Paul Helquist , University of Notre Dame
Enantioselective methods of synthesis have become very important in the industrial manufacturing of chiral organic compounds. Most prominent among the users of these methods is the pharmaceutical industry due to the fact that the two enantiomers of a chiral drug molecule can elicit widely different physiological responses. In many cases, one enantiomer has a desired therapeutic effect whereas the other enantiomer is inactive or, in the worst scenario, toxic. A case in point is the very commonly used anti-inflammatory drug, naproxen (AleveTM). Therefore, the production of one specific enantiomer becomes crucial for the pharmaceutical manufacturer. Among the most commonly used production methods are some of the very simplest organic reactions, among which metal-catalyzed hydrogenations of alkenes are prime examples. Specific applications include reactions used to produce alpha- and beta-amino acids, L-dopa, alpha-tocopherol, and Lotrafiban. Although the use of these reactions has become quite common, an ongoing challenge is the selection of chiral ligands for the catalysts to maximize the enantiomeric excess (e.e.) for a specific application. Traditionally, this optimization was accomplished by experimental screening of libraries of catalysts, which can be a very time-consuming and expensive operation. Furthermore, to use this method to assay new, previously unknown catalysts requires that they are first synthesized, which itself may be a challenging task. Our principal goal has been to develop a computational procedure, which would permit fast and reliable prediction of e.e. for any given combination of chiral catalyst and substrate. A procedure of this type would permit in silico screening of a virtual library of catalysts, whether they were previously known or not, and only the catalysts for which high e.e.'s were predicted would be selected for experimental validation. The sometimes difficult task of synthesizing new ligands would be assumed only if high e.e.'s are predicted for them.
We have achieved proof of principle of this approach in the case of enantioselective hydrogenation of certain enamide derivatives, which serve as industrially important precursors of amino acids. The key computational tool that was used to reach this goal is an implementation of the Q2MM method developed by Professor Per-Ola Norrby of Gothenburg University in Sweden. In a three-way collaboration of the Helquist and Wiest laboratories at Notre Dame and the Norrby group, the Q2MM method was employed for the automated generation of molecular mechanics force fields for the stereochemical determining transition state of the targeted reaction employing chiral rhodium catalysts. The force fields were derived from the transition state structures computed using quantum mechanical (QM) methods. Whereas the initial QM computations are very time-intensive, molecular mechanics (MM) calculations using the resulting force fields are very fast. They permit the rapid calculation of the difference in energy of the diastereomeric transition states that lead to the production of the two product enantiomers from a given substrate/chiral catalyst combination. These differences in energy then straightforwardly translate into the prediction of e.e.'s based upon the difference in rates for production of the two enantiomers. Once optimized, this procedure permits the virtual screening of hundreds of substrate/catalyst combinations in just a few hours with a multiple node computer cluster.
The initial success with enamide derivatives was the direct result of a joint effort with two of the world's largest pharmaceutical companies. Data on previous hydrogenation reactions from these companies were employed in training sets and test sets for developing the required force fields. The resulting force fields from this first effort led to the prediction of e.e.'s having a statistical correlation factor of over 95% with experimental values. We have performed a virtual screen of combinations of six substrates with over one hundred different chiral catalysts. In the cases of a few outliers, we have done follow-up computational studies to probe the differences between the predicted and experimental values. In these cases, the additional computations suggested subtle changes in the hydrogenation mechanism for some of the substrate/catalyst combinations. In this sense, our computational methods may have a second application as a mechanistic probe. In other words, if an otherwise reliable computational method leads to a grossly inaccurate prediction of e.e. in a given case, something needs to serve as the source of the error. The source could be either a fundamental problem with the computational method or a change in the mechanism. Further study of such a case would be aimed at differentiating these two explanations.
Following the initial success of our approach, we have extended our studies to additional types of substrates. including other classes of alkenes such as acrylate derivatives and ketones. The selection of specific systems is made in close interaction with our large corporate partners. A main criterion for these choices is their importance in "real world" industrial applications. Another aspect of more recent studies is the extension of the force field development to additional types of chiral ligands for which such data have not been available previously. Prime examples are diphosphine ligands based upon a ferrocene scaffold. Such ligands, as exemplified by Josiphos, are used in a number of large-scale industrial processes. The new force fields that will arise from these further computational studies will permit the fine-tuning of these types of ligands for future manufacturing applications.