK knockout animals did not develop such disease, even while consuming amounts of alcohol (and for amounts of time) that should have led to histopathological changes. (Previous work has shown that administering sugar and alcohol at the same time to rodent models increases markers of liver damage more than either one on its own). The authors also have a theory involving aldose reductase levels being modulated by the osmolarity of the hepatic portal vein, as well as another involving FGFR1 signaling - both of these tie into the above, but I won’t go into them here. But the take-home messages, if this work holds up, would be these: (1) people who are concerned about their alcohol intake should also try to cut down on their sugar intake, and vice versa, since it appears that the metabolic pathways (and metabolic sequelae) for these two are thoroughly entangled. It also suggests (2) that ketohexokinase inhibitors (such as one designated CRP427 that was generated by this research group) might be useful therapies for people with alcohol use disorder. This might reduce the consumption of alcohol while at the same time showing protective effects in the liver. This paper has drawn a good amount of attention in the field, and I very much look forward to seeing where these ideas go. Overconsumption of alcohol is a plague on individuals, their families, and society in general, and it would be very good news to have a new route to dealing with it!
Show full content
This paper advances a metabolic hypothesis that I certainly didn’t see coming. One of the great mysteries about alcoholism/alcohol use disorder is its underlying biochemical drivers. There’s obviously a behavioral and psychological component, but there are physical and metabolic ones too, and trying to untangle those has been an effort of many decades.
The authors here note a long-running series of observations about alcohol and sugar consumption. For one thing, laboratory rats that are given access to alcohol will in turn consume more sugar, and those that are more dependent on sugar will in turn consume more alcohol. This may not just be a laboratory curiosity: there have been several reports over the years that humans with alcohol use disorder have a greater preference for sucrose, although this finding has been complicated by differences in rating “pleasure” versus “intensity” of the sucrose effects, and a more recent study says that the reported preferences did not reproduce in their hands (although the idea does not seem to be going away). But you can produce entire lineages of high-alcohol-drinking rodent models, and these seem to show quite substantial preferences for sweetness compared to controls.
The reasons behind all this have been unclear, and as you’d expect there have been ideas about some sort of hedonic phenotype with a disrupted risk/reward system underlying these behaviors. The paper linked in the first line of this post is an attempt to pull known metabolic pathways into the discussion, as well as the idea that there may be some pathways in common. They’ve uncovered evidence that alcohol consumption may (at least in part) be driven by fructose metabolism. Ketohexokinase is the enzyme that processes fructose to fructose-1-phosphate (the first step in fructose metabolism), and animals that have had this enzyme knocked out systemically (or who have been treated with a known inhibitor of it) consume less alcohol under free-choice conditions. And they also consume significantly less sugar under such conditions as well.
One reason for that could be that fructose metabolism in the liver stimulates the production of the alcohol dehydrogenase enzyme, and if this enzyme is not upregulated, drinking becomes a much less pleasant experience. The experience of some Asian populations where ALDH levels are low serve as an example (consumption of alcohol causes skin flushing and headaches, among other symptoms, due to increased amounts of acetaldehyde being formed). But it’s not a simple story: as the authors show (and have shown in earlier work), tissue-specific knockouts of ketohexokinase can produce very different effects. KO in hepatocytes only lowers alcohol consumption to a degree, and these animal retain a strong preference for sugar. But KHK knockout in the gut tissue decreases alcohol use much more robustly, quite possibly through a mechanism involving GLP-1 (which is of course a hot topic of its own in gut/CNS mechanisms!) Those gut-knockout KHK animals also consume markedly less sugar when given a choice.
There also appears to be a connection with alcohol-associated liver damage. Remarkably, the hepatocyte KHK knockout animals did not develop such disease, even while consuming amounts of alcohol (and for amounts of time) that should have led to histopathological changes. (Previous work has shown that administering sugar and alcohol at the same time to rodent models increases markers of liver damage more than either one on its own). The authors also have a theory involving aldose reductase levels being modulated by the osmolarity of the hepatic portal vein, as well as another involving FGFR1 signaling - both of these tie into the above, but I won’t go into them here.
But the take-home messages, if this work holds up, would be these: (1) people who are concerned about their alcohol intake should also try to cut down on their sugar intake, and vice versa, since it appears that the metabolic pathways (and metabolic sequelae) for these two are thoroughly entangled. It also suggests (2) that ketohexokinase inhibitors (such as one designated CRP427 that was generated by this research group) might be useful therapies for people with alcohol use disorder. This might reduce the consumption of alcohol while at the same time showing protective effects in the liver.
This paper has drawn a good amount of attention in the field, and I very much look forward to seeing where these ideas go. Overconsumption of alcohol is a plague on individuals, their families, and society in general, and it would be very good news to have a new route to dealing with it!
Here’s a phenomenon - yet another one - that never crossed my mind before. It’s long been known that enzymes that catalyze proteolysis (cleavage of peptide bonds) can, under certain circumstances, catalyze the reverse reaction of peptide bond formation. Folks who have had to think about chemical kinetics will immediately realize that those conditions would include high concentrations of the two cleavage products and low concentrations of the longer protein substrate, an example of Le Chatlier’s principle in action. It’s also an example of the principle of Microscopic Reversibility in action, too: the chemical steps are the same whether you run things forwards or backwards. That doesn’t mean those steps are always thermodynamically feasible, of course - the energies involved (with both enthalpic and entropic contributions) might be too great a barrier to run backwards very easily, as in unburning a piece of wood back from a cloud of soot and hot gases. Fire is not a good example of an equilibrium process, but peptide bond breakage and formation is a lot closer to balancing on a knife edge than combustion is. This recent preprint suggests, though, that this “reverse proteolysis” is happening under physiological conditions, particularly with cysteine-based cathepsin enzymes. And it’s not just re-formation of the proteins that have just been cleaved (although that must be happening, too). No, you get mix-and-match combinations of various proteins to generate species that were certainly never coded for in the genome. And on top of that, you can even spot chimeras between human proteins and bacterial or viral ones (!) Now, some species of this sort have been reported before (in reports going back to at least 2004) but this new work suggests that it’s a much more common process than anyone realized, one with implications for immunity and perhaps other cellular processes as well. Recall that antigen proteins are displayed to the immune system via the major histocompatibility complex, and that these antigens are cleaved from larger proteins via degradation. Displaying weirdo newly assembled protein sequences from this chemical splicing route could cause some real effects downstream. This could, for example, be one of the links between prior infections and later autoimmune disease, through those human/pathogen hybrid proteins. The authors here shore up that connection by showing that auto-antigenic peptides implicated in Type I diabetes can be produced by cathepsins running in reverse, and that proteins that have been modified by citrullination (on arginine residues) seem to undergo the process more readily. That sort of Arg modification is already known to be over-represented in autoimmune antigens. In addition, the cathepsin enzyme subtypes that are most dominant in immune tissues (such as inside macrophages) seem to be the best at producing such splicing hybrids. These reverse reactions are also more prevalent at closer to neutral pH, which suggests that lysosomal dysfunction (where cathepsins and other enzymes normally work in an acidic environment) might be a source of increased neo-peptides. Overall, it seems that we’re going to have to learn to deal with these species, and to study them in the context of both normal conditions and in infectious disease. Acute viral infections might well be producing waves of human/viral protein hybrid species, and we can’t expect them all to be silent!
Show full content
Here’s a phenomenon - yet another one - that never crossed my mind before. It’s long been known that enzymes that catalyze proteolysis (cleavage of peptide bonds) can, under certain circumstances, catalyze the reverse reaction of peptide bond formation. Folks who have had to think about chemical kinetics will immediately realize that those conditions would include high concentrations of the two cleavage products and low concentrations of the longer protein substrate, an example of Le Chatlier’s principle in action. It’s also an example of the principle of Microscopic Reversibility in action, too: the chemical steps are the same whether you run things forwards or backwards. That doesn’t mean those steps are always thermodynamically feasible, of course - the energies involved (with both enthalpic and entropic contributions) might be too great a barrier to run backwards very easily, as in unburning a piece of wood back from a cloud of soot and hot gases. Fire is not a good example of an equilibrium process, but peptide bond breakage and formation is a lot closer to balancing on a knife edge than combustion is.
This recent preprint suggests, though, that this “reverse proteolysis” is happening under physiological conditions, particularly with cysteine-based cathepsin enzymes. And it’s not just re-formation of the proteins that have just been cleaved (although that must be happening, too). No, you get mix-and-match combinations of various proteins to generate species that were certainly never coded for in the genome. And on top of that, you can even spot chimeras between human proteins and bacterial or viral ones (!)
Now, some species of this sort have been reported before (in reports going back to at least 2004) but this new work suggests that it’s a much more common process than anyone realized, one with implications for immunity and perhaps other cellular processes as well. Recall that antigen proteins are displayed to the immune system via the major histocompatibility complex, and that these antigens are cleaved from larger proteins via degradation. Displaying weirdo newly assembled protein sequences from this chemical splicing route could cause some real effects downstream. This could, for example, be one of the links between prior infections and later autoimmune disease, through those human/pathogen hybrid proteins.
The authors here shore up that connection by showing that auto-antigenic peptides implicated in Type I diabetes can be produced by cathepsins running in reverse, and that proteins that have been modified by citrullination (on arginine residues) seem to undergo the process more readily. That sort of Arg modification is already known to be over-represented in autoimmune antigens. In addition, the cathepsin enzyme subtypes that are most dominant in immune tissues (such as inside macrophages) seem to be the best at producing such splicing hybrids. These reverse reactions are also more prevalent at closer to neutral pH, which suggests that lysosomal dysfunction (where cathepsins and other enzymes normally work in an acidic environment) might be a source of increased neo-peptides.
Overall, it seems that we’re going to have to learn to deal with these species, and to study them in the context of both normal conditions and in infectious disease. Acute viral infections might well be producing waves of human/viral protein hybrid species, and we can’t expect them all to be silent!
This is a very useful article on phenotypic screening, and is well worth a read. And if you haven’t done this sort of screen before but are looking to try it out, I’d say it’s essential. The authors (both with extensive industrial experience) go into detail on the factors that can make for successful screens, and the ones that can send you off into the weeds. There are quite a few of the latter! For small molecule screens, you need to be aware that you’re only going to be covering a fraction of the proteome/genome to begin with, no matter how large your library might be under current conditions. And of course as those libraries get larger, the throughput of your assay becomes a major issue. You can cast your net broadly and lower the number of compounds screened, or you can zero in on One Specific Thing and screen them all, at the risk of missing important and useful stuff. Your call! And there are other problems that the paper provides specific examples of - the way that your compounds will (probably) not distinguish well between related proteins in a family, and the opposite problem of how some of them distinguish so sharply between (say) human and rodent homologs that your attempts at translational assays break down. For genomic-based screens, you have to be cognizant of the time domain you’re working in. One the one hand, the expression of a particular gene may be a rather short-lived phenomenon (and only under certain conditions which you may not be aware of), and on the other hand you might have a delayed onset of any effects of your compounds as they have to work their way through the levels of transcription, translation, protein stability, and so on. You can definitely run into genetic redundancies that will mask the activity of some compounds, so take the existence of false negatives as a given. And you should always be aware that the proteins whose levels or conditions you’re eventually modifying probably have several functions in addition to whatever their main “active site” function might be - partner proteins, allosteric effects, scaffolding, feedback into other transcriptional processes, and more. Another consideration: it may be tempting to focus on gene knockouts or knockdowns, and you can often get a lot done that way, but that ignores the whole universe of activation mechanisms. There are more! And in general, you’re going to have to ask yourself - be honest - what your best workflow is and what you mean by “best”. Is what you’re proposing going to fit well with cellular or animal models of disease, or are you going to be faced with bridging that, too (not recommended)? Do you really have the resources (equipment and human), the time, and the money to do a reasonable job of it all? Another large-scale question, if you’re really thinking of drug discovery by this route, is whether you (or your organization, or your funders) have the stomach for what is a fairly common outcome: you find hits, you refine them, you end up with a list of interesting compounds that do interesting things. . .and no one has the nerve to make the jump into the clinic if there isn’t a well-worked-out translational animal model already in place. You’re not going to discover and validate one of those from scratch along the way, so if there isn’t such a model out there already you’d better be ready for a gut check at the end of the project. I like to say that a good phenotypic assay is a thing of beauty. But I quickly add quickly that those are hard to realize, and that a bad phenotypic assay is just about the biggest waste of time and resources that you can imagine. Unfortunately, the usual rules apply: there are a lot more ways to do this poorly than to do it well, and many of those done-poorly pathways are temptingly less time- and labor-intensive than the useful ones.
Show full content
This is a very useful article on phenotypic screening, and is well worth a read. And if you haven’t done this sort of screen before but are looking to try it out, I’d say it’s essential.
The authors (both with extensive industrial experience) go into detail on the factors that can make for successful screens, and the ones that can send you off into the weeds. There are quite a few of the latter! For small molecule screens, you need to be aware that you’re only going to be covering a fraction of the proteome/genome to begin with, no matter how large your library might be under current conditions. And of course as those libraries get larger, the throughput of your assay becomes a major issue. You can cast your net broadly and lower the number of compounds screened, or you can zero in on One Specific Thing and screen them all, at the risk of missing important and useful stuff. Your call! And there are other problems that the paper provides specific examples of - the way that your compounds will (probably) not distinguish well between related proteins in a family, and the opposite problem of how some of them distinguish so sharply between (say) human and rodent homologs that your attempts at translational assays break down.
