The Jamison group at MIT have a big paper out in Science (commentary here) on their drug-synthesis-in-flow project, which I last wrote about here. It’s impressive – they show on-demand production of USP-standard diphenhydramine hydrochloride (a classic antihistamine), lidocaine hydrochloride (the topical anaesthetic), diazepam (known to the world under its brand name of Valium), and fluoxetine hydrochloride (famous as Prozac).
These are ingenious, because they set themselves the task of doing them in an apparatus of fixed (and rather compact) size. It’s modular and reconfigurable for each synthesis, since they each have somewhat different requirements. DARPA, who put a lot of funding into this work, initially had even wilder ideas about the size – I’ve heard Jamison speak on this topic, and I believe that their initial call was for a hand-held device, which is not going to happen with present-day technology, or next year’s, either, for that matter (see below!) But just getting things down to this level is a real achievement, because there are, as you can well imagine, a lot of issues that needed working out.
The machine is roughly divided into a front side and a back side: the former synthesizes the compound, and the latter purifies and crystallizes it. The syntheses take advantage of flow chemistry conditions (high temperatures for short times, high pressure and concentration, etc.), with some unusual features. For example, the first step in the diphenhydramine synthesis is a high-temperature neat reaction of dimethylaminoethanol and chlorodiphenylmethane (edit – fixed the name!), 180C at high pressure for 15 minutes. The hydrochloride product is a liquid under those conditions as well (it melts at 168), so it was pumped over neat into the next stage. Here’s the rest, and it should give you a good feel for how this system handles things:
The molten salt was then treated with a stream of preheated (140°C) aqueous NaOH (7). An inline purification and extraction process employing a packed-bed column to increase mass transfer, a gravity-operated liquid-liquid separator with automatic level control (fig. S1), and an activated charcoal filter to remove the colored impurities produced the diphenhydramine API as a solution in hexanes in 82% yield.
In the downstream section, the API was precipitated with HCl (10), and the resulting salt was filtered, washed, and dried in a specially constructed device with a Hastelloy filtration membrane (fig. S6) (27). After redissolving in isopropyl alcohol at 60°C, the diphenhydramine hydrochloride (1) was recrystallized in a crystallizer, while being cooled to 5°C. Upon filtering and drying, the crystals were dissolved in water. Real-time monitoring using an ultrasonic probe yielded the final dosage concentration (5 ml at 2.5 mg/ml). High-performance liquid chromatography analysis determined that the purity of the product conformed to USP standards (fig. S12) (28). Overall, the system capacity based on the optimal yield observed in each step was 4500 doses per day.
All of these steps, as will be clear to anyone with lab experience, needed a good deal of work. As someone who’s done a reasonable number of flow reactions, I would not even want to think about the number of times they must have clogged up this apparatus at one point or another while working out these four drug syntheses, and I suspect that there may have been one or two leaks from time to time. Just a guess. There are a lot of interesting mechanical innovations in this process for liquid separation and the like, and the paper is well worth a close read for anyone with an interest in this area just for these alone.
Fluoxetine is the most involved of the four – the cycle time for producing lidocaine hydrochloride in a final dosage form is about twelve hours, while fluoxetine is closer to 48. (DARPA’s next requirement might be for a machine where someone hits the start button and goes away for the whole 48 hours, because I’m willing to bet that’s not how they did this one!) Most of that time is in the downstream processing. Several of the reaction steps themselves are done in something like five minutes, but precipitation, purification, dissolution and crystallization are harder to force, as many in the audience will know from long experience. The group also spent some time configuring the system for easier switchover to the other synthetic routes, which means standard sizes for the modules, procedures for flushing and cleaning the lines, and so on. The most difficult switchover can be done in about two hours.
Experienced pharma folks will already have a host of questions. Several of these can be answered by noting that each of these processes were designed to produce solutions, for either oral or topical dosing (all approved forms for these drugs). As the paper says:
The current system focused on liquid oral and topical dosage formulations commensurate with the on-demand approach. A complete alternative platform to current batch manufacturing would inevitably have to produce pharmaceuticals in the common dosage forms of tablets and capsules as well as sterile injectable solutions, which would require advances in downstream processing.
