Recognition of a Conserved Minimal Catalaphile Explains Catalyst Generality in Enantioselective Nitroalkene Reductions

Abstract

In catalyst development for asymmetric synthesis, generality is most often an aspirational goal rather than a reality, driven by the direct relationship between catalyst utility and product access. Strategies and tools have been developed to prospect and identify catalyst generality, but the mechanistic underpinnings for generality remain unclear. Here we investigate the selectivity-generality paradox for chiral Amidine Amide (AmA) catalyzed nitroalkene reductions, identifying the origin of generality using both experimental and computational approaches. Use of an Independent Gradient Model (IGMH) and Non-Covalent Interaction maps (NCImaps) reveal a stark contrast in recognition modes: while limited-scope catalysts rely on adaptive Van der Waals interactions that vary with substrate structure, the general AmA catalyst utilizes a conserved hydrogen-bonding network that recognizes the nitroethylene moiety—the minimal catalaphile. This understanding provides a framework for elucidating how early development focused on generality can be propagated through rational design, supported by computation, and translated to the broadest possible application. We posit that the minimal catalaphile concept, and its prioritization in development, could be a coalescing principle in hypothesis-driven development of privileged catalysts.

Graphical Abstract

Introduction

A ‘holy grail’ within catalytic asymmetric synthesis is the discovery of catalysts that are highly reactive, selective, and broad in substrate scope.^{[1, 2]} Despite extensive empirical efforts spanning over three decades^[3–9] and recent advances in high-throughput experimentation (HTE),^{[10, 11]} machine learning-guided discovery,^[12] and machine learning tools to assess, quantify, and visualize generality,^[13–15] the molecular origins of catalyst generality remain poorly understood, hampering rational design approaches.^[16–18] Conceptually, high selectivity and broad scope are at odds insofar as the design of substrate engagement pockets must fit one enantiomeric transition state much better than its mirror image. Enzymatic catalysts use tailoring to readily achieve high selectivity for specific substrates by leveraging their size and residue diversity, accelerated by directed evolution and high throughput screening technology. ^[19] Small molecule metal and non-metal catalysts,^[20–22] however, use far fewer atoms to achieve similar goals. Considering noncovalent catalysts specifically, enzymatic^[23] and sterically-confined catalysts^[24] illustrate this design across biomolecular and small molecule catalysis, respectively.

A prevailing design principle for asymmetric catalysis is to directly engage a functional group (Figure 1, top) while considering the proximal structure to achieve high selectivity.^[25] However, a single catalyst that achieves high selectivity with an uncharacteristically broad substrate scope is paradoxical because the peripheral environment changes significantly across all possible substrates. Miller and coworkers demonstrated that peptide-based catalysts can achieve generality through 'adaptive recognition' where the catalyst flexibly adopts distinct binding modes to accommodate different substrates.^[26] In contrast, we hypothesized that rational solutions to this paradox might be generated by the discovery and development of catalyst-substrate interactions conceptualized by a minimal catalaphile (Figure 1, bottom) ^[27] The ideal minimal catalaphile for a reaction is defined as the minimum number of atoms required for activation by the catalyst, as well as the minimal geometric plane required for prochiral face discrimination by the chiral pocket. A limited influence by atoms beyond the minimal catalaphile would then be a hallmark defining minimal catalaphile recognition. the For example, a Lewis acid-catalyzed aldol reaction would be most general if the minimal number of atoms is recognized (e.g. -CHO of an aldehyde) but still favoring one π-face during the 1,2-addition step. We posit that catalyst designs realizing minimal catalaphile engagement could aspire to privileged status^[28] while advancing the principles of sustainability.^[29–34]

Figure 1.

Illustration of minimal catalaphile recognition as a strategy for achieving high selectivity and generality.

