Experimental Lineage and Computational Analysis of a General Aminoxyl-based Oxidation Catalyst: Generality from Substrate-Specific Interactions

Abstract

Traditionally, catalyst optimization in asymmetric catalysis is approached as an iterative, heuristic-based process where point modifications of a hit catalyst are tested against one (or a small number of) model substrates. While optimization to high levels of selectivity is sometimes successful, catalyst generality with respect to substrate scope (i.e. its ability to deliver a diverse set of target products with consistently high levels of selectivity) is more elusive. This work describes and models computationally a successful peptide catalyst optimization campaign carried out on a verifiably diverse set of substrates, which delivered a highly selective and general chiral catalyst. The success of every generational improvement of the catalyst design is now rationalized with atomistic resolution by ab initio modeling of the individual substrate that mostly benefitted from that generational improvement. Structural and topological insights about the non-covalent interaction networks orchestrating both the catalyst conformation and the substrate-catalyst interactions were dissected, culminating on the underpinnings of the optimized catalyst’s generality. Surprisingly, and significantly, the generality of high selectivity for many substrates was found to be consistent with alternative, and indeed substrate-specific interactions, suggesting that functional generality need not be a result of mechanistic homology.

Graphical Abstract

Introduction

Asymmetric catalysis is now a central activity in the heart of organic synthesis.¹ Catalyst optimization, however, remains a bottleneck due to selectivity or generality challenges, especially as substrate diversity increases.² Optimization towards highly differentiated diastereomeric transition states, as required for high selectivity, can often conflict with structural features of diverse substrates.³ Accordingly, the pursuit of generality for chiral catalysts has been a daunting challenge,² but momentum in the field is growing as generality-based catalyst optimization approaches are now emerging.^{2, 4-13}

We recently reported a peptide-based catalyst that achieves high enantioselectivities in an oxoammonium-mediated desymmetrization of diols.⁸ The best catalyst was found to be pentameric structure P7 (Figure 1a) that possesses many of the desirable features of peptide-based catalysts.^14,15 Among these, the capacity for multivalent non-covalent interactions and conformational flexibility, as a matter of speculation, were thought to be potentially operative a priori. Accordingly, our screening campaign began with well-known conformationally biased β-turn peptides,¹⁶ applied previously in many catalytic reactions across a variety of mechanistic manifolds.¹⁴ Yet, the optimization campaign rapidly diverged to unprecedented catalyst structures. Described herein is (a) the iterative decision process that resulted in catalyst evolution, not discussed in our previous manuscript,⁸ and (b) an expansive retrospective computational study that illuminates possible mechanistic underpinnings of the catalyst’s generality. This analysis resulted in the conclusion that the observed substrate generality involved substrate-specific interactions, which are not homologous across the scope – a highly unexpected outcome.

Figure 1.

a) Catalyst evolution: from hit (P1) to most general catalyst (P7). b) Optimization conditions and substrate set for the meso-diol desymmetrization.

Results and Discussion

Catalyst optimization campaign.

Catalyst hit identification within the Miller group often relies on use of prior optimized peptide scaffolds and their adaptation to new systems. Due to both their historical reliability and their tendency for preorganization, β-turn biased peptides are often surveyed first. In the study of meso-diol desymmetrization, a new aminoxyl-containing catalytic residue, Azc, was utilized (3-methyl-2-azaadamantane carboxylate N-oxyl). Screening against dibenzyl ketal S1, a peptide with the Azc residue appended to the N-terminal “n”-position¹⁷ (P1, Table 1) demonstrated high selectivity (75% ee) and was selected for further optimization efforts. Three key features were identified for a series of single point modifications to answer significant questions for each element. First, the optimal diastereomeric configuration of the catalyst was examined - the four amino acids comprising peptide P1 were in the LDXL configuration, where X denotes an achiral residue. Second, the β-turn biasing elements at n+2 and n+3 were modulated. Finally, a subsequent evaluation of both esters and amides was surveyed at the peptide C-terminus.

Table 1.

Initial rounds of catalyst optimization to catalyst P4.