That it would. I suspect that most of the press coverage this paper gets will gloss over this issue, giving people a mental image of a machine that spits out pills, but that will be a significantly harder project. Solid dosage forms are, of course, a big part of the drug industry, but they come with a whole list of their own issues (polymorphs, excipients, stability, coatings and more). Companies are certainly interested in streamlining these processes and making manufacturing in general more responsive (and less inventory- and capital-intensive), but it ain’t easy, especially when every batch needs to meet some pretty tight specs. And each batch sure does – the FDA does not take a live-and-let-live attitude toward manufacturing violations, as well they shouldn’t, and they hand out some pretty eye-watering fines when things go wrong, as well they should.
From an engineering viewpoint, I was very happy to read this paper. There’s a lot of ingenious work here, and it’s really exemplary stuff. Outside of the technical aspects, though, I still have some of the same questions that occurred to me the first time I saw this work presented. Let’s start with the rationales given in the paper. Here’s one:
. . .production of a finished dosage form can require up to a total of 12 months, with large inventories of intermediates at several stages. This enormous space-time demand is one of a myriad of reasons that has led to increased interest in continuous manufacturing of APIs and drug products, as well as in the development of integrated processes that would manufacture the drug product from raw materials in a single end-to-end process.
This, to me, is the strongest case for this work. It’s absolutely true that continuous manufacturing has gotten a lot of interest (and investment), for just the reasons stated. Flow chemistry, in fact, was a well-known technique in the manufacturing side of the industry long before it ever became popular back on the benchtop, for just these reasons, but the kinds of reactions that Jamison’s group describes are still really advancing the state of the art. One worry is that some of factors that were very important in the DARPA conditions (such as portability) may be some of the ones that are least important in the commercial manufacturing side of things, but I’m certainly willing to believe that useful things will still come out of meeting those strictures. A second rationale, though, is given right after this one:
Another major challenge facing the pharmaceutical industry is drug shortages; the U.S. Food and Drug Administration (FDA) has reported well over 200 cases per year during 2011–2014 (6). The root causes of these shortages often trace back to factors reflective of the limitations of batchwise manufacturing, such as variations in quality control and supply chain interruption. Moreover, the small number of suppliers for any particular medicine further exacerbates the challenges faced by batchwise manufacturing to respond to sudden changes in demand or need, such as in epidemic or pandemic instances of influenza outbreak.
Actually, to the best of my knowledge, the root causes of these shortages often trace back to regulatory and legal issues. Many (most) of the drug shortages reported are for older generic compounds with few manufacturers, and the issue really isn’t trouble in making the drugs. Those routes are cheap and well worked out. The issue is who gets to make them and who gets to sell them, and those are completely regulatory problems.
It’s in the commentary to this article that things, to my mind, stop making so much sense. Here’s how it starts out:
Imagine it is the middle of the night. A severe snow storm has hit the region, and your 3-year-old’s fever is rising. You suspect a serious infection and cannot wait until the next morning to go to the pharmacy, yet the roads are impassable. No problem—you were recently granted access to a prototype machine no larger than a kitchen microwave that allows the user to synthesize their pharmaceutical of choice. You start up a smartphone app that has access to the family medical history and other relevant parameters, such as allergies and body weight. A few minutes later, you have consulted an emergency pediatrician through the app, inserted the relevant capsule, and pressed the start button. After a short wait, a single, personalized dose of the necessary antibiotic is ready for use. All this may sound like science fiction, but the report by Adamo et al. on page 61 of this issue (1) proves that this scenario might become reality in the not-too-distant future.
No. I’m not seeing it, and I think that I can clearly explain why. Let’s look at the list of starting materials and solvents you need for the four drugs described in this paper: 2-dimethylaminoethanol, chlorodiphenylmethane, aqueous sodium hydroxide, isopropanol, hydrogen chloride in diethyl ether, hexane, N-methylpyrrolidone, potassium hydroxide, diethylamine, chloracetyl chloride, 2,6-xylidene, sodium chloride, ammonium chloride, acetone, bromoacetyl chloride, 5-chloro-2-methylaminobenzophenone, ammonia in methanol, ethyl acetate, 4M aqueous HCl, ammonium hydroxide, DMSO, ethanol, 3-chloropropiophenone, toluene, diisobutylaluminum hydride, aqueous methylamine, THF, 4-fluorobenzotrifluoride, potassium t-butoxide, 18-crown-6, and t-butyl methyl ether. I’m not listing all those to be annoying, or to say that there are too many of them – no, that’s exactly the sort of collection that I or any other medicinal chemist would use while making them in our fume hoods, and probably more besides. This is all rock-solid organic synthesis, and (as with many rock-solid routes) the reaction schemes should be perfectly comprehensible to anyone in their second semester of sophomore organic chemistry.
