Drug Repurposing: How Often Does It Work?

By Derek Lowe

Here’s an article that will not be popular among some constituencies. It’s in a special issue of the Journal of Chemical Information and Modeling, devoted to how these disciplines have responded to the coronavirus pandemic. And in it, Aled Edwards of the Structural Genomics Consortium surveys past attempts at drug repurposing and arrives at a conclusion that others in the field have as well: it rarely works the way you’d hope.

Indeed this paper says that it has been “unable to document a single instance of a drug approved for clinical use where the idea for the clinical trial derived first from a virtual or lab-based screen of old drugs“. And the other linked paper above noted that despite all the press, there have been “impressively few data on success rates” for the approach (which it goes on to try to provide). And as is always the case, success depends on how you define it. I agree with Edwards that using the stringent definition given above, that the success rate is likely zero. Unfortunately, that definition is what popular press stories about drug repurposing tend to play up.

The successes that have come along are more modest. Look at remdesivir, for example. Although its real utility in the pandemic is still being defined, I’m willing to stipulate that it’s a useful drug (although, sadly, not the “game changer” that everyone has been looking for). But trying it against the coronavirus did not require a random screen – remdesivir is expected to have some level of activity against basically every RNA virus that comes up, due to its mechanism targeting viral-RNA-dependent RNA polymerase. The same goes for the Emory/Ridgeback/Merck compound (MK-4482) that’s going into trials now – it’s shown strong activity against a whole range of RNA viruses, because it causes an “error catastrophe” in the same viral RNA replication step.

Compounds like this are great to have on the shelf, because of that broad-spectrum activity. At the same time, there’s a definite “jack of all trade, master of none” problem with broad-spectrum antiviral drugs. When you look at the field, the only time we’ve ever really been able to control or cure any viral infections with small molecules, it’s been with a cocktail of drugs that hit several mechanisms at once (HIV, HepC). Which makes perfect sense, given viral mutation rates. I don’t see how a single small molecule drug is ever going to be an effective antiviral by itself, at least not if it’s working by the mechanisms we know now. And in the case of remedesivir, or ribavirin, or AZT or any other broad-spectrum-ish antiviral you can name, the chances of a single agent hitting a knockout blow are basically zero. They’ll do some good, and they’ll do even more good if they can be combined with a drug with a completely different mechanism such as a viral protease inhibitor. (Note also that broad-spectrum viral protease inhibitors are a lot thinner on the ground, at least ones that aren’t cytotoxic at the same time!) All that is to say that calling remdesivir a “repurposed” drug isn’t quite accurate. It’s being used for its intended purpose: to mess up viral RNA replication. Next RNA virus that causes trouble, we’ll try it on that one, too.

And not just with antivirals – if we have some mechanistic understanding of a drug’s actions, it makes perfect sense to keep an eye out for similar applications that might turn out (you see this a lot in oncology, for example). Likewise, if a known drug has unexpected side effects and unusual activities when dosed in human patients, it makes perfect sense to try to figure out what’s causing these and add that information to the mechanistic understanding pile. That’s exactly what happened with thalidomide and led to its use as an anticancer drug – for that matter, intensive study of why thalidomide caused the disaster it did in human usage is what’s led to the modern field of targeted protein degradation, opening up a whole new area of medicinal chemistry and chemical biology. So one could call this repurposing as well, but it’s still very different from deliberately screening a collection of known drugs, hypothesis-free, and hoping for something interesting to happen. Because it rarely (if ever) does.

But mechanistic understanding is a sliding scale, too. All too often, we don’t understand the diseases understudy well enough to be sure about what we’re seeing. This example from today’s new paper is (sadly) very instructive:

In the early 2000s, the NIH’s Neurodegenerative Drug Screening Consortium launched what was among the first systematic repositioning initiatives.(9) Ahead of their time, they assembled approximately 1000 FDA-approved compounds, including many antibiotics, with the idea of identifying new uses for old drugs, and described the effort in a set of influential papers.(10)

The experiments were designed well; compounds were sent to many investigators blinded for testing in their various models. When the assay results were unblinded, several antibiotics, including minocycline, an antibiotic that had previously shown activity in other models of neurological diseases, and ceftriaxone, a cephalosporin antibiotic, showed great promise in a number of assays for amyotrophic lateral sclerosis (ALS). It was a brilliant concept, a beautiful experiment, and a beautiful result. . .

. . .Fast forward to 2014, when the results from the clinical testing of ceftriaxone in ALS patients were published,(11) and that story did not end well. Not only was ceftriaxone ineffective in ALS, but it actually caused serious adverse events. The clinical testing of minocycline in ALS also failed to show efficacy in patients.(12) However, this lesser known part of the story had little impact on the repositioning horse, which had long since left the stable.

Repurposing is hard because drug discovery is hard, because understanding human biology and human disease is hard. There are no shortcuts.

https://blogs.sciencemag.org/pipeline/archives/2020/09/11/drug-repurposing-how-often-does-it-work