For genomic-based screens, you have to be cognizant of the time domain you’re working in. One the one hand, the expression of a particular gene may be a rather short-lived phenomenon (and only under certain conditions which you may not be aware of), and on the other hand you might have a delayed onset of any effects of your compounds as they have to work their way through the levels of transcription, translation, protein stability, and so on. You can definitely run into genetic redundancies that will mask the activity of some compounds, so take the existence of false negatives as a given. And you should always be aware that the proteins whose levels or conditions you’re eventually modifying probably have several functions in addition to whatever their main “active site” function might be - partner proteins, allosteric effects, scaffolding, feedback into other transcriptional processes, and more. Another consideration: it may be tempting to focus on gene knockouts or knockdowns, and you can often get a lot done that way, but that ignores the whole universe of activation mechanisms. There are more!
And in general, you’re going to have to ask yourself - be honest - what your best workflow is and what you mean by “best”. Is what you’re proposing going to fit well with cellular or animal models of disease, or are you going to be faced with bridging that, too (not recommended)? Do you really have the resources (equipment and human), the time, and the money to do a reasonable job of it all? Another large-scale question, if you’re really thinking of drug discovery by this route, is whether you (or your organization, or your funders) have the stomach for what is a fairly common outcome: you find hits, you refine them, you end up with a list of interesting compounds that do interesting things. . .and no one has the nerve to make the jump into the clinic if there isn’t a well-worked-out translational animal model already in place. You’re not going to discover and validate one of those from scratch along the way, so if there isn’t such a model out there already you’d better be ready for a gut check at the end of the project.
I like to say that a good phenotypic assay is a thing of beauty. But I quickly add quickly that those are hard to realize, and that a bad phenotypic assay is just about the biggest waste of time and resources that you can imagine. Unfortunately, the usual rules apply: there are a lot more ways to do this poorly than to do it well, and many of those done-poorly pathways are temptingly less time- and labor-intensive than the useful ones.
As regular readers well know, I get very frustrated when people use the verb “to reason” in describing the behavior of large language models (LLMs). Sometimes that’s just verbal shorthand, but both in print and in person I keep running into examples of people who really, truly, believe that these things are going through a reasoning process. They are not. None of them. (Edit: for a deep dive into this topic, see this recent paper). To bring this into the realm of medical science, have a look at this paper from earlier this year. The authors evaluated six different LLM systems in their ability to answer 68 various medical questions. The crucial test here, though was that the question was asked twice in two different ways. All of them started by saying “You are an experienced physician. Provide detailed step-by-step reasoning, then conclude with your final answer in exact format Answer: [Letter]” The prompt was written in that way because the questions would be some detailed medical query, followed by a list of likely options/diagnoses/recommendations, each with a letter, and the LLM was asked to choose among these. The first time the question was asked, one of the five options was “Reassurance”, i.e. “Don’t do any medical procedure because this is not actually a problem”. Any practicing physician will recognize this as a valid option at times! But the second time the exact same question was posed, the “reassurance” option was replaced by a “None of the other answers” option. Now, the step-by-step clinical reasoning that one would hope for should not be altered in the slightest by that change, and if “Reassurance” was in fact the correct answer, then “None of the above” should be the correct answer when phrased the second way (rather than the range of surgical and other interventions proposed in the other choices). Instead, the accuracy of the answers across all 68 questions dropped notably in every single LLM system when presented with a “None of the above” option. DeepSeek-R1 was the most resilient, but still degraded. The underlying problem is clear: no reasoning is going on, despite some of these systems being billed as having reasoning ability. Instead, this is all pattern matching, which presents the illusion of thought and the illusion of competence. This overview at Nature Medicine covers a range of such problems. The authors here find that the latest GPT-5 version does in fact make fewer errors than other systems, but that’s like saying that a given restaurant has overall fewer cockroaches floating in its soup. That’s my analogy, not theirs. The latest models hallucinate a bit less than before and breaks their own supposed rules a bit less, but neither of these have reached acceptable levels. The acceptable level of cockroaches in the soup pot is zero. As an example of that second problem, the authors here note that GPT-5, like all the other LLMs, will violate its own instructional hierarchy to deliver an answer, and without warning users that this has happened. Supposed safeguards and rules at the system level can and do get disregarded as the software rattles around searching for plausible text to deliver, a problem which is explored in detail here. This is obviously not a good feature in an LLM that is supposed to be dispensing medical advice - as the authors note, such systems should have high-level rules that are never to be violated, things like “Sudden onset of chest pain = always call for emergency evaluation” or “Recommendations for dispensing drugs on the attached list must always fit the following guidelines”. But at present it seems impossible for that “always” to actually stick under real-world conditions. No actual physician whose work was this unreliable would or should be allowed to continue working. LLMs are text generators, working on probabilities of what their next word choice should be based on what has been seem in their training sets, then dispensing answer-shaped nuggets in smooth, confident, grammatical form. This is not reasoning and it is not understanding - at its best, it is an illusion that can pass for them. And that’s what it is at its worst, too.
Show full content
As regular readers well know, I get very frustrated when people use the verb “to reason” in describing the behavior of large language models (LLMs). Sometimes that’s just verbal shorthand, but both in print and in person I keep running into examples of people who really, truly, believe that these things are going through a reasoning process. They are not. None of them. (Edit: for a deep dive into this topic, see this recent paper).
To bring this into the realm of medical science, have a look at this paper from earlier this year. The authors evaluated six different LLM systems in their ability to answer 68 various medical questions. The crucial test here, though was that the question was asked twice in two different ways. All of them started by saying “You are an experienced physician. Provide detailed step-by-step reasoning, then conclude with your final answer in exact format Answer: [Letter]” The prompt was written in that way because the questions would be some detailed medical query, followed by a list of likely options/diagnoses/recommendations, each with a letter, and the LLM was asked to choose among these.
The first time the question was asked, one of the five options was “Reassurance”, i.e. “Don’t do any medical procedure because this is not actually a problem”. Any practicing physician will recognize this as a valid option at times! But the second time the exact same question was posed, the “reassurance” option was replaced by a “None of the other answers” option. Now, the step-by-step clinical reasoning that one would hope for should not be altered in the slightest by that change, and if “Reassurance” was in fact the correct answer, then “None of the above” should be the correct answer when phrased the second way (rather than the range of surgical and other interventions proposed in the other choices).
Instead, the accuracy of the answers across all 68 questions dropped notably in every single LLM system when presented with a “None of the above” option. DeepSeek-R1 was the most resilient, but still degraded. The underlying problem is clear: no reasoning is going on, despite some of these systems being billed as having reasoning ability. Instead, this is all pattern matching, which presents the illusion of thought and the illusion of competence.
This overview at Nature Medicine covers a range of such problems. The authors here find that the latest GPT-5 version does in fact make fewer errors than other systems, but that’s like saying that a given restaurant has overall fewer cockroaches floating in its soup. That’s my analogy, not theirs. The latest models hallucinate a bit less than before and breaks their own supposed rules a bit less, but neither of these have reached acceptable levels. The acceptable level of cockroaches in the soup pot is zero.
As an example of that second problem, the authors here note that GPT-5, like all the other LLMs, will violate its own instructional hierarchy to deliver an answer, and without warning users that this has happened. Supposed safeguards and rules at the system level can and do get disregarded as the software rattles around searching for plausible text to deliver, a problem which is explored in detail here. This is obviously not a good feature in an LLM that is supposed to be dispensing medical advice - as the authors note, such systems should have high-level rules that are never to be violated, things like “Sudden onset of chest pain = always call for emergency evaluation” or “Recommendations for dispensing drugs on the attached list must always fit the following guidelines”. But at present it seems impossible for that “always” to actually stick under real-world conditions. No actual physician whose work was this unreliable would or should be allowed to continue working.
LLMs are text generators, working on probabilities of what their next word choice should be based on what has been seem in their training sets, then dispensing answer-shaped nuggets in smooth, confident, grammatical form. This is not reasoning and it is not understanding - at its best, it is an illusion that can pass for them. And that’s what it is at its worst, too.
I wanted to mention a couple of recent papers about a field that’s had a lot of interest over the last decade: engineered two-dimensional materials. These things are (at their theoretical best) only one layer of atoms or molecules thick, and can have a great many exotic and useful properties. Some of those are still a bit more theoretical than actual, but they range from optical and electronic behavior all the way to sheer physical processes like molecule-scale filtration. For an example of this, albeit not with a true single-molecular-layer filter, see here. Another example which is in use around the world is the “reverse osmosis” technology for desalination plants, where water can be pushed through the membranes but salty ions cannot. I was writing last month about the Nobel for MOFs, metal-organic frameworks, and there’s a two-dimensional aspect to those, too. Related are “covalent organic frameworks”, which are made up, as the name implies, of nothing but covalently linked groups without the metal-ion-coordination in MOFs. You can think of them as very specialized and very orderly 2D and 3D polymers. Here’s a paper that looks closely at some of the 2D ones to see what happens when they’re exposed to water. They’re only looking at three different ones with very similar structures. They all form large hexagonal pores (about 3 nm across), but the three have either no functional groups therein, phenols, or methoxys (all in the same positions. but they find that the capillary forces of the bulk water are strong enough to distort the structures greatly. That is, the strong hydrogen-bonding network of the water molecules can pull the frameworks into new arrangements (or just pull them apart!) Two of the three (the unsubstituted and the OH) basically just collapse irreversibly under the strain, actually. The methoxy-decorated one survives (and can go around for more desorption/absorption cycles) but its crystalline form is altered a bit along the way, so it is still feeling the strain. Calculations suggest that this one actually produces less “sorption pressure” as the water fills its cavities - it’s not that the channels themselves are so much more robust. As the authors put it, “capillary forces and their effects should not be ignored when using 2D COFs in liquid environments”, and they now have the data to prove it. At the other end of the scale is the material described in this paper. The authors have prepared a 2D polyamide material which is not quite a full-fledged covalent organic framework (it doesn’t have long-range order, from its powder X-ray spectrum). But what it is, is impermeable. Its layers seem to pack in a staggered arrangement with almost no free volume, so instead of having various sized-channels and cavities like most MOFs and COFs, it has so such gaps at all. That makes the permeability of this material extraordinarily low for both solvents and gases - in fact, around ten thousand times lower than the current record holders (!) The material can be formed as thin films with dozens or even hundred of such molecular layers, and the chances of anything getting through are very low indeed. This goes not only for relatively bulky stuff like (say) sulfur hexafluoride, but also for methane. And for oxygen and nitrogen. And for argon, and even for helium. For hydrogen, its permeability is roughly that of a pristine graphene layer. This obviously can be a very useful material, since there are many applications where you’d want (for example) an oxygen-impermeable barrier that doesn’t degrade. The authors demonstrate this by coating an oxygen- and water-sensitive perovskite sample with just 60nm of this material, and they show that it dramatically extends its lifetime under ambient conditions. So there you have two very different areas of modern materials science: things with huge channels (that need to be taken care of!) and others with no channels whatsoever! Plenty of uses are waiting for these and for all sorts of materials in between. . .
Show full content
I wanted to mention a couple of recent papers about a field that’s had a lot of interest over the last decade: engineered two-dimensional materials. These things are (at their theoretical best) only one layer of atoms or molecules thick, and can have a great many exotic and useful properties. Some of those are still a bit more theoretical than actual, but they range from optical and electronic behavior all the way to sheer physical processes like molecule-scale filtration. For an example of this, albeit not with a true single-molecular-layer filter, see here. Another example which is in use around the world is the “reverse osmosis” technology for desalination plants, where water can be pushed through the membranes but salty ions cannot.
I was writing last month about the Nobel for MOFs, metal-organic frameworks, and there’s a two-dimensional aspect to those, too. Related are “covalent organic frameworks”, which are made up, as the name implies, of nothing but covalently linked groups without the metal-ion-coordination in MOFs. You can think of them as very specialized and very orderly 2D and 3D polymers. Here’s a paper that looks closely at some of the 2D ones to see what happens when they’re exposed to water.
They’re only looking at three different ones with very similar structures. They all form large hexagonal pores (about 3 nm across), but the three have either no functional groups therein, phenols, or methoxys (all in the same positions. but they find that the capillary forces of the bulk water are strong enough to distort the structures greatly. That is, the strong hydrogen-bonding network of the water molecules can pull the frameworks into new arrangements (or just pull them apart!)
Two of the three (the unsubstituted and the OH) basically just collapse irreversibly under the strain, actually. The methoxy-decorated one survives (and can go around for more desorption/absorption cycles) but its crystalline form is altered a bit along the way, so it is still feeling the strain. Calculations suggest that this one actually produces less “sorption pressure” as the water fills its cavities - it’s not that the channels themselves are so much more robust. As the authors put it, “capillary forces and their effects should not be ignored when using 2D COFs in liquid environments”, and they now have the data to prove it.
At the other end of the scale is the material described in this paper. The authors have prepared a 2D polyamide material which is not quite a full-fledged covalent organic framework (it doesn’t have long-range order, from its powder X-ray spectrum). But what it is, is impermeable. Its layers seem to pack in a staggered arrangement with almost no free volume, so instead of having various sized-channels and cavities like most MOFs and COFs, it has so such gaps at all. That makes the permeability of this material extraordinarily low for both solvents and gases - in fact, around ten thousand times lower than the current record holders (!) The material can be formed as thin films with dozens or even hundred of such molecular layers, and the chances of anything getting through are very low indeed.
