Skip to main content


Your Cancer Targets May Not Be Real

I wrote here about a paper from Cold Spring Harbor labs that invalidated MELK as a cancer target. That was straightforward enough: knocking it out via CRISPR across a whole range of cancer cell lines had no effect on their growth at all, so it’s kind of hard to make the case that it’s an important thing to go after. At the same time, the paper showed that a MELK inhibitor with activity in cancer cells continued to show such activity after the target had been deleted.

Now the same team is back with an even larger look at the CRISPR-validation question, and it’s more food for thought. It’s also a look at the RNA-interference-validation question, because in that earlier work (and in this paper), the targets under discussion had been validated by siRNA techniques. These are HDAC6, MAPK14/p38-alpha, PAK4, PBK, PIM1, and caspase-3 (that last one targeted by activators) and the paper references 180 publications that indicate that these are essential in one or more cancer subtypes. The great majority of these assignments have been made through siRNA or shRNA experiments. Identification of these targets has led to the development of a number of small-molecule drug candidates, which have collectively been the subject of 29 clinical trials.

And they are probably not the targets that we thought they were. Because (as this paper shows) when you use CRISPR instead of the RNA techniques to knock down these proteins, the cancer cells that are supposedly sensitive to their loss (32 different lines!) don’t seem to mind any more. And the compounds – which certainly are ligands for the stated proteins – continue to work in the cells where that target has been deleted. For example, the caspase-3 activators continue to kill cancer cells that don’t even have caspase-3 expressed any more, which is a problem.

The group looked carefully to make sure that homologs of these genes weren’t being upregulated instead in the CRISPR experiments, and that doesn’t seem to be the explanation. Such CRISPR screens do identify targets such as BRAF or Aurora B; but they don’t identify these. Notably, even some earlier RNA screens don’t validate them, either: the team re-analyzed hundreds of reported genome-wide shRNA screens and found that these don’t really point at these targets, either. Switching to the CRISPR interference (CRISPRi) system also showed that you can’t explain the results just by assuming different responses to partial versus complete loss of function. It’s worse than that.

There’s also combination therapy to consider. The paper notes that HDAC6 inhibition is being combined with paclitaxel in trials, since its inhibition is thought to sensitize tumor cells to interference with the microtubule system. But the CRISPRed HDAC6 cell lines don’t show any particular sensitivity to paclitaxel (in contrast to past siRNA cell experiments), and similarly, a whole list of putative p38-alpha interactions doesn’t validate, either.

It comes down to effects of the siRNA protocol, apparently. The paper looks at some of the RNA constructs used for those earlier experiments and shows that when you apply those to cells where their putative target gene has already been knocked out by CRISPR, that you still see effects on proliferation and viability. Now that’s bad news. As the paper says, “Our results therefore suggest that these drug targets have advanced to clinical testing due, at least in part, to promiscuous RNAi constructs“. Oy.

The paper goes on to take a closer look at a reported PBK inhibitor (OTS964), which is indeed on the list of compounds that still work (somehow) when there’s no PBK around any more. Developing several resistance cell lines and doing whole-exome sequencing on them identified CDK11B as likely the real target in such cells, and which also makes OTS964 the first selective ligand known for CDK11. It turns out, on closer inspection, to be a 40 nM inhibitor of that subtype, with good selectivity. Not that anyone realized that until now, of course. Knocking in the mutant form of the kinase into tumor cells instantly made them resistant, too. Further work showed that CDK11 is important for mitotic entry, and does indeed appear to be a valid cancer target in its own right – as opposed, say, to PBK. Or the others on the list above.

Our results indicate that many cancer drugs in clinical trials kill cells independently of their reported targets“, say the authors, and I’d agree that they’ve backed that statement up pretty thoroughly. This doesn’t mean that these compounds are necessarily going to be ineffective in the clinic – polypharmacology can work, and hitting a specific target that you didn’t know about (as with the CDK11 story) can work, too. After all, these compounds had enough preclinical efficacy to make the case for going into man. It’s just that the reasons that people had attached to them were wrong, and that’s something that we really should know about. A serious disconnect can emerge if the patients picked for the clinical trials are selected based on the status of the wrong target protein, of course!

These results also make an even stronger argument against drawing too many conclusions from RNA-level knockdown experiments in cancer cells. Many researchers have already come to similar conclusions about CRISPR experiments versus siRNA/shRNA ones based on their own experiences, but this is an excellent chapter-and-verse demonstration of just what the problems are and how deep they can run. I would say that any paper proposing a cancer target where the main line of evidence is siRNA/shRNA should be double-checked in just this manner. . .

23 comments on “Your Cancer Targets May Not Be Real”

  1. Sken says:

    Shots on goal am I right?

    1. Anon says:

      When each shot has a billion-dollar price tag, maybe not…

  2. darts_are_my_stock_broker says:

    How many MD simulations were done, 3D pharmacophores were constructed using those putative protein targets?

    1. jim says:

      It doesn’t say that the compounds do not hit the targets they were designed for. Just that they can kill cancers without hitting it.
      So all in-silico, and just normal med-chem design as well – no reason why this was just a computational problem, was done in good faith and delivered for what was asked.

