Skip to Content

Analytical Chemistry

A Small-Molecule CRISPR Inhibitor

The number of stories and journal articles about how CRISPR DNA-editing technology works, has worked, and is planned to work are beyond counting. How about an article about how to stop it in its tracks? That’s this one, just published in Cell from a multicenter team in Cambridge and New York. It describes a screening program for small-molecule inhibitors of S. pyogenes Cas9 (spCas9), because one would want some ways (not all of which currently exist) to turn its effects off in given places and at given times. There are already protein-based inhibitors, but to the best of my knowledge, this is the first report of small-molecule ones. And in a line I like very much, the paper states “Unsurprisingly, the pharmacological inhibition of intracellular proteins is usually accomplished using small molecules.”

Yes, it is! So how did they find these? Because setting up a Cas9 inhibitor screen is not so straightforward. For one thing, the enzyme binds its substrate with picomolar affinity. It’s also (obviously) a DNA-binding protein, and finding small-molecule inhibitors of that interaction has been. . .well, let’s say “challenging”, with all the baggage that adjective implies. What’s more, a direct inhibitor would need to deal with two nuclease domains, and the overall structure of the Cas9 protein has some unusual protein folds in general, so your chances of getting hits from a standard screening collection could be diminished for that reason, too. Finally, as the authors note, no one has yet reported a Cas9 screen that’s even suitable for high-throughput screening (in terms of miniaturization, signal/noise, readout, etc.) Thus all the co-authors, no doubt.

The protein-based inhibitors, though, are known to target the binding of the Cas9 protein to the protospacer adjacent motif (PAM) on DNA, making it an obvious choice for a screen (the affinity there is much less daunting, while still vital for CRISPR function). The team set up a fluorescence polarization assay, a DSF (differential scanning fluorimetry) assay, and a BLI (biolayer interference) assay as a suite of ways to look for inhibitors, and for people who don’t do this stuff for a living, I’ll briefly go into what the heck those are. All such assays are built around some readout that changes when a molecule binds to a protein, but man, are there ever a lot of ways to skin that particular cat.

FP assays depend on fluorescently labeled species, which emit light at specific wavelength when another specific wavelength excites them. Think of a fluorescent sticker, which might give off bright green light when ultraviolet light hits it. If you used polarized light to do this excitation, you get polarized fluorescent light back out – but as it happens, the kind you get varies. If you shoot vertically polarized light in and measure both vertical and horizontally polarized light on the way out (at the new fluorescent-emitter wavelength) you’ll see a mixture of the two: and that mixture depends on the physical rotation, in solution, of the fluorescent species. So if something has bound to it in your assay, its rotation rate will slow down, and the larger the binding species, the more it’ll slow. Changes in fluorescent polarization are thus a measure of binding events.

DSF is a variety of thermal shift assay. You add a particular type of dye to a sample of your protein target of interest, a dye that is fluorescent but whose fluorescence is quenched when the dye molecules are floating around in water solution. You warm up the sample while watching for the particular fluorescence wavelength of your dye to appear. At the melting point of the protein, you’ll see a pretty sharp increase in the signal, because the protein has now unfolded and thus exposed some new hydrophobic surfaces for the dye molecules to bind to – and now that they’re out of aqueous solution, they will fluoresce. The key thing is, when a ligand has bound to the protein, the resulting complex is generally more stable, and has a higher melting point. You test small molecules against the protein/dye system, looking for the ones that make the complex melt (and hence fluoresce) at higher temperatures. In other words, you’re looking for a thermal shift, and generally the larger the better. You can, of course, also look for compounds that are disrupting some complex that’s already bound (of two proteins, say, or of protein/DNA), in which case you’ll be watching for lower melting points.

