About this Author
College chemistry, 1983
The 2002 Model
After 10 years of blogging. . .
Derek Lowe, an Arkansan by birth, got his BA from Hendrix College and his PhD in organic chemistry from Duke before spending time in Germany on a Humboldt Fellowship on his post-doc. He's worked for several major pharmaceutical companies since 1989 on drug discovery projects against schizophrenia, Alzheimer's, diabetes, osteoporosis and other diseases.
To contact Derek email him directly: firstname.lastname@example.org
August 21, 2014
Edward Zartler ("Teddy Z" of the Practical Fragments blog) has a short piece in the latest ACS Medicinal Chemistry Letters on fragment-based drug discovery. He applies the term "fragonomics" to the field (more on this in a moment), and provides a really useful overview of how it should work.
One of his big points is that fragment work isn't so much about using smaller-than-usual molecules, as it is using molecules that make only good interactions with the target.. It's just that smaller molecules are far more likely to achieve that - a larger one will have some really strong interactions, along with some things that actually hurt the binding. You can start with something large and hack pieces of it off, but that's often a difficult process (and you can't always recapitulate the binding mode, either). But if you have a smaller piece that only makes a positive interaction or two, then you can build out from that, tiptoeing around the various landmines as you go. That's the concept of "ligand efficiency", without using a single equation.
He also emphasizes that having a simpler molecule to work on means that the SAR can be tested and expanded quickly, often without anyone hitting the lab bench at all. You can order things up from the vendors or raid your own screening collection for close analogs. This delays the entry of the medicinal chemists to the project, which (considering that their time is always in demand) is a feature to be happy about.
The article ends up by saying that "Fragonomics has won the field. . .The age of the medchemist is over; now is the time of the biophysicist." I don't know if that's quite the way to win friends and influence people, though. Medicinal chemists are rather sensitive to threats to their existence (with good reason), so my worry is that coming on like this will make chemists who haven't tried it leery of fragment-based drug design in general. I'm also not thrilled with "fragonomics" as a term (just as I'm not thrilled with most of the newly-coined "omics" terms). The word doesn't add anything; it's just a replacement for having to say "fragment-based drug discovery" or "FBDD" all the time. It's not that we don't need a replacement for the unwieldy phrase - it's just that I think that many people might (by now) be ready to dismiss anything that's had "omics" slapped on it. I wish I had something better to offer, but I'm coming up blank myself.
+ TrackBacks (0) | Category: Drug Assays
August 19, 2014
Here's a very good review article in J. Med. Chem. on the topic of protein binding. For those outside the field, that's the phenomenon of drug compounds getting into the bloodstream and then sticking to one or more blood proteins. Human serum albumin (HSA) is a big player here - it's a very abundant blood protein that's practically honeycombed with binding sites - but there are several others. The authors (from Genentech) take on the disagreements about whether low plasma protein binding is a good property for drug development (and conversely, whether high protein binding is a warning flag). The short answer, according to the paper: neither one.
To further examine the trend of PPB for recently approved drugs, we compiled the available PPB data for drugs approved by the U.S. FDA from 2003 to 2013. Although the distribution pattern of PPB is similar to those of the previously marketed drugs, the recently approved drugs generally show even higher PPB than the previously marketed drugs (Figure 1). The PPB of 45% newly approved drugs is >95%, and the PPB of 24% is >99%. These data demonstrate that compounds with PPB > 99% can still be valuable drugs. Retrospectively, if we had posed an arbitrary cutoff value for the PPB in the drug discovery stage, we could have missed many valuable medicines in the past decade. We suggest that PPB is neither a good nor a bad property for a drug and should not be optimized in drug design.
That topic has come up around here a few times, as could be expected - it's a standard med-chem argument. And this isn't even the first time that a paper has come out warning people that trying to optimize on "free fraction" is a bad idea: see this 2010 one from Nature Reviews Drug Discovery.
But it's clearly worth repeating - there are a lot of people who get quite worked about about this number - in some cases, because they have funny-looking PK and are trying to explain it, or in some cases, just because it's a number and numbers are good, right?