Optimization against a single substrate failed to provide a catalyst that was general across a range of diverse substrates, although some promising results were obtained (P1/S1, 75% ee, Table 1). Thus, a multi-substrate screening approach was undertaken, wherein each new catalyst prepared was screened against all (or most) of a set of 15 algorithmically selected diverse prochiral diols. The aim of this approach was to optimize towards general catalyst-substrate interactions as opposed to the specific interactions that may be unearthed using a single optimization substrate. This strategy led to larger amounts of data to further inform catalyst optimization.

The LDXD diastereomeric configuration of catalyst P2 proved to be slightly more selective than the initial P1 hit (Table 1), with small but noticeable increases in selectivity across many substrates (S3, S4, S7, S10, S13, S14, and S15). The other two diastereomeric configurations (P1a and P1b) failed to provide high levels of selectivity across the selection of diols - although it must be said that several individual pairings do indeed show increased selectivity (both P1a and P1b show increased selectivity for S8). Individual results such as these highlight the advantages of a generality-based optimization approach in circumventing local, narrow selectivity maxima in favor of optimizing towards higher, more global ones.

Analyzing the β-turn biasing elements (^DPro-Aib), it was quickly noted that substitution of Aib for the cyclopropanebearing Acpc (P1c) or Gly (P1d) failed to provide enhanced selectivity. Curiously, however, substitution of ^DPro for ^DPip (P1e) showed comparable or better selectivities than P1. At the time, this observation failed to elicit deeper thinking into the conformational changes this homologation would impart. Both Pro and Pip are well known for inducing turn motifs in small peptides; however, their propensity to nucleate one specific secondary structure over another is markedly different. Proline has been shown, in particular with an adjacent disubstituted residue linked to the C-terminus, to nucleate β-turns with Pro at the i+1 position.¹⁸ In contrast, Pip demonstrates a strong tendency to nucleate β-turns with itself at i+2 position. The fact that Pip behaved comparably in this system should have immediately alerted us that the conformation of the peptide in some bound state may not lie as the previously depicted β-turn, but perhaps one where Pro and Pip occupy the i+2 position relative to the β-turn.

Finally, analysis of the C-terminus caused rather significant selectivity perturbations for what we considered a minor change in the linear sequence. Modulation of the N,N-dimethyl C-terminus to a methyl ester (P1f; Table 1) slightly diminished the median selectivity. Conversely, modification to the C-terminal N-methyl secondary amide (P3), led to significant enhancements in enantioselectivity for individual substrates (S1, S5, S6). Although the median ee did not improve significantly, the individual increases led us to believe that the C-terminus would have an unexpectedly central role to play in peptide optimization. Encouragingly, we also observed similar increases utilizing C-terminal N-phenyl (P3a), N-α-methyl benzyl (P3b) and N-isobutyl (P3c) caps, with these peptides demonstrating modest increases in selectivity for a broader range of substrates as well (S8, S9, S10, S14). Although we found enhanced selectivity for C-termini beyond methyl, analog synthesis proved to be a challenge for many peptides, requiring alternative synthetic routes.

With a breadth of information on the catalyst behavior, the next phase of catalyst optimization was selection of a new starting point for single point modifications. The selectivity-unlocking nature of P3 was encouraging; if the contacts leading to high selectivity for S1, S5 and S6 could be generalized to more substrates, this campaign would prove highly successful. Subsequently, we determined that the C-terminal secondary amide would be used in conjunction with the changes that resulted in both P1e and P2. Retaining the C-terminal secondary amide, in combination with swapping ^DPro for ^DPip did not lead to substantial enhancements in individual selectivities but provided quantifiable increases for nearly every substrate examined (P1e). These incremental improvements, however, paled in comparison to altering the configuration of the sequence from LDXL to LDXD, providing P4 from P3. Dramatic increases in selectivity were observed for nearly every substrate analyzed, with an increase in median ee from 31% to 69%.

Further optimization: more specialized residues.

