Corin Wagen^†, Spencer McMinn^†, Eugene Kwan*, Eric Jacobsen*

The paper in Nature is here.

What do the most useful reactions in organic chemistry have in common?  Think of hydrogenation, cross-coupling, or Sharpless epoxidation.  Each reaction works reliably for a broad range of substrates and can be counted on in synthetic planning.  Despite substrate generality being a widely appreciated goal in reaction discovery, we typically screen for high enantioselectivity with only a single model substrate.  Unsurprisingly, we get what we screen for, and most reactions end up working well only for a narrow range of substrates similar to the model.

A better way to develop general reactions would be to look for enantioselectivity across multiple substrates at once.  This is an old and perhaps obvious idea, but multi-substrate screening has been impractical until now because of the burden of chiral analysis.  Typically, enantioselectivity is determined by chiral chromatography with ultraviolet (UV) detection, which is sensitive to interfering impurities and therefore impractical for complex mixtures.  Here, we built on recent developments in faster chromatography by increasing the number of measurements per injection.

By using mass spectrometry (MS) as the detection method instead of UV, different products with unique masses can be pooled into one vial and analyzed together (Figure 1).  Reactions are run normally, with one substrate per reaction vessel.  When the reactions are complete, samples from multiple reaction vessels are combined into a single vial.  Chiral supercritical fluid chromatography (SFC) is performed and then the enantioselectivity for each product is determined on its own mass channel.  This trick works well even in complex mixtures, is accurate enough for screening work (±7% ee), and speeds up chiral analysis a lot.

Figure 1: Multi-Substrate Screening via Sample Pooling

We used multi-substrate screening to study known catalysts for the Pictet–Spengler reaction (Figure 2) across 14 diverse substrates.  No one model system is representative: even when a reaction works well in one system, it often works poorly in others.  Thus, despite the fact that each of these systems has been reported to give over 90% ee in multiple publications (including our own), the substrate scope of each system is narrow.

Nonetheless, the SPINOL xii stands out as being the most general.  When xii was originally studied using a single model substrate (23), the authors found that toluene was the optimal solvent.  However, across multiple substrates, the optimal solvent turns out to be 2-methyl tetrahydrofuran.  Additionally, when these condensation reactions were translated from a high-throughput format to a singleton format, where molecular sieves could be used, the enantioselectivity improved further.  8 of the 14 substrates gave over 70% ee, which is a good starting point for targeted single-substrate optimization.

Reassuringly, this modified catalytic system generalizes to substrates that are outside the training set of 14 molecules.  When we evaluated a test set of three substrates, we found that each substrate exceeded 70% ee.  The fact that xii is exceptionally general was not detected by routine single substrate screening, and illustrates the potential of multi-substrate screening.  Our method uses existing equipment and can immediately be used for the discovery of catalytic, asymmetric reactions with broad substrate scope.