The virtues and shortcomings of total synthesis are both on display in this new paper from the Baran group at Scripps, “Total Synthesis of Maoecrystal V”. Actually, they’re on display in not only that paper, but in it and in the four previous total syntheses of the molecule, but we’ll get to that.
Maoecrystal is definitely an appealing molecule for academic synthesis, in that it’s a collection of tightly bunched ring systems and dense functionality. It was also reported to show very good activity in cancer cell lines, which for decades now has been enough to get things underway. The existing routes to it (reported starting in 2010) all go through a Diels-Alder reaction, as the new paper shows, but Baran and co-workers decided to try to follow something more like the proposed biosynthesis. This led to an eleven-step synthesis (discussion here from one of the co-authors) where the known routes are between 18 and 35 steps, although there are some telescoped two-for-one-pot transformations in there, so if you’re picky you can argue about that figure. (I regard those no-workup transformations as a desirable feature, no matter how you count them, and any process chemist will tell you the same).
An interesting feature of the synthesis is that many of the intermediates have small-molecule X-ray structures. That’s a good idea, since the ring systems involved are not trivial to establish by NMR and other methods. Nothing will show that your 2.2.2 bicyclic system now has third and fourth rings attached to it in the right configurations like good solid X-ray data, though. The other reason that they were able to get so many structures has to be that this synthesis actually produced 80mg of the synthetic natural product, which by the standards of the field is like scooping flour out of a fifty-pound sack.
When you look at the synthesis in detail, though, you see the price of that good a route. Some of these steps took a horrific amount of optimization. Let’s just use the first step as an example in detail:
The synthesis of 1 commenced (Scheme 1) with a highly enantioselective conjugate addition of an allyl silane to cyclohexenone to deliver 7 in 80% isolated yield (99% ee). Among the many ligands explored, the TADDOL-derived phosphine-phosphite L1 designed by Schmalz was singularly successful. The use of CuI·0.75DMS was also critical to minimize dimerization of the Grignard reagent. A profound solvent effect was also observed with a mixture of PhMe/MeTHF being essential to obtain consistently high yield and enantioselectivity on 20 g scale.
Right. You don’t even have to be an organic chemist to notice that none of these conditions would have been the first things that the group tried, or the fourth, or the ninth. And note that since it’s the first step in the synthesis, you’re going to be trying these things out pretty quickly on 20-gram scales, since showing that a new combination works on 50 milligrams will be of limited interest. The next step, an oxidative alpha-acetoxy formation, also needed a lot of beating on according to the paper, and for the next step (a Sakurai reaction) it’s noted that over 50 Lewis acids were screened. And that takes us up to the hard parts of the synthesis – no, that’s exactly the case, because all the ring-forming stuff is yet to come. One of those steps includes a description that we’re going to be hearing about for years to come:
Roughly 1000 experiments were conducted changing every conceivable variable from the base used to deprotonate, the solvent employed, additives, and the electrophile. Emerging from this exhaustive study was the remarkable finding that the addition of LaCl3·2LiCl to the extended sodium enolate of 3, followed by quenching with freshly prepared formaldehyde gas led to the desired adduct 11 in 84% yield as a 2:1 diastereomeric mixture favoring 11 (3 g scale).
Exhaustive is right. That’s a ferocious amount of work, chewing through piles of advanced intermediates which are in themselves no fun to prepare. Process chemists in industry have run synthetic variations on this scale (or approaching it) when the situation warrants, but the key to successful industrial work is to avoid putting yourself into such situations at all. The molecules in those cases are less complex, so there are generally more robust routes that can be found without having to go this far. In natural products synthesis, though, you have little choice: every single step is like juggling blown-glass sculptures on a tightrope.
The next step, a stereoselective reduction, needed only about a hundred variations to be optimized, and again the conditions settle down onto a wildly picky protocol that has never been used before in the literature: zinc triflate, which has to go in first, followed by lithium borohydride. “Zinc borohydride”, you think, but you’re wrong: that reagent itself does squat when you try it as such. And so on. Skipping forward a bit, the last step of the synthesis is particularly audacious, with what are formally about seven reactions all happening in the same flask in sequence. And there you have maoecrystal V!