This goes not only for relatively bulky stuff like (say) sulfur hexafluoride, but also for methane. And for oxygen and nitrogen. And for argon, and even for helium. For hydrogen, its permeability is roughly that of a pristine graphene layer. This obviously can be a very useful material, since there are many applications where you’d want (for example) an oxygen-impermeable barrier that doesn’t degrade. The authors demonstrate this by coating an oxygen- and water-sensitive perovskite sample with just 60nm of this material, and they show that it dramatically extends its lifetime under ambient conditions.
So there you have two very different areas of modern materials science: things with huge channels (that need to be taken care of!) and others with no channels whatsoever! Plenty of uses are waiting for these and for all sorts of materials in between. . .
Here’s another example of biochemistry being weird, one that I had never come across until recently. Did you know that there are enzymes that are dependent on tungsten (of all things?) As far as we know, they aren’t found in higher organisms, but they are scattered across a number of bacteria and archaea. There’s a whole related family of molybdenum-dependent enzymes that I had heard about, but there is indeed a group that can only use tungsten. Both of these types tend to have similar environments immediately around the metal, with a heterocycle called “molybdopterin” involved in complexing the metal ions themselves (though the two thiols in the structure). Several magnesium ions help to hold things in place. There is even a rare human genetic disorder that traces back to an inability to produce this cofactor, a deficiency with very severe results. The enzymes that use these metallo-centers are largely oxidoreductases, but the obligate tungsten forms in the bacteria and archaea also include such exotica as acetylene hydratase. These tend to be found in anaerobic organisms, and these reactions are unusual enough to be of potential industrial importance. For example, Clostridium autoethanogenum (as its name implies!) is able to produce ethanol from the unlikely starting material of syngas (a mixture of hydrogen, carbon monoxide, and carbon dioxide), and this is actually being scaled up for large-scale use. And it now turns out that a key enzyme in this pathway is one of the tungsten-dependent variety, reducing acetate to acetaldehyde in a step that you probably would have modeled as thermodynamically unfeasible if you didn’t know the whole system it’s imbedded in (which features a key electron-transport step involving very low-potential electrons of the sort that you don’t usually see in biochemical systems). The immediate reduction to ethanol of the acetaldehyde thus produced (via a nearby alcohol dehydrogenase enzyme) helps keep things moving as well, while other factors are an intracellular pH that’s notably less acidic than the outside medium and a high tolerance for acetate buildup. Knowing these details is of course important if you’re trying to optimize the bacteria to do your syngas valorization for you, and it’s also something you have to understand if you want to shoot for a cell-free process using only the enzymes. As the paper explains, though, there are still a number of mysteries, especially around the details of the tungsten center’s changing oxidation state and coordination chemistry as the reaction cycle goes along. Personally, I’m very happy to find that there might be another biotechnology route to remediating industrial gas emissions, and I’m also pleasantly surprised that tungsten, of all things, is a necessary metal for life in some species. The way that these two subjects intersect is what science is all about - you never could have predicted it!
Show full content
Here’s another example of biochemistry being weird, one that I had never come across until recently. Did you know that there are enzymes that are dependent on tungsten (of all things?) As far as we know, they aren’t found in higher organisms, but they are scattered across a number of bacteria and archaea. There’s a whole related family of molybdenum-dependent enzymes that I had heard about, but there is indeed a group that can only use tungsten.
Both of these types tend to have similar environments immediately around the metal, with a heterocycle called “molybdopterin” involved in complexing the metal ions themselves (though the two thiols in the structure). Several magnesium ions help to hold things in place. There is even a rare human genetic disorder that traces back to an inability to produce this cofactor, a deficiency with very severe results.
The enzymes that use these metallo-centers are largely oxidoreductases, but the obligate tungsten forms in the bacteria and archaea also include such exotica as acetylene hydratase. These tend to be found in anaerobic organisms, and these reactions are unusual enough to be of potential industrial importance. For example, Clostridium autoethanogenum (as its name implies!) is able to produce ethanol from the unlikely starting material of syngas (a mixture of hydrogen, carbon monoxide, and carbon dioxide), and this is actually being scaled up for large-scale use. And it now turns out that a key enzyme in this pathway is one of the tungsten-dependent variety, reducing acetate to acetaldehyde in a step that you probably would have modeled as thermodynamically unfeasible if you didn’t know the whole system it’s imbedded in (which features a key electron-transport step involving very low-potential electrons of the sort that you don’t usually see in biochemical systems). The immediate reduction to ethanol of the acetaldehyde thus produced (via a nearby alcohol dehydrogenase enzyme) helps keep things moving as well, while other factors are an intracellular pH that’s notably less acidic than the outside medium and a high tolerance for acetate buildup.
Knowing these details is of course important if you’re trying to optimize the bacteria to do your syngas valorization for you, and it’s also something you have to understand if you want to shoot for a cell-free process using only the enzymes. As the paper explains, though, there are still a number of mysteries, especially around the details of the tungsten center’s changing oxidation state and coordination chemistry as the reaction cycle goes along.
Personally, I’m very happy to find that there might be another biotechnology route to remediating industrial gas emissions, and I’m also pleasantly surprised that tungsten, of all things, is a necessary metal for life in some species. The way that these two subjects intersect is what science is all about - you never could have predicted it!
Some weeks back I wrote about a paper suggesting that for many organic chemistry reactions stirring is not very necessary. That one sure set off a lot of comments (here on the blog, on social media, in my own email and conversations, etc.)! I wanted to revisit the topic in light of two more publications since then. First off is this preprint, a swift response to the original, from a group of industrial chemists. They do a good job of reviewing the principles behind reaction mixing, and they emphasize that larger-scale reactions are going to be much more sensitive to its effects. I don’t see how anyone who’s done scale-up can disagree! I do have to note that the first paper did most cover smaller-scale reactions and that the authors did note that fact. The response paper is written by people who think about larger-scale reproducibility all the time, and under those conditions I don’t think anyone can argue about taking another variable out of the system by using reproducible stirring and mixing protocols. There are too many considerations of mass transfer, concentration changes, temperature gradients and more to ignore. But on a small scale, how much do these matter? I think that’s what the first paper was getting at. So this new analysis of that paper’s actual data is certainly worth a look. The author takes the extensive supplementary data from that paper and goes back over it statistically, coming to somewhat different conclusions: . . .where the lowest-yielding reaction group shows the largest overall increase in isolated yield (4.5%; 95% CI: 2.6 to 6.4%; p < 0.0001) with stirring. Stirring of moderate yield reactions is associated with a 1.6% (95% CI: 0.4 to 2.9%; p = 0.0082) increase in yield, while the highest yielding reactions show smaller effect sizes in stirred versus unstirred reactions (0.8% increase; 95% CI: 0.4 to 1.3%; p< 0.0001). These results suggest that while the overall change in yields with stirring in these data for many reactions are modest, there is extensive variation of impact, and stirring is associated with real and practically important increases in isolated yields across a wide range of substrates and reaction manifolds. So (at least on a certain scale) reactions that work well are going to work well whether you pay attention to stirring them or not. But in other cases, it seems that you can get a slight (or sometimes more than slight) edge by stirring. I’d be willing to bet that these yield increases are well below what a random set of bench chemists would have assumed, though, so in one sense the original point stands (so long as it’s “stirring isn’t as important as you thought” rather than “stirring isn’t important at all”). And for larger scale reactions, I think that “stirring is important” is still undisputed (at least it had better be in any lab I’m working near!) And the reproducibility angle is a good one to remember as well. I do mostly exploratory reactions on small scale, so I really will never notice if my yields vary between 72% and 78%. But a process chemist will notice that for sure (and will also be working hard to find a way to not run such crappy reactions to start with!) I’m in the world (as I’ve said before) where there are basically two yields for a reaction, “enough” and “not enough”, and as long as I’m over the threshold and can move on I’m probably happy. But that is certainly not everyone’s world! Looking over the re-analysis of the reaction yields, it seems that the reaction types that showed the biggest pro-mixing effects were metal-catalyzed cross-couplings, electrochemical reactions, polar mechanisms with a lot of anion/cation character, and (weirdly, to me) rearrangements like the Claisen and Schmidt. The first two categories I can well believe, but the last one strikes me as odd because the Claisen (and many other sigmatropic rearrangements) are unimolecular, and you wouldn’t expect big mixing effects. The Schmidt is different, since you’re having to form an acyl azide (through a polar mechanism, a reaction class already noted) before the expulsion of nitrogen, and even after that you have to have water adding in, etc. So I can imagine mixing effects on that one, but the Claisen is a harder sell for me, unless the starting diene is having to form in situ prior to rearranging. Perhaps that’s what people are seeing with things like the Eschenmoser or Johnson variations of it? Those have more polar-leaving-group mechanisms tacked on the front end of the process, which could fit. So that’s the current state of debate, but I am willing to be that we haven’t heard the last of this topic. I’m interested in seeing where this all goes, but I would advise people to watch closely for signs of people talking past each other, in the “Elephants are big!” “No they’re not, they’re gray!” fashion. It'll be easy for that to happen. I’ll report back as things develop!
Show full content
Some weeks back I wrote about a paper suggesting that for many organic chemistry reactions stirring is not very necessary. That one sure set off a lot of comments (here on the blog, on social media, in my own email and conversations, etc.)! I wanted to revisit the topic in light of two more publications since then.
First off is this preprint, a swift response to the original, from a group of industrial chemists. They do a good job of reviewing the principles behind reaction mixing, and they emphasize that larger-scale reactions are going to be much more sensitive to its effects. I don’t see how anyone who’s done scale-up can disagree! I do have to note that the first paper did most cover smaller-scale reactions and that the authors did note that fact. The response paper is written by people who think about larger-scale reproducibility all the time, and under those conditions I don’t think anyone can argue about taking another variable out of the system by using reproducible stirring and mixing protocols. There are too many considerations of mass transfer, concentration changes, temperature gradients and more to ignore.
But on a small scale, how much do these matter? I think that’s what the first paper was getting at. So this new analysis of that paper’s actual data is certainly worth a look. The author takes the extensive supplementary data from that paper and goes back over it statistically, coming to somewhat different conclusions:
. . .where the lowest-yielding reaction group shows the largest overall increase in isolated yield (4.5%; 95% CI: 2.6 to 6.4%; p < 0.0001) with stirring. Stirring of moderate yield reactions is associated with a 1.6% (95% CI: 0.4 to 2.9%; p = 0.0082) increase in yield, while the highest yielding reactions show smaller effect sizes in stirred versus unstirred reactions (0.8% increase; 95% CI: 0.4 to 1.3%; p< 0.0001). These results suggest that while the overall change in yields with stirring in these data for many reactions are modest, there is extensive variation of impact, and stirring is associated with real and practically important increases in isolated yields across a wide range of substrates and reaction manifolds.
So (at least on a certain scale) reactions that work well are going to work well whether you pay attention to stirring them or not. But in other cases, it seems that you can get a slight (or sometimes more than slight) edge by stirring. I’d be willing to bet that these yield increases are well below what a random set of bench chemists would have assumed, though, so in one sense the original point stands (so long as it’s “stirring isn’t as important as you thought” rather than “stirring isn’t important at all”). And for larger scale reactions, I think that “stirring is important” is still undisputed (at least it had better be in any lab I’m working near!)
And the reproducibility angle is a good one to remember as well. I do mostly exploratory reactions on small scale, so I really will never notice if my yields vary between 72% and 78%. But a process chemist will notice that for sure (and will also be working hard to find a way to not run such crappy reactions to start with!) I’m in the world (as I’ve said before) where there are basically two yields for a reaction, “enough” and “not enough”, and as long as I’m over the threshold and can move on I’m probably happy. But that is certainly not everyone’s world!
Looking over the re-analysis of the reaction yields, it seems that the reaction types that showed the biggest pro-mixing effects were metal-catalyzed cross-couplings, electrochemical reactions, polar mechanisms with a lot of anion/cation character, and (weirdly, to me) rearrangements like the Claisen and Schmidt. The first two categories I can well believe, but the last one strikes me as odd because the Claisen (and many other sigmatropic rearrangements) are unimolecular, and you wouldn’t expect big mixing effects. The Schmidt is different, since you’re having to form an acyl azide (through a polar mechanism, a reaction class already noted) before the expulsion of nitrogen, and even after that you have to have water adding in, etc. So I can imagine mixing effects on that one, but the Claisen is a harder sell for me, unless the starting diene is having to form in situ prior to rearranging. Perhaps that’s what people are seeing with things like the Eschenmoser or Johnson variations of it? Those have more polar-leaving-group mechanisms tacked on the front end of the process, which could fit.
So that’s the current state of debate, but I am willing to be that we haven’t heard the last of this topic. I’m interested in seeing where this all goes, but I would advise people to watch closely for signs of people talking past each other, in the “Elephants are big!” “No they’re not, they’re gray!” fashion. It'll be easy for that to happen. I’ll report back as things develop!
