Julie Fournier

The authors discuss the challenges and future directions in the development of covalent allosteric modulators.

Drawn to science

Drawn to science, Published online: 03 October 2018; doi:10.1038/d41586-018-06832-0

Teachers do not need training in the arts to create useful drawing experiences for science students, says Bethann Garramon Merkle.

I wrote up this year’s Nobel Prize awards in chemistry yesterday, and there’s no arguing that they’re significant achievements worthy of a prize at this level. For many chemists, though, I think that this year’s award will join the 2015, 2012, 2009, 2008, 2006, 2004, 2003, 1997, and 1993 ones (and there are arguably even more) as years when molecular biology was redefined by the Nobel Committee as chemistry in order to recognize its achievements. I’ve written about many of these awards on this blog over the years, and have often made the case that (1) there’s more chemistry involved in them than one would think and (2) that chemists themselves need to broaden their own definitions of what their field encompasses. Both of those are true, but at this point, there’s also little doubt that the character of the chemistry prize has changed. And I speak as someone whose own career has, over the years, moved from pure organic synthesis to more chemical biology, so I hope I can’t be accused of sour grapes in this regard.

Those achievements in molecular biology are real and absolutely worth the recognition that the prize confers. But Alfred Nobel’s will recognizes no such area of science. The prizes do not even recognize an area of science called “biology”. Nobel’s will stipulated certain definite categories and made no provision for changes, and Swedish law is (I’m told) quite rigorous about this sort of thing, so we are stuck with what Alfred Nobel thought (in 1895) to be the most appropriate categories of scientific progress. Like every other prediction of that sort from the late 19th century, this one has inevitably gone off the rails, and the grinding noises are nowhere louder than at the border between chemistry and biology.

The prizes for science, for all their difficulties are still in better shape than the ones for Peace and for Literature. The former has mixed worthy recipients in with some wild misfires over the years, diluting its impact. The latter, which in its early years distinguished itself by failing to recognize Joyce, Tolstoy, and Proust, has in more recent times put itself in danger of becoming a not-very-amusing joke. (There isn’t even an award this year because the committee itself is in such disarray). That’s not to say that there haven’t been some odd awards in the sciences. Johannes Fibiger won in 1926 for supposedly showing the infectious nature of cancer in experiments with rats and cockroaches (and that one was another mess, with no actual award in 1926 and a catch-up retroactive award in 1927, which is what the literature prize is planning on for 2019). His experiments are, though, completely erroneous. Another famous one was the Moniz award in 1949 for developing the surgical technique of prefrontal lobotomy, which in retrospect does not look like as much of a benefit to mankind as it must have at the time. In the chemistry prize, one of the stranger awards (in restrospect) was Artturi Vitanen‘s 1945 prize for (largely) an improved method of storing cattle fodder. Beyond these, if we get into the various arguments about credit, both false positives and false negatives, there will be no end to it.

But any prize, administered in any fashion, will have its controversies. Criticizing the Nobels for these is actually besides the point, or at least besides mine at the moment. I think that the problem with the prizes is the inflexibility of Nobel’s will and the subsequent adherence to it. An email correspondent of mine favored the idea that anyone setting up a foundation in their will should have a review provision in it beginning a certain number of years after their death, with the administrators being authorized to make increasingly large changes as time went on (or to dissolve the thing altogether). I find some merit in that idea, but we certainly don’t have that with the Nobels, so that’s a separate problem.

But even the adherence to Nobel’s will has been inconsistent. The Literature prize is supposed to be for “the most outstanding work in an ideal direction”. What that means is open to debate, but there are numerous winners that it seems hard to apply to. The Economics prize is not, strictly speaking, a Nobel at all, but an add-on memorial prize started in 1969. It attracted criticism at the time, and still does. But instead of complaining about how the Economics prize is a violation of Nobel’s intent, why not violate it some more by starting “Nobel Memorial Prizes” in biology, in mathematics, in whatever areas we see fit? And allow ourselves to revisit those new categories over time?

This is not a new proposal. It will probably never happen. The existing prizes have too long a history by now, especially in the popular imagination. That’s the biggest difficulty with the science prizes, perhaps: the bulk of the population does not understand what any particular science Nobel is for, because they don’t have the background for it. For many of them, this may be the only time they ever hear about ubiquitination, phage display, palladium-catalyzed coupling, or the like. It’s good to have something that puts great achievements in these fields into the public eye once a year, but messing around with the awards risks that. The world could get along fine without hearing about these things every October, I fear, and if we add more prizes and rearrange things it might decide to do just that.

Back to chemistry, then. As it stands, I see no real solution to the “That prize isn’t chemistry!” problem. It’s a large enough field that molecular biology can be jammed into it without (much) shame, and that brings us to another difficult subject: there have been more world-impacting discoveries in that field over the last decades than there have been in chemistry. I know I’m going to get roasted in the comments for saying that – and let me make clear up front that I think that there have indeed been great discoveries in chemistry itself. But molecular biology has been having a tremendous run, and its implications for human health and the nature of life itself give it a very high profile.

