Screening for Generality in Asymmetric Catalysis

Summary Paragraph

Research in the field of asymmetric catalysis over the past half century has resulted in landmark advances, enabling the efficient synthesis of chiral building blocks, pharmaceuticals, and natural products.^1–3 A small number of asymmetric catalytic reactions have been identified that display high selectivity across a broad scope of substrates; not coincidentally, these are the reactions that have the greatest impact on how enantioenriched compounds are synthesized.^4–8 We postulate that substrate generality in asymmetric catalysis is rare not simply because it is intrinsically difficult to achieve, but also because of the way chiral catalysts are identified and optimized.⁹ Typical discovery campaigns rely on a single model substrate, and thus select for high performance in a narrow region of chemical space. Here, we put forth a practical approach for using multiple model substrates to select simultaneously for both enantioselectivity and generality in asymmetric catalytic reactions from the outset.^10,11 Multi-substrate screening is achieved by conducting high-throughput chiral analyses via supercritical fluid chromatography-mass spectrometry (SFC-MS) with pooled samples. When applied to Pictet–Spengler reactions, the multi-substrate screening approach revealed a promising and unexpected lead for the general enantioselective catalysis of this important transformation, with high enantioselectivities obtained for substrate combinations both within and outside of the screening set.

Introduction

Since the discovery that chiral phosphine-rhodium(I) complexes catalyze the highly enantioselective hydrogenation of certain dehydroamino acids,^1,2 asymmetric synthesis with small-molecule catalysts has been demonstrated in a dazzling variety of contexts.³ It is now widely appreciated that high enantioselectivity, typically defined as >90% enantiomeric excess (ee), is often attainable for specific model substrates in a reaction of interest. While these model substrates may be the targets of a specific synthetic campaign, they are more frequently chosen on the basis of accessibility, ease of chiral analysis, lack of peculiar structural features, or similarity to substrates studied successfully in related reactions.

To this day, high enantioselectivity remains the sine qua non of asymmetric catalysis development efforts; it is only after that condition is met with a model substrate that the substrate scope and limitations of the reaction are evaluated (Fig. 1A). Given the truism that “you get what you screen for,”⁹ an unintended consequence of this paradigm is that it fundamentally selects for success with substrates that are similar to the model. In rare cases, asymmetric transformations with broad substrate scope do emerge, and those are typically the ones that are most impactful as they can be applied predictively in new contexts.^4–8 But such generality is effectively accidental. In the vast majority of cases, the scope is limited, researchers strain to identify enough high ee examples to fill the “Substrate Scope” table requisite for publication, and the methods remain underutilized because synthetic practitioners shy away from trying unproved or unpredictable chemistry.

Fig. 1. Approaches to the discovery and analysis of enantioselective reactions.

(A) The standard approach to discovery of new asymmetric catalytic reactions involves optimization around a single model substrate. The scope of the method is then examined in a separate exercise, often resulting in methods that are only highly effective for substrates similar to the model. (B) Optimization via multi-substrate screening across a broad cross-section of substrate space improves the chances of identifying more general catalysts and conditions. (C) Conventional ee determination of single isolated products via chromatography on CSPs and detection by UV-Vis. (D) High-throughput ee determination of multiple products via SFC-MS. As long as each product possesses a unique mass and because enantiomers have equal response factors, multiple enantioselectivity values can be measured simultaneously by direct integration of the extracted ion chromatograms (EICs).

Optimization against multiple, diverse substrates simultaneously rather than against a single model substrate would shift the focus in asymmetric catalysis discovery efforts from identifying circumscribed examples of high enantioselectivity to revealing more general solutions (Fig. 1B). This approach was proposed in 1999 by Kagan et al.^10–11 and articulated compellingly in recent studies by List et al.,^12,13 but its adoption has been largely precluded by the challenges associated with conducting large numbers of chiral analyses on a variety of products. Even though methods for ee determination have advanced substantially, their development and application remain so laborious and time-consuming that researchers will typically only commit to developing assays for multiple products after success with a model substrate has been achieved.