+ TrackBacks (0) | Category: Drug Assays | Drug Development | Pharmacokinetics
July 24, 2014
The topic of phenotypic screening has come up around here many times, as indeed it comes up very often in drug discovery. Give your compounds to cells or to animals and look for the effect you want: what could be simpler? Well, a lot of things could, as anyone who's actually done this sort of screening will be glad to tell you, but done right, it's a very powerful technique.
It's also true that a huge amount of industrial effort is going into cancer drug discovery, so you'd think that there would be a natural overlap between these: see if your compounds kill or slow cancer cells, or tumors in an animal, and you're on track, right? But there's a huge disconnect here, and that's the subject of a new paper in Nature Reviews Drug Discovery. (Full disclosure: one of the authors is a former colleague, and I had a chance to look over the manuscript while it was being prepared). Here's the hard part:
Among the factors contributing to the growing interest in phenotypic screening in drug discovery in general is the perception that, by avoiding oversimplified reductionist assumptions regarding molecular targets and instead focusing on functional effects, compounds that are discovered in phenotypic assays may be more likely to show clinical efficacy. However, cancer presents a challenge to this perception as the cell-based models that are typically used in cancer drug discovery are poor surrogates of the actual disease. The definitive test of both target hypotheses and phenotypic models can only be carried out in the clinic. The challenge of cancer drug discovery is to maximize the probability that drugs discovered by either biochemical or phenotypic methods will translate into clinical efficacy and improved disease control.
Good models in living systems, which are vital to any phenotypic drug discovery effort, are very much lacking in oncology. It's not that you can't get plenty of cancer cells to grow in a dish - they'll take over your other cell cultures if they get a chance. But those aren't the cells that you're going to be dealing with in vivo, not any more. Cancer cells tend to be genetically unstable, constantly throwing off mutations, and the in vitro lines are adapted to living in cull culture. That's true even if you implant them back into immune-compromised mice (the xenograft models). The number of drugs that look great in xenograft models and failed out in the real world is too large to count.
So doing pure phenotypic drug discovery against cancer is very difficult - you go down a lot of blind alleys, which is what phenotypic screening is supposed to prevent. The explosion of knowledge about cellular pathways in tumor cells has led to uncountable numbers of target-driven approaches instead, but (as everyone has had a chance to find out), it's rare to find a real-world cancer patient who can be helped by a single-target drug. Gleevec is the example that everyone thinks of, but the cruel truth is that it's the exceptional exception. All those newspaper articles ten years ago that heralded a wonderful era of targeted wonder drugs for cancer? They were wrong.
So what to do? This paper suggests that the answer is a hybrid approach:
For the purpose of this article, we consider ‘pure’ phenotypic screening to be a discovery process that identifies chemical entities that have desirable biological (phenotypic) effects on cells or organisms without having prior knowledge of their biochemical activity or mode of action against a specific molecular target or targets. However, in practice, many phenotypically driven discovery projects are not target-agnostic; conversely, effective target-based discovery relies heavily on phenotypic assays. Determining the causal relationships between target inhibition and phenotypic effects may well open up new and unexpected avenues of cancer biology.
In light of these considerations, we propose that in practice a considerable proportion of cancer drug discovery falls between pure PDD and TDD, in a category that we term ‘mechanism-informed phenotypic drug discovery’ (MIPDD). This category includes inhibitors of known or hypothesized molecular targets that are identified and/or optimized by assessing their effects on a therapeutically relevant phenotype, as well as drug candidates that are identified by their effect on a mechanistically defined phenotype or phenotypic marker and subsequently optimized for a specific target-engagement MOA.
I've heard these referred to as "directed phenotypic screens", and while challenging, it can be a very fruitful way to go. Balancing the two ways of working is the tricky part: you don't want to slack up on the model just so it'll give you results, if those results aren't going to be meaningful. And you don't want to be so dogmatic about your target ideas that you walk away from something that could be useful, but doesn't fit your scheme. If you can keep all these factors in line, you're a real drug discovery scientist, and no mistake.