A few years ago here, I wrote about an interesting hypothesis involving the TrkB receptor and the action of antidepressant drugs. The short form of that one is that TrkB is important in the signaling and action of the BDNF neuronal growth factor, and BDNF in turn has been the subject of several theories about major depression. The claim was that the BDNF/TrkB complex produces a small molecule binding site that accommodates many known antidepressant molecules, a use for them that their developers never anticipated, and that this might be a main mode of their actions in vivo. The same group believes that this might be a mechanism for efficacy of psychedelic drugs in treating some major depression patients as well, although this seems to require endogenous BDNF to be present, while some other ligands might be able to work without it. Here’s a recent review of the overall idea, and another is here. That’s a bold statement on a therapeutic area of great interest, so I wanted to highlight some recent papers that bear on it. This recent J. Med. Chem. paper describes a phenotypic screen for antidepressant compounds using the “chronic unpredictable mild stress” rodent model. They identified a few compounds with what is for antidepressants unusually fast onset - typically in humans you need weeks of dosing to start seeing effects (and indeed, this is what this paper noted with the widely used antidepressant fluoxetine). The authors went on to do compound optimization using rodent assays, and found a compound (their B11) that seemed promising. Further evaluation in neuronal cellular assays showed that the compound seemed to have notable growth-factor-like effects (similar to BDNF treatment, in fact), and brain tissue from treated mice seemed to confirm this. B11 treatment in fact upregulated BDNF levels within 15 minutes of treatment, and a set of further experiments (not detailed here for reasons of time!) all pointed to this being its likely mechanism of antidepressant action. The authors report that preclinical selectivity studies show no particular activities on a number of other CNS targets, and also say that toxicity testing has been performed in rats and dogs up to 1g/kg (!) So that would make a person very interested in seeing what this or a related compound would do in human patients! I would hope that this is in the works. The authors are all from a pharma company in Chengdu, so there is a reasonable possibility that we might see something like this. But if not them, then it’ll be someone (or something!) else, because this hypothesis is too interesting not to be investigated. Meanwhile, this recent paper zooms back out in this area to propose another related idea. The authors are also working the neuroplasticity-as-therapy idea for depression and other disorders, but they are proposing hypoxia (!) as a common thread. In their view, the hallucinogenic states brought on by hypoxia at high altitude or in near-death experiences might be setting off the same downstream neuronal adaptability mechanisms as those seen with hallucinogenic drugs themselves. There actually has been research in controlled intermittent hypoxia as a neurological therapy, which is something I hadn’t known. Here’s an overview. (As a personal note, if this is true then I've been through some seminars that might have done me a lot of good. . .) The hypoxia folks are also referencing BDNF pathways in their proposed mechanisms of action, and it’s not clear to me (and I’m sure that it’s not clear to anyone!) whether these TrkB approaches are a way to jump right through that without having to experience hypoxia or whether hypoxia provides added benefits that a TrkB-BDNF targeted approach would lack. Either way, as wooly as some of this is, I’m very glad to see such approaches being studied. Our treatment of depression (and many other cognitive disorders) is very rudimentary and has huge room for improvement. Show some real, reproducible benefit and let’s talk!
Show full content
A few years ago here, I wrote about an interesting hypothesis involving the TrkB receptor and the action of antidepressant drugs. The short form of that one is that TrkB is important in the signaling and action of the BDNF neuronal growth factor, and BDNF in turn has been the subject of several theories about major depression. The claim was that the BDNF/TrkB complex produces a small molecule binding site that accommodates many known antidepressant molecules, a use for them that their developers never anticipated, and that this might be a main mode of their actions in vivo. The same group believes that this might be a mechanism for efficacy of psychedelic drugs in treating some major depression patients as well, although this seems to require endogenous BDNF to be present, while some other ligands might be able to work without it. Here’s a recent review of the overall idea, and another is here.
That’s a bold statement on a therapeutic area of great interest, so I wanted to highlight some recent papers that bear on it. This recent J. Med. Chem. paper describes a phenotypic screen for antidepressant compounds using the “chronic unpredictable mild stress” rodent model. They identified a few compounds with what is for antidepressants unusually fast onset - typically in humans you need weeks of dosing to start seeing effects (and indeed, this is what this paper noted with the widely used antidepressant fluoxetine). The authors went on to do compound optimization using rodent assays, and found a compound (their B11) that seemed promising.
Further evaluation in neuronal cellular assays showed that the compound seemed to have notable growth-factor-like effects (similar to BDNF treatment, in fact), and brain tissue from treated mice seemed to confirm this. B11 treatment in fact upregulated BDNF levels within 15 minutes of treatment, and a set of further experiments (not detailed here for reasons of time!) all pointed to this being its likely mechanism of antidepressant action. The authors report that preclinical selectivity studies show no particular activities on a number of other CNS targets, and also say that toxicity testing has been performed in rats and dogs up to 1g/kg (!)
So that would make a person very interested in seeing what this or a related compound would do in human patients! I would hope that this is in the works. The authors are all from a pharma company in Chengdu, so there is a reasonable possibility that we might see something like this. But if not them, then it’ll be someone (or something!) else, because this hypothesis is too interesting not to be investigated.
Meanwhile, this recent paper zooms back out in this area to propose another related idea. The authors are also working the neuroplasticity-as-therapy idea for depression and other disorders, but they are proposing hypoxia (!) as a common thread. In their view, the hallucinogenic states brought on by hypoxia at high altitude or in near-death experiences might be setting off the same downstream neuronal adaptability mechanisms as those seen with hallucinogenic drugs themselves. There actually has been research in controlled intermittent hypoxia as a neurological therapy, which is something I hadn’t known. Here’s an overview. (As a personal note, if this is true then I've been through some seminars that might have done me a lot of good. . .)
The hypoxia folks are also referencing BDNF pathways in their proposed mechanisms of action, and it’s not clear to me (and I’m sure that it’s not clear to anyone!) whether these TrkB approaches are a way to jump right through that without having to experience hypoxia or whether hypoxia provides added benefits that a TrkB-BDNF targeted approach would lack. Either way, as wooly as some of this is, I’m very glad to see such approaches being studied. Our treatment of depression (and many other cognitive disorders) is very rudimentary and has huge room for improvement. Show some real, reproducible benefit and let’s talk!
I recommend anyone who wants to learn more about generic drug manufacturing to read this article at the New York Times (it’s a gift link, free to read). There’s been a lot of coverage of drug manufacturers “on-shoring” production and packaging in response to pressure from the Trump administration, and there is definitely some of that happening (although it doesn’t happen as quickly as it might seem to). But this is another world entirely. That’s because the generic drug business is so different from the patented prescription drug business that it might as well be a separate industry. Generics generally compete on price, for one thing, which is what (famously) patented drugs hardly do. And while prescription drug prices are high here in the US, our generic drugs have generally been some of the cheapest in the world. Competition is fierce, and there are a lot of manufacturers of all sizes around the world in the game, from really huge ones (Teva, Dr. Reddy’s, Sandoz and more) to little outfits in places that you would hardly believe. There are generally a whole list of possible ways to make these older drugs, but most of those routes are in the “now superseded” category as cheaper ones were found. Some of those, though, are really only cheaper on scale or if you have the equipment, so for those small local producers it’s whatever works with whatever’s available. And there’s a lot of confusion even on the large scale, because there are often a number of actual producers of these drugs and even more resellers and repackagers. Trying to figure out just where a given batch of pills was made, all the way down to its various intermediates and reagents along the way, can be very difficult - and by the time you’ve worked it out, those pathways might have changed (!) See this post from back in May for an example. What’s for sure is that not much of this sort of drug manufacturing takes place in the US, and there doesn’t seem to be much sign of it returning. That NYT article makes this painfully clear with a look at a former Dr. Reddy’s plant in Shreveport. It started out owned by Boots (UK) and then by BASF before being bought by Dr. Reddy’s in 2009, but it shut down earlier this year after losing millions of dollars a year for the company. They were mostly formulating and packaging drug substances that were themselves made overseas, but even that wasn't enough of a foothold. This sort of work has been on a downwards trend here a good twenty years now, both at the manufacturing end and at the formulation/packing end of the business, and the reasons are the same: very small margins in the generic business mean that every possible cost savings will be sought out. And it costs too much do it that sort of thing here as opposed to somewhere else. That’s it, in one sentence. Which means that the main way the administration would be able to make these things come back to the US would be to make the foreign drugs even more expensive, which is where tariffs come in. The end result is that the consumer pays higher prices for the drugs, whether they even get made in the US or not. I don’t see how you’re going to suddenly make the US manufacturing more competitive under those thin-margin conditions: this isn’t something that you’re going to solve by throwing AI at it. There are machines and production lines, pill presses and coating machines and sorters and cartons and shrink-wrapping machines before things are loaded into boxes and taken away by fork lifts. The people who do this in India (to pick a major example), who run those machines and fill those bottles and load those boxes, cost five to ten times less than American workers cost to do those jobs. Which is why that plant in Shreveport is sitting there, with dust gathering on its increasingly outmoded equipment.
Show full content
I recommend anyone who wants to learn more about generic drug manufacturing to read this article at the New York Times (it’s a gift link, free to read). There’s been a lot of coverage of drug manufacturers “on-shoring” production and packaging in response to pressure from the Trump administration, and there is definitely some of that happening (although it doesn’t happen as quickly as it might seem to). But this is another world entirely.
That’s because the generic drug business is so different from the patented prescription drug business that it might as well be a separate industry. Generics generally compete on price, for one thing, which is what (famously) patented drugs hardly do. And while prescription drug prices are high here in the US, our generic drugs have generally been some of the cheapest in the world. Competition is fierce, and there are a lot of manufacturers of all sizes around the world in the game, from really huge ones (Teva, Dr. Reddy’s, Sandoz and more) to little outfits in places that you would hardly believe.
There are generally a whole list of possible ways to make these older drugs, but most of those routes are in the “now superseded” category as cheaper ones were found. Some of those, though, are really only cheaper on scale or if you have the equipment, so for those small local producers it’s whatever works with whatever’s available. And there’s a lot of confusion even on the large scale, because there are often a number of actual producers of these drugs and even more resellers and repackagers. Trying to figure out just where a given batch of pills was made, all the way down to its various intermediates and reagents along the way, can be very difficult - and by the time you’ve worked it out, those pathways might have changed (!) See this post from back in May for an example.
What’s for sure is that not much of this sort of drug manufacturing takes place in the US, and there doesn’t seem to be much sign of it returning. That NYT article makes this painfully clear with a look at a former Dr. Reddy’s plant in Shreveport. It started out owned by Boots (UK) and then by BASF before being bought by Dr. Reddy’s in 2009, but it shut down earlier this year after losing millions of dollars a year for the company. They were mostly formulating and packaging drug substances that were themselves made overseas, but even that wasn't enough of a foothold. This sort of work has been on a downwards trend here a good twenty years now, both at the manufacturing end and at the formulation/packing end of the business, and the reasons are the same: very small margins in the generic business mean that every possible cost savings will be sought out. And it costs too much do it that sort of thing here as opposed to somewhere else.
That’s it, in one sentence. Which means that the main way the administration would be able to make these things come back to the US would be to make the foreign drugs even more expensive, which is where tariffs come in. The end result is that the consumer pays higher prices for the drugs, whether they even get made in the US or not. I don’t see how you’re going to suddenly make the US manufacturing more competitive under those thin-margin conditions: this isn’t something that you’re going to solve by throwing AI at it. There are machines and production lines, pill presses and coating machines and sorters and cartons and shrink-wrapping machines before things are loaded into boxes and taken away by fork lifts. The people who do this in India (to pick a major example), who run those machines and fill those bottles and load those boxes, cost five to ten times less than American workers cost to do those jobs. Which is why that plant in Shreveport is sitting there, with dust gathering on its increasingly outmoded equipment.