The enantioselective nitroalkene reduction has progressed significantly in recent decades, but only within a relatively narrow scope if scrutinizing against all possible structural classes of nitroalkenes.^[35] The reduction of electron-deficient alkenes is a broadly useful paradigm, and prochiral trisubstituted nitroalkenes can deliver chiral nitroalkane products that are precursors to amines,^[36] aldehydes,^[37] alcohols,^[38] and acids.^[39] We posited that nitroalkenes serve as an ideal platform to test the minimal catalaphile hypothesis. A catalyst capable of exclusively recognizing the conserved nitroethylene core rather than adapting to the variable periphery could solve this generality paradox. Building upon our recent discovery of the most general catalyst for enantioselective nitroalkene reductions, ^[35] we report the first minimal catalaphile hypothesis-driven mechanistic investigation using DFT and machine learning, experimental mechanistic studies, and 76 examples. We identify an apparent bifurcation in recognition models: whereas less general catalysts rely on “adaptive recognition” (variable dispersive forces with the substrate periphery), the AmA catalyst utilizes “conserved recognition” of the minimal catalaphile (the nitroethylene core). The minimal catalaphile hypothesis might transfer to other reactions and catalysts, with the hope that the intense search-and-discover paradigms to identify privileged catalysts might be increasingly replaced by rational design.^{[40, 41]}

Methods

Generation of Virtual Catalog of Nitroalkene Substrates

Chemically Reasonable Mutations (CReM) framework was used to generate the virtual nitroalkene substrate collection. Using a precompiled database obtained by fragmentation of the whole ChEMBL with a maximum synthetic complexity of 2, substrates were generated with the enumerate compounds utility from CReM. The base structure was defined as nitroethylene itself and scaffold mode was used to add fragments onto nitroethylene considering a radius of 1, and with no limits on number of replacements allowed per iteration for two iterations. A threshold of 30 was set for the frequency of fragments included in order to emphasize molecules that better represented in ChEMBL and thus more likely to be of interest to researchers. See additional details in the supporting information.

General Workflow for Classification of Nitroalkenes and Assignment of Generality

Uni-Mol2 was used to generate embeddings for the nitroalkene scope. This yielded molecule-level vector embeddings with 1536 dimensions for the 3476 substrates. The autoencoder architecture was based on what has been reported in the literature for clustering tasks, which is a fully connected Multi-Layer Perceptron with the dimensions d-500–500-2000–30, where d is the dimensionality of the data, and the decoder structure is a mirror of the encoder. All layers use ReLU activation and the optimizer is Adam. By first annotating the substrates based on element and hybridization of atom adjacent to β-carbon, StratifiedKFold from sklearn was used to split the dataset for 5-fold validation in order to determine the best batch size (16). The autoencoder was trained for 200 epochs for each split and batch size tested as it was observed that training loss plateaued quickly after about 100 epochs. With this, the Uni-Mol embeddings were re-embedded to 30 dimensions, and UMAP was used to learn a more clusterable manifold from the latent space of the autoencoder. Supervised UMAP based on the previously mentioned annotations was performed to learn a manifold that has the same dimensions (30) but emphasizes information relevant to the reaction site, that is the more immediate local surroundings of the hydride transfer, while preserving the structure of information in the rich 3D-based embeddings from Uni-Mol. From this, unsupervised HDBSCAN was performed to cluster nitroalkenes based on the learned manifold. Hyperparameter optimization for UMAP and HDBSCAN was performed in tandem with a grid search over several of the most important parameters for the two algorithms. Additional details are provided in the supporting information.

DFT methods

All DFT calculations were performed using the ORCA^[42] 6.0.1 software package along with the xtb program package for GFN2-xTB^[43] level calculations. Conformational searches of the catalyst/nitroalkene/Hantzsch ester complex were performed with the Conformer–Rotamer Ensemble Sampling Tool (CREST) developed by Grimme.^[44] Initial conformational sampling using CREST was performed at the semiempirical GFN2-xTB level with ALPB(toluene) as the solvent model using the iMTD-GC algorithm which combines iterative metadynamics simulations with Cartesian root-mean-square-deviation (RMSD) between generated structures as collective variables and genetic Z-matrix crossing. Default settings for sampling of non-covalent complexes were used. Conformers within 3 kcal/mol of the lowest energy structure were retained for further optimization at the r2SCAN-3c^[45]/SMD^[46] (toluene) level. Transition states were identified using relaxed surface scans (RSS) by systematically decreasing the C-H···C distance between the Hantzsch ester hydride and the electrophilic carbon of the nitroalkene. The identified transition states were confirmed by the presence of a single imaginary vibrational frequency and validated through intrinsic reaction coordinate (IRC) calculations. These transition states were also compared with literature reports of thiourea-catalyzed nitroalkene reduction to further judge their plausibility.^[47] All structures in the transition state study were optimized using r2SCAN-3c/SMD(toluene) level of theory, and single point energies were calculated at the wB97M-D4^[48]/def2-TVZPP level of theory with SMD(toluene) as the implicit solvent model. Gibbs free energy corrections were calculated at temperature and pressure corresponding to standard reaction conditions. Independent Gradient Model based on Hirshfeld partition (IGMH) analysis was applied to visualize non-covalent interactions and Natural Bond Orbital (NBO) second-order perturbation analysis was employed to quantify the non-covalent interactions. Additional detail is provided in the supporting information.

