I’m glad to see this paper – it’s a look from Pfizer about how they were able to scale up nirmatrelvir (the antiviral component of the Paxlovid combination) so quickly. The authors are surely correct when they say that this 17-month timeline is a record for the drug industry, and it’s very much worth seeing how this all happened.
I’ve already written about one of the big reasons the compound emerged at that speed – the fact that Pfizer had already been doing research in these types of protease inhibitors and had a lot of very targeted chemical matter ready to test. The paper doesn’t address this directly, but you can see this effect in the part of the paper where they go into the synthesis of the three key chemical building blocks (previous post on the subject here). The synthesis of the dimethylcyclopropyl proline core of Fragment 6 was first reported in 1999, as part of a study on proline structure in proteins. Building off of that, Fragment 6 itself was first reported in 2002, and was part of Merck’s boceprevir, a hepatitis C protease inhibitor. So this one had already had serious process chemistry work done on it and had proven itself as part of the structure of another antiviral protease drug. The Pfizer team used the exact synthesis that Merck had used years before (an enzymatic route starting from ethyl chrysanthemate, the “alpha raw” in this route, developed with their partners at Codexis). Boceprevir itself is no longer on the market, but the paper notes that the existence of this industrial route allowed them to restart that supply chain quickly. They also lined up an alternate route via hydroxyproline in case the supplies of ethyl chrysanthemate hit a snag.
The second key intermediate, Fragment (11) is just the trifluoroacetamide of tert-leucine. There is no shortage of reagents to do that (ethyl trifluoroacetate, for example), and t-leucine has been an article of commerce for a long time and is available on large scale. So this was the least troublesome of the bunch.
Then you have Fragment 3, and that one is yet another piece that has shown up before in protease inhibitor research (ruprintrivir and lufotrelvir, and a potential protease inhibitor drug for SARS-CoV-1 in 2003. All of those were/are Pfizer research programs, and the synthesis was first reported over twenty years ago and has had a great deal of process improvement done on it since then – for example, cleaning up the route so no chromatography was needed at a key step. The lufotrelvir program had already prepared significant quantities of the key starting material for this fragment, and two different routes were used to convert it to 3.
You can see how things came together, then, and how all of this built on years (decades) of antiviral protease research, both in academia and (crucially) on an industrial chemistry scale. It should go without saying that you cannot produce a new antiviral in 17 months without such foundations. And even so, this was a major challenge. The paper points out that the supply chains for these three fragments end up using over seventy raw materials and reagents – everything from stuff basic commodities like sodium chloride, zinc metal, and potassium phosphate, common solvents like acetonitrile, acetone, and methanol, up through industrial-scale reagents like Boc anhydride, dimethyl sulfate, toluenesulfonic acid, bromoacetonitrile, and pivaloyl chloride.
And then you hit the ones whose supplies were limited, like 2,2-dibromopropane (for making that cyclopropyl group in Fragment 1) and methyl cyclopentyl ether. Even lithium aluminum hydride became a worry at these scale and under the disrupted supply chain conditions, and good ol’ LAH is one that organic chemists are generally used to being available at all times and in more or less any amount you might need. The paper notes that the usual procedure for a new product is to have two suppliers ready for each starting material at the time you file with the FDA (and maybe a third for contingency, in some cases). But for nirmatrelvir, there were five to seven suppliers for each of the starting materials, companies from all over the world, and I can only imagine what a fiesta it must have been managing that sort of situation under time and materials pressure.
Outside of the chemical synthesis (which is the focus of this paper) another huge time savings was to run as much of the preclinical and clinical work in parallel as possible, on an at-risk basis. Preclinical studies on formulation, stability, and toxicology were going at the same time as process development. Past that, work on the Phase II trial was already well underway before Phase I had completed, and the large-scale production for eventual product launch was running while the Phase II and III trials were still going. You do indeed save time that way – the “normal” route is to finish each of these processes off before moving completely on to the next phase. After all, there’s no need to gear up for clinical trials if your compound fails tox, is there? Likewise, you’d usually want to make sure that your compound has made it through Phase I (first-in-human dosing and safety, blood levels and pharmacokinetics) before starting the efficacy trial, because those Phase II trials are significantly more expensive. Keep in mind as well that infectious disease trials are among the fastest-moving of any therapeutic area, since the diseases themselves move quickly. A 17-month timeline for an Alzheimer’s drug (for example) is physically impossible no matter how many whips you crack.
And it’s important to realize this: if every drug were developed on this compressed, telescoped crash basis, the resulting prices of the ones that made it through would make your eyes water – and yes, I mean compared to the prices of what we have now. Pfizer had, as mentioned, a very solid foundation for nirmatrelvir, but it most certainly could have failed or been delayed at several points along the way, which would have bonfired a lot of that at-risk development money. This was a sprint across a tightrope, and it worked. But we can’t run the industry on that basis.