But you’re not going to keep all this in your house. And you’re not going to pack all this into your box-sized flow synthesis machine, either. Take a look at that list – it’s been carefully optimized, as well it should be, and that means that there are several closely-equivalent reagents that still can’t quite be substituted for each other. Moreover, these things all come from outside the unit, and are loaded into it (and offloaded) by the people standing around it. (There are a few other things that are as well, such as seed crystals, but that’s another issue). The commentary completely passes over this issue, which becomes wildly more complex as you start to picture more versatile drug-o-matic machines of the sort it invokes. Just how many reagents is a microwave-sized device going to carry inside it if it’s ready to up and synthesize your “pharmaceutical of choice” at 3 AM in a snowstorm? How long are some of them going to be sitting around before they’re ever used? Do they ever evaporate, precipitate, clog up or decompose? Apparently not. No, the opening paragraph of that commentary is a fantasy; I don’t see any other way to describe it. That’s particularly true if you’re making an antibiotic from scratch, since most of those structures are a lot more complex than the compounds in the original paper, which (as you can see by reading it) are certainly enough of a challenge as they stand. If you need vancomycin for a resistant bacterial infection from a machine at 3 AM, you are hosed – heck, if you need plain old ampicillin for an earache at 3 AM, you’re hosed, too, because we don’t make that one by organic synthesis, anyway. I don’t think this is a lack of vision on my part; that microwave-sized box idea can’t stand up to simple engineering constraints, and it’s important to note that the authors of the original paper are proposing nothing of the kind.
The rest of the commentary is reasonable, but then you get to the last paragraph:
The work by Adamo et al. offers tantalizing prospects of on-demand drug delivery in emergency and specialized cases, such as epidemics of rare diseases in refugee camps or after natural disasters. Issues remain with regard to accessibility of cartridge refills, waste disposal, and how to avoid the production of illicit substances. However, the major advance in this work lies not in the delivery of a desktop drug synthesis machine, but in demonstrating the great potential of flow chemistry in drug production.
And we’re back to the same problems as before, and they’re a lot bigger than cartridge refills and waste disposal. If you’re trying to help people in a refugee camp, you would surely be a lot better off sending them a helicopter full of the actual drug then sending them a drug synthesis machine, several crates of reagents, and some technicians – and a 15 kW diesel generator, because that’s the power consumption of the current device. You’re almost certainly going to get the drug faster by having an existing plant make it somewhere else and flying it in, as you would with any other high-value-added material, in the same way that you’re going to send them truckloads of gasoline, too, not build them a refinery.
The counterargument is “What if there’s not enough of that drug to start with?”, but that supposes that someone has done the (labor-intensive) work of optimizing the synthesis for the machine. The modules in it are plug-and-play, but the chemistry isn’t, and that’s what I think that people might not understand. That’s why there’s chloracetyl chloride in one route, and bromoacetyl chloride in another, and why the chloracetyl chloride reacts with a 1.43 M solution of the xylidene in a specific solvent, and so on. These syntheses have been intensively worked on by a very capable team of chemists and engineers, and that’s what every production flow protocol needs. That, in fact, is why it’s been more a manufacturing technique than a benchtop one: it makes sense to do this kind of work for a specific product that’s going to be in production for years, but it does not makes sense to work something out like this for every drug in the pharamacopeia. (That’s also why the Burke synthesis machine is interesting, because it does the same carbon-carbon bond forming reaction, the same way, every time, and makes the most out of that process – more here and here – but it will not make you any ampicillin, either).
I don’t want to give the impression that I didn’t like this paper, and the work behind it. I liked it very much. I do think, though, that there’s a lot of room for people who don’t know organic chemistry, or the drug business, or economics, for that matter, to get some odd and unworkable ideas about its implications.