I have lost count of the number of times over the years that I’ve said “Huh, I didn’t think mass spec could do that”. So you’d think that I would be used to this by now, but apparently not, because that was my exact reaction to this new paper. It’s from a team of groups at Leiden, Utrecht, and Jena, and they report a “self-encoded library” technique for some pretty large-scale screening. It should be noted up front that there are no tags, labels, or isotopic enrichments involved in this. The paper demonstrates screening libraries up to about 500,000 compounds in their native state. These are produced by pretty straightforward solid-phase synthesis techniques, and the paper shows several of them using reactions like amide formation, SnAr amine displacement, heterocyclic condensations, palladium-catalyzed couplings and so on. As the world well knows by now, you can make an awful lot of compounds rather quickly through such techniques, and since this isn’t DNA-encoded library technology (where you of course have to use reactions whose conditions are compatible with the oligonucleotide barcode tags), your toolbox is rather large. These compound libraries are cleaved from the solid beads and then screened against immobilized protein targets using pretty standard affinity-driven techniques to winnow down to the most potent binders. That takes you up to the hit ID step, and this is where traditionally you’ve needed some sort of Secret Sauce to be able to figure out what your hits really are. Any reasonably sized library made by these combinatorial techniques is going to produce a lot of compounds with overlapping molecule weights, so just straight mass spec will not be enough to decide that question. You can use various isotopic labeling schemes to make the signal/noise better, and of course there’s the aforementioned DNA encoding. With DEL you have a unique DNA sequence attached to every single compound, and you use the terrifying powers of PCR and modern sequencing to track down and reveal very minute quantities of the best binders. (That’s why DEL collections can easily get up in to the tens of millions of possibilities per run). In this case, though, the authors do some preliminary work on a few hundred typical compounds from their libraries, and look at MS-MS fragmentation patterns. They build up a flow-chart library of the most common fragment events for given compound classes, and the resulting software (COMET, the Combinatorial Mass Encoding Tool) looks for hits that have both one of the predicted product masses and an associated fragment mass that matches the software’s predictions. As is usual with such systems, you’re generally willing to throw away a certain number of real hits if you can eliminate a much greater pile of false positives by doing so. But this is the part that I’m surprised works as well as it does! The combination of mass spec hardware throughput and ferocious computational power on the back end comes through again. The paper demonstrates the effectiveness of this system with a test case (good ol’ carbonic anhydrase) and a much more challenging target (flap endonuclease 1). The former seems to have worked very well, but their initial libraries came up blank against that latter target. That's not really a surprising result, since it doesn’t have much of a small-molecule binding pocket, but a focused 4000-compound library incorporating structural features with a better chance of hitting such a dsDNA binding site yielded two sub-micromolar hits. Overall, it looks like you really can get away with this! This idea looks particularly promising for focused-library type screening, I would think. The wide variety of chemisty available and the ability to quickly retool to different scaffolds is an advantage - you can get moving on this sort of thing without even having to work up a traditional assay based on FRET or other fluorescent/luminescent methods, and the readout is a very direct one. The authors estimate that under favorable conditions you could push the screening library size up past a million, which isn’t bad, but remember that these are going to be a million broadly related compounds. For that sort of screening you might still be better off with DEL, especially if you already have some of those libraries built (and most especially if you already have experience with that screening technique, which does have its idiosyncracies). A real advantage here is that the chemistry opportunities are much wider and there are far fewer steps in the hit ID part of the process. You do have to put in some work training the COMET system to recognize fragmentation patterns, but that doesn’t look too bad (and that data will be useful for the lifetime of that particular screening set).
Show full content
I have lost count of the number of times over the years that I’ve said “Huh, I didn’t think mass spec could do that”. So you’d think that I would be used to this by now, but apparently not, because that was my exact reaction to this new paper. It’s from a team of groups at Leiden, Utrecht, and Jena, and they report a “self-encoded library” technique for some pretty large-scale screening.
It should be noted up front that there are no tags, labels, or isotopic enrichments involved in this. The paper demonstrates screening libraries up to about 500,000 compounds in their native state. These are produced by pretty straightforward solid-phase synthesis techniques, and the paper shows several of them using reactions like amide formation, SnAr amine displacement, heterocyclic condensations, palladium-catalyzed couplings and so on. As the world well knows by now, you can make an awful lot of compounds rather quickly through such techniques, and since this isn’t DNA-encoded library technology (where you of course have to use reactions whose conditions are compatible with the oligonucleotide barcode tags), your toolbox is rather large.
These compound libraries are cleaved from the solid beads and then screened against immobilized protein targets using pretty standard affinity-driven techniques to winnow down to the most potent binders. That takes you up to the hit ID step, and this is where traditionally you’ve needed some sort of Secret Sauce to be able to figure out what your hits really are. Any reasonably sized library made by these combinatorial techniques is going to produce a lot of compounds with overlapping molecule weights, so just straight mass spec will not be enough to decide that question. You can use various isotopic labeling schemes to make the signal/noise better, and of course there’s the aforementioned DNA encoding. With DEL you have a unique DNA sequence attached to every single compound, and you use the terrifying powers of PCR and modern sequencing to track down and reveal very minute quantities of the best binders. (That’s why DEL collections can easily get up in to the tens of millions of possibilities per run).
In this case, though, the authors do some preliminary work on a few hundred typical compounds from their libraries, and look at MS-MS fragmentation patterns. They build up a flow-chart library of the most common fragment events for given compound classes, and the resulting software (COMET, the Combinatorial Mass Encoding Tool) looks for hits that have both one of the predicted product masses and an associated fragment mass that matches the software’s predictions. As is usual with such systems, you’re generally willing to throw away a certain number of real hits if you can eliminate a much greater pile of false positives by doing so. But this is the part that I’m surprised works as well as it does! The combination of mass spec hardware throughput and ferocious computational power on the back end comes through again.
The paper demonstrates the effectiveness of this system with a test case (good ol’ carbonic anhydrase) and a much more challenging target (flap endonuclease 1). The former seems to have worked very well, but their initial libraries came up blank against that latter target. That's not really a surprising result, since it doesn’t have much of a small-molecule binding pocket, but a focused 4000-compound library incorporating structural features with a better chance of hitting such a dsDNA binding site yielded two sub-micromolar hits. Overall, it looks like you really can get away with this!
This idea looks particularly promising for focused-library type screening, I would think. The wide variety of chemisty available and the ability to quickly retool to different scaffolds is an advantage - you can get moving on this sort of thing without even having to work up a traditional assay based on FRET or other fluorescent/luminescent methods, and the readout is a very direct one. The authors estimate that under favorable conditions you could push the screening library size up past a million, which isn’t bad, but remember that these are going to be a million broadly related compounds. For that sort of screening you might still be better off with DEL, especially if you already have some of those libraries built (and most especially if you already have experience with that screening technique, which does have its idiosyncracies). A real advantage here is that the chemistry opportunities are much wider and there are far fewer steps in the hit ID part of the process. You do have to put in some work training the COMET system to recognize fragmentation patterns, but that doesn’t look too bad (and that data will be useful for the lifetime of that particular screening set).
It is a longstanding dream of mine that some day, some year I will no longer feel obliged to write disparaging blog posts about Sarepta and their Duchenne muscular dystrophy drug portfolio. I have been doing that for a loooooong time now (twelve years?) as the company, with what from one perspective is admirable persistence, has continued to develop and test various small-molecule drugs and gene therapies. My misgivings have come from the clinical trials of these agents, which have been very small and (to my eyes) very inconclusive. And from the FDA’s decisions over the years to go ahead and approve them anyway, despite what is (to my eyes, and not just to my eyes) a near complete lack of evidence for any real-world benefit. FDA committee members have resigned in protest over these decisions, and in other cases foreign regulatory authorities (like the EMA) have refused authorization entirely. The first of the Sarepta “exon-skipping” drugs approved in this fashion was eteplirsen (Exondys 51). The FDA let this one through in 2016 (more here) and a follow-up trial reported results in 2021, with what I would call very equivocal and overall unimpressive results. The second drug, golodirsen (Vyondys 53) was approved in 2019 (after having first been rejected!) for a different Duchenne population with a different mutation (exon 53, thus the name). And a third, casimersen (Amondys 45) was approved in 2021 for a third Duchenne population. That last one was comparatively uneventful: by that point the FDA had clearly established a precedent for approving these things based on biomarker data on the disease-relevant protein without reference to whether that translated into clinical efficacy. All three of these approvals were/are in theory contingent on the result of follow-up efficacy studies among the paying customers, although (as mentioned) the eteplirsen/Exondys results in 2021 did not exactly seal that deal. The studies for the other two have now, these many years later, finally read out. I realize that recruitment is difficult in this area, but hey, at least the company has been selling the drugs and booking the revenues over all that stretch, right? And the result is. . .well, gosh, what do you think? Neither of them work. They never had. They never did. They never will. Sarepta says that it does not believe that the drugs will lose their marketing authorization as a result of these readouts, but why the hell shouldn’t they? The drugs do not work. Sarepta, though, is apparently filing for full approval nonetheless, which should get them an award for sheer nerve if nothing else. As mentioned above, the company also has some gene therapies in the works. The first of these (Elevidys) was. . .wait for it, rejected by the FDA until that decision was overruled. Unfortunately there have been patient deaths directly linked to this one. The viral vector used for the gene therapy is very hard on the liver, and these events have been designated as treatment-related fatalities. So in this case you’d be waiting years for yet another confirmatory trial that might possibly show some benefit - although there’s not really any sign of that in the earlier data - but all the time exposing the Duchenne patients to what appears to be a real, provable risk of liver failure. This is not a promising situation. Reasonable people can disagree on how to characterize things at this point. Perhaps Sarepta has just been in there giving it their best shot over and over against a very difficult disease, but not making the progress that they had hoped for. Perhaps the FDA has been trying to give hope to a basically untreated-because-currently-untreatable patient population by approving these therapies on a risk basis. But it’s also the case that dreams often have to come to an end. Sarepta’s drugs have had multiple chances to show that they can improve the lives of Duchenne’s sufferers, and as far as I can see they have failed to do so every single time. It’s time to stop wishing and hoping that this isn’t, that this wouldn’t be the case. It’s time to stop selling these patients drugs that don’t help them and (in the case of Elevidys) are actually harming them. It’s time for the FDA to do its job - and it’s just our bad luck that this comes at a time when the FDA has never been less able to do that.
Show full content
It is a longstanding dream of mine that some day, some year I will no longer feel obliged to write disparaging blog posts about Sarepta and their Duchenne muscular dystrophy drug portfolio. I have been doing that for a loooooong time now (twelve years?) as the company, with what from one perspective is admirable persistence, has continued to develop and test various small-molecule drugs and gene therapies. My misgivings have come from the clinical trials of these agents, which have been very small and (to my eyes) very inconclusive. And from the FDA’s decisions over the years to go ahead and approve them anyway, despite what is (to my eyes, and not just to my eyes) a near complete lack of evidence for any real-world benefit. FDA committee members have resigned in protest over these decisions, and in other cases foreign regulatory authorities (like the EMA) have refused authorization entirely.
The first of the Sarepta “exon-skipping” drugs approved in this fashion was eteplirsen (Exondys 51). The FDA let this one through in 2016 (more here) and a follow-up trial reported results in 2021, with what I would call very equivocal and overall unimpressive results. The second drug, golodirsen (Vyondys 53) was approved in 2019 (after having first been rejected!) for a different Duchenne population with a different mutation (exon 53, thus the name). And a third, casimersen (Amondys 45) was approved in 2021 for a third Duchenne population. That last one was comparatively uneventful: by that point the FDA had clearly established a precedent for approving these things based on biomarker data on the disease-relevant protein without reference to whether that translated into clinical efficacy. All three of these approvals were/are in theory contingent on the result of follow-up efficacy studies among the paying customers, although (as mentioned) the eteplirsen/Exondys results in 2021 did not exactly seal that deal.
The studies for the other two have now, these many years later, finally read out. I realize that recruitment is difficult in this area, but hey, at least the company has been selling the drugs and booking the revenues over all that stretch, right? And the result is. . .well, gosh, what do you think? Neither of them work. They never had. They never did. They never will. Sarepta says that it does not believe that the drugs will lose their marketing authorization as a result of these readouts, but why the hell shouldn’t they? The drugs do not work. Sarepta, though, is apparently filing for full approval nonetheless, which should get them an award for sheer nerve if nothing else.
As mentioned above, the company also has some gene therapies in the works. The first of these (Elevidys) was. . .wait for it, rejected by the FDA until that decision was overruled. Unfortunately there have been patient deaths directly linked to this one. The viral vector used for the gene therapy is very hard on the liver, and these events have been designated as treatment-related fatalities. So in this case you’d be waiting years for yet another confirmatory trial that might possibly show some benefit - although there’s not really any sign of that in the earlier data - but all the time exposing the Duchenne patients to what appears to be a real, provable risk of liver failure. This is not a promising situation.
Reasonable people can disagree on how to characterize things at this point. Perhaps Sarepta has just been in there giving it their best shot over and over against a very difficult disease, but not making the progress that they had hoped for. Perhaps the FDA has been trying to give hope to a basically untreated-because-currently-untreatable patient population by approving these therapies on a risk basis. But it’s also the case that dreams often have to come to an end. Sarepta’s drugs have had multiple chances to show that they can improve the lives of Duchenne’s sufferers, and as far as I can see they have failed to do so every single time. It’s time to stop wishing and hoping that this isn’t, that this wouldn’t be the case. It’s time to stop selling these patients drugs that don’t help them and (in the case of Elevidys) are actually harming them. It’s time for the FDA to do its job - and it’s just our bad luck that this comes at a time when the FDA has never been less able to do that.