While crystallization of catalyst P3 proved unattainable, we were able to obtain a single crystal X-Ray structure of analog P3a. This catalyst adopted a 3₁₀-helix conformation in the solid state, which also features the expected β-turn motif nucleated by Pro-Aib. It was then hypothesized that this conformation would be retained in peptide P4, with only an inversion of stereochemistry at n+4 present relative to P3. At the time, we hypothesized that this 3₁₀-helix might be the active conformation of the substrate-bound peptide and would be responsible for the high levels of selectivity. Modifications for catalyst optimization were then selected on the basis of their potential to 1) stabilize the 3₁₀-helix conformation or 2) destabilize the β-hairpin conformation. Two parameters were then prioritized to be tuned in the next round of optimization starting from P4: the backbone Pro-Aib segment and the C-terminal cap. It must be noted that the absolute stereochemistry of the catalysts was inverted at this point, due to increased availability of “^L” amino acids. The configuration of the n+1 residue remained fixed as ^Dtert-leucine (^DTle).

The n+3 position was not surveyed at this stage, due to diminished selectivities when replacing Aib for other disubstituted residues in the initial optimization (vide supra). However, the n+2 residue was analyzed with mixed success. The substituted analogue α-methyl Pro (P4a) which has shown success in several other campaigns, dramatically reduced selectivity in substrates that showed improvement with P4 (Table 2). Utilizing the previously beneficial Pip (P5) provided a noticeable boost in enantioselectivity overall with several substrates being particularly responsive to this change (S4, S7, and S10). Substitution for the benzan-nulated tetrahydroisoquinoline (Tic, P4b) also showed improved selectivity over P4 (Pro), however less advantageous than P5 (Pip), which was chosen as the replacement for Pro going forward.

Table 2.

Optimization of catalyst P4 towards P5+6.

Attention then turned to tuning the C-terminus. We hypothesized that increasing the acidity of the NH group would lead to increased selectivity, as we had seen in the earlier round of optimization (P3a compared to P3; Table 1). An enhanced H-bond donor could potentially reinforce a rigidified catalyst conformation. It was also noticed that a slightly increased steric profile seemed to enhance the selectivity (P3b compared to P3; Table 1) therefore, several peptides with larger, more electron deficient C-terminal secondary amide caps were prepared.

Two aniline-capped peptides were successfully prepared, with a simple N-phenyl cap (P4c) providing good boosts in enantioselectivity for several previously recalcitrant substrates (S10, S11). Increasing steric bulk at the cap also provided an increase in selectivity, as an N-adamantyl capped peptide (P4d) provided similarly enhanced selectivities for S10 and S11. Most notably, the combination of both a more electron deficient and slightly larger cap (N-trifluoroethyl capped P6) caused the greatest improvement of the series, providing an increase in selectivity for every single substrate examined. Largely, we attributed the increase in selectivity to an increased electron withdrawing character as opposed to a significative steric contribution. C-terminal caps with similar steric profiles to P6 but with simple aliphatic (P4e) or aromatic (P4f) substituents failed to provide substantive increases in selectivity, even over P4.

Finally, the n+4 residue was analyzed in relation to P4; an array of residues resembling phenylalanine were assayed. Direct analogs to Phe were analyzed, with O-benzyl tyrosine (P4g) providing a mix of increased and decreased selectivities, and diphenylalanine (P4h), 1-napthylalanine (P4i), and 2-napthylalanine (P4j) diminishing selectivity in almost every case. Large aliphatic substituents similarly provided no increase in selectivity: 3-hydroxyadamantyl glycine (P4k) and cyclohexylalanine (P4l) substituted peptides diminished selectivity for all but one peptide-substrate combination.

Final optimization: fine-tuning.