The most general and commonly applied analytical methods for ee determination involve high-performance liquid chromatography (HPLC) or supercritical fluid chromatography (SFC) with chiral stationary phases (CSPs) using isocratic elution and UV-Vis detection (Fig. 1C). Such methods are capable of separating most products, but are inherently limited in throughput because baseline separation is required and interfering analytes must be excluded. While there has been long-standing interest in developing high-throughput methods for chiral analysis, the techniques identified to date have been tailored for repeated analyses of specific analyte classes of interest. For instance, the Anslyn and Wolf groups have developed circular-dichroism-based sensors that enable ee determination for compounds containing chelating functional groups.^14,15 Other approaches have employed chiral ¹⁹F-NMR shift reagents,¹⁶ fluorescent DNA biosensors,¹⁷ mass-tagging with pseudo-enantiomers,¹⁸ and selective enzymatic oxidation.^19,20 Although these methods can provide high accuracy and sample throughput, more general analytical strategies are needed to conduct effective multi-substrate screening across a range of chemical space.

We envisioned that rapid and simultaneous ee determination of diverse panels of products might be achieved by combining sample pooling and supercritical fluid chromatography (SFC), with mass spectrometry (MS) as the detection method (Fig. 1D).²¹ SFC already enjoys widespread application in academic and industrial laboratories for rapid analysis, and recent advances in immobilization strategies have given rise to commercially available CSPs with excellent separation properties and improved robustness that enable the analysis of crude mixtures.^22,23,24 Application of MS as the detection method would allow generation of extracted ion chromatograms (EICs) enabling accurate ee determinations of products of different masses even when they co-elute with other products or residual reaction components.²⁵ Furthermore, while conventional UV-Vis methods require isocratic elution to maintain flat baselines and equal response factors for the separated enantiomers, MS signal magnitudes are solvent-independent and allow solvent gradients to be used. Thus, a single, generic gradient could be applied to analytes possessing a wide range of polarities, greatly simplifying analytical method development. Overall, we anticipated that pooled SFC-MS would raise the throughput of ee determination to the point where it becomes practical to optimize directly for generality using multiple substrates.

Methods

The performance of SFC-MS for rapid ee determination of single and pooled samples was first assessed with a set of commercially available compounds using standard chromatographic columns. When pure samples of known enantiomeric composition were analyzed at high concentrations (10 mM), a root-mean-square error (RMSE) of 22% resulted, with the greatest deviations for scalemic mixtures (Fig. S2). Most of the error originates from the nonlinear relationship between detector response and concentration, which results in reduced intensities for larger peaks and therefore underestimates of ee (Fig. S2). However, the signal dependence on concentration approaches linearity at lower sample concentrations (<0.1 mM, Fig. S3) and significantly reduces the RMSE to 3%, which is comparable to the errors encountered using standard HPLC-UV conditions.²⁶ Consequently, all subsequent samples were analyzed at the lowest concentrations that still afforded good signal-to-noise (typically 0.1 mM, assuming 100% theoretical yield), a practice that was convenient for high-throughput assays of reactions carried out at micromole-scale.

Given the intention to carry out analyses of multiple compounds simultaneously with short run times and the fewest possible injections, we recognized that baseline separation of all enantiomeric pairs would not always be achieved. We therefore sought to develop a general peak-fitting method to extract accurate integrations from partially separated enantiomers. Fitting the experimental data of baseline-resolved peaks produced from scalemic mixtures of known composition to linear combinations of peak functions revealed that the SFC-MS peak shapes were not well reproduced by any single known peak functions. However, a “Frankenstein” model involving a piecewise combination of Gaussian and Voigt functions was found to provide excellent fits (Fig. S1). We combined this peak model with parameter-fitting methods in a convenient web-based application for analyzing chromatographic data. Accurate ee determinations were obtained when the protocol was applied to the analysis of the same scalemic standards on columns that produced only partial separation of the enantiomers, even in the challenging cases of high-ee samples where the first major peak tailed into the second minor peak (Fig. 2A).

