Skip to Content

Bright Chemical Matter?

When the topic of “frequent hitters” in drug screening comes up (PAINs, promiscuous compounds, whatever you’d like to call them), the literature divides into two general camps. I don’t think they’re of equal size, but still. . .the larger one is the one that I caucus with, the one that says “These things need to be watched carefully, because they have an excellent chance of wasting your time”. But there’s another group, typified by the paper discussed here, who look at these structures as a real opportunity. So much activity! Hitting targets that nothing else can reach!

A new paper in Drug Discovery Today makes just this case: “Filtering promiscuous compounds in early drug discovery: is it a good idea?” Their big objection is that many compounds act polypharmacologically, and that filtering out PAINs only makes sense if you’re thinking one-drug-one-target, or perhaps they’d say still stuck thinking that way. They especially recommend not filtering if you’re doing phenotypic screening, and point out a number of known drugs that would have been filtered out.

I’m unconvinced. Doxorubicin, for example, is brought in to make the case for quinones. And while it’s certainly true that it’s a valuable chemotherapy drug, it’s also true that cancer chemotherapy is one of the few places where it would be part of the pharmacopeia. It’s toxic stuff. The other therapeutic areas where you find quinones also tend to be massive-attack situations (antiinfectives) or topical agents. It’s very hard to dose something like that systemically and not run into trouble. Not impossible, to be sure – just very difficult, and you have to weigh those chances of success versus what else you might have to work with.

The authors also take care to distinguish many forms of assay interference from such in vivo problems, which is fine. Fluorescent interfering compounds, aggregators, and the like are trouble no matter what, and you need to run the appropriate controls to make sure that you don’t go chasing after them. (The lower the intrinsic hit rate of your assay, the greater the danger of this kind of thing). Still, some of these mechanisms are related. Redox-active compounds can be trouble in the assay well, and trouble in the whole animal, too, for similar reasons.

This paper calls frequent-hitter chemical classes “Bright Chemical Matter”, referring  to the “dark chemical matter” discussed here. Those are the compounds that hit very rarely in screens – which, interestingly, are hard to distinguish from a lot of the stuff that does hit.

It was suggested that DCM is in fact made of nonpromiscuous molecules that can occasionally demonstrate activity in specific biological assays. Thus, DCM could potentially have unique activity and clean safety profiles, making DCM a valuable starting point for lead optimization efforts [25]. The reciprocal is true: molecules that frequently have a biological effect on diverse assays, but do not act as bioassay interference compounds, although challenging, can be optimized into innovative and safe drugs under the premises of new conceptual frameworks such as systems biology and polypharmacology. Here, we propose referring to these privileged molecules with high biological activity as bright chemical matter. . .

The paper goes on to say that (in the authors’ view) such compounds can be optimized into useful drugs, especially if you’re looking for (or not selecting against) polypharmacology. I’m not denying this, but I am pointing out that this approach does give you a greater risk for failure. I suppose there are two ways of looking at this: you could say that our failure rates are so high already, that what’s a bit more? Or you could say that our failure rates are so high already that piling on even more risk is the last thing you’d want to do. My worry is that giving these compounds a name like “bright chemical matter” (which I really hope doesn’t catch on) makes them actually seem desirable instead. Even the authors here don’t go that far.

An interesting contrast to all this can be found in another new paper, from Jonathan Baell (who originated the PAINs acronym and has published widely on the concept). He’s looking at natural product structures that overlap with known promiscuous-activity motifs. As he points out, these quinones, catechols, Mannich bases and so on don’t lose their reactive character just because they happen to be found in natural products. In fact, many of them have probably been evolutionarily selected for just those shotgun properties. (They’re also typically sequestered in the cell, excreted, or produced in transient amounts if they’re used as signaling molecules). Says Baell:

As was the case for catechol-containing drugs, the quinone PAINS moiety can still display PAINS behavior even if it is embedded in a drug, and whether or not it is a natural product, and this behavior may contribute paradoxically to both efficacy and toxicity. Discovery of quinone-containing drugs once again arose from early observation of useful in vivo efficacy or potent cell-based activity at therapeutically relevant concentrations or close to thereof, prior to knowledge of mechanism of action. So once again, the utility of these compounds in humans does not represent a modern and rational progression from upstream assays to downstream assays and eventually to clinical use. No connection should be made between a quinone-containing screening hit and a quinone-containing drug. The PAINS behavior more or less universally exhibited in quinones should render them deemed to be unprogressable as low micromolar potency screening hits, whether in a target-based assay such as Kv1.3 described here, or a phenotypic assay.

Here’s where these two papers overlap. But Baell’s makes the point that compounds like doxorubicin were discovered in phenotypic assays already acting at the potency needed. If you’ve got a red-alert PAINs-type structure as a hit, it had better blow out your assay. The huge majority of the drugs containing these structures were found by direct in vivo screening, and were active at clinically relevant levels right from the start, not as micromolar leads that were going to be the foundation for a development program. But that’s where you see the great majority of PAINs in the literature now: micromolar stuff presented as promising leads against difficult targets. Optimizing these is not an impossible task (as the authors of the Drug Discovery Today article would be quick to say), but neither is it a very good bet (as Baell would say, and so would I). Just because doxorubicin is a valuable drug doesn’t mean that the 30-micromolar quinone that just came out of your protein-protein screen is a worthwhile lead compound. You’re facing some of the longest odds in the business if you try it.