How hard this is can be seen from the paper's Table 1, where they look over the oncology approvals since 1999, and classify them by what approaches were used for lead discovery and lead optimization. There's a pile of 21 kinase inhibitors (and eight other compounds) over in the box where both phases were driven by inhibition of a known target. And there are ten compounds whose origins were in straight phenotypic screening, with various paths forward after that. But the "mechanism-informed phenotypic screen" category is the shortest list of the three lead discovery approaches: seven compounds, optimized in various ways. (The authors are upfront about the difficulties of assembling this sort of overview - it can be hard to say just what really happened during discovery and development, and we don't have the data on the failures).
Of those 29 pure-target-based drugs, 18 were follow-ons to mechanisms that had already been developed. At this point, you'd expect to hear that the phenotypic assays, by contrast, delivered a lot more new mechanisms. But this isn't the case: 14 follow-ons versus five first-in-class. This really isn't what phenotypic screening is supposed to deliver (and has delivered in the past), and I agree with the paper that this shows how difficult it has been to do real phenotypic discovery in this field. The few assays that translate to the clinic tend to keep discovering the same sorts of things. (And once again, the analogy to antibacterials comes to mind, because that's exactly what happens if you do a straight phenotypic screen for antibacterials. You find the same old stuff. That field, too, has been moving toward hybrid target/phenotypic approaches).
The situation might be changing a bit. If you look at the drugs in the clinic (Phase II and Phase III), as opposed to the older ones that have made it all the way through, there are still a vast pile of target-driven ones (mostly kinase inhibitors). But you can find more examples of phenotypic candidates, and among them an unusually high proportion of outright no-mechanism-known compounds. Those are tricky to develop in this field:
In cases where the efficacy arises from the engagement of a cryptic target (or mechanism) other than the nominally identified one, there is potential for substan- tial downside. One of the driving rationales of targeted discovery in cancer is that patients can be selected by pre- dictive biomarkers. Therefore, if the nominal target is not responsible for the actions of the drug, an incorrect diagnostic hypothesis may result in the selection of patients who will — at best — not derive benefit. For example, multiple clinical trials of the nominal RAF inhibitor sorafenib in melanoma showed no benefit, regardless of the BRAF mutation status. This is consistent with the evidence that the primary target and pharmacodynamic driver of efficacy for sorafenib is actually VEGFR2. The more recent clinical success of the bona fide BRAF inhibitor vemurafenib in melanoma demonstrates that the target hypothesis of BRAF for melanoma was valid.
So, if you're going to do this mechanism-informed phenotypic screening, just how do you go about it? High-content screening techniques are one approach: get as much data as possible about the effects of your compounds, both at the molecular and cellular level (the latter by imaging). Using better cell assays is crucial: make them as realistic as you can (three-dimensional culture, co-culture with other cell types, etc.), and go for cells that are as close to primary tissue as possible. None of this is easy, or cheap, but the engineer's triangle is always in effect ("Fast, Cheap, Good: Pick Any Two").
+ TrackBacks (0) | Category: Cancer | Drug Assays | Drug Development
July 22, 2014
So, when you put some diverse small molecules into cellular assays, how many proteins are they really hitting? You may know a primary target or two that they're likely to interact with, or (if you're doing phenotypic screening), you may not have any idea at all. But how many proteins (or other targets) are there that bind small molecules at all?
This is a question that many people are interested in, but hard data to answer it are not easily obtained. There have been theoretical estimates via several techniques, but (understandably) not too much experimental evidence. Now comes this paper from Ben Cravatt's group, and it's one of the best attempts yet.
What they've done is to produce a library of compounds, via Ugi chemistry, containing both a photoaffinity handle and an alkyne (for later "click" tagging). They'd done something similar before, but the photoaffinity group in that case was a benzophenone, which is rather hefty. This time they used a diazirine, which is both small and the precursor to a very reactive carbene once it's irradiated. (My impression is that the diazirine is the first thing to try if you're doing photoaffinity work, for just those reasons). They made a small set of fairly diverse compounds (about 60), with no particular structural biases in mind, and set out to see what these things would label.
