I was speaking to a university audience the other day (over Zoom, of course) and as I often do I mentioned the studies that have looked at what kinds of reactions medicinal chemists actually use. The cliché is that we spend most of our time doing things like metal-catalyzed couplings and amide formation, and well, there’s a reason that got to be such a cliché, because there’s a lot of truth in that.
At the same time, there’s some evidence that innovative drug molecules come with innovative structures, more often than you’d expect by chance. It’s for sure that some of the hottest research areas right now (such as bifunctional protein degraders) can produce some rather off-the-beaten-track structures. So how do we reconcile these? Can we be making innovative drugs using a bunch of boring reactions?
This new paper (open access) says that yes, we sure can. The authors (from AstraZeneca) first note that about a third of all the reactions in AZ’s electronic notebooks are amide couplings, which sounds about right. They assembled two random sets of 10,000 compounds that had been made and screened in at least two assays, with one of them featuring amide formation and the other with it specifically excluded. These sets (Amide Formation and Other Reactions) were then evaluated by various techniques to roughly measure structural complexity, diversity, and novelty, and in addition the targets that they had hit in past AZ screens were examined.
And as it happens, the Amide Formation set had similar, but slightly higher complexity than the Other Reactions set. The two sets were virtually identical in lipophilicity and percent of saturated carbon atoms, but the amide group was slightly higher in molecular weight and the the number of chiral centers. As for molecular diversity, two different measurements broadly agreed: the Other Reactions set covered more diversity space, but the two sets also had significant non-overlapping regions. That is, the Amide Formation set was not just contained inside the larger diversity space carved out by the Other Reactions set, but had space all its own as well. And there was no real difference in novelty between the two sets, as measured by the number of structures that already occurred in databases such as ChEMBL. And when historical assay behavior was examined, the Amide Formation set had more active compounds in it, while the Other Reactions set covered a slightly wider range of assays themselves. But the two sets had a large overlap in the actual targets covered, so there was, in the end, not a significant difference between the two in “target space”.
The authors suggest that one reason that so simple a reaction as amide formation can hold its own (versus so many other possibilities) is that there are more and more unique amines available for such reactions. They looked through the ELNs for one-step amide couplings that made compounds for testing and examined the amines involved. On average, 8,000 different amines were used each year for such reactions, and every year about 2,000 of them were new. The authors:
In practice, building-block availability is one of the main determining factors. If the desired building blocks are unavailable, the chemist is faced with the decision whether to invest in new route development, or to make analogs with established routes, or to avoid making the target molecule at all. Given the uncertain nature of drug design, investing more time and resources in making a compound does not guarantee improved molecular quality. . .
. . .In medicinal chemistry, we have now reached a state where millions of building blocks have previously been engineered and can now be used in molecular design and synthesis. In addition to the increase in the number of new amines, boronic acids have been another fast-expanding reagent class since the introduction of the Suzuki coupling method
That really has been a change over my career. There are just so many more neat little functionalized compounds available now; it’s become an entire business of its own. As the paper notes, you even have setups such as Enamine’s REAL compound set, which is a virtual-but-easily-made collection via mixing and matching their available building blocks. That one would come out to well over a billion compounds if someone placed an order for the whole collection.
And if we can get our work done via such easy reactions – plenty of experience in doing the reactions, relatively easy purifications, existing scaleup expertise, and so on – then why shouldn’t we? (I should note that the paper under discussion has a lot of good references to past arguments about this issue). That gets to another point I was emphasizing to my university audience: medicinal chemistry is a means to an end. The end, of course, is the discovery of useful drug molecules, and if the synthetic chemistry can (as much as possible) get out of the way of all the other tricky steps in that process, then so much the better.
That’s not to say that we shouldn’t try new reactions or new technologies. Among other things, these can lead to even more new building blocks that can feed into the easy reactions themselves. And God knows, as you develop the SAR of a compound series you may find yourself unavoidably being pushed into difficult chemistry, where you will need all the help you can get and throwing amide couplings and Suzukis at the problem will avail you not. No, we definitely need our skills and our imaginations – but we need them for the times we need them, and when we don’t need them we should speed drug discovery along with the best tools we have for it. To paraphrase Einstein about physical theories, a synthetic route should be as simple as possible, but not any simpler. Getting as much done as you can with the easy methods leaves you more time to tackle the hard stuff. Get flashy only when you have to.