Fig. 2: Development of the SFC-MS method.

(A) Combining Gaussian and Voigt functions provides accurate integration of poorly separated peaks (“det’d” refers to ee determined by SFC-MS). (B) Heatmap visualization of the total ion chromatogram. (C) Analysis of 20 pooled racemic compounds by SFC-MS (RMSE=7%, n=58). (D) Pairwise experiments with racemates reveals that co-elution can result in ion suppression, leading to inaccurate ee values. When columns and pooling combinations that minimize co-elution with strong ionizers are used, typical errors are 5–10% ee.

With robust methodology for single sample analysis in hand, we proceeded to evaluate the accuracy for pooled samples. A mixture of 20 commercially available racemic compounds (mostly pharmaceuticals; Table S3) at low concentration was analyzed with five chromatographic columns (Fig. 2A). While most compounds were resolved effectively and registered ee values of 0±5%, a few outlying ee values of >10% were also apparent. Inspection of the total ion chromatogram (Fig. 2B) revealed that strongly ionizing species can reduce the intensity of co-eluting ions.^27,28 If one enantiomer falls under such a band and its partner does not, the apparent ee is distorted in favor of the unaffected peak. This effect is illustrated in pairwise experiments with racemates (Fig. 2C). Nonetheless, at low concentrations, this ion suppression phenomenon is minimized, with overall ees of −1±7% in this experiment.

The screening-for-generality workflow detailed here involves three steps: substrate selection, reaction screening, and high-throughput chiral analysis. First, the reaction of interest is defined and a set of products representing the targeted scope is identified. Next, racemic samples of each product are prepared and chiral separations are attempted using all available columns. In general, no single column will be capable of separating every pair of enantiomers, but useful separations of a representative subset of the candidates can usually be achieved with a small set of different columns. To avoid injecting compounds onto columns that do not separate them and minimize ion suppression due to co-elution, the selected subset is sorted by intended column. In the screening phase, reactions are run in independent vials and aliquots from each reaction are diluted and pooled appropriately. Finally, the pooled samples are subjected to chiral SFC-MS, and the enantioselectivity values are determined by peak fitting.

Results and Discussion

We selected asymmetric catalysis of the Pictet–Spengler reaction as a timely and synthetically relevant platform to illustrate the application of the new high-throughput enantioselectivity determination methodology and the “screening for generality” concept. The condensation of tryptamines with aldehydes or ketones to generate tetrahydro-β-carbolines (Fig. 3A) is a venerable reaction with crucially important applications in laboratory and biological synthesis.^29,30 The reaction has inspired intensive research to find asymmetric catalytic variants, with more than a dozen catalytic systems reported thus far.^21–45 Each study identified highly enantioselective reactions, relying on optimization of catalyst and conditions around limited numbers of model substrates. By traditional standards, this output of new catalysts, high ee examples, and publications in high-impact journals can certainly be viewed as a major success. Yet despite this apparent progress, none of the published methods has found widespread application, and the chemist interested in carrying out an enantioselective catalytic Pictet–Spengler reaction on a never-before-tested substrate combination would be hard-pressed to know which system to try. We sought to establish whether screening across broad stretches of substrate space might prove informative in that regard and possibly enable the identification of general systems.

Fig. 3. High throughput ee-determination of enantioselective catalytic Pictet–Spengler reactions.

(A) The 14-member panel of products (left) used to study the Pictet–Spengler reaction and a map (right) of potential products (grey) with previously reported products from the literature (blue) vs. our panel (red). Test products (yellow) excluded from the initial screens were evaluated after all optimization studies were completed (Fig. 4C). Map generated by UMAP dimensionality reduction of product molecular fingerprints. (B) Enantioselectivity screen using 14 previously reported organocatalysts against the 14-member panel. Reactions with weakly acidic H-bond-donor catalysts i–ix and xiii–xiv were run with benzoic acid as a co-catalyst. Empty squares represent low-yielding reactions. A metric (g) was constructed to quantify the degree of generality exhibited by each catalyst.

