Friday’s news about the real-world effects of PCSK9 drugs continue to reverberate. It’s worth going into the topic again, but from a wider view of genomics-driven medicine, because this is currently one of the biggest strategies for drug discovery in the industry.
The reasoning behind this is pretty compelling. If you comb through the human population, you’re going to find a lot of genetic variation. Some people are going to have loss-of-function mutations, and some (probably fewer) will have gain-of-function ones (it’s generally easier to mess a protein or pathway up than it is to make it work more than it already does). If you can assign these mutations to various phenotypes in the people who carry them, or alternatively, if you can spot various interesting phenotypes and work back to particular mutations that brought them on, you have data that cannot be obtained any other way. You can generate animal genetic models all day long (and people do), but no one’s going to deliberately produce crops of human babies with selective gene alterations just to see what happens. Nature, though, has run such experiments for us.
That, by the way, includes embryonic lethal mutations (which is what you get a fair amount of the time when you do knockout gene animal experiments). A good number of miscarriages and failed attempts at pregnancy are likely due to such mutations, although these studies are still in their early stages. But looking at living adults still gives you plenty of genetic data to deal with, since (at a fine enough level) every single one of us is a mutant compared to someone else. There are, of course, a number of rare diseases that are absolutely driven by genetic mutations, often in a single glaring important unmissable gene, and these are the subject of a lot of interest these days. The rare-disease model is that if you can address these, the small (often extremely small) patient population can be offset by a very expensive drug, so long as said drug has major meaningful effects. And the chance of getting such effects is improved greatly by knowing exactly what kinds of cells to screen in, which exact patients to enroll in your clinical trials, etc. (The offsetting factor is that the drug targets suggested by many (most? all?) of these rare disease can be very difficult to attack by traditional drug discovery means, so you may increase your chances in the clinic at a cost of decreasing your chances of ever getting there).
But outside of the obvious Mendelian diseases, the great majority of such mutations are silent, or nearly so; the challenge is to pick out the meaningful ones. That is nontrivial. GWAS (genome-wide association studies) have been run in all kinds of populations, comparing patients with a known disease versus controls, in the hopes of picking out such mutations. It is a fearsome statistical challenge, with many opportunities to get things wrong, because there are very few mutations that have a large effect (and the patients with such mutations tend to be very rare). Most of the time, what you pull up is a net full of genes that seem to have some association, but none of which can be said to be causative (or protective) on its own. The hope in the early days was that we’d find some heavy-duty signals for drug discovery against things like Alzheimer’s or diabetes, but there just aren’t many such discoveries to be had. For every disease that has a real genetic smoking gun (like the prototype genetic disease, sickle-cell anemia), there are far more of them that have fuzzy multivariate changes in relative risk, at best.
But PCSK9 by itself is indeed meaningful – from a drug discovery standpoint, it’s probably the most actionable of the mutations discovered by modern sequencing. The handful of patients who are double-recessive for nonfunctional PCSK9 have extraordinarily low levels of LDL (the so-called “bad cholesterol). You may know your levels from your own blood work, but whatever they are, I’ll bet that your LDL isn’t (say) 14, which is what you’d see in a human PCSK9 knockout. These patients also have greatly reduced chances (up to 90%) of heart disease versus “normal” human controls. It’s no wonder that the industry jumped on these results as soon as they came out.
Jumping on them wasn’t all that easy, though. PCSK9 is one of those (many) targets that doesn’t have a small-molecule binding site on it that controls its function. The most straightforward way to lower its functional activity was to raise an antibody against it and use that as a therapy. These days, a significant portion of the best-selling drugs in the world are antibodies, and that number has done nothing but grow over the years. Several companies were able to come up with good PCSK9 antibody candidates, and Regeneron (joined later by Sanofi) and Amgen made it into the clinic first. Pfizer was there as well, for some time, but had to drop out of the race due to a problem that shows up in the antibody business – raising immunity to the therapeutic antibody itself. It’s rather hard to predict when and if that’s going to happen, as Pfizer’s example shows – they certainly wouldn’t have plowed ahead into such large trials if they’d known that would happen to them.
But as with any drug target, you have to ask whether or not you can hit it to clinical effect, but also what the size of the results will be once you do. It can be a bit like the old story about the guy who’d lost three cars while betting on filling inside straights in poker – twice because he hadn’t made his hand, and once because he had. The human data from PCSK9 looked pretty compelling, though, so the results of the first major outcomes trial (which Amgen reported on Friday) were eagerly awaited. And awaited not only by the medical community, but by insurance companies, because the newPCSK9 antibodies present an unusually stark contrast in costs compared to their alternative (cheap generic statins). Are they worth it? A lot of prescriptions for them are getting rejected by insurance companies, and Amgen is arguing that these rejections are more arbitrary than evidence-based.
That’s where the arguing is now, because the Amgen study, while successful, was not all that compelling. Yes, the relative risk fpr the composite cardiovascular endpoints used in the study went down, but not by as much as observers were hoping for (15% reduction versus 20 or 25%). And when you get down to overall mortality, there was no change at all, which has to be a disappointment. Amgen has been arguing that this was a relatively short study, and that the first measurements were also taken at a relatively early point in the treatment, and that the overall trend is for better numbers as the treatment goes on (which may well continue). But while these points may be valid, it’s a little rich for Amgen to be making them, because they designed this trial themselves, presumably to generate the most compelling results in the shortest amount of time. The fact that they’re having to make such arguments at all is a sign that the trial definitely did not come out the way that they’d hoped – you can be sure that the plan was not to have to say “Well, gosh, it’s really not bad if you look closely”.
So here’s a big question: did the genetic basis for the whole PCSK9 discovery and development effort lead people down the wrong path? Given the human mutation data, should we have expected better data? I’d argue, from Amgen’s trial design, that they did indeed expect more than they got. Recapitulating a true genetic knockout pharmacologically is not easy. An antibody is probably your best shot, at least under present conditions, but it still isn’t going to flatten things down like a defective gene will, and perhaps that’s what we’re seeing here. Update: and keep in mind, having a gene knocked down all through development is never going to be quite the same as hitting that same pathway in an adult.
There’s some irony here, because the effect of statins on overall mortality and morbidity seems to have more going on than just their effect on LDL. Statins have a number of other effects in humans, some beneficial and some not, and some of the beneficial ones are not well understood. So we have a classic attempt at a targeted small-molecule drug that works better than it should, probably because it’s not quite as selective as planned, versus an antibody, which is probably very selective indeed, but may not be quite delivering on its promise. That’s the drug business for you – and note that determining both of these has taken a great many years and a great huge pile of money, time, and effort.
What does this tell us, then, about genomics-based drug discovery? PCSK9 is about as compelling a story as we’re likely to see in this space, and if it has indeed come up a bit short, that’s food for thought. To be fair to Amgen, they may well be right about the continued improvements over longer-term therapy, in which case this story may have a better ending. But the slam-dunk game-winning ending is already gone. That may be the main lesson we can draw for now – here’s a terrific case, and it still didn’t blow everyone away when it finally got to multiyear human therapy. Everyone who’s following genetic-based clues to human therapy (a big crowd indeed) should keep this in mind.