They treated PC-3 cells (human prostate-cancer derived) with each member of the library at 10 µM, then hit them with UV to do the photoaffinity reaction, labeled with a fluorescent tag via the alkyne, and fished for proteins. What they found was a pretty wide variety, all right, but not in the nonselective shotgun style. Most compounds showed distinct patterns of protein labeling, and most proteins picked out distinct SAR from the compound set. They picked out six members of the library for close study, and found that these labeled about 24 proteins (one compound only picked up one target, while the most promiscuous compound labeled nine). What's really interesting is that only about half of these were known to have any small-molecule ligands at all. There were proteins from a number of different classes, and some (9 out of 24) weren't even enzymes, but rather scaffolding and signaling proteins (which wouldn't be expected to have many small-molecule binding possibilities).
A closer look at non-labeled versions of the probe compounds versus more highly purified proteins confirmed that the compounds really are binding as expected (in some cases, a bit better than the non-photoaffinity versions, in some cases worse). So even as small a probe as a diazirine is not silent, which is just what medicinal chemists would have anticipated. (Heck, even a single methyl or fluoro isn't always silent, and a good thing, too). But overall, what this study suggests is that most small molecules are going to hit a number of proteins (1 up to a dozen?) in any given cell with pretty good affinity. It also (encouragingly) suggests that there are more small-molecule binding sites than you'd think, with proteins that have not evolved for ligand responses still showing the ability to pick things up.
There was another interesting thing that turned up: while none of the Ugi compounds was a nonselective grab-everything compound, some of the proteins were. A subset of proteins tended to pick up a wide variety of the non-clickable probe compounds, and appear to be strong, promiscuous binders. Medicinal chemists already know a few of these things - CYP metabolizing enzymes, serum albumin, and so on. This post has some other suggestions. But there are plenty more of them out there, unguessable ones that we don't know about yet (in this case, PTGR and VDAC subtypes, along with NAMPT). There's a lot to find out.
+ TrackBacks (0) | Category: Chemical Biology | Drug Assays
July 9, 2014
Here's a look at Emerald Biotherapeutics (a name that's unfortunately easy to confuse with several other former Emeralds in this space). They're engaged in their own drug research, but they also have lab services for sale, using a proprietary system that they say generates fast, reproducible assays.
On July 1 the company unveiled a service that lets other labs send it instructions for their experiments via the Web. Robots then complete the work. The idea is a variation on the cloud-computing model, in which companies rent computers by the hour from Amazon.com, Google, and Microsoft instead of buying and managing their own equipment. In this case, biotech startups could offload some of their basic tasks—counting cells one at a time or isolating proteins—freeing their researchers to work on more complex jobs and analyze results. To control the myriad lab machines, Emerald has developed its own computer language and management software. The company is charging clients $1 to $100 per experiment and has vowed to return results within a day.
The Bloomberg Businessweek piece profiling them does a reasonable job, but I can't tell if its author knows that there's already a good amount of outsourcing of this type already. Emerald's system does indeed sound fast, though. But rarely does the quickness of an assay turn out to be the real bottleneck in any drug discovery effort, so I'm not sure how much of a selling point that is. The harder parts are the ones that can't be automated: figuring out what sort of assay to run, and troubleshooting it so that it can be reliably run on high-throughput machines are not trivial processes, and they can take a lot of time and effort. Even more difficult is the step before any of that: figuring out what you're going to be assaying at all. What's your target? What are you screening for? What's the great idea behind the whole project? That stuff is never going to be automated at all, and it's the key to the whole game.
But when I read things like this, I wonder a bit:
While pursuing the antiviral therapy, Emerald began developing tools to work faster. Each piece of lab equipment, made by companies including Agilent Technologies (A) and Thermo Fisher Scientific (TMO), had its own often-rudimentary software. Emerald’s solution was to write management software that centralized control of all the machines, with consistent ways to specify what type of experiment to run, what order to mix the chemicals in, how long to heat something, and so on. “There are about 100 knobs you can turn with the software,” says Frezza. Crucially, Emerald can store all the information the machines collect in a single database, where scientists can analyze it. This was a major advance over the still common practice of pasting printed reports into lab notebooks.