General procedure for enantioselective reduction of nitroalkenes

A flame-dried vial was charged with nitroalkene and AmA catalyst 3 (5–10 mol%), and dry toluene (0.1 M). The catalyst used is (R,R) in all cases unless specified otherwise. The solution was stirred at the specified temperature (0–60 °C) for an hour, then Hantzsch ester (1.6–2.6 equiv) was added to the vial. The vial was evacuated and backfilled with argon and stirred at the specified temperature for 48–96 h. After the completion of the reaction, the reaction mixture was filtered through a silica plug, eluting with hexanes/DCM or hexanes/EtOAc to yield the desired nitroalkane. The enantiomeric excess was determined by chiral HPLC. The absolute configuration of 2a was assigned by literature reference^[49] as (R) and 2b-2bx were assigned by analogy.

Results and Discussion

Probing the Limits of the Minimal Catalaphile Hypothesis

Having previously established the baseline generality of the Amidine-Amide (AmA) organocatalyst (3), we sought to stress-test the minimal catalaphile hypothesis by probing its recognition scope with more a comprehensive and potentially challenging rang of substrates. To achieve this, we doubled the prior substrate scope (from 37 to 76 examples). Remarkably, catalyst 3 maintained consistently high enantioselectivity across this expansive chemical space, averaging 90% ee (<5% standard deviation). A curated selection of highly hindered substrates is highlighted in Table 1A, while the comprehensive 76-example scope is detailed in the Supporting Information (Extended Table1). Notably, the highly enantioselective reduction of β,β-diaryl and tetrasubstituted nitroalkenes has only been achieved using this AmA platform; examination of other common organocatalysts yielded either no reactivity or low selectivity. ^[35] Furthermore, Benaglia reported several thiourea-based organocatalysts where selectivity ranged from 0–67% ee for tetrasubstituted nitroalkenes,^[50] highlighting the severe steric challenges associated with these substrates. The scope also successfully encompasses substrates with potentially competitive hydrogen bond acceptors, including cyano, carbonyl, and heterocyclic moieties. The maintenance of excellent yields and enantioselectivities across these subclasses provides empirical support that dual-point hydrogen bonding to the minimal catalaphile thermodynamically outcompetes binding to these alternative hydrogen bonding acceptors. Beyond these structural extremes, we also demonstrated a new functional capability of this system by achieving catalytic turnover of the organic reductant (Table 1B). In several cases, particularly those with neutral or electron-deficient aryl substituents, a procedure using sub-stoichiometric Hantzsch ester (40 mol %)^[51] and sodium cyanoborohydride^[52] as the stoichiometric reductant could be used (Table 1B). Yield and selectivity were comparable in these cases, while electron-rich aryl substituents were lower yielding and less selective. Control experiments using sodium cyanoborohydride alone resulted in no reduction on the time scale of the reaction, while the comparable levels of yield and enantioselection between stoichiometric and catalytic Hantzsch ester reductions indicate that turnover of this reductant is possible. Focused studies will be needed to realize the full potential of this design.

Table 1.

A) Catalyzed enantioselective reductions of nitroalkenes. General experimental conditions: Nitroalkene (1 equiv.), Hantzsch ester (1.6 equiv.) and catalyst (5–10 mol%) were stirred under Ar at 0–60 °C for 48 h. All yields are isolated yields and enantiomeric excess (ee) was determined by HPLC using a chiral stationary phase. See details in SI; B) Demonstration of turnover in the enantioselective reduction of nitroalkenes substoichiometric in Hantzsch ester. Modified from the general method: Hantzsch ester (40 mol %) and NaBH₃CN (2 equiv) were used.