      1. CADD-UK says:

        Also not all these targets and their compounds were originally worked on for cancer.

  3. Frank says:

    I think RNAi dirtiness is well-know effect in industry since early 2010. But papers relying only RNAi still get published on “top-tier” journals. A few biotech continued to go after ShRNA-based targets.

    A key question for the field, though, there might be very few true cell-autonomous targets and we may have exhausted it (think the CRISPR screen vs RNAi screen).

    1. Tommysdad says:

      These weren’t just biotechs or academic labs. CELG, PFE , etc

  4. Anonymous says:

    Time for a new poll on the value of phenotypic screening versus target-based screening? (Previously discussed In The Pipeline; link in my handle.)

    1. John Wayne says:

      I’d argue that everybody really does phenotypic screening, but the compounds have to also be active in the flagship project assay to be tested further.

      1. Derek Lowe says:

        Cynical but hard to refute. . .

  5. John Wayne says:

    A nonzero percentage of all your targets may not be real; it’s probably worse in cancer than other areas.

  6. Jakob says:

    It has been suggested that every presently known drug has approximately 6 target proteins:

    Data completeness–the Achilles heel of drug-target networks. Nat Biotechnol 2008;26:983–4
    Mestres J, Gregori-Puigjane E, Valverde S, Sole RV

    1. Cb says:

      And in addition their metabolites which also hit a bunch of different targets

  7. Anon says:

    Great headlines come from simple statements, but the authors have oversimplified their analysis and overstated the impact of their findings. I know one of these inhibitors was only ever claimed to be ~10x selective over other family members, and I can’t imagine any of these compounds are thought to be “completely” selective. So yes, depending on the concentration of any small molecule it will be more or less selective. In one of their assays they show cell killing in knockout cell lines, but with IC50’s of 10uM! I’m not aware of any small molecules that retain perfect selectivity for their primary target at those kinds of concentrations… Oh, and sure RNAi has issues, but since when does CRISPR have no off target effects?

    1. Probe master says:

      Agreed. I love it when folks refer to any chemical probe as “specific” for a certain target.
      Where’s the chemist in your team, and why are you not listening to them?

    2. MrXYZ says:

      Although the key finding of the paper had to do with mis-understanding of mechanism of action. I think it’s clear that many potential drugs (both small molecules and biologics) can bind multiple targets and, probably, in many cases this doesn’t matter one bit. Off-target binding is a problem when it leads to adverse effects or severely effects the PK of the molecule.

      This paper add another layer by reminding us that just because a molecule targets a specific protein and has an positive effect on cells/animals/people is not proof that targeting the protein is what led to that effect (sorry, that was a convoluted sentence). You always have to dig deeper.

      Definitely not an uncommon finding although this paper is a nice case study.

  8. Barry says:

    in 2018, the first siRNA drug was approved by FDA, two decades after the phenomenon was first characterized. Specific targeting was billed as a revolution relative to small molecules. As we learn more about siRNA and its limitations, it looks less revolutionary

    1. Mol biologist says:

      May be everyone need a little bit more time? 10 or 20 more years? Since, Derek is chronically refusing a redundancy of information in human genome and keeps telling us that the target strategy is approachable.

    2. Mol biologist says:

      Since, Derek is chronically refusing a redundancy of information in human genome and keeps telling us that the target strategy is approachable. 10 or 20 more years?

  9. MLM says:

    Reminds me about my personal off-target frustration with siRNA work… One of the main issues with reports on siRNA-mediated phenotypes is that many studies frequently report a phenotype that is based on observations with 1 siRNA or 1 pool of siRNA’s. Even though pooling might improve the efficacy, the used concentrations are frequently high enough that each individual siRNAs within a pool of 4 siRNAs still has the potential to elicit off-target activities > 4 x risk at off-target.

    The level of confidence in siRNA work could be greatly improved if the phenotype would be repeated by parralel experiment with a different individual siRNA against the same target (or rescue).

  10. FTLast says:

    Even when genetic manipulations correctly identify a target protein with enzymatic activity as being important for cancer cell growth, that does not mean that a small molecule inhibitor will necessarily be effective- a lot of enzymes have other cellular functions unrelated to enzymatic activity.

    William Kaelin had a nice discussion of the issues involved in Nature Reviews of Cancer in 2017. The gold standard for confirming an enzyme’s activity as being a possible drug target should be an active site mutation.

    Or, as others have noted, maybe just do phenotypic screening in the first place. Death is a pretty straightforward phenotypic endpoint.

  11. Stifen says:

    Small molecule inhibitor is different from RNAi, since target protein has several other domain not target by the same inhibitor. Otherwise, PROTAC molecules might be less off-target than RNAi or CRISPR.

  12. Steve says:

    It seems silly to expect siRNA to have only one target when microRNA routinely have hundreds. At this rate, developing microRNA drugs might make the most sense.

Leave a Reply

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

Time limit is exhausted. Please reload CAPTCHA.