Here’s some pure process chemistry from the group at Amgen, looking at the synthesis of their well-known KRAS (G12C) drug, sotorasib. That one of course targets the mutant Cys residue in that variation of the protein, but that covalent warhead isn’t the subject of this paper. Instead, it’s looking at what would seem to be a much easier part of the structure, a Suzuki-Miyaura coupling that adds a fluorophenol group onto a pyridine ring.
It’s not like the Amgen group couldn’t get this reaction to work. They mention that the current manufacturing process takes five steps and goes in 65% overall yield. That’s an average of about 92% yield for each step (those add up quickly!), so that’s really respectable. But rare is the chemical process that can’t be improved. In this case, they had a 2-chloropyridine in the starting material, commercially available boronic acid (actually the cyclic boroxine), and a commercially available Pd catalyst (Pd(DPEphos)Cl2). They used 0.6% catalyst loading in 2-methylTHF/water at 70C, with potassium acetate as base.
There’s nothing obviously wrong or weird about those conditions at all, but they needed some improvement. The catalyst had to be added once things had come up to the reaction temperature, for one, and an even bigger problem was the the boroxine coupling partner had to be added slowly. If it went in too quickly, you would get protodeborylation as a side reaction and made plain ol’ 3-fluorophenol instead, with unreacted chloride starting material standing around watching it happen. In this case, the reaction could be rescued with an addition of more catalyst and more boroxine, but that’s not ideal (not least because of the purification needed afterwards).
The team used in situ infrared to watch the aryl chloride going away (and to see the undesired deborylated side product forming), and did this under several different conditions to try to understand what the problems were. It turned out that the product formation was zero-order with respect to the starting materials, zero-order with respect to everything except the Pd catalyst (that is, its rate of formation didn’t depend on the concentration of anything except the catalyst). But the protodeborylation was first-order with respect to the concentration of the boroxine and the catalyst, which is why it had to be added slowly. If you heaved it all in at the start, you maximized the yield of the useless side product and minimized the yield of the desired drug substance.
But heaving it all in at once is what the Amgen team would much rather have been able to do, of course, and it would be better to suppress that protodeborylation pathway completely. The team found that the catalyst itself was being changed to a nonproductive species that was only making the protodeboylation side product (which is why just adding more boroxine by itself wasn’t enough to push the reaction forward – you needed to add fresh catalyst, too). Trying catalyst species with one or both P atoms oxidized strongly suggested that it was the monoxide form that was the active species for making the desired product, but was not involved in the protodeborylation. That was apparently coming from formation of the bis-oxide species, which then decomposed to Pd black with formation of side product.
Now that’s good to know, but another observation from the mechanistic experiments pointed toward a way out of this problem. Just switching the base from potassium acetate to potassium carbonate caused the product formation to become first-order with respect to the boroxine, which indicates a complete change in the rate-determining step. Instead of the final reductive elimination step of the catalytic cycle being the key, going to “K-carb” made the initial transmetalation the important one, and this was accompanied by a big increase in selectivity for the product over the protodeborylation. That side product barely even formed until the starting pyridyl chloride had been consumed and there was only some excess boroxine sitting around with nowhere else to go.
Boron NMR provided the explanation for this good news. It turns out that with potassium acetate as base, all the boron species were in the organic phase of the biphasic reaction mixture (mostly as the monomeric boronic acid). But with potassium carbonate, the boron was almost entirely in the aqueous phase instead, as the “ate” four-coordinate form with three OH groups on the boron atom of the fluorophenol species. But the catalyst (which again was apparently active in its monoxide form) is almost entirely in the organic phase, and having the boron reagent sitting around with it under the KOAc conditions is an invitation to trouble. Fortunately, mass-transfer between the layers wasn’t a limitation – as the boron reagent was consumed and converted to the desired product in the organic layer, fresh reagent was able to cross over and take its place without slowing the process.
The team was then able to successfully model the reaction and predict how the system would respond to various changes. The catalyst loading could be decreased, and the boroxine could be added all at once at the beginning of the reaction before things heated up. Once things got up to 70C, added the now-smaller charge of Pd catalyst got things going, with a 99% yield of desired product and no side product at a 200g scale. That’s more like it.
This kind of detailed understanding of the mechanism (or mechanisms!) of your reaction is just what is needed to really troubleshoot a reaction, and in a drug company it’s really only the process chemists who do this sort of thing. It takes a lot of time, thought, and effort, and it’s the sort of thing you’ll only devote to high-value synthetic steps that are being run on large scale. Back in the med-chem labs where I work, we’re generally just looking for “good enough”, some reaction that can produce enough of our compound ideas to test. Most of those compounds don’t work so well in the assays, to put it mildly, so there’s no need to optimize their synthesis beautifully. In fact, there’s less than no need: doing that compulsively will put you to fixing things that don’t need to be fixed and cut into your ability to explore more chemical territory. But if you’re a “Anything worth doing is worth doing well” type, process chemistry is your home. Because in that part of the business, it’s absolutely true.
Addendum: it’s true in med-chem as well, of course. But the disconnect comes when someone gets confused about what it is that they’re doing! In a discovery med-chem lab, your job is actually not to do elegant, well-optimized chemistry. Your job is to make new compounds and explore the landscape of how they interact with your target (and with the other parameters you need to fix – pharmacokinetics, toxicity, and more). That’s best accomplished by making those compounds with the least amount of effort under the most broadly applicable conditions, giving you a chance to make a wide variety of interesting things without having to treat each of them as an individual artisinal project. That is what we call good chemistry in a discovery effort. I have over the years worked with a few people who had difficulties adapting to this mindset at first, if you’re wondering!