Pictet–Spengler reactions can proceed via N-acyl-, N-protio-, and N-alkyl-iminium ion intermediates, and enantioselective catalytic variants have been identified for each of these manifolds.^31–45 We selected reactions between N-benzyl tryptamines and aldehydes as a particularly convenient platform to survey the substrate/catalyst landscape. In particular, reactions with N-benzyl tryptamines were found to remain homogenous, enabling the reactions to be run in a format that is highly amenable to parallel screening with standard equipment (96 well plates, no stirring, 0.01 mmol scale). Acceptable agreement between the enantioselectivity values obtained by single sample SFC-UV and pooled sample SFC-MS was achieved for the resulting N-benzyl-tetrahydro-β-carboline products (8% RMSE, Fig. S32).

To identify representative aldehyde/tryptamine combinations, we constructed an in silico library of 340 potential tetrahydro-β-carboline products, generated molecular fingerprints, and performed UMAP dimensionality reduction to generate a chemical space representation (Fig. 3A).⁴⁶ We then balanced the recommendations of k-means clustering with practical considerations, including commercial availability and reactivity of the requisite substrates, to generate a panel of 14 representative products (divided across two groups for separation). Notably, these products encompass regions of chemical space that are largely unexplored by reported methodologies.

We evaluated a set of previously reported chiral Brønsted-acid and H-bond-donor catalysts across the selected panel. The resulting ee data (Fig. 3B) reveal that different catalyst classes respond very differently to variations in substrate. In some cases, such as thioureas i-vii, moderate (e.g., 20–40%) enantioselectivity was observed fairly consistently across most substrates. Other catalysts afford higher (e.g., >60%) enantioselectivity, but only for specific subsets of the substrate space; for example, Miller squaramide ix is particularly effective in catalyzing the reaction between neutral indoles (X=Y=H) and aryl aldehydes, whereas the SPINOL-phosphoric acid xii reported by Lin and Wang et al. stood out in reactions of the electron-deficient indole (X=Cl, Y=H).^36,45

Tabulation of the data as a heatmap (Fig. 3B) provides a visually straightforward tool for identifying correlations in behavior between related catalysts. For instance, chiral phosphoric acids xi and xii perform poorly with the electron-rich indole (X=H, Y=OMe), and generally much better with the electron-poor or neutral tryptamine analogs. Similarly, comparison of ee values between rows reveals correlations between substrates. For example, compound 23, a prototypical minimally functionalized model product derived from N-benzyl tryptamine and benzaldehyde, responds similarly to catalyst effects as do products 24 and 28 also possessing the neutral indole. However, product 23 is a very poor model for products possessing electron-rich indoles such as 11, 12, 16, or 19. These findings highlight the risks associated with optimizing around a single model substrate combination and the value of employing multi-substrate screening for the accurate assessment and optimization of a given methodology.

To help assess the level of generality for each catalyst, we constructed a generality metric g that summarizes a collection of enantioselectivity values into a number between 0 and 1, where 1 represents a completely general catalyst that induces 100% ee in every reaction (see SI Section 3.9). By this analysis, catalyst xii stands out among those surveyed as the most promising from a generality standpoint.