Chemistry, as a science, has always been stuck in an awkward position when you try to explain its importance. Biology, as just mentioned, can often point to direct connections to medicine. All you have to do in the press release for any big prize in that area is mention “cure for cancer” or the like. Physics often alternates between being stupendously removed from everyday life and being right on top of it. By the latter, I mean things like the development of nuclear weapons, or of new energy sources. For the former, there are phrases like “the God particle” or “how the Universe is put together”. You can write a headline with this stuff.

What’s the magic phrase for chemistry? There isn’t one; there never is. We always find ourselves two or three layers of explanation away. The Haber-Bosch process makes ammonia, and people say “Big deal”, but then you have to explain that it actually feeds the world, because of nitrogen this and fertilizer that. Palladium-catalyzed coupling reactions form carbon-carbon bonds, and people say “So what”, and you have to explain that this is how many of the medicines that they’re taking get made these days. And some chemistry prizes are so many layers down in explanation that you just have to tell people to take your word for it, that the discovery of a new form of carbon or fivefold symmetric quasicrystals were a big deal, trust us. (Physics has some of these too, naturally, but next year there’s always the key to the universe coming back around again).

So what I’m saying is that as far as I can see, Chemistry as a field is just going to have to deal. We have little or no leverage in the Nobel space, and complaints about how the prizes need to get back to “real chemistry” are, I’m afraid, a waste of time. The Nobels are flawed, yes indeed, in several key ways. But that realization and a dollar will buy you a bag of potato chips.

As many had expected, the Nobel prize in medicine/physiology this year recognizes advances in immuno-oncology: James Allison (for CTLA4) and Tasuku Honjo (PD-1). For some years now, that has been a huge, massive, unstoppable wave in cancer research, and I would not want to try to estimate how much time, effort, and money has gone into it. But it’s been worthwhile, very worthwhile, and the good part is that the story is still going on. Here’s the Nobel scientific summary, which as always is very well put together.

People have been trying to get the immune system enlisted into cancer treatment for at least a hundred years. That involves both figuring out ways to activate an immune response, and understanding how tumors largely manage to evade such a response in the first place. But as anyone who has looked into the field for fifteen seconds can tell you, immunology is one of the most fiendishly complex areas of medicine. It would have to be: left unchecked, a full-blown immune crisis can kill you where you stand, within minutes. That’s one reason why I roll my eyes when I hear ads for “dietary supplements” and the like promising to “activate my immune system”. You want to be really sure about what you’re asking for, because an activated immune system is capable of fearsome amounts of damage if it gets even slightly mis-aimed.

But that’s all the more reason to try to aim it at tumor cells. This gets to the intricate question of “self and non-self” recognition in immunology, which has been the subject of research for many decades now. The medical implications for better control of this process are immediately clear – to start with, organ transplants (keeping your immune system from attacking foreign tissue) and cancer therapy (making your immune system recognize some defined tissue as foreign) and treatment for autoimmune diseases (persuading your immune system to realize that your own tissues are not, in fact, foreign after all). It’s an intricate process, though – it would have to be. There are multiple, overlapping checks and balances, switches and regulators everywhere you look. We are clearly the descendants of creatures who found it beneficial to have a complex, decentralized immune response. Our ancestors are the ones who managed to balance things better than the ones who were killed off more easily (or made less likely to reproduce) either by insufficient responses to infection or by too-vigorous immune responses when they weren’t needed.

Today’s Nobel recognizes the discovery of some of these regulatory systems. Allison’s work involved CTLA-4, a protein found on the surface of T-cells. It was discovered in 1987, and by 1995 it had been worked out that it was a negative regulator of T-cell function. Many labs were working in this area, with the more clinically-oriented ones going after just those sorts of applications mentioned in the previous paragraph. Mutations in CTLA-4 are very strongly associated with autoimmune diseases (a whole list of them), so a lot of work was directed at this area (trying to make the inhibitory pathway more active). But Allison’s lab concentrated on the possibility of cancer therapy – which frankly, was considered by some to be less likely to work. Immune approaches in this area had a history of failure, or at the very least underperforming greatly, along with a sprinkling of interesting, hard-to-replicate individual responses.

In this application, you’d want T-cells that were more active. Allison and co-workers developed an antibody to CTLA-4, blocking the blocker in a very direct approach. Interestingly, the detailed mechanism by which CTLA-4 inhibits T-cell activity remains a matter for debate (the standard immunology response of sighing deeply and saying “Well, it’s complicated. . .” works just fine here as well). But absolutely jamming it up on the cell surface with an antibody, you’d figure that would certainly do something. The first crucial experiments were done with mouse xenografts (transplanted tumors) in 1994, and they were dramatic indeed. See the data at right – now, xenografts are not endogenous tumors, and mice sure aren’t humans. But still. That’s the sort of thing you want to see!

It was still not easy to get this idea into humans. Kicking out the jams on CTLA-4 was a scary prospect, and there were not-unreasonable worries that you might shrink a patient’s tumors while killing them in an entirely new way via an autoimmune response. Medarex (then a small company that few knew much about) was interested, though, and development of a humanized antibody began. It was not a smooth and uneventful path to approval. There were indeed autoimmune side effects in human patients, and  there was the usual problem that a broad-spectrum oncology program has: where to start? It can be quite difficult to figure out which tumor types (and which patients) will benefit the most, and really useful therapies can be obscured if you go down the wrong path. I wrote here about ipilimumab, which is the antibody that came out of this work, which it produced hard-to-interpret results in prostate cancer patients (an area where it still has had difficulty showing a survival benefit). But that blog post mentions that results were already better in melanoma trials, and that’s where the drug has been approved (as Yervoy). Bristol-Myers Squibb bought Medarex for Yervoy and for the company’s antibody platform, in what has been called “one of the best biotech acquisitions of all time”.