Cheminformatics Analysis: Defining and Quantifying Generality

The generality of the AmA catalyst with all results to-date was quantified using a data-driven approach that involved parameterization of nitroalkene substrates using cheminformatics descriptors. Although generality is considered during catalyst development, it is not typically defined clearly. Catalyst generality is often used to refer to satisfactory yield, particularly in non-enantioselective reactions,^[53] and/or achieving satisfactory ee in enantioselective reactions.^[10] Here, we employ definitions for both reaction generality and selection generality (Figure 2A): reaction generality is defined as high yield realized across diverse groups of substrates, while selectivity generality is high enantioselectivity (>80% ee, as defined by Reid) realized by diverse groups of substrates. Depending on catalyst performance for each class of substrate, selection generality was assigned.

Figure 2.

A) Mathematical definition of reaction generality and selection generality. N: number of substrate (N>0); K: number of substrate types; success: enantioselectivity threshold for catalyst to be assigned as successfully applied with each type. B) UMAP clusters using 3476 virtual nitroalkenes generated by a fragment-based method. A representative substrate (medoid) is shown for each cluster. The substrates for this work (Table 1) are labeled as triangles. C) Comparison of reaction and selection generality between all reported asymmetric catalysts. Generality = 1 considers all possible substrates/clusters; reaction: clusters where average yield is greater than threshold value for substrates reported; selection: clusters where average selectivity is greater than threshold value for substrates reported; % clusters tested: clusters covered by reported outcomes that did not pass the threshold; % clusters untested: clusters not covered by reported outcomes. See SI for detailed information on all data sets.

Neither the quantity nor diversity of the nitroalkene substrate scope present in the literature could adequately define a sufficiently comprehensive chemical space. As a result, a fragment-based method of molecule generation^[54] was employed to create a more exhaustive virtual scope of nitroalkene substrates with two unique β-substituents (3476 nitroalkenes generated in total, see SI for details). For these substrates, we generated embeddings using Uni-Mol, which is pretrained on 3D conformer ensembles and demonstrates state-of-the-art performance on molecular property prediction and other downstream tasks.^[55] Recent developments in deep clustering methods^[56] drove the training of an autoencoder that learns efficient codings of the high dimension Uni-Mol embeddings. We then applied supervised Uniform Manifold Approximation and Projection (UMAP) on the latent space to learn the underlying manifold in a manner that is consistent with knowledge of reaction site (see SI for details). Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) was applied to cluster substrates based on the learned representation ^{[57, 58]} resulting in 32 clusters (Figure 2B). HDBSCAN instead of k-means clustering was used for defining substrate classes as it demonstrated better clustering capability for non-spherical clusters and could learn the number of clusters to choose from the structure of the data itself. The quality of clustering was quantitatively evaluated using various metrics and by comparison of mean Tanimoto coefficients for compounds from the same cluster versus those in different clusters (see SI).^[59] The AmA catalyst fulfilled the criteria of being general (>80% average ee) for 23 clusters (selection generality = 0.72) while the best reported catalyst in the literature was general for only 14 (Figure 2C). To mitigate concerns that catalysts reported in the literature could have low generality due to limited substrate scope, similar catalysts were grouped together with the assumption that they may exhibit similar reactivity and selectivity for a generality benchmark. The AmA catalyst alone was found to be more general than the best performing catalyst type (see details in SI).

Mechanistic Studies

A hypothesis to explain the unusually broad and diverse substrate scope requires an understanding of the catalyst-substrate recognition step. The mechanism of the AmA-catalyzed reduction was therefore studied using spectroscopic and kinetic tools. The reaction order for each component of the reduction was determined by variable time normalization analysis (VTNA) using NMR spectroscopy (See SI, Fig. S5-S7).^[60] It shows that the AmA catalyst, nitroalkene, and Hantzsch ester reductant, are each first order in the rate determining step (RDS). A key underlying feature of the catalyst is dual activation, first as a hydrogen bond donor to the nitroalkene, and second as a hydrogen bond acceptor for the Hantzsch ester.^[61–63] The enantiomeric excess (ee) is irreversibly determined by the chirality at the β-carbon during the hydride transfer (RDS). The subsequent tautomerization establishes the α-carbon stereocenter, which is non-selective due to the multiple proton sources, including the pyridinium or amidinium moieties from both intramolecular and intermolecular donors. The observation of low dr for tetrasubstituted nitroalkenes (2bq-2bx) supports this hypothesis. It is further supported by the finding that reduction of (E)-nitrostilbene yields racemic product (See SI Fig. S8).

