I’ve written a number of times here about bifunctional protein degraders, which have been a big topic in drug discovery for the past few years. There’s a new paper that illustrates some of the challenges in this area, and it’s worth using as an example.
For those outside the field, the idea behind these things is pretty straighforward, at least in principle. You find a protein that you think is involved in a disease process, one whose activity you would like to dial down. You find a small-molecule ligand that binds that protein – you may already have some inhibitors around, in house or from the literature, and for these purposes your small molecule doesn’t even have to be an inhibitor, just a binder. (Of course, the way we run assays means that most of the time we’re not set up to detect silent binders, so those are thinner on the ground). Now you break out your synthetic organic chemistry skills and build out a linker group from that known ligand, and at the other end of that linker you attach a known ligand for an “E3 ligase” enzyme. There are several possibilities, but so far the well-established ones for the enzymes cereblon and VHL are the ones that get used the most, by far.
And what happens when this new bifunctional molecule is dosed is that the “ligand end” goes and binds to your target protein, while the “E3 ligase end” goes and binds to its target. This brings the E3 ligase and its associated partner proteins in close proximity to the target protein – your linker isn’t very long compared to the size of the proteins involved. And the E3 ligase complex starts doing what it does, which is sticking ubiquitin groups all over the surface of the nearby protein. Those are a signal for protein degradation, the cellular equivalent of a red waste disposal tag, and your target protein is then dragged off to the cellular recycling center, aka the proteasome, to be shredded, while your bifunctional molecule skips away to do the whole thing all over again. End result: the cellular levels of your target protein drop sharply, weirdly, non-physiologically, and its functions are thus greatly impaired, just like you wanted.
OK, that’s how it works in theory. In practice, there are a lot of things to slow you down. Maybe you don’t have a ligand for your target protein, for example – maybe no one has ever reported one. Maybe the process of building a linker group off of it kills its activity for its target, and you have to keep searching for different ways to have one coming off of a different part of the molecule (some of which may turn out to be rather difficult, chemically). Maybe your resulting bifunctional compound just doesn’t work when you dose it in cells! Could be it doesn’t penetrate cell membranes well, or it goes in and gets pumped right back out, or it doesn’t do a good job (for some reason) of forming a productive ternary complex (target protein/bifunctional/E3 ligase). For that last process, see this recent review. In that case, you’ll need to try a few more linkers. And probably a few more after that! This would be a good time to mention that in the absence of other painstaking experiments, you won’t really be able to tell if your problem is cell permeability or lack of activity once it’s inside. Or both, of course, why not. And who knows – maybe neither cereblon nor VHL are up to doing the job on your particular target protein, but another more exotic E3 ligase would work? No good way of knowing that beforehand, either.
And that’s the subject of the paper linked in the first paragraph. It’s from a group at Janssen/J&J, and they’re detailing a sped-up way to try out new linkers in cellular assays as quickly as possible. That “for some reason my bifuncional doesn’t seem to lower protein levels” problem is a common one, but at the same time there are many examples now of such cases that turned around and worked well once a better linker was found. The tough part is, we’re not quite sure about how to make a better linker. That answer seems to vary from system to system, and there are a lot of possibilities. You can change the length of the tether between the two ends of your bifunctional, for sure. You can change its polarity and solvation (introducing oxygen or nitrogen atoms into the chain). You can change its geometry and flexibility entirely, too, by incorporating rings (aliphatic or aromatic), double or triple bonds, and so on. You can see how the list of possibilities becomes alarmingly large. You’d think that you could sort through them by checking how your new molecules bind to the target and then how they bind to the E3 ligase, but that generally doesn’t help you. It’s the formation of that active ternary complex that’s often the key, and a cell assay is the most realistic place to answer such questions. There are plenty of cases where bifunctionals that had the best binding constants at each end didn’t work so well, while others with apparently lower affinity did the job in the real systems.
And unfortunately, a lot of the time we’re still at the “try ’em and see” stage of optimization. That’s why you see papers like this being written: if we’re going to have to run through a bunch of different linkers, empirically banging on cellular doors and looking under enzymatic rocks over and over, then we need to streamline that make-them-and-put-them-into-cells process as much as possible. It would be much, much nicer if we could stand at the whiteboard or look at a screen, pursing our lips thoughtfully and then pointing purposefully at The Compound To Make, but we ain’t there yet. In this paper, the team really went for it, doing high-throughput chemistry on small scales and testing the resulting bifunctionals without even really purifying them by chromatography. They did run them across scavenger resins to try to strip out unreacted starting materials and reagents, which certainly helps, but it’s still a bareback approach compared to the usual way we’d do things.
The linkers themselves were simplified chemically to diamines, and both the protein ligand and the E3 ligase ligand had carboxylic acids on them, so it’s just a bis-amide formation all the way. That still leaves you with hundreds, thousands of diamine choices to stretch across the middle, but there will always be the (unanswerable) question of What If Amides On Both Ends Are Bad? In this case, they ended up calculating properties on about 2800 possible candidates and narrowing those down to 91 to synthesize as a scatter across chemical property space. In this case, the ligand end was always a molecule that’s known to bind the BRD4 protein, whose degradation by bifunctional molecules is well established, and these were made with two different standard cereblon ligands on the E3 side. And these were run through both regular live cells and “permeabilized” ones, to try to distinguish these reasons for inactivity.
As for cell permeability, the paper found that overall having fewer hydrogen bond donors and acceptors in the linkers helped cell permeability. Low cLogD values (indicating relatively more polar compounds) correlated with worse cell permeability and both of these trends follow the accepted med-chem wisdom, for what it’s worth. But looking at pairs of similar compounds showed some weird effects. One of their flexible-ether-linkers performed far better with one of the cereblon-targeting ligands than the other, even though that particular compound had only rather weak cereblon target engagement in cells when you just looked at that alone. More confirmation that you have to run the experiments! If you’d tried to toss compounds based on the E3 target engagement assays you’d have chucked one of your best compounds before you even knew about it. But it’s not just because it was flexible – another shorter and more rigid linker also showed a big split between the two cereblon-targeting possibilities. Meanwhile, other linkers could act almost identically between the two. The linkers with basic amines in them tended to behave that way, for example. The detailed reasons for all of these end results are unclear indeed.
So until we get all this straightened out, and I wouldn’t hold my breath while waiting for that, we’re just going to have to keep whacking away in as efficient a manner as possible. Eventually we will pile up enough data – such is the hope – to start making sense of these trends, but it won’t be simple. There are clearly a number of different factors at work here: how the two protein ends are brought together, in distance and in orientation, where the ubiquitination sites are on the target protein and how they’re presented to the E3 ligase complex, what the kinetics are of that whole process (on- and off-rates), and more. The only thing we can say for sure now is that it’s a complex situation with a lot of room for surprises. Go make some more linkers!