The other half of today’s award is PD-1. Honjo’s group discovered this cell-surface protein in the early 1990s while working on dying mouse cells. It was thought to have a role in such cell death pathways (thus the name, from Programmed cell Death), but knocking it out in mice led only to an apparently mild phenotype that seemed to have something to do with the immune system. The animals had enlarged spleens, and developed lupus-like symptoms late in life. Further work (by groups all around the world) helped establish that PD-1 was in the same general family as CTLA-4, and was part of yet another regulatory pathway to keep T cells from going wild. Honjo’s group and collaborators discovered twos endogenous ligand for the receptor, the PD-L1 and PD-L2 proteins, and also found clues that this pathway might be important to tumor cells.

The first report of the use of a PD-1 antibody against tumors came in 2005 from Honjo’s group and from the group of Lieping Chen, who had also made fundamental discoveries in this area (and who is a plausible candidate, one of several, for the “How come there aren’t three people on this prize” question). At right is the effect of that antibody on mice who had had susceptible tumor cells introduced, and this is another one of these no-contest graphs that tells you that a hypothesis has been nailed. These papers went into the details of what’s been driving the field ever since: whether or not a given tumor expresses PD-1, and to what degree, and the idea of using antibodies against either the receptor or the ligand protein. Ono Pharmaceuticals in Japan got into the field and partnered with Bristol-Myers Squibb (again!) to advance nivolumab (known as Opdivo) into the clinic. It has performed very well, at times spectacularly, and there are plenty of other PD-1 based therapies out there with it by now (Merck’s Keytruda/pembrolizumab being the most famous).

The whole PD-1 field is too large and too fast-moving to be capable of easy review. It seems that every couple of months there’s another study in another tumor type, or another combination. But I do want to mention the intersection of PD-1 and CTLA-4, because there’s no obvious reason why they might not work at the same time. That is beginning to be decided right now in the clinic, but it’s too early to say what’s going to happen overall. I have little doubt that there will be a number of situations where that combination will be better than either agent alone, though. Frankly, one of the biggest problems of the whole immuno-oncology area is that there are just too many things to try, which is quite a situation to be in. The number of I/O clinical trials is arguably already into hold-on-a-minute-here-dudes territory, and there’s an immense amount of dust in the air.

But what’s clear is that this has been a revolution in oncology. For all the false starts, missed endpoints, fights over credit and struggles for market share, let there be no doubt: immuno-oncology has been pulling people out of their graves. Cancer cells being what they are, tumors can eventually turn around and mutate their way past the existing therapies in many cases. But there are people with advanced cancers who had been told to get their affairs in order who are still walking around, watching their children grow up. Many of these combinations of drugs and specific tumor types are too new to even be sure what the overall survival benefits are, other than being (in many cases) very substantial compared to the options that anyone had before. Those curves are still being drawn. One other point: I’ve mentioned some of the companies involved along the way in this post, and it should be emphasized that industrial development of these discoveries has been absolutely crucial in getting them available to the world.

And the good part, as I said at the beginning, is that the story is still going on. Today’s prize was widely anticipated, and that’s because it’s so well-deserved. The only regret I have about it – and it’s a regret that shows up every year during Nobel season – is that the prize tends to make discoveries like these seem like the work of far fewer people than they really are. This one has absolutely taken a scientific army, and the campaign continues.

Angewandte Chemie International Edition, EarlyView.

Time is short for blogging today, but I wanted to take a moment to point out people for whom time for writing things up is (apparently) never, ever short. This is a study on prolific authorship, and the high end of that cohort is pretty terrifying. At least 9,000 authors have been on 72 papers or more in at least one year between 2000 and 2016 (that’s a paper every few days). Some of these, to be sure, are giant collaborations – the great majority of the authors are in physics, where that’s the norm. But what about the ones who aren’t? Read the article for more interesting thoughts on authorship, productivity, and what happens when both of those concepts get stretched. . .

Biologist Dan Gibson edits and programs DNA, just like coders program a computer. But his "code" creates life, giving scientists the power to convert digital information into biological material like proteins and vaccines. Now he's on to a new project: "biological transportation," which holds the promise of beaming new medicines across the globe over the internet. Learn more about how this technology could change the way we respond to disease outbreaks and enable us to download personalized prescriptions in our homes.

Jakob Magolan is here to change your perception of organic chemistry. In an accessible talk packed with striking graphics, he teaches us the basics while breaking the stereotype that organic chemistry is something to be afraid of.

Herschel Mukherjee | Neil P Grimster

Working out of her garage, Anushka Naiknaware designed a sensor that tracks wound healing, becoming the youngest winner (at age 13) of the Google Science Fair. Her clever invention addresses the global challenge of chronic wounds, which don't heal properly due to preexisting conditions like diabetes and account for billions in medical costs worldwide. Join Naiknaware as she explains how her "smart bandage" works -- and how she's sharing her story to inspire others to make a difference.