With two major advances for selectivity in hand, the substitutions leading from P4 to both P5 (Pro to Pip) and P6 (methyl to trifluoroethyl C-terminus) were combined to provide an intermediate peptide P5+6 that would be fine-tuned to provide the final catalyst. P5+6 expectedly provides excellent selectivities across a diverse substrate set (Table 2), and only minor structural modifications were considered further. Namely, two positions were considered at this stage: the identity of the polyfluorinated C-terminal cap and the residue at the n+4 position. First, pentafluoropropyl- (P5+6a), heptafluorobutyl- (P5+6b) as well as the sterically encumbered (1-trifluoromethyl)cyclopropyl- (P5+6c) amines were appended to the C-terminus, however, only minor changes were observed (Table 3). Peptide P5+6a showed a slight increase in performance against substrate S10, and therefore the pentafluoropropyl tail was selected as the most advantageous C-terminus. Finally, tuning the n+4 residue was attempted once again. Previous attempts at optimizing this position led to highly selective catalysts only when a phenylalanine or a derivative thereof (Nal, Tyr, etc.) was used. Unfortunately, substitution with homophenylalanine P5+6d, O-benyzl tyrosine P5+6e, or diphenylalanine P5+6f failed to provide any improvement over P5+6. In a final attempt to increase selectivity, a single derivative of P5+6a was prepared with biphenylalanine at n+4. This peptide, P7, fortunately, provided slight enhancements across all substrates tested and was selected as the final, fully optimized catalyst. P7 provided high selectivities for substrates beyond the screening set, as even prochiral 1,3- and 1,5- diols were efficiently desymmetrized. This concluded the optimization campaign and led to our publication on enantioselective meso-diol oxidation.

Table 3.

Final rounds of optimization.

Throughout the later stages of optimization, obtaining conformational information via single crystal X-ray analysis became a feature sought largely to confirm the hypothesis of a 3₁₀ helix motif being responsible for selectivity. To our surprise, upon obtaining single crystal X-ray analysis of P7, we observed a double β-turn motif, with the most prominent turn nucleated with the catalytic residue (n) at the i-position. Computational analysis of peptides P1-P7 revealed an unexpected prominence of this β-turn in earlier generations of the catalyst, prompting a fuller computational investigation as to what conformational changes and alterations to interaction networks were elicited by residue-by-residue changes.

Computational Methods

Computational modeling of enantioselective catalysis is known to be a complicated enterprise for different factors, including catalyst flexibility, additivity of numerous weak interactions, and of course the computational cost of performing numerous calculations at a high level of theory.^19-24 Nevertheless, very significant progress has been made in the cost-to-accuracy ratio in recent years, with accurate geometrical and energetic descriptions of relatively small organic molecules coming at a computational cost that allows larger-scale modeling efforts to be carried out with reasonable time and resources.²² We present here a self-consistent and rational basis for the success of the catalyst optimization described above, encompassing the ground state free energy characterization of ~6000 different structures across the optimization campaign.

The computational workflow was designed to rationalize what stereoelectronic effects enabled the development of the highly general catalyst for meso-diols desymmetrization. Our goal was to understand what role each generational modification of the catalyst played towards reaching the final structure (P7), ideally with atomic-level resolution of the interaction mode landscape. In this workflow, pairs of peptide catalysts were analyzed for a single substrate. A substrate was chosen for each peptide pair such that the generational advancement of the catalyst would induce the greatest improvement in selectivity for that substrate (as measured by enantiomeric excess), with a few substrate choices also affected by computational cost considerations. Once we defined the pairs, peptide-substrate covalent adducts were modeled as four possible stereoisomers, as alcohol binding to the oxoammonium moiety generates two new chiral elements: the first arising from the choice of which of the two meso alcohol groups is bound to the oxoammonium moiety, and the second from which oxoammonium face was used for coordination.

Modeling of these covalent adducts consisted in a multi-level computational workflow to characterize their conformational landscape (see Supporting Information for detailed computational procedures). In short, an initial metadynamics-based, unrestricted conformational search via CREST²⁵ was followed by layers of geometrical and similarity refinement via XTB²⁶, FIRECODE²⁷ and ORCA^{28, 29}, up to the ωB97M-V³⁰/def2-TVZPP³¹/SMD³²(CH₂Cl₂)//R²SCAN-3c³³/CPCM³⁴(CH₂Cl₂) level, from which free energy data was composed.