Despite the positive results obtained with catalyst xii, it is apparent from the data in Fig. 3B that reactions of the electron-rich indole are particularly challenging for that, and indeed all catalysts in the screen. We zeroed in on that substrate class by evaluating several solvent–catalyst combinations for product 16 (Fig. S40). Improved results were obtained with polar aprotic compounds such as 2-methyl THF (2MT) and ethyl acetate (EA), prompting us to evaluate these solvents across the entire substrate panel (Fig. 4A). Notably, the solvent effects were highly catalyst-dependent, with enhanced enantioselectivities observed in reactions performed in 2MT for chiral phosphoric acid xii, but no systematic improvement observed with H-bond donor catalyst i. The substantial benefits of 2MT were possibly missed in the original work that led to catalyst xii³⁶ because the authors effectively relied on a single model (23 and related N-protected analogs). While changing from PhMe to 2MT leads to a small decrease in the enantioselectivity for 23, most other substrates benefit substantially, with the increase from 1% to 62% ee for product 11 serving as a particularly notable example. These results illustrate how performing screens on multiple substrates simultaneously can provide valuable and otherwise elusive insights.

Fig. 4. Further reaction optimization and validation.

(A) Solvent screening reveals the beneficial effects of 2-methyl THF (2MT) compared to toluene (PhMe) and ethyl acetate (EA) for most substrate combinations. (B) Rerunning the optimal conditions with conventional glassware with molecular sieves results in improved enantioselectivity for some substrates, and diminished enantioselectivity for others. (C) Assessment of previously untested substrate combinations leading to products 14, 20, and 36 under the high-throughput screening conditions with four selected catalysts. In each case, catalyst xii displayed the highest enantioselectivities, consistent with the results of the generality screen.

The substrate panel was re-evaluated with catalyst xii in a standard, rather than high-throughput, experimental format. Running all 14 substrates under air-free conditions with activated 4Å molecular sieves³⁶ resulted in increased enantioselectivity for several products and restored reactivity to products 1 and 9 (Fig. 4B). However, product 11 saw a substantial decrease in ee, which may be due to the susceptibility of that product to racemization (SI Section 3.7). This outlier further underscores the importance of multi-substrate screening: selection of product 11 as a unique model substrate would lead to the erroneous conclusion that molecular sieves are deleterious. The optimal conditions were further adapted to generate product 6 on 0.1 mmol scale with good yield and enantioselectivity (SI Section 3.8), thereby demonstrating the successful translation of the high-throughput screens and analyses to laboratory-scale reactions. To assess the transferability of the conditions optimized for the 14 model products, we tested a subset of the catalysts against 3 previously untested substrate combinations (Fig. 4C). In each case, catalyst xii proved to be the most effective, and afforded useful enantioselectivities under synthetically relevant conditions.

Relative to traditional ee-determination methods, the analytical workflow outlined in this study trades a modest decrease in accuracy for the ability to analyze multiple crude reaction mixtures simultaneously with substantial reductions in method development and analysis time. As a result, optimization of catalyst structure and reaction conditions is more readily performed across a variety of substrate combinations, thereby increasing the chances of identifying general protocols. If no broadly general protocols are found for a given reaction of interest, the broad survey of the catalyst/substrate landscape can be used to ensure that highly effective catalyst-substrate combinations are not missed, and to identify “islands” of high enantioselectivity involving specific catalyst structures and subsets of the substrate space.

The survey of the substrate/catalyst landscape in the asymmetric Pictet–Spengler reaction reveals how the standard approach of basing optimization campaigns on a single model system has indeed resulted in substrate-specific islands of high enantioselectivity for this reaction. However, an unexpected finding from this study is that there is already a family of catalysts (exemplified by xii) that displays very promising levels of generality across a wide range of substrate combinations. The uniquely interesting features of xii were likely masked by the fact that the catalyst was identified through an evaluation of a narrow region of substrate space.³⁶ Similarly, the beneficial effect of 2MT in reactions with catalyst xii was masked by the anomalous behavior of the prototypical model substrates for that reaction. Thus, the use of multi-substrate screening revealed a promising lead for the development of a highly general and enantioselective Pictet–Spengler reaction, and will hopefully prove useful in the identification and discovery of general chiral catalyst systems for other transformations of interest.

Data and materials availability:

All data are available in the main text or the supplementary materials.