Angewandte Chemie International Edition, EarlyView.

Irina Kareva translates biology into mathematics and vice versa. She writes mathematical models that describe the dynamics of cancer, with the goal of developing new drugs that target tumors. "The power and beauty of mathematical modeling lies in the fact that it makes you formalize, in a very rigorous way, what we think we know," Kareva says. "It can help guide us to where we should keep looking, and where there may be a dead end." It all comes down to asking the right question and translating it to the right equation, and back.

When you take an NSAID (naproxen, ibuprofen, aspirin, etc.), how does it work? This is one of those questions that improves on further inspection – or deteriorates, according to your point of view, because it just keeps on getting more complicated. For decades, there was no good answer at all, but then there was “It reduces signaling in the inflammation pathways”, followed by “OK, it seems to do that by decreasing this prostaglandin signaling molecule” and “Ah, it does that by inhibiting cyclooxygenase, a key enzyme in the pathway to make those”. Then we moved to “Hold it, there’s more than one cyclooxygenase”, which is how we ended up with the COX-2 inhibitors like Vioxx, which led to “You know, the functions of these two are more complicated than we even thought”.

And on top of all these, there are clearly effects of NSAID drugs that have nothing to do with the COX enzymes at all. Blood clotting you can pretty much chalk up to the COX isoforms, but there are more things going on, and they’re very poorly defined. Which brings up a more general question: how do you ever know about all the things that a small-molecule drug might be doing in a living system? The answer is, you don’t. That’s not generally appreciated outside the biomedical professions, but it’s the truth. We don’t have some way to track things around and watch them interact at a molecule level through out a cell or an organism.

Well, maybe. Mapping compound-protein interactions is a big part of chemical biology, and this new paper from the Woo lab at Harvard shows an attempt to track small molecules in just this way. It uses photoaffinity labeling, via the now-nearly-standard combination tag of a diazirine and an acetylene, which lets you form a covalent bond to (most) of the thing the molecule of interest is spending time next to, and on the back end of the process, it uses isotope-labeled tags on the acetylene-click-conjugation step to give the mass spec detection a much higher signal/noise. Just digging through the whole cellular proteome without the kind of validation that isotope tagging gives you is a tall order, but the deliberately-weird-mass-combination effect of the stable isotope tags moves things up out of the noise.

Three NSAIDs are given the treatment: naproxen, celecoxib (Celebrex), and indomethacin. It’s a good spread of activity: naproxen hits both COX-1 and COX-2 pretty much equally, celecoxib is of course a COX-2 selective compound, and indomethacin is in a structurally distinct class that’s already known to have some other modes of action. (Update: here’s work from another group on an alkyne-tagged aspirin derivative, which also interacts with a long list of cellular proteins). All the photolabel-derivatized compounds were still active against COX enzymes (albeit with somewhat lower affinity), and all of them could be displaced by the parent compounds in competition assays. All of them photo-labeled COX-2 in an in vitro experiment, as they certainly should, and this experiment was used to validate the mass spec analysis methods for the whole-cell experiments. (For example, each of these compounds produces six or seven different photoadducts, depending on which nearby amino acid gets snagged by the reactive carbene that forms from the diazirene, and the group was able to see and account for all of these without difficulty).

So what happens in cells? They tried the dosing/photolabeling experiment in Jurkat cells, along with experiments with photoactive negative control (non-NSAID) compounds. About 700 proteins were labeled in all, with quite a bit of overlap between the three NSAIDs (40% of them are hit by all three), and that figure alone should make a person stop and think a bit: as simple and widely used a molecule as naproxen binds to hundreds of proteins well enough to photolabel them. No, we don’t know the details of what’s going on in there, do we? Admittedly, that’s at a high concentration (250 micromolar), but (1) patients take whopping doses of this stuff in the real world, and (2) the majority of these interactions also showed up at a 50 micromolar dose as well.

This same set of experiments was also run in K562 cells, a leukemia line. In this case, there were 513 enriched protein targets and only 206 of them overlapped with the set from the Jurkat cells. And that’s also something to think about – not only do small, well-known compounds like these interact with a long list of proteins, that list can change dramatically in different cell backgrounds. Only about 30% of the total list of proteins have ever been report to interact with any small molecules at all.

The distribution of these doesn’t show any particular compartmentalization effects in the cells; it pretty much tracks protein abundance across the cytoplasm, various organelles, nucleus and so on. A closer look at the proteins and their interaction sites can be found in the paper, but one thing worth mentioning is a “hot spot” in histone H2A, which is certainly not a target that anyone had been thinking about for NSAIDs, to my knowledge. Assigning all the specific sites of labeling across the proteome is still beyond current technology, although you can think of some experiments that would help narrow things down a lot (different protein digestion before the mass spec, and so on). But even at this level – which is more detailed than we’ve ever had for such compounds – there’s a lot to deal with here.

So if you needed a reminder about just what a complicated mixing bowl we’re throwing our compounds into, here you go. The hope is that such techniques and the ungodly huge piles of data that they generate will help up build up a clearer picture of drug action. But what a picture that’s going to be!