Well, that may be common in some places, but in my own experience, that paste-the-printed-report stuff went out a long time ago. Talking up the ability to have all the assay data collected in one place sounds like something from about fifteen or twenty years ago, although the situation can be different for the small startups who would be using Emerald (or their competitors) for outsourced assay work. But I would still expect any CRO shop to provide something better than a bunch of paper printouts!
Emerald may well have something worth selling, and I wish them success with it. Reproducible assays with fast turnaround are always welcome. But this article's "Gosh everything's gone virtual now wow" take on it isn't quite in line with reality.
+ TrackBacks (0) | Category: Drug Assays
June 25, 2014
A look through some of the medicinal chemistry literature this morning got me to thinking: does anyone have any idea of which drug target has the most different/diverse chemical matter that's been reported against it? I realize that different scaffolds are in the eye of the beholder, so it's going to be impossible to come up with any exact counts. But I think that all the sulfonamides that hit carbonic anhydrase, for example, should for this purpose be lumped together: that interaction with the zinc is crucial, and everything else follows after. Non-sulfonamide CA inhibitors would each form a new class for each new zinc-interacting motif, and any compounds that don't hit the zinc at all (are there any?) would add to the list, too. Then you have allosteric compounds, which are necessarily going to look different than active-site inhibitors.
My guess is that some of the nuclear receptors would turn out to win this competition. They can have large, flexible binding pockets that seem to recognize a variety of chemotypes. So maybe this question should be divided up a bit more:
1. What enzyme is known to have the widest chemical variety of active-site inhibitors?
2. Which GPCR has the widest chemical variety of agonists? Antagonists? (The antagonists are going to win this one, surely).
3. And the the open field question asked above: what drug target of any kind has had the widest variety of molecules reported to act on it, in any fashion?
I don't imagine that we'll come to any definitive answer to any of these, but some people may have interesting nominations.
Update: in response to a query in the comments, maybe we should exempt the drug-metabolizing enzymes from the competition, since their whole reason for living is to take on a wide variety of unknown chemical structures.
+ TrackBacks (0) | Category: Chemical News | Drug Assays
June 19, 2014
Here's a discussion I got into the other day, in which I expressed some forceful opinions. I wanted to run it past a wider audience to see if I'm grounded in reality, or out on my own island (which has happened before).
Without getting into any details, we were talking about an area of potential drug research that has to do with transcriptional regulation. This one is clearly complicated - what part of transcription isn't complicated? But it's known that you can get things to happen by using things like epigenetic tool compounds (bromodomains, HDAC inhibitors, methyltransferases), and nuclear receptor ligands. None of these give you everything you want to see, by any means, but you do see some effect.
My take on this was that an effort to follow up with more epigenetic compounds and nuclear receptor ligands might well be a case of the classic "looking under the lamp-post because that's where the light is" syndrome. We don't have many small-molecule handles for affecting transcription, went my reasoning, and although such things are bromodomains, HDAC inhibition, and nuclear receptor signaling are wide-ranging, there's a lot more than the compounds in these spaces surely don't cover. In fact, given the wide range of these mechanisms, seeing a little tickling of any given transcriptional mechanism is about what I would expect from almost any of them, applied to almost anything. But that, to my mind, didn't necessarily mean that it was a lead worth following up.
My recommendation was for a phenotypic screen, if a good one could be worked up. There must be plenty of stuff going on with this system that we don't have any idea about, went my thinking. In the same way that the matter we can see through a telescope is only a tiny fraction of what appears to be really out there in the universe, I think that there's a vast amount of "dark biology" that we don't know much of. And the overwhelming majority of it has to be considered dark if we only consider the parts that we can light up with small molecules. For something that has to involve a huge array of protein-protein interactions, protein-nucleic acid interactions, and who knows what ancillary enzymes and binding sites, I wondered, what are the odds that the things that we happen know how to do with small molecules are the real answer?
