Skip to Content

PAINs by X-Ray

It’s safe to say that the concept of “pan-assay interference compounds” evokes some strong feelings in medicinal chemists. And those feelings run in several directions: some people are very glad to have a tool with which to winnow down their screening hit lists (or at least to prioritize them), while others are infuriated by the idea of tossing out what could actually be useful compounds on the basis of structural similarities alone. (Here’s a recent overview (open access version) from the originators of the concept). I myself land somewhere in the middle of this one. I think the idea of such a list is a sound one, as long as it’s seen in shades of trouble rather than a sharp dividing line. If you want to try to develop one of these structures, you’d better have a good reason, and you should be prepared to give it extra scrutiny. It’s a sliding scale – some of the PAINS structures are particularly prone to give false positives with some assay technologies but not so much with others, for example, so things might not be so bad. But if you want to proceed with (say) a rhodanine that has a quinone hanging off it, you have to realize that you’re running a huge risk of wasting your time.

This new paper examines well-known PAIN motifs in the X-ray crystallographic literature, which is an angle that hasn’t gotten as much attention. Digging through the PDB, the authors find 2784 structures of such ligands – but those represent 1107 separate molecules, which tells you that some of them are showing up in several proteins. That’s one flag, binding promiscuity at the X-ray structure level, but it’s not the whole story. The paper also identified a number of structures where it appears that reactive groups have formed covalent adducts with the protein binding sites – but that’s not the whole story, either.

That’s because there are also many “normal”-looking structures, where the compounds are making reasonable interactions with their binding pockets. In many cases, these interactions don’t actually involve the problematic parts of the structures – it’s not the PAINful part of the molecule that’s feeling the protein. This emphasizes that you can’t take the fact that a compound’s (sub)structure is on a PAINs list and just cross it off immediately as an artifact. (Nor, of course, should you proceed blithely ahead with it!) There are marketed drugs with some of these structure in them, remember.

But there are a lot more marketed drugs that don’t have them. I very much believe that if you take a randomly selected similarly-sized list of functional groups and run it through a Protein Data Bank ligand search, you will find far fewer problematic results. And remember that an X-ray structure search certainly doesn’t cover all the mechanisms that can make a compound troublesome. For example, as the paper shows, you can find aminoacridines making perfectly legitimate binding interactions in X-ray structures – but that doesn’t mean that they’re not interference hits in fluorescent assays, because believe me, they are.

This paper, at least to me, reinforces the “yellow caution light” approach to these structures. It’s lazy and bad practice to immediately throw all such structures into the compost pile, but it’s just as bad (or worse) to ignore the potential problems and carry on as if everything is going to be fine.

21 comments on “PAINs by X-Ray”

  1. Jason says:

    Sounds like a politician:

    some people are very glad to have a tool… while others are infuriated… I myself land somewhere in the middle of this one.

    1. Derek Lowe says:

      Oh, now them’s fightin’ words. . .

    2. UKPI says:

      Same here. I find the complete tools we have running the country quite infuriating…

      1. Anon says:

        Only a bad workman blames his tools…

    3. Chris Phoenix says:

      Oh, if only we had more politicians who were willing to land in the middle!

      The idea that an extreme position is the only valid one is part of the dysfunction we’re facing today. Any complex system is not going to have simple answers.

      “There are only two genders!” Um, not factually correct, even at the chromosome level.
      “GMOs are ____!” Likely true for some GMOs, almost certainly untrue for others.
      “Guns should _____!” Where is the politician who will say, “Even if that’s true, it won’t solve the bigger problem?”

      Strict party-line votes are a symptom of a very deep problem with our democracy. When our politicians are unable to vote as they actually think, what kind of government can we have? Where is the wisdom in that?

      So being called a politician may be fighting words, but being called middle of the road is probably a badge of honor and wisdom these days. We need more of it.

      1. Pennpenn says:

        It’s just a pity that so many thinking in such a strict dichotomy are going to see people trying to hold the middle ground as being on the other side (and/or traitors to the cause) and discount or attack them, further polarising whatever situation we’re talking about. That’s even leaving out situations where the correct position to hold on something is demonstrable and people will still reject it out of hand because it conflicts with their worldview.

  2. John Wayne says:

    I am in complete accord with Derek’s philosophy on this subject. Problematic structures and substructures have been known to medicinal chemistry for decades; dealing with them requires a balanced approach. If you don’t pay any attention to these issues you run the risk of wasting a lot of time. If you over correct you become guilty of a sort of molecular stereotyping that will limit your ability to make real discoveries; a majority of the drugs that really help people are non ideal if approached from the perspective of looking at screening hits.

  3. HTSguy says:

    So once these are found during lead ID, you recommend “proceed with extreme caution” at best (probably because there isn’t anything better that has been found)? I’m curious about your and your readers (hopefully Medicinal Chemists) gut feelings about a related issue: when assembling a (of necessity very limited relative to chemical space) compound screening collection, should molecules like these be avoided?

    1. Barry says:

      There have been compounds (organomercurials*…) that I have yanked out of our screening deck. But others–even if I would not treat them as med-chem leads–stay in. If they give x-ray co-crystal structures, I may be able to exploit those contacts (and the induced conformation of the target) with a very different small molecule.

      *they did fill some “small molecule diversity space” that the computational guys thought was worth filling in the screening set

    2. tommysdad says:

      There are so many molecules out there of beautiful diversity and no triage issues — why would you pollute a screening deck with molecules that will be artefacts a significant percentage of the time?