Let’s think for a bit about how proteins bind to each other. After all, messing around with that is what keeps everyone in the drug industry employed, and the unmessed varieties of such binding events are what keep us all vertical and above room temperature, so it’s a worthy subject.

The mental picture is of two proteins adopting complementary shapes along some kinds of binding surfaces. “Complementary” is doing a lot of work in that sentence, though, because we could be talking about hydrophobic interactions (whatever those are), hydrogen bonding, or outright charged residues that are pairing up positive/negative. In fact, since we’re talking about proteins here, all of these can be operating at the same time, and probably are. And there are all sorts of entropic/enthalpic things going on, too – things are happening to water molecules at the surface of the proteins as they come together (as well as to the bulk water that used to be between them), parts of each protein are probably moving around and getting more or less constrained, other internal interactions within each partner are adjusting, etc. It’s a mess.

But it’s still a somewhat orderly mess. In the end, these two complex three-dimensional shapes have found some sort of defined relationship that’s overall lower-energy than what they started with, and now this is their new shape. What if that’s not the case, though? This unnerving thought is brought on by this paper, published late last month in Nature. We’re talking intrinsically disordered proteins again – those beasts, rather more common than was once thought, that have large sections of them that have no particular defined shape at all. (Indeed, some of them are disordered from snout to tail). I’ve generally thought of these, though, as flopping around in that way until they encounter a binding partner, at which point they settle down into some defined shape and slot themselves obligingly into my weltanschauung.

As should have been quite clear by now, though, proteins don’t care what I think about them. This paper shows a particular protein interaction (between histone H1 and prothymosin-alpha) that is down in the picomolar affinity range. The histone protein has a small structured region in the middle, but the N- and C-terminals head off into complete disorder. Prothymosin is disordered all the way through. If you’d asked me about this at one time, I would have been certain that this sort of binding required the formation of a solid, well-defined structure with plenty of clear interactions. But that’s not what’s going on. The paper shows that both proteins are still disordered even as a complex. In the NMR, you can see the structured globular part of the histone, but that’s the only order in sight. The circular dichroism spectra reflect this as well – the complex, in fact, is just the CD spectra of the two partners added on top of each other, with no sign of induced helicity, etc. The team did a whole series of FRET experiments, attaching the partner groups on a number of different residues, and there’s really no pattern to it at all.

What’s the interaction? Sheer charge. The partners are strongly positive/strongly negative, but they don’t seem to care what residues associate with what. Doubling the ionic strength of the buffer decreases the binding constant by six orders of magnitude, so yeah, it’s pretty much an ionic thing all the way. They’re just shifting around unfolded on a 100ns-timescale, with no apparent need for anything more organized.

There were already signs that something like this was going on. Such histone protein tails had already been shown, most disconcertingly, to bind their protein partners even when their sequences had been scrambled, and they’re not the only proteins that have demonstrated such behavior. It makes sense: if you don’t got no defined structure, you don’t need no defined sequence, right? Here’s a good try at classifying protein binding along scales of static/dynamic and order/disorder, with this latest example falling thoroughly into the “dynamic disordered” quadrant.

How should we think about this stuff? Well, for now, I’m modeling this in my head as “proteins have all sorts of binding modes that fit different needs in the cell”. There are some, obviously, that need pretty hard, defined structures both at their interface and in the other parts of the protein. There are some where a protein’s ordered regions bind other ordered regions, with disordered parts still boogieing around, and others where a totally disordered protein folds one end of its structure up into an ordered complex while still leaving the far end loose. All of that I’ve been able to handle without much problem. But, as this latest paper shows, we have to stretch this concept to include “disordered binding” itself. In these cases, I suppose that the key event is just bringing these proteins together somehow, without so much need for three-dimensional perfection.

The thing is, it looks like these sorts of disordered binding modes may be a lot more common in the proteome than any of us thought. We’re all going to have to accommodate that reality, apparently. Are we going to be able to attack such things with small molecules, though? I have to say that given the choice, I would try something else first. With a disordered protein we’ve always been able to make the argument that binding something to it in its unstructured state might throw its conformational manifold off enough to disturb its function – and of course, targeting it in a structured binding complex is always theoretically possible right up front. But picomolar-level binding that has no defined structure at the binding interface? I will read about this with interest, and I will think about its implications, but I would not like to lead a drug discovery project against it.

I blogged here last year about some really interesting work from the Cravatt group at Scripps. It’s sort of an intersection between fragment-based screening and screening in cells, which is an intersection that I’d previously never thought existed. That’s because fragment screening typically involves biophysical methods (NMR, SPR, DSF and others) that can pick up relatively weak binding events, ones that would probably not drive much (or any) functional readout in a cell assay. And functional readout is what screening in cells is all about: you fire in the compounds and look for what happens (ideally, a well-controlled group of effects that allow you to distinguish what could be useful compounds from things that are banging on the big piano with both hands, as it were).

