I remember the time before anyone did fragment-based drug discovery, when I heard high-micromolar binding ligands described as “carpet lint” or “stuff from the bottom of your shoe”. And you only heard then when the assays themselves even read out at that level, which certainly wasn’t a given. When I started, a suggestion that one should go back this screen that didn’t read out any reproducible hits below 1 micromolar and start from a selection of the smallest molecules out below (say) fifty micromolar would have been met with raised eyebrows and some concern for your state of mind. No one had really heard of ligand efficiency back then (note that link for some advanced thoughts on the subject).
Now, of course, it’s a standard method for drug discovery, grown well past its controversial years – one more tool in the collection. But it’s still an open question how far you can push the idea. In other words, how weakly bound a ligand can you start with and still optimize to useful potency? Intuitively, you feel that it has to get harder the further out you go, but how much harder, and how far out? These thoughts are brought on by this new paper from Penn, which describes a new variation on the initial fragment screening step.
As people discovered when fragments became popular, sometimes you got pretty lively hits from the technique and sometimes you didn’t. Fragment collections absolutely have a higher hit rate by percentage; that’s from the whole idea of covering simpler binding motifs more comprehensively in chemical space. But you can still run your entire fragment collection over a target and get nothing that seems worth advancing. Now, “run you fragment collection over” is kind of a slippery phrase, because there are many methods of fragment screening. Picking up low affinities accurately and reproducibly is a different sort of problem than sorting through a giant compound deck looking for the nanomolar hits, and it needs different approaches. Probably the best way to do it is to have several orthogonal biophysical techniques (for example, NMR, SPR, a thermal-shift assay like DSF, direct X-ray crystal soaking, etc). You run the collection all the different ways and see what the overlap is, for starters, prioritizing the compounds that hit by more than one technique.
But what if you do that and you *still* don’t have anything particularly interesting? One thing to think about is the detection limit of those various assays. Detecting single-digit micromolar binders should be easy, double-digit micromolar ones a bit less so as you start heading out. But what about approaching a thousand micromolar? Millimolar binders are pretty weak by drug discovery standards, even for diehard fragment screeners. Double digit millimolar? Not many people will have experienced that, not least because it’s really difficult to get an assay that can even read out at such weak binding affinities. At some point you start to wonder what the binding constants really are for that stuff off the bottom of your shoe.
What this new paper does is increase the concentration of the fragments around the protein of interest by using reverse micelles, as shown at right. Inside a bubble of aqueous solvent that’s floating in an alkane bulk phase (and surrounded by an amphipathic layer), the molarity of the fragments is increased to levels that you really shouldn’t try to reach in bulk solution. Interestingly, the authors report that the proteins under these conditions don’t show the sorts of artifacts that appear when you actually try that latter experiment. And you use less material as well – amounts of compound that would correspond to 0.8 mM in bulk are actually about 40 mM inside the reverse micelles. You do have to play around with the buffers and concentrations to get conditions under which your protein still looks happy, but once you’re there you can start in with the fragments and see what you get. The reverse micelles themselves are inside a bubble of N,N-dimethyldodecylamine N-oxide in the examples shown.
One of those examples is interleukin 1-beta, which is certainly a representative of proteins that don’t give you a lot to work with as far as small-molecule binding sites. As you’d figure, you need to bias yourself towards polar fragments (cLogP < 0.5), so they’ll want to partition into the aqueous compartments. That cuts the Maybridge 2500-fragment library down to about 233 compounds. Screening mixtures of these against IL-1beta under standard conditions (100 micromolar protein, 800 micromolar fragment) gave no hits whatsoever. But under the reverse-micelle conditions, 31 compounds showed binding. Fitting dissociation constants to the chemical shift changes showed that these were binding at 50 mM to 1M (!) And the group validated these by more traditional NMR experiments in low-volume capillaries – the binding constants reproduced, albeit with uglier spectra than seen in the micelles (which were actually pretty good).
Ten of the 31 were too nonspecific to proceed with, on closer inspection. Now, almost all of the remaining 21 compounds turned out to have multiple binding sites, which is what you’d probably expect from such extremely simple interactions, but they were at least discrete binding events. And these cover about two-thirds of the entire surface of the protein; you’d have to assume that pretty much any part of it that can recognize a small molecule has been given its chance to do so. (Note that many of the specific-binding compounds also had at least one nonspecific interaction with some other region of the protein, though). When you consider that under standard screening techniques IL-1beta is basically a featureless cue ball, this is pretty interesting.
But are these things useful – that is, can they be developed into more potent species? There are some examples from the fragment literature of this succeeding (here’s one, although I’d guess that there are plenty of examples that didn’t work that will never be heard of in the literature). You can also take potent ligands and break them back down into plausible fragments, only to find that these all have very low affinity. And the authors here demonstrate that a hit from the list does show apparent SAR when a set of analogs is screened (some improve, some get worse, etc.) So there may well be paths forward – and as the paper mentions, the necessity for polar fragments with this technique is actually a useful complement to the more traditional methods and the chemical starting points they provide.
So this could be an interesting addition to the fragment arsenal, particularly in those break-glass-in-case-of-emergency situations when you have an important target that’s a well-structured protein that nonetheless ignores your entire screening collection. You’d probably only start back at these binding potencies for such an important target, for sure, and you’re going to want some experienced fragment folks and a real time commitment, because it’s likely going to be a long road back up, with no shortage of dead ends. But there are certainly targets that are worth it.