So if you're going to dive into such waters (and many of you out there might be swimming around in them right now), by all means test whatever epigenetic and nuclear receptor compounds you might have around. Maybe you'll get a strong response. But if it all comes back as a little bit of this and a tiny bit of that, I'd say that these are unlikely to be convertible into robust drug mechanisms - the odds are that if there even is a robust drug mechanism out there, that you haven't hit it yet and that it will announce itself a bit more clearly if you manage to. A well-designed phenotypic screen might well be the best way to find such things, always keeping in mind that a badly designed phenotypic screen is the tar pit itself, the worst of both worlds.
So, am I too gloomy? Too jaded? Or simply a well-meaning realist? Thoughts welcome.
+ TrackBacks (0) | Category: Drug Assays
June 13, 2014
See what you think about this PDF: Cyclofluidics is advertising the "Robot Medicinal Chemist". It's an integrated microfluidics synthesis platform, assay/screening module, with software to decide what the next round of analogs should be:
Potential lead molecules are synthesised, purified and screened in fast serial mode, incorporating activity data from each compound as it is generated before selecting the next compound to make.
To ensure data quality, each compound is purified by integrated high pressure liquid chromatography (HPLC), its identity confirmed by mass spectrometry and the
concentration entering the assay determined in real time by evaporative light scattering detection (ELSD). The compound's IC50 is then measured in an on-line biochemical assay and this result fed into the design software before the algorithm selects the next compound to make – thus generating structure-activity relationship data. The system is designed to use interchangeable design algorithms, assay formats and chemistries and at any stage a medicinal chemist can intervene in order to adjust the design strategy.
I can see where this might work, but only in special cases. The chemistry part would seem to require a "core with substituents" approach, where a common intermediate gets various things hung off of it. (That's how a lot of medicinal chemistry gets done anyway). Flow chemistry has improved to where many reactions would be possible, but each new reaction type would have to be optimized a bit before you turned the machine loose, I'd think.
The assay part is more problematic. There are assays suitable for in-line evaluation like this, but there are plenty of others that aren't. (I would think that SPR would be particularly well-suited, since it operates in flow, anyway). But that prior optimization that the chemistry needs is needed even more here, to make sure that things are robust enough that the machine doesn't generate crappy numbers (and more swiftly than you could do by hand!)
The software is the part I'm really wondering about. How is this thing picking the next round of analogs? Physiochemical descriptions like logD? Some sort of expert system with med-chem lore in it? Does it do any modeling or conformational analysis? Inquiring minds want to know. And I'd also like to know if they've sold any of these systems so far, and to hear some comments from their users.
Update: here's one.
+ TrackBacks (0) | Category: Chemical News | Drug Assays
June 9, 2014
X-ray crystallography is great stuff, no doubt about it. But it's not magic. It takes substantial human input to give a useful structure of a ligand bound to a protein - there are decisions to be made and differences to be split. It's important to emphasize, for those of us who are not crystallographers, that unless you have resolution down below 1Å - and I'll bet you don't - then your X-ray structures are not quite "structures"; they're models. A paper several years ago emphasized these factors for chemists outside the field.
About ten years ago, I wrote about this paper, which suggested that many ligand-bound structures seemed to have strain energy in them that wouldn't have been predicted. One interpretation is that there's more to ligand (and binding site) reorganization than people tend to realize, and that ligands don't always bind in their lowest-energy conformations. And while I still think that's true, the situation is complicated by another problem that's become more apparent over the years: many reported X-ray structures for ligand-bound proteins are just messed up.
Here's an editorial in ACS Medicinal Chemistry Letters that shows how bad the problem may well be. Reviews of the crystallographic databases have suggested that there are plenty of poorly refined structures hiding in there. But I didn't realize that they were as poorly refined as some of these. Take a look at the phosphate in 1xqd, and note how squashed-out those oxygens are around the first phosphorus. Or try the olefin in 4g93, which has been yanked 90 degrees out of plane. It's bad that there are such ridiculous structures in the literature, but the larger number of semi-plausible (but still wrong) structures is even worse.
Those structures at the left illustrate what's going on. The top one is an old PDB structure, 3qad, for an IKK inhibitor. It's a mess. Note that there's a tetrahedralish aromatic carbon (not happening), and a piperazine in a boat conformation (only slightly less unlikely). The structure was revised after this was pointed out to the middle version (3rzf), but that one still has some odd features - those two aromatic groups are flat-on in the same plane, and the amine between them and the next aryl is rather odd, too. Might be right, might be wrong - who's to know?
