Thread by @JSheltzer

Our new paper is out today. We used CRISPR to uncover some really striking findings with several drugs and drug targets in clinical trials. Also, we accidentally found the first-ever inhibitor of the cyclin-dependent kinase CDK11. stm.sciencemag.org/content/11/509…

This project started with a statistic that was really shocking to us: 97% of drug-indication pairs tested in human cancer patients fail during clinical trials and never receive FDA approval.

(Source: academic.oup.com/biostatistics/…)

This suggested to us that there were some big problems in how new cancer drugs were being studied. We thought that one potential factor that could contribute to this high failure rate was if drugs were entering trials with an incorrect understanding of their mechanism-of-action.

To test this, we selected a set of drugs and drug targets at various stages of clinical development, and we set out to confirm their mechanism-of-action.

We chose genes that had been reported to be cancer-essential, and drugs that had been reported to act through those specific targets. We excluded drugs that had a published resistance-granting mutation, which we think is the gold-standard for proving an MOA.

We then set out to verify that these genes were cancer-essential. Surprise #1: using a variety of CRISPR approaches (CRISPR competitions, CRISPR-interference, CRISPR-KO clones), we found that we could eliminate each of these genes without a detectable loss of cancer fitness.

So, right now, there are on-going clinical trials directed against proteins that so far as we can tell have no effect on cancer cell growth or proliferation.

Why were these genes previously believed to be essential? Largely due to RNAi experiments. When we re-tested the constructs that had been previously used, we found that they continued to kill clones in which their putative target had been deleted.

This indicates that RNAi promiscuity has contributed to the mis-identification of cancer-essential genes and the initiation of clinical trials against non-essential targets.

What about the drugs being used in these trials? First, some positive controls.

In every instance in which a drug had a published resistance-granting mutation, we were able to use CRISPR to verify its mechanism-of-action. Recapitulating work in yeast, if you eliminate FKBP12, you get rapamycin resistance, for instance.

However, for every drug that we tested that lacked a known resistance-granting mutation, we found that the drug kills cells through a target-independent effect.

For instance, the drug PAC-1 is reported to work by activating caspase-3. It’s currently being used as a caspase activator in multiple clinical trials. But, we totally eliminated caspase-3 using CRISPR in four different cell lines, and we saw no effect on PAC-1 resistance:

So, a drug that’s in clinical trials as a caspase-3 activator kills cells in a caspase-3-independent manner.

There are lots of clinical trials right now with HDAC6 inhibitors. But… we deleted HDAC6, and we saw no change in how cells respond to the drugs:

We end up with a list of potent anti-cancer drugs, except it turns out that they don’t act through the mechanism-of-action that they were previously believed to. So… what do they actually do?

We thought that the key to figuring out a drug’s MOA is genetic evidence. So, we took one compound, called OTS964, and we raised cancer cells that were resistant to it. Our hope was that the resistant clones would harbor a mutation in its true target.

We got some OTS964-resistant clones, and using whole-exome sequencing, we found that every single resistant clone had a mutation in the kinase domain of CDK11B!

Using CRISPR, we showed that this single amino acid substitution was both necessary and sufficient for OTS964 resistance.

We did some follow-up in vitro experiments to confirm… OTS964 is the first specific inhibitor of CDK11 that has ever been found, and we found it by profiling a mischaracterized drug.

So, where does this leave us? I think, that if we want “precision medicine” to be successful, we need to know exactly what genes are cancer-essential, and we need to be able to develop drugs that can target those proteins.

I think that stringent genetic testing – particularly using multiple orthogonal CRISPR approaches – and the isolation of drug-suppressor mutations can validate gene-essentiality and confirm on-target effects.

We’re also very excited to keep studying CDK11. Other members of the cyclin-dependent kinase family have been successfully targeted in breast cancer, and we think that a therapeutic window may also exist for CDK11 inhibitors (with the right predictive biomarkers).

And, of course, there are caveats: we couldn’t test every cancer cell line in existence, so it’s possible that these genes are important in a rare lineage that we didn’t examine.

Secondly, we performed fitness experiments in vitro to specifically test the hypothesis that these genes represented cell-autonomous genetic dependencies. We can’t rule out the possibility that these genes have some role in an in vivo-specific process like immune recognition.

To sum up: CRISPR turns cancer cells into yeast. You can make clean knockouts, you can do suppressor screens, and you can make really precise genetic alterations to deeply interrogate hypotheses about gene and drug function.

I think that applying these genetic technologies in a preclinical setting will decrease the number of new drugs that get tested in cancer patients but fail to provide any clinical benefit.