So the two worlds typically don’t overlap much. When you’re screening fragments, you have no idea if they (or the compounds derived from them) are going to have the effects you want in cells, because you’re way back at the beginning, just observing binding to a purified protein, a defined target. Cell activity will come later, maybe much later, and it’s going to be up to you to look for it and try to optimize it. And when you’re doing cell screening, you have no assurances about what target or targets you might be hitting. If you’re screening as part of a larger program where you already have lots of defined selectivity data on proteins, you can be pretty sure of your readout (although never 100%), but on the other hand, if you’re phenotypically screening and just hunting for the readout you like, who knows? Target elucidation is up to you, too.

What the Cravatt paper did was to build a particular library of fragments, with two key characteristics. Every one of them had a diazirene attached to it on a side chain, which is a group that chemical biologists know and love, since under strong light it turns into a reactive species that tends to form a covalent bond with whatever might be next to it. And each fragment also had an acetylene, also beloved ever since the Sharpless group unleashed click chemistry on the world, since that gives you a handle to later attach a fluorescent group or whatever other tag you like in order to find what proteins the covalent group attached to. This library of diazirene/acetylene fragments was turned loose on cells to see what they’d label – if this sounds interesting, go back and read the blog post, and most definitely read the paper it’s referencing, because it’s a real tour de force.

Near the end of it came a further development, though – a move out of fragment space and into larger, more drug-like chemical matter. That takes us into an interplay of affinity and reactivity: a red-hot covalent labeling group has a better chance of just sticking onto everything it sees, as you’d figure. On the other end of the scale, a really unreactive group is only going to stir itself when everything lines up perfectly: relatively tight binding that brings the reacting partner group in just so, right distance, right angle, room to maneuver, and so on. In general, that means that covalent fragment screening will probably call for somewhat more reactive groups, since their affinities will seldom be very high. But as you get up to more complex chemical matter, some of these things could have really significant binding, and you could well see less reactive groups in play. (This effect has been documented many times in the literature).

Of course, the odds of finding fragment-level binding are a lot better than the odds of finding this more complex binding, which is why people screen fragment libraries in the first place (and why they use the hit rates from them as proxies for “druggability” in general”). But if you’re not aiming at a particular target, but are instead just shotgunning away through the cells looking for something interesting, that’s not as much of a problem. Odds are small that you would hit one particular target with a small-to-medium library of covalent/phtotoaffinity tagged drug-like molecules, but the odds that you’ll hit something are pretty good. And if you keep profiling with diverse chemical matter, you could end up with a collection of compounds that (1) selectively hit a lot of known disease-relevant proteins and (2) hit a lot of others that are less well characterized and might lead to new biology or target ideas entirely. In both cases, you end up with a covalent probe compound ready to take you further.

And that’s what this latest paper follows up on, calling the whole process “Inverse Drug Discovery”. It goes further by bringing in compounds with aryl fluorosulfates, a functional group recently popularized by (again!) the Sharpless and Fokin groups. This is a good example of a weakly reactive covalent warhead – to get these things to react, not only do you need a nucleophile that’s able to come in at the correct angle to the sulfur-fluorine bond, but you also need an environment on the other side that will stabilize the transition state where that fluorine is in the act of leaving and give that developing fluorine anion a good home. Otherwise the energy of that TS will be too high for the reaction to ever get over the hump, and no covalent labeling will take place.

Such a functional group would be expected to label an unusual suite of proteins, and so it proved. Incubating either live cells or cell lysates (which, as you’d expect, gave more protein hits) with three different fluorosulfates gave several labeled proteins, and the mechanism was confirmed by competition with the same probe molecules minus the alkyne group (so they wouldn’t be picked up by the eventual labeling). The group narrowed down on 12 proteins for which structures are available and whose labeling seemed robust, and 11 of them also labeled when the experiment was tried on recombinant protein alone. (The outlier just points out that there could be differences in the in vivo protein – post-translational modifications? – that made it different from the pure-protein case). These 11 proteins were labeled at specific sites, mostly tyrosines but with some lysines as well.

Some of them (such as biliverdin reductase A) were labeled in their active site, shutting the enzyme down completely, while others still labeled near the binding pocket (thiopurine S-methyltransferase). There were also typically cationic residues present nearby in these pockets, which could be expected to both make the side chains involved more nucleophilic and to possibly stabilize the departing fluoride ions. Several of the enzymes (NME1, CRABP2, e.g.) are of potential drug development interest, and several of them have never been picked up in covalent screens before, either.

This is just a glance into a whole landscape of drug-compound interactions that we’ve hardly begun to explore. It’s especially interesting to consider the less-explore functional groups, as was done here, and there are plenty of these that have never been investigated at all. Such compounds could potentially turn into drug candidates (after optimization) with their covalent warheads still as part of their mechanism, or one could try to pick up enough binding energy during that optimization to dispense with the covalent binding later on. (That second idea is theoretically possible, but I’ve never seen an example of it so far, and would be very interested in one). New territory!

Well, here I was the other day going on about automated chemistry when this paper was waiting in my RSS feed. It’s from a group at Pfizer, and they’re using an automated microscale flow chemistry rig for reaction optimization. Inspired by this work from Merck, which demonstrated evaluation of 1536 reactions in a plate-based system, this new paper moves from microscale batch mode to continuous flow.