21 comments on “Bright Chemical Matter?”

  1. Rule (of 5) Breaker says:

    This paper seems like an attempt to get attention. Some academics without real-world drug discovery experience think these compounds are good leads? Have at them I say! Those of us who have been unfortunate enough to come across these and have some of our time wasted by them know better. Also, it is not as if these compounds are always hitting a bunch of targets. They are often interfering with the assay – you know the “I” in “PAINs”

    So maybe some could make good leads, but best of luck figuring out what activity is real and what is an artifact in your assays.

    1. hn says:

      No, these kinds of projects take up money and resources that could be better used for other projects. They don’t provide good training for students.

    2. Peter Kenny says:

      If there is hard evidence that a compound is behaving badly from the perspective of drug action then I fully agree that it should be put out of its (and our) misery as quickly and mercifully as possible. Frequent hitter behavior by a class of compounds is clearly a cause for concern but you need to observe it using different assay technologies before you can damn an entire compound class. For example, frequent hitter behavior by catechols in AlphaScreen may reflect efficient singlet oxygen quenching/scavenging. It is also important to make a distinction between assay interference(compound does something unpleasant to the assay) and promiscuity (compound engages multiple targets) . Computational filters can sometimes lull you into thinking that your decisions are evidence-based but prejudices can just as easily be encoded as substructural targets

      My own experience in working up HTS output was that there were many compounds that we really didn’t like the look of but it was often difficult to obtain hard evidence that these were inherently bad. In my Wilmington days (1997-1999) we used to look at correlations between screens to see whether screens with the same assay technology (or targets sharing a mechanistic feature like catalytic cysteine) had a disproportionate number of hits in common.

      I’ve linked the first of series of blog posts on PAINS as the URL for this comment and, at the top of each of these, is a link to the next post.

      1. Golden Unicorn says:

        Peter, why do you need hard evidence of compound misbehaviour? It isn’t like medchem is a real science!

        1. Peter Kenny says:

          Golden Unicorn, We have to make decisions in MedChem with incomplete information and one measure of the ‘reality’ of a scientific field is how objectively the scientists in question handle uncertainty. I have argued that part of design is assembling the information that we need in order to make predictions. I explore this theme in the blog post linked as the URL for this comment and you’ll also find a link to an interesting Curious Wavefunction in that.

      2. HTSguy says:

        Peter Kenney, to take up your example, we’ve seen catechols as frequent hitters in biochemical assays using a range of detection technologies (none of them Alphascreen), so it’s not simple detection technology interference.

        1. Peter Kenny says:

          HTSGuy, I’m sure that some catechols will engage multiple protein targets but there are dangers in focusing on a substructure and assuming that all instances of that substructure will behave badly. A number of rhodanines in the original PAINS study don’t show as up hits in any of the 6 assays in the screening panel. Do you have any evidence that any of the catechols to which you refer actually interacts directly with the relevant proteins (or renders them non-functional)?

          If we’re serious about addressing the ‘false positive’ problem then we’re going to need to be a lot more open with chemical structures and associated screening data. If the compounds really as crap as everybody says they are then what’s to be lost? Personally, I do not regard evidence obtained by analysis of proprietary data to be evidence although I concede that others might consider this view to be extreme. One of the big challenges in MedChem is that opinion occasionally masquerades as fact and I’ve linked a blog post on this theme as the URL for this comment.

  2. Hap says:

    Perhaps they should be called “assay glitter” – because they shine brightly in assays and tend to get lots of attention (but not because they’re useful), and sometimes obscure what might be useful. Their use by musicians with likely future careers as trainwrecks might also be relevant. They also fit the well-noted predilection for fishers to get boats with lots of shinies on them, although fishers tend to actually catch some edible fish.

  3. Bioorganic Chemist says:

    “Assay glitter” is the best phrase I’ve seen in a while – bravo! And just like glitter, once they show up (in the literature) they’re almost impossible to completely get rid of (because somebody will keep using it for its claimed effects, generating pathway conclusions that are further elaborated on (with the role of the original compound in “establishing” this connection obscured with time)).

  4. Hap says:

    The optimism sounds like the authors read “McElligot’s Pool” too many times, though that could be a cautionary point both ways, because of Dr. Seuss.

  5. JAB says:

    As far as natural products go, the producing organisms make them for a reason, usually defensive, and for those purposes, nonspecific and promiscuous protein binders such as plant tannins are quite useful antifeedants, but lousy drug candidates. At the other end of the spectrum are very specific and potent compounds such as pilocarpine, which acts on the muscarinic M3 receptor and not much else. In between there are things like artemisinin, which hits many targets, but mostly in the malaria parasite. Lots of stuff in between!