I’d like to take a bit of time to note this paper and its authors. It was published in May of this year in JACS, and it was about the conversion of carbon dioxide to methane. That’s certainly of great interest - it’s basically “reverse combustion”, and as you can imagine there are a lot of people interested in taking industrial carbon dioxide emissions and sending them back around by such a process. You could even imagine a technology that could strip it from the Earth’s atmosphere, although that’s a truly difficult problem for reasons of scale and because of the comparatively low levels of CO2 that you’d be trying to target (low, but unfortunately rather meaningful for the Earth’s climate!) The paper specifically was looking at electrochemical reduction of dilute carbon dioxide, though, which is good. Was good. You’ve probably guessed the reason for all that switching in the past tense. The paper has now been retracted by its authors. They had been working further on their reaction system and found inconsistent results, prompting them to go back over things. One of the crucial pieces of evidence was from the use of 13C-labeled CO2 (which was shown at the time to produce labeled methane, which would be a very strong piece of evidence indeed). The mass spec equipment they were using, though, was not calibrated correctly, and when that was fixed the methane produced from their system had no isotopic label at all (!) So that makes you wonder where it was coming from then, doesn’t it? The story gets more painful: the THF solvent they were using turned out to have very low levels of dichloromethane in it as an impurity, and that’s what was getting electrochemically reduced to methane. Cleaning up the THF led to no methane being produced at all. Now that is a painful result, and I have a great deal of sympathy for the UC-Irvine researchers who had to experience this. There are a lot of variables in the chemistry business, and once in a while you get wrong data that all point in the same (wrong) direction, perfectly mimicking what successful experiments look like. I’ve experienced this, and I’ll bet a number of readers here have as well. One big difference of course is that I had not published a full paper in JACS before discovering the problems, but to be fair being in industry insulates you against that to a degree, since publication in the journals is not a priority for us. But I wanted to congratulate everyone involved anyway. They realized something was off, and they kept working until they found out exactly what it was even as it undermined the entire idea behind the whole effort and the published paper. And they quickly and publicly corrected the record: the paper came out in May, and the retraction in October. This is how science is supposed to work, and this is the work of real scientists, people of honestly and integrity, and I am glad they are doing this research and doing it with such respect for accuracy and truthfulness. I only wish that more people would live up to these standards. There are a lot of papers out there that have deserved to be retracted for years. In many of these cases, readers have flagged obvious problems at post-publication sites like PubPeer. In a subset of those, journal editors and institutions have tried to take action. But what is rare, rare indeed, is when the authors themselves look at their own work and agree that yes, this needs to be retracted. Many of them ignore all such attempts, actively resist any inquiry into their work, or impugn the motives of anyone who dares to suggest it. But in this case the authors were the ones who caught their own (honest) mistakes and did the right thing as quickly as possible. That’s how it’s supposed to work. I wish them every success in the next iteration of this research.
Show full content
I’d like to take a bit of time to note this paper and its authors. It was published in May of this year in JACS, and it was about the conversion of carbon dioxide to methane. That’s certainly of great interest - it’s basically “reverse combustion”, and as you can imagine there are a lot of people interested in taking industrial carbon dioxide emissions and sending them back around by such a process. You could even imagine a technology that could strip it from the Earth’s atmosphere, although that’s a truly difficult problem for reasons of scale and because of the comparatively low levels of CO2 that you’d be trying to target (low, but unfortunately rather meaningful for the Earth’s climate!) The paper specifically was looking at electrochemical reduction of dilute carbon dioxide, though, which is good. Was good.
You’ve probably guessed the reason for all that switching in the past tense. The paper has now been retracted by its authors. They had been working further on their reaction system and found inconsistent results, prompting them to go back over things. One of the crucial pieces of evidence was from the use of 13C-labeled CO2 (which was shown at the time to produce labeled methane, which would be a very strong piece of evidence indeed). The mass spec equipment they were using, though, was not calibrated correctly, and when that was fixed the methane produced from their system had no isotopic label at all (!) So that makes you wonder where it was coming from then, doesn’t it? The story gets more painful: the THF solvent they were using turned out to have very low levels of dichloromethane in it as an impurity, and that’s what was getting electrochemically reduced to methane. Cleaning up the THF led to no methane being produced at all.
Now that is a painful result, and I have a great deal of sympathy for the UC-Irvine researchers who had to experience this. There are a lot of variables in the chemistry business, and once in a while you get wrong data that all point in the same (wrong) direction, perfectly mimicking what successful experiments look like. I’ve experienced this, and I’ll bet a number of readers here have as well. One big difference of course is that I had not published a full paper in JACS before discovering the problems, but to be fair being in industry insulates you against that to a degree, since publication in the journals is not a priority for us.
But I wanted to congratulate everyone involved anyway. They realized something was off, and they kept working until they found out exactly what it was even as it undermined the entire idea behind the whole effort and the published paper. And they quickly and publicly corrected the record: the paper came out in May, and the retraction in October. This is how science is supposed to work, and this is the work of real scientists, people of honestly and integrity, and I am glad they are doing this research and doing it with such respect for accuracy and truthfulness.
I only wish that more people would live up to these standards. There are a lot of papers out there that have deserved to be retracted for years. In many of these cases, readers have flagged obvious problems at post-publication sites like PubPeer. In a subset of those, journal editors and institutions have tried to take action. But what is rare, rare indeed, is when the authors themselves look at their own work and agree that yes, this needs to be retracted. Many of them ignore all such attempts, actively resist any inquiry into their work, or impugn the motives of anyone who dares to suggest it.
But in this case the authors were the ones who caught their own (honest) mistakes and did the right thing as quickly as possible. That’s how it’s supposed to work. I wish them every success in the next iteration of this research.
The “nocebo” effect is something that makes a lot of sense when you think about it, but it still seems weird. Everyone has heard of the placebo effect, where some interventions tend to have a beneficial effect if you think that they’re having (or going to have) a beneficial effect. There is no doubt that this is real, although its magnitude varies a great deal depending on circumstances, as it well should. The nocebo effect is just that with the sign flipped: there are things that have a real negative effect on people when they believe that a negative effect is there. Physical pain has been a great proving ground for both of these effects, and there’s a great deal of literature on this. That’s well-summed-up in this new paper, which has some disconcerting new results to add. As the authors note, the nocebo literature is much smaller than the placebo one, in no small part due to the latter effect being recognized much earlier. One of the open questions has been whether people who have been shown to be more responsive to placebo effects are more so to nocebo ones as well (or vice versa). This work looked at pain (from experimentally applied heat) in over a hundred healthy volunteers, and they were looking at both the perception of pain reduction (placebo) and pain aggravation (nocebo) at temperatures that were the same difference off the control temperature. And they do a pretty exhaustive comparison between all three conditions. One thing that came out first was that the nocebo effect seemed to be stronger than the placebo one in the initial round of experiments (Day 1). The same subjects came back for an identical round of testing a week later, and the nocebo effect was still stronger than placebo. The strength of the two had been significantly correlated on Day 1, but not on Day 8, interestingly. On both days, looking at “expectancy” in the volunteers showed that the expected pain relief was noticeably stronger than the expected pain worsening. But neither set of expectations were linked to the real experiences on Day 1. It’s possible that the “stronger nocebo effect” setting in human psychology is driven by evolutionary adaptation (avoidance of threats and negative consequences, which is a feature of human behavior that’s been demonstrated in many model systems). One thing this paper should serve as a warning for, though, is around the persistence of these nocebo effects: experimenters should not assume that they’re going to fade away over time! This especially applies to experiment-subject relationships, with positive/negative framing, overall relationship (trust, perceptions of competence), and amount of time spent dwelling on possible side effects or other negative consequences. I hope that these findings are extended to the pharmaceutical side of things!
Show full content
The “nocebo” effect is something that makes a lot of sense when you think about it, but it still seems weird. Everyone has heard of the placebo effect, where some interventions tend to have a beneficial effect if you think that they’re having (or going to have) a beneficial effect. There is no doubt that this is real, although its magnitude varies a great deal depending on circumstances, as it well should. The nocebo effect is just that with the sign flipped: there are things that have a real negative effect on people when they believe that a negative effect is there.
Physical pain has been a great proving ground for both of these effects, and there’s a great deal of literature on this. That’s well-summed-up in this new paper, which has some disconcerting new results to add. As the authors note, the nocebo literature is much smaller than the placebo one, in no small part due to the latter effect being recognized much earlier. One of the open questions has been whether people who have been shown to be more responsive to placebo effects are more so to nocebo ones as well (or vice versa).
This work looked at pain (from experimentally applied heat) in over a hundred healthy volunteers, and they were looking at both the perception of pain reduction (placebo) and pain aggravation (nocebo) at temperatures that were the same difference off the control temperature. And they do a pretty exhaustive comparison between all three conditions. One thing that came out first was that the nocebo effect seemed to be stronger than the placebo one in the initial round of experiments (Day 1). The same subjects came back for an identical round of testing a week later, and the nocebo effect was still stronger than placebo. The strength of the two had been significantly correlated on Day 1, but not on Day 8, interestingly. On both days, looking at “expectancy” in the volunteers showed that the expected pain relief was noticeably stronger than the expected pain worsening. But neither set of expectations were linked to the real experiences on Day 1.
It’s possible that the “stronger nocebo effect” setting in human psychology is driven by evolutionary adaptation (avoidance of threats and negative consequences, which is a feature of human behavior that’s been demonstrated in many model systems). One thing this paper should serve as a warning for, though, is around the persistence of these nocebo effects: experimenters should not assume that they’re going to fade away over time! This especially applies to experiment-subject relationships, with positive/negative framing, overall relationship (trust, perceptions of competence), and amount of time spent dwelling on possible side effects or other negative consequences. I hope that these findings are extended to the pharmaceutical side of things!
Here’s another example of an idea that has been kicking around for years in medicinal chemistry without ever really breaking through: substituting a silicon atom for a carbon. To be fair, most of the time this doesn’t seem to do all that much, while introducing various uncertainties around ADME and toxicity (since we don’t have all that much experience with organosilanes as drugs). So you can see why we’re not overrun with “silyl switch” compounds. But at the same time, there really do seem to be instances where it can help. For instance, there was (is?) a camptothecin derivative, known variously as karenitecin, cositecan, or BNP1350, that had an alkylsilyl side chain that was claimed to help it be less prone to being removed by efflux pumps. As far as I can tell, this one has kicked around in a number of Phase I and II trials without ever advancing. And a silane analog of haloperidol did indeed show a different (and quite possibly beneficial) metabolic profile, but I don’t think that one even made it to the clinic. As I mentioned in that blog post linked in the first paragraph, I sent in a trimethylsilyl-for-t-butyl switch compound one time in an analoging program, and I have to say that the response from the project team was not a favorable one. But as often happens, there seemed to be no particular advantage to the TMS analog, so it didn’t become an issue, other than in the “Please don’t do that again” way. This new paper (first link in the post) is a silyl-containing KIF18A kinesin inhibitor, which is class of compounds with several representatives, some of which are in the clinic already for susceptible cancers. Like the example mentioned above, this switch (a silapiperidine for plain piperidine) seems to have improved efflux stability. I’m not completely sure how this occurs, though - the silicon analogs are a big less hydrophilic, but what efflux transport proteins like and dislike is still a mystery to me (and no, not just to me!) I find it hard to believe that “silicon slows down efflux pumping” will turn out to be a general rule, but I think it’s an idea that’s worth testing if your particular project is having that sort of trouble. Just be ready for some pushback! We’ll see if this compound (ATX020) advances. The company behind it (Accent Therapeutics) is calling it a “tool compound”, but we’ll see if they have the nerve (or the need!) to take a similar organosilane into human trials. . .
Show full content
Here’s another example of an idea that has been kicking around for years in medicinal chemistry without ever really breaking through: substituting a silicon atom for a carbon. To be fair, most of the time this doesn’t seem to do all that much, while introducing various uncertainties around ADME and toxicity (since we don’t have all that much experience with organosilanes as drugs). So you can see why we’re not overrun with “silyl switch” compounds. But at the same time, there really do seem to be instances where it can help.
For instance, there was (is?) a camptothecin derivative, known variously as karenitecin, cositecan, or BNP1350, that had an alkylsilyl side chain that was claimed to help it be less prone to being removed by efflux pumps. As far as I can tell, this one has kicked around in a number of Phase I and II trials without ever advancing. And a silane analog of haloperidol did indeed show a different (and quite possibly beneficial) metabolic profile, but I don’t think that one even made it to the clinic. As I mentioned in that blog post linked in the first paragraph, I sent in a trimethylsilyl-for-t-butyl switch compound one time in an analoging program, and I have to say that the response from the project team was not a favorable one. But as often happens, there seemed to be no particular advantage to the TMS analog, so it didn’t become an issue, other than in the “Please don’t do that again” way.
This new paper (first link in the post) is a silyl-containing KIF18A kinesin inhibitor, which is class of compounds with several representatives, some of which are in the clinic already for susceptible cancers. Like the example mentioned above, this switch (a silapiperidine for plain piperidine) seems to have improved efflux stability. I’m not completely sure how this occurs, though - the silicon analogs are a big less hydrophilic, but what efflux transport proteins like and dislike is still a mystery to me (and no, not just to me!)
I find it hard to believe that “silicon slows down efflux pumping” will turn out to be a general rule, but I think it’s an idea that’s worth testing if your particular project is having that sort of trouble. Just be ready for some pushback! We’ll see if this compound (ATX020) advances. The company behind it (Accent Therapeutics) is calling it a “tool compound”, but we’ll see if they have the nerve (or the need!) to take a similar organosilane into human trials. . .