These mechanistic experiments provided a foundation for computational studies to complement prior literature reports.^[47] The transition state of AmA 3 with substrate 1a was optimized at the r2scan-3c level of theory, and single point energies were calculated at the ωB97M-D4/def2-TZVPP level of theory (see SI for details). Analysis of these transition states (Figure 3A) revealed that (R)-TS1 (ΔG^‡_(cal) = 10.4 kcal/mol) was energetically favored over (S)-TS1 (ΔΔG^‡_(cal) = 2.6 kcal/mol), consistent with the experimentally observed preference for the (R)-product (ΔΔG^‡_(exp) = 1.7 kcal/mol). Facilitating this process, the AmA catalyst engages in two-point hydrogen bonding to the nitro group,^{[64, 65]} serving to activate the nitroalkene electrophile, while simultaneously positioning the Hantzsch ester through hydrogen bonding to the catalysťs amide carbonyl. To visualize these hydrogen-bonding interactions, an Independent Gradient Model analysis based on Hirshfeld partition (IGMH) ^{[66, 67]}was performed using Multiwfn^{[68, 69]} for each enantiomeric transition state (Figure 3B).

Figure 3.

A) Optimized transition state structures and energetics for substrate 1a with catalyst 3 (r2scan-3c/ωB97M-D4/def2-TZVPP/SMD(toluene)). B) IGMH analysis comparing non-covalent interactions in favored vs. disfavored transition states (δg^inter = 0.005), revealing that enantioselectivity originates from differential steric interactions between the Hantzsch ester and the catalysťs adamantyl group. Atom-pair δg indices (δG_pair (%)) quantify relative NCI contributions. Hydrogen bonding strengths were quantified using second-order perturbation energies (E⁽²⁾) derived from Natural Bond Orbital (NBO) theory.

To further quantify the electronic strength of these contacts, Natural Bond Orbital (NBO) second-order perturbation analysis (E⁽²⁾) was employed. The hydrogen bonds between catalyst and both reactants are prominent (∑E⁽²⁾ = 45.2 and 14.5 kcal/mol, respectively), alongside other non-covalent interactions (NCIs). Quantitative assessment using the atom-pair δg index (δG_pair (%)) revealed that hydrogen bonding between the nitro group and amidinium remains nearly constant between transition states. The key difference could be found in the Hantzsch ester delivery of hydride in competing transition states, where a repositioning was necessary in the disfavored transition state to alter unfavorable steric interactions between its methyl group and the cyclohexyl group. This resulted in decreased hydrogen-bonding contribution (2.3% reduction) and increased Van der Waals interactions (1.8% increase) in the disfavored transition state. This topological observation is corroborated by NBO analysis of the disfavored state (S)-TS1, which reveals a significant attenuation of the key Hantzsch ester interaction (14.5→7.5 kcal/mol) relative to the favored pathway.

Critically, this NBO analysis reveals a hierarchical recognition model that bridges catalyst generality and enantioselectivity. The dual hydrogen bonding network between the amidinium and the nitro group is thermodynamically dominant (ΣE² ≈ 40 kcal/mol) compared to the secondary Hantzsch ester (HE)-amide interaction (ΣE² ≈ 15 kcal/mol). Because the transition state demands a rigid C···H···C trajectory for hydride transfer, the HE possesses minimal independent freedom and its position is dictated by the nitroalkene. Thus, the minimal catalaphile interaction supports generality by rigidly anchoring the nitroethylene core in a conserved spatial coordinate. This primary anchoring forces the HE into a consistent geometry, where differential steric interactions in the disfavored transition state contribute to consistently higher energies.