The key objectives were to develop a fully automated system for HTE screening with flow chemistry technology that would (i) integrate inline high-resolution LC-MS analysis for real-time reaction monitoring; (ii) use diverse volatile and nonvolatile solvents; (iii) use ~0.05 mg of substrate per reaction to enable broad parameter space exploration with minimal material consumption; (iv) enable the preparation and analysis of up to 1500 reaction segments in a 24-hour period; (v) establish the capacity of the platform to directly scale up preferred conditions via multiple injections to produce 10- to 100-mg quantities of a specific compound; and (vi) show translation of nano-HTE conditions to both larger-scale batch and flow synthesis.

Objective met, it looks like. They’ve worked up an ingenious system to assemble five-component mixtures (two reactants, a metal catalyst, a ligand for it, and a base) for evaluating Suzuki-style coupling conditions – the five are combined in concentrated form and then injected into a carrier solvent (which can be varied as well), and checked at different temperatures. Injecting in a bolus like this also gives you the ability to send the next one in after a suitable delay, instead of waiting for the first one to get all through the tubing. Five or six dimensions gives you a lot of potential variations, which (unfortunately!) is just what you need for metal-catalyzed couplings. On the plus side, it’s long been a belief of mine that any such coupling can be optimized to a high yield if you’re just willing to spend enough of your life doing so. This setup takes that idea and runs with it.

It’s not a cheap assembly – there are two Agilent systems waiting at the far end of the flow apparatus, to handle the reaction bolus injections that are coming out every 45 seconds. The whole idea took a lot of careful validation as well, to figure out what concentrations to use for the injection stocks, what solvents to have them in, their behavior on hitting the carrier solvent and being diluted by it, the extent to which the injections would spread out as they came down the reaction tubing, the efficiency and accuracy of the LC/MS analysis at the far end, and so on. Without all this groundwork, though, it would have been easy to use the fancy robo-rig to generate pile upon pile of crappy, hard-to-reproduce data, which is a temptation that has to be avoided. Measure fifty-three times, cut once, as the old saying goes.

They ended up screening a total of 5760 reactions, which evaluated Pd couplings across eleven ligands and seven bases (organic and inorganic), all in four different solvents, for most combinations of a matrix of four electrophile and four nucleophile partners. As mentioned above, the machine ran 1500 reactions per 24-hour period, on a 0.4 micromole scale per reaction. A heat map of the results are shown, although don’t feel as if you have to work your way through all of it. You can see immediately, though, that there are some combinations that have a much better success rate than others. Xantphos, for example, seems to consistently underperform the “No ligand” control category for these transformations, whereas good ol’ triphenylphosphine is your friend. The 6-chloroquinoline is a tough customer. The trifluoroborate partner (2c) is just not reactive enough under these conditions, unless you use methanol (probably because it has to hydrolyze before it reacts?).

It’s possible to plot these conditions out in several different ways, naturally – the authors show, for example, that if you’re looking for conditions that always seem to deliver product, no matter the combination, then X-Phos or S-Phos in acetonitrile is the way to go. So if you’re going to turn around and set up a big parallel synthesis run, you’ll be very glad to have scouted all these things to improve your success rate. The paper shows that they could use the exact same conditions (but just injecting the same combination over and over) to provide 50 to 100mg of a specific product within 75 minutes, and translating the same conditions to a larger traditional batch reaction worked just as expected. (They tried a few winning conditions and a few losing ones in batch, actually, and the trend held up every time, which is encouraging).

Interestingly, the group then turned to Pfizer’s traditional parallel synthesis efforts, taking a particularly challenging aryl bromide/pinacolborane combination and optimizing it (the standard parallel conditions for such couplings had given no product at all). Running 576 different combinations across 8 hours showed that there were painfully sharp cutoffs in the reaction landscape. Only two catalysts seemed to give any product at all: CataCXium A was unusually effective, particularly in in THF/water, AmPhos showed some reactivity, but the rest of the catalysts (including the XPhos and SPhos combinations in the main screening run) were completely useless.

That is indeed metal-catalyzed coupling as I have experienced it, and until we get an utterly thorough understanding of the reaction details – don’t hold your breath – the only way to deal with this situation is this sort of brute force experimentation. And this is the best brute-force technique I have yet seen. Believe me, setting up five hundred and seventy-six Suzuki couplings in a row is not fit work for a human being, not when there’s a machine that can do it instead. Great stuff.

I’ve been meaning to link to this article, which is the best overview I know of for kinase inhibitors. The authors (a large multicenter team led out of Munich) characterize 243 (!) kinase inhibitors that have made it into human trials across a very wide range of the known kinase enzymes, and the result is a mass of data that’s finally available in one place. (I should also note that the authors have incorporated the data into their online open-access ProteomicsDB tool). Kinase inhibitors get tested against other kinases as they’re developed, but not against lists this long (and not under the same conditions, as you start to compare compounds from different organizations against each other).

Many clinical KIs (kinase inhibitors) are claimed to be potent and selective; however, this is often not the case, resulting in failure of clinical trials and obstacles with laboratory research. Assessing selectivity of a compound for a target or target class is not a trivial undertaking, because the full range of targets (and their cellular expression levels or concentrations) is often unknown and the complete compound dose range is rarely measured. All KIs in our study were profiled in a dose-dependent manner and at near thermodynamic equilibrium in cellular lysates. Thus, this large body of binding data enabled the development of a new selectivity metric termed CATDS (concentration- and target-dependent selectivity) that goes beyond previously published selectivity scores (3, 5, 19–21) in that it also captures aspects of target engagement and drug MoA.