I think that I can guarantee that you haven’t heard this phrase before: “ballistic microscopy”, the subject of this recent preprint. What the authors describe a combination of near-medieval technology on the one hand and cutting-edge analytical work on the other. They are bombarding cells with focused streams of gold nanoparticles (which range from 50 to 1000 nm diameter). These things are traveling at speeds up to 1 km/sec (over 2000 miles per hour (edit: fixed!)) and blast straight through their cellular targets. That’s a thickness of 2 to 4 microns for something like a HEK cell, and those velocities mean that the transit takes only a few picoseconds. They come out the other side of the cell and splat into a hydrogel matrix on the other side. They’ve already slowed down a bit from their passage through the cell, and the hydrogel brings them to a halt. But when you examine them there, you find that they have carried along small amounts of the cellular material with them. It’s only a few attoliters, but by gosh that’s enough for current proteomic, nucleic acid, and cryo-EM techniques to get a handle on what’s in there. So what you get is an instantaneous snapshot of the cellular contents from a very small, very well defined needle-stick through a living cell. (People have actually done that, sampling cells with micro-needles and micro-straws, but this seems to be a step further). You can tell that the authors are enjoying themselves: the technique itself is abbreviated BaM, and the hydrogel sample obtained is referred to as a “SPLAT-MAP”. (If that’s an acronym it seems to be undefined in the manuscript!) You get a lot of information from doing fluorescent imaging while the bombardment is underway - location of the particle stream hitting the cell (complete with streaks through the cytoplasm in high-speed side views), xy spatial distribution on the hydrogel itself, and depth (z) which is dependent on the size of the particles involved. The group tested this in lysate from cells that had been expressing GFP-labled actin protein, and sure enough: the particles entrained fluorescent bits of cell material that corresponded to the labeled protein. And those particles penetrated less into the hydrogel braking material than control particles that were shot in directly, showing that they had experienced drag from schlorking through the cellular contents (my term, which all are welcome to if this technique catches on). Moving on to real cells, HEK293 cells were stained for nuclear membrane and cell membrane (to aid in IDing the now-fluorescent particles after capture), and they could be cultured right on top of the hydrogel surface. If the fluorescent label was applied instead to another protein, then everything around that protein could be checked out. This was done with the known condensate-former CLIP170, and the nanoparticles pulled condensate droplets right out of the cell. Proteomic analysis showed 641 proteins (with a large number of them annotated as RNA binders, which fits with previous condensate work). One was keratin-18, which hadn’t been seen in these before but which seems to form filaments inside the droplets. But about 17% of them are unannotated, which is just the sort of thing you’d like to dredge up with a method like this. Electron microscopy of the particles and their associated cellular samples showed that the cell contents that were brought along tended to be bunched up on the high-curvature edges of the gold particles (and not wetting the entire surface) and that they tended to be membrane-enclosed, sometimes with more than one membrane layer. There’s going to have to be more work done to interpret that, but it does seem significant (and might represent a type of sampling bias with this technique?) There are a lot of things to be done in general! Zapping all sorts of cellular substructures, in both healthy and diseased or stressed cells, is an obvious set of experiments, and it’ll be interesting to see if some protein distribution maps can be produced from such runs. It’s certainly a new label-free assay technique, and I urge everyone interested in it to fire away and collect piles of data!
Show full content
I think that I can guarantee that you haven’t heard this phrase before: “ballistic microscopy”, the subject of this recent preprint. What the authors describe a combination of near-medieval technology on the one hand and cutting-edge analytical work on the other. They are bombarding cells with focused streams of gold nanoparticles (which range from 50 to 1000 nm diameter). These things are traveling at speeds up to 1 km/sec (over 2000 miles per hour (edit: fixed!)) and blast straight through their cellular targets. That’s a thickness of 2 to 4 microns for something like a HEK cell, and those velocities mean that the transit takes only a few picoseconds.
They come out the other side of the cell and splat into a hydrogel matrix on the other side. They’ve already slowed down a bit from their passage through the cell, and the hydrogel brings them to a halt. But when you examine them there, you find that they have carried along small amounts of the cellular material with them. It’s only a few attoliters, but by gosh that’s enough for current proteomic, nucleic acid, and cryo-EM techniques to get a handle on what’s in there. So what you get is an instantaneous snapshot of the cellular contents from a very small, very well defined needle-stick through a living cell. (People have actually done that, sampling cells with micro-needles and micro-straws, but this seems to be a step further).
You can tell that the authors are enjoying themselves: the technique itself is abbreviated BaM, and the hydrogel sample obtained is referred to as a “SPLAT-MAP”. (If that’s an acronym it seems to be undefined in the manuscript!) You get a lot of information from doing fluorescent imaging while the bombardment is underway - location of the particle stream hitting the cell (complete with streaks through the cytoplasm in high-speed side views), xy spatial distribution on the hydrogel itself, and depth (z) which is dependent on the size of the particles involved.
The group tested this in lysate from cells that had been expressing GFP-labled actin protein, and sure enough: the particles entrained fluorescent bits of cell material that corresponded to the labeled protein. And those particles penetrated less into the hydrogel braking material than control particles that were shot in directly, showing that they had experienced drag from schlorking through the cellular contents (my term, which all are welcome to if this technique catches on). Moving on to real cells, HEK293 cells were stained for nuclear membrane and cell membrane (to aid in IDing the now-fluorescent particles after capture), and they could be cultured right on top of the hydrogel surface.
If the fluorescent label was applied instead to another protein, then everything around that protein could be checked out. This was done with the known condensate-former CLIP170, and the nanoparticles pulled condensate droplets right out of the cell. Proteomic analysis showed 641 proteins (with a large number of them annotated as RNA binders, which fits with previous condensate work). One was keratin-18, which hadn’t been seen in these before but which seems to form filaments inside the droplets. But about 17% of them are unannotated, which is just the sort of thing you’d like to dredge up with a method like this.
Electron microscopy of the particles and their associated cellular samples showed that the cell contents that were brought along tended to be bunched up on the high-curvature edges of the gold particles (and not wetting the entire surface) and that they tended to be membrane-enclosed, sometimes with more than one membrane layer. There’s going to have to be more work done to interpret that, but it does seem significant (and might represent a type of sampling bias with this technique?)
There are a lot of things to be done in general! Zapping all sorts of cellular substructures, in both healthy and diseased or stressed cells, is an obvious set of experiments, and it’ll be interesting to see if some protein distribution maps can be produced from such runs. It’s certainly a new label-free assay technique, and I urge everyone interested in it to fire away and collect piles of data!
This is a rather unexpected article that suggests that some mRNA vaccines can potentiate the actions of some immune-checkpoint therapies used in oncology. Specifically, the authors find that the mRNA coronavirus vaccines significantly increased overall survival rates in those patients who were getting anti-PD-1 antibodies as immune checkpoint inhibitor therapy (!) In a welcome reverse of the usual way we end up studying cancer therapies, this effect could be replicated in mice. There was a strong synergistic effect of mRNA lipid nanoparticle vaccination and anti-PD-1 monoclonal antibody treatment in mouse cancer models, well beyond what either could achieve on their own. And this seems to be driven through Type 1 interferon pathways (which may remind you of the cancer vaccine work that I blogged about here the other day). It wasn’t the spike-protein mRNA that did this specifically - the authors replaced that with mRNA against a cytomegalovirus antigen and this vaccination had the same effect. So it’s the immune-potentiating effect of the mRNA dose itself that’s helping, not anything to do with the specific antigen that’s it’s aimed at. The authors went to some trouble to try to figure out the mechanism behind this. They rigorously purified out any remaining double-stranded RNA that might have been in the vaccine doses, but that had no effect. On the other hand, changing the lipid nanoparticle formulation did wipe out the benefits - coupling that with other literature reports, they hypothesize that the LNPs form higher-molecular-weight secondary RNA structures that activate some of the innate immune receptors (like MDA5) that are watching for double-stranded RNA as a sign of viral infection. They were able to demonstrate strong immune activation in the mice after these doses through multiple interferon-driven pathways, but (interestingly) could only partially replicate it by dosing them with interferon-alpha itself. Both innate and adaptive immune responses kicked into gear. So there’s a lot going on in there, and it’s safe to assume that we don’t know the whole story yet, but what seems clear is that the RNA-LNP vaccination primes the immune system for activation, presentation, and recognition of many tumor-associated antigen proteins, and that these effects are strongly potentiated if you’re simultaneously being treated with immune checkpoint inhibitors like the anti-PD1 monoclonal antibodies. Another interesting effect was that in human volunteers the cytokine response to the vaccines was stronger with the Moderna shot (mRNA-1273) than it was with the Pfizer-BioNTech one (BNT162b2), and this appears to be a sheer effect of the former vaccine having more total mRNA in it (!) The mRNA vaccines seem to amplify tumor PD-L1 across a wide range of tumor types, and this effect even seems to extend to immunologically “cold” tumors that would otherwise have been considered not to be responding to treatment. This is really, really good news, and it immediately suggests that patients who are eligible for immune-checkpoint therapy (things like Keytruda, Opdivo, Tecentriq, Bavencio, Imfinzi, etc.) should be dosed with some sort of LNP-mRNA shot right along with it. And it also suggests that there are patients who might not be thought to benefit from such ICI treatment who should be given a crack at it with this protocol, since the enhancement seems so robust. There’s obviously a lot more to be worked out as far as doses of mRNA, timing, formulations, tumor types and all sorts of other factors, but the survival curves in this paper argue against waiting for patients who are being treated right now. The benefits seem to be very real, and the risks - particularly compared to the underlying cancers - seem to be very low. Go for it! I made this point the other day, and I want to make it again, because it cannot be overemphasized: there is still so much that we don’t know about immunology. We literally don’t know when something great like this will pop up, so we have to keep banging away in these areas and discovering as we go along. And while I’m at it, there is another thing that cannot be overemphasized: if you are spreading doubts and fears about vaccines and about mRNA research, you are killing people who could benefit. Killing them by not protecting them from the diseases being vaccinated against. Killing them by making them suspicious of advances in medical science in general. Killing them by driving them away from things like this new work that could fight completely unrelated diseases like cancer. That’s you, Robert F. Kennedy, Jr. That’s you, Jay Bhattacharya. That’s you, Marty Makary. And all the rest of your minions and gofers and toadies. You have chosen the side of death and human suffering, and in a better world than this one you would be disgraced and shunned for it.
Show full content
This is a rather unexpected article that suggests that some mRNA vaccines can potentiate the actions of some immune-checkpoint therapies used in oncology. Specifically, the authors find that the mRNA coronavirus vaccines significantly increased overall survival rates in those patients who were getting anti-PD-1 antibodies as immune checkpoint inhibitor therapy (!)
In a welcome reverse of the usual way we end up studying cancer therapies, this effect could be replicated in mice. There was a strong synergistic effect of mRNA lipid nanoparticle vaccination and anti-PD-1 monoclonal antibody treatment in mouse cancer models, well beyond what either could achieve on their own. And this seems to be driven through Type 1 interferon pathways (which may remind you of the cancer vaccine work that I blogged about here the other day).
It wasn’t the spike-protein mRNA that did this specifically - the authors replaced that with mRNA against a cytomegalovirus antigen and this vaccination had the same effect. So it’s the immune-potentiating effect of the mRNA dose itself that’s helping, not anything to do with the specific antigen that’s it’s aimed at. The authors went to some trouble to try to figure out the mechanism behind this. They rigorously purified out any remaining double-stranded RNA that might have been in the vaccine doses, but that had no effect. On the other hand, changing the lipid nanoparticle formulation did wipe out the benefits - coupling that with other literature reports, they hypothesize that the LNPs form higher-molecular-weight secondary RNA structures that activate some of the innate immune receptors (like MDA5) that are watching for double-stranded RNA as a sign of viral infection.
They were able to demonstrate strong immune activation in the mice after these doses through multiple interferon-driven pathways, but (interestingly) could only partially replicate it by dosing them with interferon-alpha itself. Both innate and adaptive immune responses kicked into gear. So there’s a lot going on in there, and it’s safe to assume that we don’t know the whole story yet, but what seems clear is that the RNA-LNP vaccination primes the immune system for activation, presentation, and recognition of many tumor-associated antigen proteins, and that these effects are strongly potentiated if you’re simultaneously being treated with immune checkpoint inhibitors like the anti-PD1 monoclonal antibodies.
Another interesting effect was that in human volunteers the cytokine response to the vaccines was stronger with the Moderna shot (mRNA-1273) than it was with the Pfizer-BioNTech one (BNT162b2), and this appears to be a sheer effect of the former vaccine having more total mRNA in it (!) The mRNA vaccines seem to amplify tumor PD-L1 across a wide range of tumor types, and this effect even seems to extend to immunologically “cold” tumors that would otherwise have been considered not to be responding to treatment.
This is really, really good news, and it immediately suggests that patients who are eligible for immune-checkpoint therapy (things like Keytruda, Opdivo, Tecentriq, Bavencio, Imfinzi, etc.) should be dosed with some sort of LNP-mRNA shot right along with it. And it also suggests that there are patients who might not be thought to benefit from such ICI treatment who should be given a crack at it with this protocol, since the enhancement seems so robust. There’s obviously a lot more to be worked out as far as doses of mRNA, timing, formulations, tumor types and all sorts of other factors, but the survival curves in this paper argue against waiting for patients who are being treated right now. The benefits seem to be very real, and the risks - particularly compared to the underlying cancers - seem to be very low. Go for it!
I made this point the other day, and I want to make it again, because it cannot be overemphasized: there is still so much that we don’t know about immunology. We literally don’t know when something great like this will pop up, so we have to keep banging away in these areas and discovering as we go along. And while I’m at it, there is another thing that cannot be overemphasized: if you are spreading doubts and fears about vaccines and about mRNA research, you are killing people who could benefit. Killing them by not protecting them from the diseases being vaccinated against. Killing them by making them suspicious of advances in medical science in general. Killing them by driving them away from things like this new work that could fight completely unrelated diseases like cancer.
That’s you, Robert F. Kennedy, Jr. That’s you, Jay Bhattacharya. That’s you, Marty Makary. And all the rest of your minions and gofers and toadies. You have chosen the side of death and human suffering, and in a better world than this one you would be disgraced and shunned for it.
