The Cravatt group (in collaboration with partners from Harvard and Bristol-Myers Squibb) has a paper out on Chemrxiv that’s a followup to a 2017 paper of theirs which (I will freely admit) is one of my favorites. That was on taking fragment-sized compounds (and a slightly higher-MW collection), each labeled with a diazirene (for photoaffinity binding to proteins) and an alkyne (to report in later readouts via click conjugation), and looking at what sorts of proteins they labeled in living cells. These things pulled out small-molecule-responsive proteins that no one had ever investigated before.
But going on to the grand project of mapping the ligandibility of the human proteome and turning out lists of new chemical probes is a big step, although this sort of approach (with FFF, “fully functionalized fragments”) could do it in theory. In practice, though, things were confounded by difficult SAR around the probes themselves:
The fragment binding elements in our initial studies were selected based on their representation in drug-like molecules and were accordingly diverse in structure and physicochemical properties. As a consequence, we found that individual FFF probes showed substantial differences in their overall proteomic interaction profiles, which made for complicated structure-activity relationships (SARs) requiring careful manual review to identify fragment-protein interactions that reflected authentic recognition events (versus simply correlating with the overall proteomic interaction profiles of the FFF probes).
Their solution is an ingenious one: they take a set of chemical probes that have chiral centers in them, and make both enantiomers. Then testing each enantiomer separately in cell experiments allows you to cancel out all the physiochemical property variables, and you’re left only with the interactions that really reflect protein interactions. Using just eight pairs of these “enantioprobes”, they’ve identified a variety stereochemistry-dependent binding examples.
These were run in good ol’ HEK293T cells, but also in primary human peripheral blood mononuclear cells. 176 proteins showed enantiomer-selective interactions with the probes (defined as at least a 2.5-fold difference between the pairs), with 119 of those in the PBMCs and 108 in the HEK293Ts. In general, the stereoselective interaction profiles were similar in both cell types, although the PBMCs did show some immune-related proteins popping up (PARP10, IRAK3) that didn’t hit in the HEK203s. From the chemistry end, the number of proteins labeled varied quite a bit from one scaffold to another (as you’d imagine) and between enantiomeric pairs in a given scaffold (which is good news for picking up real hits). Around 80% of the 176 proteins only hit with one compound out of the whole list of 16.
And just as with the original paper, the types of proteins being picked up are of great interest. There are, naturally, quite a few enzymes, with binding sites that are presumably ready in many cases to accommodate small molecule probes. But there are also scaffolding proteins and transcriptional regulators, things that are considered very difficult to target, and here they are showing enantiodiscrimination in photoaffinity labeling. Follow-up experiments transfecting a varied set of the protein hits into more HEK cells confirmed the protein labeling. One really, really wants to rip through some interesting cell lines with a much larger collection to see what sorts of proteins might fall out, and I’m sure that’s the eventual hope for this work: to come up with (as much as possible) Small Molecule Probes For Everything. This paper illustrates the principle with 16 compounds, but imagine the size of the screens that could be run, and how many proteins might be probed!
Interestingly, some of the proteins that came out already have one or more small-molecule ligands, so the obvious question was whether these would compete with these new photoaffinity hits. Indeed, this was the case for the lysine methyltransferase SMYD3, the lipid-binding protein UNC119B, and the sterol transporter TSPO. And in these experiments, they also noted the small-molecule competitors not only blocked the enantioprobe interactions on these target proteins, but often the enantioprobe interactions on the other proteins already found for them as hits. The binding events across these different sites must be rather similar, and these other proteins were actually unknown targets for the literature small-molecule ligands themselves.
The paper goes on to describe a higher-throughput multiplexed method of screening these enantioprobe pairs using isobaric tandem mass spectrometry. I won’t go into the details of that (folks who are really into it will already be downloading the PDF and reading about it!), but the readouts from this modification correlated very well with the lower-throughput experiments. At least 85% of the protein interactions showed up again, and 115 new ones were picked up as well (some of these were cases where both enantiomers hit a target without much stereopreference).
There are complications, for sure – for example, how well the readouts show up might also be due to differential photolabeling reactions as well as the binding events themselves (photoaffinity labeling, as anyone who has done it well knows, is not a perfectly transparent process!) And there are a number of questions to be addressed as this work goes forward. How will these enantioselective reactions perform as you elaborate the structures further? Will new interactions with the larger structures take over? How elaborate do the probe molecules have to be, anyway? How far outside of fragment size/complexity should you go, and how many more proteins will you start to uncover to make that effort worthwhile?
There are other questions that apply to photoaffinity experiments in general, and we can already make some educated guesses (although the problem is that the data for these things may be buried and scattered across a wide number of experiments across many labs). What kinds of interactions are these, anyway? How many of them are taking place in functional binding sites, versus allosteric sites, versus more “dark matter” chemical handholds on protein surfaces? Are there classes of proteins that will resist all attempts to generate probes for them (intrinsically disordered ones, for example?) What sorts of different results will you get by (say) testing at different points in the cell cycle or under different sorts of environmental stress? As the data pile up, will there be cellular compartments that are more resistant or more friendly to small-molecular probes in general? A broad-based effort to probe the whole proteome has a better chance of answering such questions in an organized fashion.
So let’s have one! Or several – there’s certainly room for a lot of people to join in, because the number of potential experiments is downright staggering. But so is the goal: find small molecule chemical matter that binds proteins we’ve never targeted before, and assemble a list of potential chemical probes like no one has ever seen. It’ll be great.