Now, BLI is the only one of these three that I haven’t been involved with. The acronym is “bio-layer interferometry“, and it depends on being able to immobilize your protein target onto a small tip. Quite a few of these biophysical techniques depend on that sort of attachment; it has its good points and its bad ones. Chief among the latter is that you may find that you can’t seem to attach your protein in a useful manner, and it’s often impossible to say just why that’s happening, which leads you to try another attachment chemistry method, and then another, and there are a lot of choices. For BLI, you’re shining light on the spot of attached protein and reading out the resulting interference pattern. If ligands bind to that protein sample, though, the optical thickness of layer increases, and shining light on it now gives you a shifted interference pattern, and you’re measuring the difference. Optical interference patterns are brutally sensitive – in my weekend hobby of astronomy, that’s how you test telescope mirrors during their production, and it’s a very unforgiving test indeed. One nice thing about this instrument is that you can monitor the binding event (and its dissociation, if you want) in real time, a feature BLI shares with another common immobilized-protein biophysical assay, SPR.

Back to CRISPR inhibitors. Running three orthogonal technologies for a screen like this is highly recommended, because each of them (like any assay) is subject to its own false positive and false negatives. What you want, ideally, are compounds that work across several very different techniques, greatly increasing your confidence that real protein binding events are at the center of it all. And indeed, when this team ran a diverse set of about 10,000 molecules across the FP assay, they found particularly enriched hits in three different chemical classes. The first class, though, tended to give an FP signal even when you left out the Cas9 protein in the assay, which is why you had better run those control experiments! The second class of compounds tended to have fluorescence of their own, which always complicates the assay interpretation (and is one of the particular banes of FP and other fluorescence-driven assay techniques), and in addition were also cytotoxic when taken on into cells. So that left the third class.

That one made it through all the assay techniques, and was active in cells as well. Control experiments suggested that it was indeed affect the Cas9/PAM interaction, and in cells, the compounds were able to disrupt a model system where dCas9 caused upregulation of particular targeted genes (shown at right). So these really do look like Cas9 inhibitors, and it appears that they’re interacting first with the complex of Cas9 and its guide RNA. I’ve left out a lot of validation experiments, I should add:

Our screening strategy involved disrupting DNA binding by SpCas9, followed by demonstrating activity in multiple mammalian cell lines using gene or protein delivery. Furthermore, we demonstrated inhibition of SpCas9 nuclease and transcription activation in assays with gain of signal (e.g., eGFP-disruption assay), loss of signal (e.g., HiBiT assay), various DNA repair pathways, and a myriad of readouts (e.g., fluorescence, luminescence, next-generation sequencing [NGS], qPCR).

What exactly is going on – how compound binding disrupts the enzyme activity in detail – remains to be seen. It could be direct competition, could be an allosteric site, who knows. The SAR of the compounds, from what I can see, looks pretty tight. Small changes in the compound structure or in the protein target can have big effects, which suggests some sort of tight, specific interaction. The hope is that structural biology techniques will be able to shed some light on the actual structures of the bound forms. Along the way, we’ll probably learn quite a bit about CRISPR inhibition (and what it can do), as well as about the CRISPR enzymes themselves.

25 comments on “A Small-Molecule CRISPR Inhibitor”

  1. MoBio says:

    I see no data on potential off-target actions of the best compound (normally this would include some sort of profiling). Also, how do they exclude aggregators? Finally, the compound is not especially potent…..

    1. Derek Lowe says:

      Probably in the next paper! The authors say that “it remains to be seen” if these molecules (or indeed, the reported CRISPR-inhibiting proteins) have other cellular targets.

    2. Kim Tai Tran says:

      High potency may not be required, since the SpCas9-PAM interaction is fairly weak.

  2. anonymouuse says:

    Why the f*#k would one plot a graph purporting to show compound potency in which the ordinate does not start at zero? and on an non-log(10) abscissa?

    1. Peter S. Shenkin says:

      The abscissa is on a log scale, though it is labeled somewhat idiosyncratically. Note that the distance between 1 and 2 is equal to the distance between 5 and 10 and to the distance between 10 and 20.

  3. Barry says:

    The inhibition seems pretty insensitive to absolute stereochemistry. That’s alarming (“Pfeiffer’s Law”) for an agent purporting to bind to a (highly chiral) protein, to a (highly chiral) DNA strand, or to both

    1. PhotoDeTox says:

      Is it really single enantiomers or do the stereobonds only describe the relative configuration of a racemate? I did not read the paper.