Ah, quantum mechanics. We’ve known about it for a hundred years now, and it’s still as weird as it ever was. This paper has a case in point: the authors are investigated a catalytic hydrogenation reaction, the reduction of a benzyldehyde to a benzyl alcohol with hydrogen gas and a palladium catalyst dispersed on a titanium particle surface. There are a lot of ways to get this sort of reaction accomplished, and they occur through a lot of different mechanisms. And one of the best ways to investigate those mechanisms is through good old fashioned physical organic chemistry - looking at solvent effects, temperature effects, isotope effects and more to see what the energetics of the reaction are like. In this case, running the reaction in methanol with hydrogen gas is an efficient way to get the product, while switching out of a protic solvent really knocks things down. The authors tried it, then, with deuterium gas instead of hydrgen, and with CD3OD rather than methanol. This is to look for a “primary kinetic isotope effect”, because if there’s a bond to one of those hydrogens that’s breaking in the rate-determining step, then it was be a bit more difficult to break when it’s a (heavier) deuterium atom as opposed to plain hydrogen. Looking to see how much the reaction rate slows down under these conditions can be quite informative (and you can get smaller but very interesting effects when it’s a D-for-H switch next to the key bond as well, a secondary kinetic isotope effect). Under classical conditions (where you can approximate the bonds as balls-on-springs harmonic oscillators), a primary deuterium KIE can send a reaction rate several-fold lower than with a hydrogen, probably maxing out at around ten-fold. That’s pretty easy to spot! (Secondary isotope effects probably max out at about 1.1 to 1.2-fold). But when the authors here switched to deuterium, the rates was over two thousand fold slower (!) And by gosh, that tells you that something non-classical is underway, specifically atom tunneling. That’s the unnerving quantum effect when a particle appears on the other side of an energy barrier despite not having enough energy to have gotten past it by classical means. It occurs when the distance covered is down near the de Broglie wavelength of the particle doing the tunneling, so you’re going to see it much more often with lighter particles than with heavier ones. And that means that isotope effects can become huge - a deuterium is far, far less likely to pull it off compared to a hydrogen, and that’s just what’s been seen here. The authors propose a three-hydrogens-at-once movement as shown at left, and the concerted nature (rather than a stepwise mechanism) appears necessary to explain the huge rate differences seen experimentally. This also suggests that the protic solvent is an important part of the reaction, and that’s why switching to deuteromethanol was a big factor. And it also suggests that if you could cause that hydrogen-bonding network to vibrate at a higher energy level, you might be able to speed up the tunneling even more, and they demonstrate with far-infrared illumination, which speeds up the reaction right in step with the intensity of the light applied. (In case you’re wondering, the reaction has very little temperature dependence at 0C where these reactions were run, and the reaction mixture was vigorously cooled while the solution was being irrradiated. There have been a lot of tunneling events seen during chemical reaction mechanisms (including enzymatic ones) and there's no doubt that it can be a big factor all by itself if conditions are right. But this is probably the most dramatic example I've seen yeat!
Show full content
Ah, quantum mechanics. We’ve known about it for a hundred years now, and it’s still as weird as it ever was. This paper has a case in point: the authors are investigated a catalytic hydrogenation reaction, the reduction of a benzyldehyde to a benzyl alcohol with hydrogen gas and a palladium catalyst dispersed on a titanium particle surface. There are a lot of ways to get this sort of reaction accomplished, and they occur through a lot of different mechanisms. And one of the best ways to investigate those mechanisms is through good old fashioned physical organic chemistry - looking at solvent effects, temperature effects, isotope effects and more to see what the energetics of the reaction are like.
In this case, running the reaction in methanol with hydrogen gas is an efficient way to get the product, while switching out of a protic solvent really knocks things down. The authors tried it, then, with deuterium gas instead of hydrgen, and with CD3OD rather than methanol. This is to look for a “primary kinetic isotope effect”, because if there’s a bond to one of those hydrogens that’s breaking in the rate-determining step, then it was be a bit more difficult to break when it’s a (heavier) deuterium atom as opposed to plain hydrogen. Looking to see how much the reaction rate slows down under these conditions can be quite informative (and you can get smaller but very interesting effects when it’s a D-for-H switch next to the key bond as well, a secondary kinetic isotope effect).
Under classical conditions (where you can approximate the bonds as balls-on-springs harmonic oscillators), a primary deuterium KIE can send a reaction rate several-fold lower than with a hydrogen, probably maxing out at around ten-fold. That’s pretty easy to spot! (Secondary isotope effects probably max out at about 1.1 to 1.2-fold). But when the authors here switched to deuterium, the rates was over two thousand fold slower (!) And by gosh, that tells you that something non-classical is underway, specifically atom tunneling.
That’s the unnerving quantum effect when a particle appears on the other side of an energy barrier despite not having enough energy to have gotten past it by classical means. It occurs when the distance covered is down near the de Broglie wavelength of the particle doing the tunneling, so you’re going to see it much more often with lighter particles than with heavier ones. And that means that isotope effects can become huge - a deuterium is far, far less likely to pull it off compared to a hydrogen, and that’s just what’s been seen here. The authors propose a three-hydrogens-at-once movement as shown at left, and the concerted nature (rather than a stepwise mechanism) appears necessary to explain the huge rate differences seen experimentally.
This also suggests that the protic solvent is an important part of the reaction, and that’s why switching to deuteromethanol was a big factor. And it also suggests that if you could cause that hydrogen-bonding network to vibrate at a higher energy level, you might be able to speed up the tunneling even more, and they demonstrate with far-infrared illumination, which speeds up the reaction right in step with the intensity of the light applied. (In case you’re wondering, the reaction has very little temperature dependence at 0C where these reactions were run, and the reaction mixture was vigorously cooled while the solution was being irrradiated.
There have been a lot of tunneling events seen during chemical reaction mechanisms (including enzymatic ones) and there's no doubt that it can be a big factor all by itself if conditions are right. But this is probably the most dramatic example I've seen yeat!
Now, transcription and translation are indeed wonders of nature. The constant reading-off of our genetic code and its expression into proteins kind of has to be at that level, you’d figure, for living cells to work at all. But it’s important to remember that not-so-exact versions of these things are important, too. I’ve written about how error-prone mechanisms can be useful for bacteria and viruses (and indeed, how switching gears to these can be an outright roll-the-dice survival mechanism). But what about us? Well, as impressive as the fidelity usually is, there are a lot of errors that still creep in just because of the scale of the process. And this recent preprint is an invitation to rethink our attitudes toward that. The authors have done painstaking sequencing across both normal and tumor tissue samples looking for alternate translation events. They’re out there, and the amounts of the resulting proteins depend (as the paper emphasizes) both on their rates of synthesis and their rates of degradation. You can’t rule that latter one out: some of these changes may actually allow variant proteins to accumulate. The authors rooted through huge piles of LC/MS data from those human samples and identified nearly nine million spectra that were identified as modified versions of known peptides. About 40% of those were good ol’ post-translational modification, and those were set aside for another day. But there were 124,000 sequences that looked to be single-amino-acid substitutions with no other modifications. There was no evidence for these at the genomic level, supporting the idea that these are mistakes in transcription and translation instead. The authors chopped that list down pretty rigorously, and by their own admission probably threw out plenty of real examples, but at the end they were left with about 9,000 unique examples, 2,000 of which were very well localized on the proteins themselves. Only about 110 of those 9,000 on the list had been previously predicted from genomic translations. Looking at the abundance of these alternate sequences compared to the “base sequences” they were derived from was revealing. Most of them were at a lower level, as you’d certainly have predicted. But about 10% were at a higher level. In fact, the paper estimates that for at least 360 proteins the alternate form is by far more abundant than the canonical one! They call this “highly unexpected”, and I have to agree. The rest of the paper features several different attempts for them to get their heads around it. That started with several attempts to explain it away due to incomplete RNA sequencing, problems with peptide ionization and detectability in the mass spec, possible origins in cleavages of other highly abundant normal precursor proteins, and more. But none of these come close to accounting for the numbers. Stipulating, then, that this result was real, they went on to analyze it across patients, proteins, and tissue types. These data strong suggest that there are underlying biological reasons for some of those alternate-to-baseline protein ratios. Factors that influence these ratios at the RNA level include the number of codon changes that have to occur, the underlying frequency of those baseline codons (rarer ones are more likely to be substituted), and the level of uracil modifications present. Up at the protein level, it seems that higher stability of the alternate versions (as mentioned above) is a big factor as well. Shorter peptides are less likely to have more-stable alternates than longer ones (probably because longer ones are less stable to start with). There are tissue differences (for example, a strong tendency for G-to-S substitutions in pancreatic samples, and overall, polar amino acids are significantly more likely to be involved in the high-ratio cases. As for tumor tissue versus normal, there are some real outliers there as well, with major differences in ratios across different tumor types and in tumor-versus-normal comparisons. A look at function suggests that signaling sequences and proteasome subunits are over-represented, too, as are proteins known to be involved in neurodegenerative diseases. Finally, comparing human versus mouse tissue strongly suggest that many (perhaps most) of these changes are even conserved across species! Well, this is all certainly something to think about, and these results open up plenty of new areas for research. The authors emphasize that their strong focus on eliminating false positives probably created plenty of false negatives along the way - that is, the data presented here, as head-scratching as they are, surely only represent part of the real situation. Mandatory snarky aside: gosh, we should have though to ask AI about all this sooner, don’t you think? Could have saved ourselves lots of time. Similarly, I wonder if the people talking about creating a “digital cell” to let AI tell us all about biochemistry will get around to incorporating this stuff any time soon. . .
Show full content
Now, transcription and translation are indeed wonders of nature. The constant reading-off of our genetic code and its expression into proteins kind of has to be at that level, you’d figure, for living cells to work at all. But it’s important to remember that not-so-exact versions of these things are important, too. I’ve written about how error-prone mechanisms can be useful for bacteria and viruses (and indeed, how switching gears to these can be an outright roll-the-dice survival mechanism). But what about us?
Well, as impressive as the fidelity usually is, there are a lot of errors that still creep in just because of the scale of the process. And this recent preprint is an invitation to rethink our attitudes toward that. The authors have done painstaking sequencing across both normal and tumor tissue samples looking for alternate translation events. They’re out there, and the amounts of the resulting proteins depend (as the paper emphasizes) both on their rates of synthesis and their rates of degradation. You can’t rule that latter one out: some of these changes may actually allow variant proteins to accumulate.
The authors rooted through huge piles of LC/MS data from those human samples and identified nearly nine million spectra that were identified as modified versions of known peptides. About 40% of those were good ol’ post-translational modification, and those were set aside for another day. But there were 124,000 sequences that looked to be single-amino-acid substitutions with no other modifications. There was no evidence for these at the genomic level, supporting the idea that these are mistakes in transcription and translation instead. The authors chopped that list down pretty rigorously, and by their own admission probably threw out plenty of real examples, but at the end they were left with about 9,000 unique examples, 2,000 of which were very well localized on the proteins themselves. Only about 110 of those 9,000 on the list had been previously predicted from genomic translations.
Looking at the abundance of these alternate sequences compared to the “base sequences” they were derived from was revealing. Most of them were at a lower level, as you’d certainly have predicted. But about 10% were at a higher level. In fact, the paper estimates that for at least 360 proteins the alternate form is by far more abundant than the canonical one! They call this “highly unexpected”, and I have to agree. The rest of the paper features several different attempts for them to get their heads around it.
That started with several attempts to explain it away due to incomplete RNA sequencing, problems with peptide ionization and detectability in the mass spec, possible origins in cleavages of other highly abundant normal precursor proteins, and more. But none of these come close to accounting for the numbers. Stipulating, then, that this result was real, they went on to analyze it across patients, proteins, and tissue types. These data strong suggest that there are underlying biological reasons for some of those alternate-to-baseline protein ratios. Factors that influence these ratios at the RNA level include the number of codon changes that have to occur, the underlying frequency of those baseline codons (rarer ones are more likely to be substituted), and the level of uracil modifications present.
Up at the protein level, it seems that higher stability of the alternate versions (as mentioned above) is a big factor as well. Shorter peptides are less likely to have more-stable alternates than longer ones (probably because longer ones are less stable to start with). There are tissue differences (for example, a strong tendency for G-to-S substitutions in pancreatic samples, and overall, polar amino acids are significantly more likely to be involved in the high-ratio cases. As for tumor tissue versus normal, there are some real outliers there as well, with major differences in ratios across different tumor types and in tumor-versus-normal comparisons. A look at function suggests that signaling sequences and proteasome subunits are over-represented, too, as are proteins known to be involved in neurodegenerative diseases. Finally, comparing human versus mouse tissue strongly suggest that many (perhaps most) of these changes are even conserved across species!
Well, this is all certainly something to think about, and these results open up plenty of new areas for research. The authors emphasize that their strong focus on eliminating false positives probably created plenty of false negatives along the way - that is, the data presented here, as head-scratching as they are, surely only represent part of the real situation.
Mandatory snarky aside: gosh, we should have though to ask AI about all this sooner, don’t you think? Could have saved ourselves lots of time. Similarly, I wonder if the people talking about creating a “digital cell” to let AI tell us all about biochemistry will get around to incorporating this stuff any time soon. . .