    I confess to recent frustration hearing a very simple synthetic quinone structure being touted as a great probe in an NIH grant evaluation – I had to disagree. At the same time a very potent and specific compound of my own was dissed by a pharma group for having too many Michael acceptors, despite demonstrated in vivo activity….so it can cut both ways.

    1. Peter Kenny says:

      I don’t think one can write off a compound as a probe just for being a quinone and I’ve taken atovaquone (one of the components of the antimalarial drug marketed as Malarone) on several occasions without noticing any side effects. No doubt, some quinones will be unacceptable as probes and it’s important that we present evidence to support the case against compounds that we claim are unacceptable. It’s also important that we demonstrate that we can make a distinction between what we know and what we believe. If we can’t do so then we run the risk that those who fund drug discovery will conclude that the difficulties we face are of our own making.

      With respect to your compound being dissed by Pharma ‘experts’ for its surfeit of Michael acceptors despite showing in vivo activity, did you get it assayed for cysteine reactivity? Also was this something the Pharma ‘experts’ asked about? Pharma types are often very conservative (sometimes showing a touching Faith in absurdities like Ligand Efficiency and property forecast index) and often equate formation of covalent bonds between ligand and protein as equivalent to irreversible. Even when binding is irreversible, clear SAR can sometimes be discerned (e.g. for cysteine protease inhibition by vinyl sulfones).

      I’ve linked by blog post on ‘Expertitis’ as the URL for this comment

      1. drsnowboard says:

        Err… you’ve done the classification thing that Dereck refers to in reverse. Atovaquone is a drug that was optimised for its biological activity, redox activity being fundamental to its mechanism of action against the parasite. And you use your anecdotal evidence (patients =1) to assert quinones are not necessarily poor start points for DD campaigns. If we as scientists can’t construct an argument without skipping from anecdote to comparing apples to oranges, we might as well ask The Food Babe what she thinks

        1. Peter Kenny says:

          drsnowboard, You don’t appear to have noticed “(one of the components of the antimalarial drug marketed as Malarone)” which is not anecdotal and suggests a number of patients a lot greater than 1. I used the anecdote about taking atovaquone in a blog post (linked as URL for this comment) to make the point that I was less worried about the ‘quinoneness’ of atovaquone than I was about potential CNS effects of mefloquine which many compound quality ‘experts’ would claim had a more beautiful chemical structure. Anecdotal it may be but is it more so than the chorus of, “but it’s useful!” that is the knee jerk response to criticism of compound quality metrics?

  6. Barry Levine says:

    laying aside the Shoichet aggregators and the compounds that mess with particular aspects of your assay format (anything displaying a biotin…) consider the case of staurosporine. It’s a small molecule, it has decent transport properties, it really hits about 500 kinases (many of them below 50 nM, if memory serves)
    To judge by just the published literature, we have collectively wasted several whole careers (100s of man-years?) trying to carve a drug out of this promiscuous binder. that seems a very poor return on investment.

    1. Design Monkey says:

      500 kinases and in all of them it hits the ATP binding site. That’s not the same as quinones , who reacts with any cysteine residue in any protein they can lay their little grabby paws on. Also kinase inhibitors generally are for cancers or other equally nasty diseases, and that field is notorious for “anything goes” and relaxed tox/side effects requirements.

  7. LeeH says:

    It sounds like the issue may be being framed in somewhat too categorical a manner. I would definitely argue that there is merit in polypharmacology for certain therapeutic areas, and that too selective an agent is counterproductive. But does that mean that you want to hit 20 targets? 50 targets? Really? I’m guessing that for those applications 5 or 6 critical targets is near the sweet spot, and more than that you’re back to the classical game of optimizing activity and having to eliminate multiple liabilities towards the end of the project. Deja vu all over again.

    1. Peter Kenny says:

      When assessing promiscuity /selectivity by counting numbers of targets that a compound show activity against, it’s important to be aware of the threshold that defines activity. One prominent study defines actives as >30% inhibition at 10 micromolar and one can’t help wondering why such a low level was chosen. Kinases are interesting case in point since different assays will be run using different ATP concentrations (useful when screening where you need to detect activity for compounds that have not been optimized for the target of interest) but most (all?) kinases ‘see’ the same intracellular ATP concentration.

  8. hn says:

    No no no. They are not privileged or bright. You are more likely, but not guaranteed to fail, with these structures.

  9. partial agonist says:

    I call them glowing red chemical matter. As in don’t touch, and run away, FAST.

    The biologists are always enamored with what appears to be the most active compounds from the screen,

    Without fail the chemist then skips over all of these Michael acceptors, quinones, polyphenols, and thiazolidine diones (& their cousins), then hopefully his/her ears and eyebrows are raised, Spock-like, as he/she gets to page 2.

    For perfectly good reasons!

    Once you’ve wasted time on crap chemical matter, you learn something.

  10. Marcos says:

    Unfortunately spending time on problematic structures is commom place in my university. I know the authors and some people the group, but their paper seems like an attempt to justify the research of some.

Comments are closed.