Using this transition state model, we performed comprehensive transition state calculations with 12 substrates bearing diverse structural features to assess whether the catalyst was recognizing a small number of atoms, i.e. a minimal catalaphile (Figure 4). Two AmA catalysts (3 and its methylene homolog 4) were prepared and tested (see Figure S10) alongside two thiourea catalysts (5, 6) reported previously for this reaction.^{[70] [47]} Both favored and disfavored transition states were computed. This comparison is possible due to the relatively straightforward mechanism of the hydride transfer step, as well as the similarity of activation mode between amidinium and thiourea in the reaction.^[71–77] These structures were further assessed by comparing the calculated ΔΔG^‡_(cal) and experimental ΔΔG^‡_(exp) values, showing good agreement (average differences: 0.5–0.9 kcal/mol, see SI Tables S1-S4).^{[78, 79]} The resulting 96 catalyst-substrate transition state conformers were overlaid using either the amidinium or thiourea functionality as the reference point with key hydrogen bonding interactions highlighted. The extent of recognition was quantified by the RMSD of nitroethylene moieties (defined as the hypothesized minimal catalaphile) across the 12 substrates. From the overlay analysis, both AmA catalysts 3 and 4 show significantly conserved recognition, with 3 being the most conserved (lowest RMSD, 0.20 Å).

Figure 4.

Comparative analysis of substrate recognition patterns across four chiral catalysts. Twelve diverse substrates were selected to combine with 4 different chiral catalysts. The optimized conformations (r2scan-3c) overlaid with the catalaphore fixed (blue color) and RMSD of nitroethylene moiety were calculated to quantify the degree to which the interactions between the catalyst and minimal catalaphile are conserved.

This can also be visualized by the high degree of overlap for the nitroethylene substructure across all 12 substrates. For disfavored conformers, the overlap is slightly worse. However, for thiourea catalyst 5, the overlap of nitroethylene moieties is considerably worse, especially for disfavored conformers, giving the largest RMSD (1.72 Å). This suggests that when catalyst 5 binds with different substrates, the nitroethylene moiety must change its relative conformation to a higher degree in the disfavored transition states. This adaptive behavior indicates that the catalaphore of 5 is more specific to a subset of substrates. With others, 5 is less accommodating as the substituents are changed on the nitroalkene. Thiourea catalyst 6 presents an illuminating 'positive control.' Previously reported by Fochi as a general catalyst, ^[80] it exhibits a significantly lower RMSD and higher NCI similarity compared to 5, approaching the conserved recognition profile of the AmA catalysts. This intermediate behavior serves to validate our central hypothesis: the degree of generality correlates directly with the conservation of the minimal catalaphile interaction, independent of the specific catalytic scaffold (amidinium vs. thiourea).

To further distinguish between conserved and adaptive recognition modes, we generated Non-Covalent Interaction Conservation Maps (NCImaps) (Figure 5A). These were generated through a comprehensive IGMH analysis of all 96 transition states (see SI Figure S5 for maps generated by standard NCI analysis). These maps visualize the spatial conservation of NCIs by overlaying IGMH isosurfaces for each catalyst across all 12 substrates. The conservation of NCIs was quantified using a similarity metric that combines spatial and energetic components.^[81] The NCImaps reveal a clear hierarchy of conservation: 3 > 4 > 6 > 5, consistent with RMSD analysis. Critically, catalysts 3 and 4 maintain their recognition pattern through three dominant hydrogen bonds (appearing as intense blue regions), while catalyst 5 shows extensive green surfaces indicating multiple variable Van der Waals interactions whose strength depends heavily on substrate structure. This corresponds to low spatial and energy similarity scores with high variance across substrates. Beyond hydrogen bonding, the interactions are classified roughly into catalyst-nitroalkene interactions and catalyst-Hantzsch ester interactions; details of identified patterns emerged (See SI Fig. S1-S4). To move beyond qualitative visual assessment, we quantified the 'rigidity' of the recognition interface using two complementary metrics (Figure 5B, see full details in SI Tables S6-S21). First, spatial similarity was assessed using the NCI atom-pair δg index, revealing that 3 maintains a highly consistent interaction topology (80% similarity for favored transition state and 69% for disfavored transition state) across the substrate scope. In contrast, 5 exhibited significantly lower spatial conservation (44%/28% for favored/disfavored transition state respectively). Second, to probe the electronic consistency, we calculated the NBO Energy Conservation Score (S_NBO). 3 achieved a high score of 0.88/0.87, indicating that the minimal catalaphile interactions remain energetically dominant regardless of steric demand. The overall conservation rank is still 3 > 4 > 6 > 5.