      1. Chirality says:

        Those are enantiomers, not racemate (Figure 4E). None of the enantiomers are active within the experimental error while their inhibitor shows dose-dependent activity. I would say the inhibition is pretty sensitive to stereochemistry as only 1 of the 4 enantiomers is active.

  4. utility says:

    Derek is clearly a fan-boy of this, but in practice how would this really help? Would it improve off-target profiles? No, the crispr protein is still integrated and active so unless someone will stay on this molecule for long after treatment, I don’t see what limiting the kinetic window of action will do. Will it make crispr more efficient? No? Will it help delivery? No. Cool case study in finding new chemical matter, but I really don’t see why this will really aid crispr efficacy.

    1. @Dev2Death says:

      Therapeutic use of Cas9 will probably be limited to ex vivo electroporation of patient derived cells. The Cas9 inhibitor could improve the off-target profile for a specific guide by decreasing the editing rate or limit activity to specific parts of the cell cycle which is important to avoid p53-induced cell death

  5. DMPK dude says:

    That pyridine group might make it a pretty good pan-CYP inhibitor… Just saying…

  6. Jb says:

    Eh, you can engineer in some tissue specific promoters. Inducible systems also exist. The biggest issue is the safety of Cas9 itself, because it is likely immunogenic and can be presented by MHCs for T cell targeting. Right now I can only see CRISPR being used ex vivo. But why would you need an inhibitor then? You could just transfect cells with the RNA or the complex itself which is transient enough.

  7. tlp says:

    since CRISPR-Cas9 is involved in bacterial protection against phages, can one make a case for this inhibitor as new class of antibiotics?

    1. Unchimiste says:

      Very smart, I like the idea.

      1. TroyBoy says:

        I bet you could set up a good phenotypic screen with phage and Cas-expressing bacteria and get better chemical starting points.

    2. loupgarous says:

      … very cool, a small molecule that shields phages from bacterial defense mechanisms (similar to β-lactamase inhibitors when used along with amoxicillin and/or ticarcilin).

      You’d still want to be vigilant for off-target activity, but you’d be assisting something Nature selected to work in its original setting, not a penicillin derivative. It should reduce the chances for misadventure.

      1. Scott says:

        That sounds like a *very* good idea to chase down for those nasty XYZcilin-resistant bugs out there.

        And also to reduce creating more resistant strains!

    3. Sriram Shankar says:

      Locus Bio, an RTP company, is engineering phages with CRISPR-Cas9 to act as targeted antibiotics, interestingly.

  8. TroyBoy says:

    I worry about non-specific double stranded DNA/RNA intercalation. This could interfere with some of their validation assays as well. How would ethidium bromide do in their assays, including their validation assays? To my untrained chemist eyes, that central ring system looks like an intercalator. More work needs to be done!

    1. Pete says:

      TroyBoy, I was thinking the same thing. That core looks like it would intercalate DNA/RNA and there were not assays shown in the paper to address this as a possible mechanism. And it makes sense that the primary assays would pick up compounds with this mechanism.

      1. Deepthinker says:

        An intercalator will non-specifically inhibit all CRISPR enzymes and inhibit transcription activation at all genomic locus in their dCas9 transcription assay. On the contrary, their inhibitors only block SpyCas9 and not its cousin Cas13a/Cpf1 (Fig 5I) and they report inhibition of only target genes in dCas9 transcription activation (Fig 5J). Also, the inactive analog of their inhibitor has the same core (fig 4C) and if the intercalation mechanism was going on, the “inactive analog” should be active.

  9. Barry says:

    CRISPR/Cas9 shields bacteria only from old ‘phage. That might keep a patented ‘phage useful longer. But usually with therapeutic ‘phage, one is betting on the rapid evolution of new ‘phage to out-evolve the pathogens

Leave a Reply

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

Time limit is exhausted. Please reload CAPTCHA.