The most recent comprehensive look (from 2012) suggests that about 25% of the reported ligand-bound structures are mangled to the point of being misleading. This new editorial goes on to mention some computational tools that could help to keep this from happening, such as this one. If we're all going to draw conclusions from these things (and that's what they're there for, right?) we'd be better off using the best ones we can.
+ TrackBacks (0) | Category: Drug Assays | In Silico
June 4, 2014
So you make a new chemical structure as part of a drug research program. What's it going to hit when it goes into an animal?
That question is a good indicator of the divide between the general public and actual chemists and pharmacologists. People without any med-chem background tend to think that we can predict these things, and people with it know that we can't predict much at all. Even just predicting activity at the actual desired target is no joke, and guessing what other targets a given compound might hit is, well, usually just guessing. We get surprised all the time.
That hasn't been for lack of trying, of course. Here's an effort from a few years ago on this exact question, and a team from Novartis has just published another approach. It builds on some earlier work of theirs (HTS fingerprints, HTSFP) that tries to classify compounds according to similar fingerprints of biological activity in suites of assays, rather than by their structures, and this latest one is called HTSFP-TID (target ID, and I think the acronym is getting a bit overloaded at that point).
We apply HTSFP-TID to make predictions for 1,357 natural products (NPs) and 1,416 experimental small molecules and marketed drugs (hereafter generally referred to as drugs). Our large-scale target prediction enables us to detect differences in the protein classes predicted for the two data sets, reveal target classes that so far have been underrepresented in target elucidation efforts, and devise strategies for a more effective targeting of the druggable genome. Our results show that even for highly investigated compounds such as marketed drugs, HTSFP-TID provides fresh hypotheses that were previously not pursued because they were not obvious based on the chemical structure of a molecule or against human intuition.
They have up to 230 or so assays to pick from, although it's for sure that none of the compounds have been through all of them. They required that any given compound have at least 50 different assays to its name, though (and these were dealt with as standard deviations off the mean, to keep things comparable). And what they found shows some interesting (and believable) discrepancies between the two sets of compounds. The natural product set gave mostly predictions for enzyme targets (70%), half of them being kinases. Proteases were about 15% of the target predictions, and only 4% were predicted GPCR targets. The drug-like set also predicted a lot of kinase interactions (44%), and this from a set where only 20% of the compounds were known to hit any kinases before. But it had only 5% protease target predictions, as opposed to 23% GPCR target predictions.
The group took a subset of compounds and ran them through new assays to see how the predictions came out, and the results weren't bad - overall, about 73% of the predictions were borne out by experiment. The kinase predictions, especially, seemed fairly accurate, although the GPCR calls were less so. They identified several new modes of action for existing compounds (a few of which they later discovered buried in the literature). They also tried a set of predictions based on chemical descriptor (the other standard approach), but found a lower hit rate. Interestingly, though, the two methods tended to give orthogonal predictions, which suggests that you might want to run things both ways if you care enough. Such efforts would seem particularly useful as you push into weirdo chemical or biological space, where we'll take whatever guidance we can get.
Novartis has 1.8 million compounds to work with, and plenty of assay data. It would be worth knowing what some other large collections would yield with the same algorithms: if you used (say) Merck's in-house data as a training set, and then applied it to all the compounds in the CHEMBL database, how similar would the set of predictions for them be? I'd very much like for someone to do something like this (and publish the results), but we'll see if that happens or not.
+ TrackBacks (0) | Category: Drug Assays | In Silico
May 30, 2014
Many drug discovery researchers now have an idea of what to expect when a fragment library is screened against a new target. And some have had the experience of screening covalent, irreversible inhibitor structures against targets (a hot topic in recent years). But can you screen with a library of irreversibly-binding fragments?