Figure 5.

A) NCImaps generated from overlaid IGMH isosurfaces of 12 substrates show highly conserved H-bonding patterns for Catalysts 3 and 4 (blue regions, >80% similarity), while Catalyst 5 exhibits variable dispersion interactions. B) Quantification of spatial consistency (from NCI) and energetic conservation (from NBO analysis). C) Detailed analysis of outlier substrates demonstrates maintained H-bond dominance for 3 (consistently E⁽²⁾ energies, inset) despite substrate variation, whereas 5 shows complete reorganization with dispersion interactions superseding H-bonding (δg^inter = 0.005).

Application to Target-Oriented Synthesis

The practical utility of a general catalyst lies in its predictability across structurally distinct targets, obviating the need for extensive substrate-specific re-optimization. Platform nitroalkene reduction has been applied to therapeutically relevant peptidomimetics, a glucokinase activator, and four proteasome inhibitors.^{[35, 82]} We targeted the synthesis of a peptide-based compound with promising activity for targeted protein degradation (TPD).^[83] Discovered through a DNA-Encoded Library (DEL) screen, compound 7 was found to be a selective small molecule binder of the human CTLH E3 ligase subunit GID4. The reported route to 7 requires a noncanonical thiophene amino acid. Although sourced commercially as a single enantiomer, the amide coupling step resulted in a 1:1 dr of epimers at the α-center, a finding that we corroborated independently. To address this, nitroalkene 1av was subjected to the enantioselective reduction catalyzed by 3 to deliver the β-amino nitroalkane (2av) in 81% yield with 92% ee. Bromination of the nitroalkane readied it for coupling to the α-amino amide that delivered 10 in moderate yield. The conservation of stereochemical integrity (96:4 er→94:6 dr) ^[84]. was expected based on the mechanistically-prohibited epimerization feature of UmAS.^{[82, 85–87]} Preparation of dipeptide precursor 10 constitutes a formal synthesis of 7, but the first that is selective for the more potent diastereomer (Figure 6).

Figure 6.

Integration of the general, enantioselective reduction to establish configuration at an epimerization-prone non-canonical α-amino amide residue.

Conclusion

As methods to prospect for general catalysts, and discoveries of catalysts that deliver unusually general scope multiply, a mechanistically-based design construct becomes increasingly possible. Using an unusually general catalyst for the reduction of nitroalkenes, we have advanced the concept of a nitroethylene minimal catalaphile that is supported by transition state calculations and NCImaps. The catalyst’s minimal catalaphile recognition is conserved throughout its differentiation of enantiomeric transition states. Furthermore, a similar analysis with a leading thiourea catalyst reveals that multiple Van der Waals interactions outcompete generally hypothesized hydrogen bonding interactions that appear responsible for its limited generality. We demonstrate that generality might be achieved through a primary interaction zone that optimizes robust minimal catalaphile-catalaphore interactions, and a secondary zone where only minor contributing interactions occur.

This insight provides a framework for understanding and developing highly general catalysts for asymmetric reactions. The principle of minimal catalaphile recognition enables a broad substrate scope, while conserved transition state differentiation ensures consistent enantioselectivity across diverse substrates. We believe that this strategy can be applied to other catalytic systems wherein a minimal catalaphile can be identified to realize high generality. ^[88] The suite of quantitative tools applied here provide a blueprint for rational design that could accelerate the discovery of next-generation catalytic systems. The mechanistic basis for catalyst generality may further empower rational catalyst design, lead to more efficient general catalyst discovery, and ultimately support generality-enabled chemical synthesis.

Supplementary Material

Supporting Information. This material is available free of charge: Complete experimental details (PDF); NMR and HPLC trace data (PDF); X-ray (CIF); code for generality calculations including database of results from literature and this work (ZIP)