But what do you have once you have it? That’s the kicker at the end of the paper. With a really useful amount of the compound in hand at last, and all the spectral data (including X-ray) to show that they really did have the exact compound, the group had it run across a large panel of cancer cell lines. And it showed no activity whatsoever. Well, now. This means, almost certainly, that the original reports of activity are incorrect: something was active in their samples, most likely, but it wasn’t maoecrystal V. Either that, or the assay was totally blown.
Now, that (from one viewpoint) this is a great advantage of total synthesis: here we have the definitive compound and producing has provided definitive information. It’s tempting to say, though, that this information could have been acquired earlier had the original work been done with more care. That’s partly unfair, though, because the original authors presumably believed that the compound was clean enough to evaluate and the assays were working and were presumably competent enough to make those calls. I’m sure that the original data (60 nM on HeLa cells!) will be revisited in light of these new findings. And this certainly isn’t the first time that a natural product, reported as biologically active, has been come up short on resynthesis.
On that other hand, though. . .maoecrystal V has been synthesized by four different research teams before this one, and we’re only now finding out that it’s not active? Here’s the Thomson group at Northwestern in 2014 (discussion here), and here are the Davies (Emory) and Zakarian (UCSB) groups in 2014, here’s the Danishevsky synthesis (with one co-author, Feng Peng!) in 2012, and here’s the first synthesis (the Yang group in Beijing) from 2010 (discussion here), with another one in 2015. In every case, there appears not to have been enough of the final product produced to actually try repeating any cell assays. If one of the points of total synthesis is to produce bioactive compounds, then something has gone awry here if the bioactivity never gets a look in over all these years and after all this work.
Make no mistake, the Baran route is an excellent piece of organic synthesis. The paper states, though, that “The primary goal of this study was to demonstrate that the notorious difficulties surrounding the exotic architecture of 1 could be solved by total synthesis in a practical way.” Fair enough, but I strongly suspect that this structure would not have set off the amount of interest it did in the synthesis community without that cancer-cell bioassay data attached to it. Those difficulties only became notorious because so many people have worked on this molecule, and its (putative) biological activity had plenty to do with that. There’s also room to raise an eyebrow at that word “practical”. It’s true that this route has delivered more maoecrystal V than anything else in the world, and has thus produced truly practical amounts of the compound. That’s a good thing. But as a look at the synthesis itself shows, it took a brutal amount of effort to get things to this state: it seems a bit odd to refer to a practical synthesis that had a single step (out of many hard ones) that needed a thousand tries to figure out the right conditions. It ends up practical, but if every chemotherapy drug had to be synthesized in this fashion, we’d be in a hell of a mess, not least because very few research groups in the world are capable of doing organic synthesis at this level at all.
There’s also the traditional (stated) goal of total synthesis of providing analogs of such natural products that can’t be obtained otherwise. That certainly does happen sometimes, but not as often as it should. None of the existing syntheses of maoecrystal V, as far as I can see, have provided any. This latest one, which can indeed deliver the original compound in quantity, illustrates the problem. A synthesis like this is (obviously) an extremely finely tuned machine. Throw in an extra methyl group somewhere, pretty much anywhere, and that machine will start spraying oil and throwing piston rods into the ceiling. Get ready for another thousand optimization runs if you change much of anything, is my bet, and that goes for pretty much any bespoke synthesis of this kind. If you want to make analogs, you have design the synthesis from the start to make analogs, and not every sort of molecule is going to be amenable to that. This one certainly isn’t – it’s not a modular synthesis where you make Part A over here and Part B over here and bring them together. It’s all in one place, with the spotlights on every single step, and I don’t see how you’re going to get around that.
So I congratulate the Baran group on a terrific piece of organic synthesis. And I offer my condolences to the Baran group for a terrific piece of organic synthesis. How do we keep those both from being appropriate at the same time?