About 10 to 20% of the 243 are quite selective, including Tykerb (lapatinib) and rabusertib. but the scores decrease pretty smoothly down to compounds that are just not selective at all, such as Rydapt (midostaurin) and XL-228. You’ll note that there are both approved and unapproved drugs at both ends of the scale – selectivity really doesn’t, in the end, correlate with clinical usefulness as much as we’d like to imagine it does. Rabusertib, for example, is an extraordinarily selective chk2 inhibitor, but guess what? Lilly has given up on it after failures in the clinic, from every indication, because that’s just not enough to show benefits in the real world. It’s also worth noting that some of the compounds in this study are listed at chemicalprobes.org, but turn out not to be as selective as previously thought (!)

The group tried profiling across mechanistic classes – for example, Type I inhibitors (which target the enzyme’s active conformation) versus Type II inhibitors (which hit inactive ones), but those two really didn’t show much of a selectivity difference across the numerous compound examples. The covalent kinase inhibitors (a smaller set) are more selective, but still not perfect. Some of them do hit other kinases, just without the covalent “warhead” coming into play (after all, they have to fit into an active site for the covalent modification to occur). So it’s difficult to generalize.

As for off-target effects in general:

As expected, the vast majority of compounds interacted with protein/lipid kinases, but our study also revealed binding to seven metabolic kinases, 19 other nucleotide binders, five FAD (flavin adenine dinucleotide) binders, and the heme-binding enzyme FECH (ferrochelatase) (Fig. 3A and table S2). These unanticipated interactions not only may lead to desired consequences but also can represent mechanisms of drug toxicity. A survey of the scientific and patent literature (using PubMed, SciFinder, or ChEMBL) revealed that many of the 243 drugs investigated in this study are surprisingly poorly characterized with regard to their target space or bioactivities.

Exactly (and there are many more details on off-target binding to other proteins with ATP binding sites, I should add. Update: see here for a review on these things). And that’s even as the authors note that there are over 110,000 papers in PubMed and over 47,000 patents and patent applications in Scifinder on kinase inhibitors. There’s a power-law distribution, as you might expect – half of those papers are on just five compound! At the other end of the scale, although everything apparently shows up in the patent literature, there are 17 kinase inhibitors that have been into human trials that still have no publications on them in PubMed. But that still leaves you with a lot of literature to cover in between (with a lot of gaps) which is why I’m glad that this new paper exists.

That’s how, for example, we know about Tafinlar (dabrafenib). It’s in the literature as a selective BRAF inhibitor, but that may not be the case:

The kinobeads data showed that the drug is a multikinase inhibitor with ~30 submicromolar targets (fig. S5B). Kinase activity assays confirmed potent inhibition of several SRC family members, and there was no apparent difference in selectivity between the three RAF family members. Moreover, wild-type (WT) BRAF and the V600E mutation for which the drug is used in the treatment of melanoma were equally well inhibited (fig. S5, C to F).

It’s not alone. As mentioned above, there’s nothing wrong with polypharmacology per se in this area, but everyone should know what the real situation is. Claiming selectivity seems to be an artifact of the selectivity-is-good mindset that all of us tend to have, but you know what’s really good? Clinical efficacy and safety. Your chances for the latter are probably improved if your kinase inhibitor is not a blunderbuss that blasts the kinome to shreds, of course, but your chances for the former are not necessarily improved by exquisite targeting. Either way, the real selectivity data need to be out there.

The paper suggests a number of older candidates for re-evaluation on more recently appreciated kinase targets, either as drugs or as starting points for new programs, and the paper suggests several specific examples of approved compounds that may well have until-now-unevaluated new indications (such as cabozantinib in FLT3-ITD–stratified AML patients). Back upstream, the authors also show how the data can be used to try to profile entire pathways in cellular assays:

One key challenge in drug discovery is to assess whether a drug molecule engages a target or associated pathway in a cell. The present resource allowed us to explore this in a novel way by analyzing the phosphoproteome of cancer cells in response to KI treatment and by integrating this information with the target spectrum of the drug(s) used. To illustrate this concept, the phosphoproteomes of BT-474 cells after treatment with the EGFR/HER2 inhibitors lapatinib, afatinib, canertinib, dacomitinib, and sapitinib were determined to a depth of ~15,000 phosphorylation sites (fig. S8C and table S9). The analysis revealed a surprisingly large number of statistically significantly regulated phosphorylation events for each drug. . .

The five drugs mentioned have about 211 protein phosphorylation events in common in that particular cell line, which tells you a lot about the EGFR/HER2 network – but on the other side, they all have somewhat different selectivity profiles against other kinases in those same cells, so you can learn from what they have in common and from the places where they differ as well.

So anyone who’s at all interested in the kinase inhibitor world needs to look over this paper. And anyone who might want a reminder of (on one side) just how messy and complex things are, or (on the other side) how many interesting opportunities remain out there, should have a look, too. This is state-of-the-art stuff.