This intersection has occurred to more than one group, but this paper marks the first published example that I know of. The authors, Alexander Statsyuk and co-workers at Northwestern, took what seems like a very sound approach. They were looking for compounds that would modify the active-site residues of cysteine proteases, which are the most likely targets in the proteome. But balancing the properties of a fragment collection with those of a covalent collection is tricky. Red-hot functional groups will certainly label your proteins, but they'll label the first things they see, which isn't too useful. If you go all the way in the other direction, epoxides are probably the least reactive covalent modifier, but they're so tame that unless they fit into a binding site perfectly, they might not do anything at all - and what are the chances that a fragment-sized molecule will bind that well? How much room is there in the middle?
That's what this paper is trying to find out. The team first surveyed a range of reactive functional groups against a test thiol, N-acetylcysteine. They attached an assortment of structures to each reactive end, and they were looking for two things: absolute reactivity of each covalent modifier, and how much it mattered as their structures varied. Acrylamides dropped out as a class because their more reactive examples were just too hot - their reactivity varied up to 2000x across a short range of examples. Vinylsulfonamides varied 8-fold, but acrylates and vinylsulfones were much less sensitive to structural variation. They picked acrylates as the less reactive of the two.
A small library of 100 diverse acrylates were then prepared (whose members still only varied about twofold in reactivity), and these were screened (100 micromolar) against papain as a prototype cysteine protease. They'd picked their fragments so that everything had a distinct molecular weight, so whole-protein mass spec could be used as a readout. Screening ten sets of ten mixtures showed that the enzyme picked out three distinct fragments from the entire set, a very encouraging result. Pretreatment of the enzyme with a known active-site labeling inhibitor shut down any reaction with the three hits, as it should have.
Keep in mind that this also means that 97 reasonably-sized acrylates were unable to label the very reactive Cys in the active site of papain, and that they did not label any surface residues. This suggests that the compounds that did make it in did so because of some structure-driven binding selectivity, which is just the territory that you want to be in. Adding an excess of glutathione to the labeling experiments did not shut things down, which also suggests that these are not-very-reactive acrylates whose structures are giving them an edge. Screen another enzyme, and you should pick up a different set of hits.
And that's exactly what they did next. Screening a rhinovirus cysteine protease (HRV3C) gave three totally new hits - not as powerful against that target as the other three were against papain, but real hits. Two other screens, against USP08 and UbcH7, did not yield any hits at all (except a couple of very weak ones against the former when the concentration was pushed hard). A larger reactive fragment library would seem to be the answer here; 100 compounds really isn't very much, even for fragment space, when you get down to it.
So this paper demonstrates that you can, in fact, find an overlap between fragment space and covalent inhibition, if you proceed carefully. Now here's a question that I'm not sure has ever been answered: if you find such a covalent fragment, and optimize it to be a much more potent binder, can you then pull the bait-and-switch by removing the covalent warhead, and still retain enough potency? Or is that too much to ask?
+ TrackBacks (0) | Category: Chemical Biology | Chemical News | Drug Assays
May 12, 2014
Here's the sort of experimental detail that can destroy a whole project if you're not aware of it. The platinum chemotherapy drugs are an odd class of things compared to the more typical organic compounds, but it's for sure that many of the people using them in research aren't aware of all of their peculiarities. One of those has been highlighted recently, and it's a sneaky one.
DMSO is, of course, the standard solvent used to take up test compounds for pharmacological assays. It's water-miscible and dissolves a huge range of organic compounds. Most of the time it's fine (unless you push its final concentration too high in the assay). But it's most definitely not fine for the platinum complexes. This paper shows that DMSO displaces the starting ligands, forming a new platinum complex that does not show the desired activity in cells. What's more, a look through the literature shows that up to one third of the reported in vitro studies on these compounds used DMSO to dissolve them, which throws their conclusions immediately into doubt. And since nearly half the papers did not even mention the solvent used, you'd have to think that DMSO showed up a good amount of the time in those as well.
What's even more disturbing is that these sorts of problems were first reported over twenty years ago, but it's clear that this knowledge has not made it into general circulation. So the word needs to get out: never dissolve cisplatin (or the related complexes) in DMSO, even though that might seem like the obvious thing to do. Editors and referees should take note as well.
+ TrackBacks (0) | Category: Biological News | Drug Assays