  4. Peter Kenny says:

    I completely agree that one should be extremely wary of compounds that have been demonstrated to behave badly in assays. Should one worry the potential of rhodanines to interfere with an AlphaScreen readout if assaying a rhodanine by SPR? I would say no and the SPR assay is going to be a lot more informative about shit hitting the fan than claims that some rhodanines were frequent hitters. Should one worry about rhodanines being electrophilic? If a rhodanine lacks the exocyclic double bond that would make it a Michael acceptor then I would worry a lot less about Michael addition (even though rhodanines lacking the exocyclic double bond still display frequent-hitter behavior in the PAINS assay panel). All that said, I would still worry about metabolic stability of rhodanines on account of the thiocarbonyl group and I would also still worry about CYP inhibition by acidic rhodanines lacking N-substituent. However, these are pharmacokinetic and not assay worries.

    I’ve linked an article DOI as the URL for this comment. This is the article on photochemically enhanced binding of rhodanines (and other compounds) to TNF-alpha. A crystal structure is reported (for one of the other compounds) and it shows that a covalent bond forms between a peptide backbone nitrogen and one of the carbon atoms of a substituted phenyl ring.

    1. tangent says:

      more informative

      Exactly the thing, the relative amounts of information from one versus the other. Could we take a usable stab at quantifying this beyond the idea of “it’s not absolute, but it’s a good guideline”?

      It’s not just bits of information, it’s more like KL divergence, that can express that a PAINS structure renders one assay meaningless but doesn’t do much against another. Not saying it’s easy. But what’s the state of the art for integrating all these pieces of information?

      1. Peter Kenny says:

        Hi tangent, Guidelines like PAINS filters are in essence simple predictive models and the simplicity of a model (or metric) does not absolve it from the requirement that it be predictive of relevant phenomena. To assess how good a guideline is you need to be able to assess the relevance of the supporting data and the strengths of the trends in that data. In my view, the most useful warning that a compound may be misbehaving in an assay is the observation that structurally similar compounds have been shown to misbehave in similar assays. I have linked my recent JCIM perspective on assay interference as the URL for this comment in which some of this is discussed in more detail.

  5. Wavefunction says:

    So just like Ro5, Ro3 and the laundry list of gospels in medicinal chemistry, we need to treat PAINS as guidelines and not rules? Funny how we keep on coming back to the same theme and make the same mistakes.

    1. Peter Kenny says:

      Hi Ash, Even guidelines need a basis and in drug discovery that means trends in data. The strength of the trends and the relevance of the data tell you how strictly you should adhere to the guidelines. With respect to PAINS filters, I would argue that an assay panel of 6 AlphaScreens is not relevant if you’re using FRET or SPR. Ro3 should not be grouped with Ro5 since details of the supporting analysis for Ro3 are yet to be disclosed and Ro3 hydrogen bond donors and acceptors remain undefined. I have linked the War and PAINS blog post as the URL for this comment.

  6. Villoutreix says:

    A first study on the topic was reported last summer, Drug Discov Today. 2017 Aug;22(8):1131-1133. The authors were for instance pointing to molecules related to Varespladib that contain a PAINS substructure. Yet, these molecules were of importance to gain structure-function insights. Varespladib was stopped due to inadequate efficacy… but this is different story …

    1. Barry says:

      Geldanamycin did not prove to be a useful lead for small-molecule inhibitors of Hsp90 (despite the work at Kosan). But it is a quinone (red flag!) that gave us an instructive x-ray diffraction structure bound to Hsp90.
      Lots of assay hits that should not be pursued as lead compounds nonetheless have bits to teach us.

  7. Calvin says:

    I’ve always considered the PAIN compounds to be things I should avoid if I possibly can. As Barry, notes, though, it depends on what you want to do. Sometimes a crap hit can still be helpful even if it is not a useful starting point for a med chem project.

    My personal bias is that I’d much rather start from a less potent compound that had fewer structural liabilities than many PAINs possess. I’ve seen far too often epic amounts of time (some of which were my own) wasted on trying to fix a crap starting point and the optimize. So personally, I would not work on a PAIN compound unless the data I had was utterly compelling and I was sure that I had a real hit with a decent chance of being optimized.

    I’d rather try to get potency from a fragment than “fix” a PAINs compound. Sure, I might miss that one in a billion one that turns out to be a drug but that’s a risk I’m willing to take. My risk appetite is high, but crappy hits are rarely are fruitful starting points for good projects.

  8. Adam Shapiro says:

    Methods have been described in the literature to identify non-specific inhibition by screening hits in enzyme assays, and to correct for interference by compounds with measurements. Hit evaluation test cascades can include secondary assays using detection methods orthogonal to the screening assay, and biophysical measurements to verify specific binding of the compound to the target. Together, these techniques can greatly simplify the hit list, eliminating most or all of the hits that are false positives or have undesirable inhibition modes.

  9. heretic says:

    The overarching rule of med chem these days that is ‘thou shalt not consider any molecule that does not fit the rules of drug-likeness’. You cant consider one as a hit, learn from its intarctions never mind, and god forbid, _make_ a molecule that does not pass the filters.

    We are throwing out useful information, especially in strcuture based prgrams, because the hit molecule has an unappealing feature even if you are miles off the desired efficacy.

    Nothing beats being told that your initial fragment hit is uselss as it conatins a motif that has an AMES risk.

  10. Ray says:

    “…well-known PAIN motifs in the X-ray crystallographic literature, which is an angle that hasn’t gotten as much attention.”

    This pun is painfully good.

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.