Sigman and AbbVie advance B.H. amination by analyzing bulky ligands

Citation:

Ickes, A. R.; Liles, J. P.; Borlinghaus, N.; Henle, J.; Swiatowiec, R.; Kaushik, N. P.; Braje, W. M.; Harper, K. C.; Shekhar, S.; Sigman, M. S. Leveraging Data Science to Elucidate Ligand Features for Pd-Catalyzed Enantioretentive N-Arylations of cyclic a-Substituted Amines in Aqueous Media. J. Am. Chem. Soc. ASAP. https://pubs.acs.org/doi/10.1021/jacs.5c07224

Summary Figure:

 

Background:

To my best approximation, Buchwald-Hartwig aminations (or N-Arylation with Ar-X like Ullmann couplings) make up about 6% of all reactions medicinal chemists run (see The Medicinal Chemist’s Toolbox: An Analysis of Reactions Used in the Pursuit of Drug Candidates | Journal of Medicinal Chemistry). That's an enormous percentage, given that the only more common classes are condensation, alkylation, and other Pd-catalyzed cross-coupling reactions are more common:

 

From Cooper, T. W. J.; Campbell, I. B.; Macdonald, S. J. F., Angew. Chem. Int. Ed. 2010, 49, 8082-8091. https://doi.org/10.1002/anie.201002238

 However, Sigman and coworkers point out that it is currently difficult to do these reactions with sterically hindered amines, such as their base case of 2-phenylpyrrolidine.

  (R)-2-phenylpyrrolidine

Another challenge with this substrate is the potential for beta-hydride elimination at the stereocenter to form an imine, which scrambles the stereocenter. The goal is to (A) optimize yield and (B) optimize enantioselectivity. The ideal catalyst is able to (A) do reductive elimination despite the steric hindrance and (B) avoid beta-hydride elimination. 

How it works:

The team set up a high-throughput screen to evaluate different ligand classes, additives, and solvents to find an initial hit. After no organic solvents provided conditions considered good enough, they hypothesized that water would be a good solvent explicitly because it would be able to coordinate to the palladium and block off coordination sites for beta-hydride elimination. This is a really interesting theory, and I appreciate how they specified this possibility. Obviously, using water as a solvent has other benefits as the most green solvent, and they used micelles to actually enable the reaction to proceed. Micellar media for palladium catalysis has been around for a while, and it's great to see it used like this. 

And all of a sudden, after a few hundred reactions in the HTE screen, several ligands popped out with high yields and very high enantiospecificity. TrixiePhos was the best of the "common" ligands, while a more specialized "NiniPhos" ended up being the best ligand overall. 

 

I'll note that these fall into the general class of "very bulky dialkylarylphosphine" ligands, and it's definitely good to include at least one in any screen for a palladium or nickel cross-coupling reaction. As described the the authors, the ligands were "sterically hindered about phosphorus" and also "displayed low vmin (the minimum electrostatic potential on phosphorus) values (i.e., strong sigma-donor P-ligands)". 

However, while this is a general descriptor of the ligands that were effective, there were many ligands that fit this description that weren't effective. 

This is where the machine learning comes in. Basically, they split the known ligand set into two categories, one for testing and one for validation, then tuned a mathematical model with known parameters until they found a model that could predict the yield for a given ligand. Then, they took the validation set and compared the prediction to make sure the data wasn't overfitted. Once the model was considered good enough (able to predict yield as output for a given ligand input), the authors analyzed what factors the model was taking into account and used it as reasoning. 

Interestingly, the factors that the model considered important were predominantly electronics, not sterics. Despite the tert-butylphosphines, the most successful class, being extremely sterically bulky, the more important point is that they make extremely electron rich phosphines. The most important descriptor was the NBO lone pair occupancy at phosphorus, so how electron rich and nucleophilic that phosphorus atom is. 

One steric parameter found to be important was Sterimol Lmin. The technical definition they provide is "the ligand length along the axis in the M-P direction defined by a dummy metal atom (M) bound datively to phosphorus". In practice, one of the key factors for reactivity is a lack of steric bulk on the opposite side of the ligand. Steric bulk near the phosphorus is beneficial, but steric bulk away from the metal center prohibits favorable rotation and maneuvering.

 

The authors do note that these observations are only true for this system in aqueous solvent. The best ligands for that system (TrixiePhos, NiniPhos) only give ~40% desired product in ethereal solvents, which makes them not the optimum ligands. 

Initial Questions and Key Findings:

1. Can an improved catalyst be found for the traditionally difficult coupling reaction between an aryl halide and a secondary amine with an enantioenriched branching chain?

A. Yes, through an HTE screen, an optimum catalyst/additive/solvent combination was found that provides the cross-coupled product in high yield while retaining high stereochemical fidelity. 

2. Can a justification based on human-identifiable chemical principles be found to explain the performance of this catalyst?

A. Yes, the optimum catalysts for this reaction (a) have very electron rich phosphorus centers (b) have high steric bulk near the phosphorus center and (c) have lower steric bulk away from the phosphorus to enable rotation. 

3. Can these observations be applied to other systems?

A. Not yet. The optimum catalyst for the aqueous system is not the same with organic solvents, meaning these findings are not universal. 

 Takeaways:

First, the power of HTE to quickly find the optimum catalyst for a reaction is incredible. I doubt that screening >100 ligands took longer than a week, when to do so via rational design would take quite a while. It also found that the best ligand is counter intuitive. I think Sigman's role in this project is in the post-hoc justification of the best ligand. I think this work is incredibly important to help push the field forward, and I'm glad they did it. While the specific finding about the need for less steric hindrance and greater rotation is not necessarily universal, if more reactions have this feature, it could begin to bias ligand designs and screening in a positive direction.