At this composite level, the DFT equilibrium geometries within 3 kcal/mol from the most stable (G_233K) for each oxoammonium alkoxide complex were compared to their stereoisomeric ensembles as well as the ensembles of the same substrate with the adjacent generation of catalyst. These interaction modes were then used to qualitatively trace the origin of the success of each new generation of catalysts through ground state conformational analysis. Mindful of the Curtin-Hammett principle,³⁵ the choice to use ground states as a proxy for transition state conformational landscapes was adopted due to the large scale of the modeling effort and the desire to avoid the computation of Hessian matrices for thousands of structures. The use of this heuristic is also justified by the very small geometrical difference between the oxoammonium alkoxide complexes and the following Cope elimination transition states, which causes the relative conformational energetics to be dominated by the backbone hydrogen bond networks of the catalyst and substrate. Moreover, electroanalytical independent rate measurements show that the ΔG^‡_obs for both diol S9 and cyclohexylmethanol (as a proxy mono-ol) is very similar, while the one-pot competition experiment between the two substrates shows a selectivity ratio of 26.8 (Figure S6). This pictures a mechanistic scenario where the energetic differentiation of the transition state structure reflects the one of the oxoammonium alkoxide ground state complexes (ΔG° ~ ΔΔG^‡_obs) and offers experimental support for the modeling approach using ground states.

To further validate the use of ground states as a proxy for transition states, full transition state modeling of a minimal system (Azc-NHMe-1,4-butanediol adduct) was carried out at the same level of theory and provided the insight that the transition states for the Cope-type elimination^36-38 (Figure 1b) are geometrically very similar to the relative ground states (<1 Å RMSD, root-mean-squared deviation of atomic positions), and that energetic trends for ground states conformers closely reflect transition state ones, allowing a qualitative analysis of activation topologies based on ground state interaction modes. The same modeling also revealed that, for the model system, all kinetically relevant transition states feature activation modes where the second alcohol interacts with another Lewis basic moiety, either the bound alcohol oxygen atom or the carbonyl oxygen of the Azc residue. This observation also supports the rationale of a two-point interaction mode of the substrate as having an influence on rate acceleration as we forwarded in our previous work. This is also in agreement with the observation that two-point interaction ground states are dominant in the modeling of pentameric peptide catalysts.

P1 to P3 and P2 to P4: NMe₂ to NHMe.

Initial hit catalyst P1 was obtained via a short screen of peptides and showed inconsistent performance across the diol test set. For this reason, we expected a complex conformational space featuring a multitude of topologically distinct activation modes. Indeed, the final ensemble for interactions of S6 (desymmetrized in only 36% ee) featured many accessible states (six within 2 kcal/mol for the major enantiomer), and interestingly the most stable ones all feature a secondary structure where Pro is not at the i+1 position of a β-turn (as expected at the time of its synthesis), but at the i+2, with ^DTle occupying i+1 (Figure 2a, 2b, 2c). This structural feature will be maintained across all catalysts throughout the optimization campaign. Modifying the terminal NMe₂ group to NHMe gave catalyst P3, which was able to desymmetrize substrate S6 in 69% ee (+0.44 kcal/mol improvement in ΔΔG^‡_obs). Tracing back this effect to the most stable interaction modes for these two catalysts, it became evident that the additional hydrogen bond present for the NHMe end cap imparted a significative improvement in the catalyst preorganization. Inspection of the most stable interaction modes showed how this change disfavors an activation mode for the minor enantiomer where the second alcohol moiety coordinates to the top face of the catalyst rather than the bottom, which will ultimately become the dominant one in the following generations.

Figure 2.

C-terminus cap effect: generational improvement from P1 to P3 (a, b c) and from P2 to P4 (d, e, f). Free energy of activation differences (ΔΔG^‡_obs) calculated from experimental selectivities (−40 °C). a) Overlay of the divergent activation modes for the minor diastereomer of P1 and P3 with S6, with the latter featuring a spurious mode from the catalyst top face. b) Diol S6 used in the modeling. c) Structural differences of P1 and P3, and their effect on the stereoselectivity in desymmetrizing S6. d) Overlay of the divergent activation modes for the major diastereomer of P2 and P4 with S9. The additional hydrogen bond donor of P4 tightens the second alcohol non-covalent interactions and reduces the complexity of the conformational space. e) Diol S9 used in the modeling. f) Structural differences of P2 and P4, and their effect on the stereoselectivity in desymmetrizing S9.

The same point change was even more beneficial on P2, a diastereomer of P1 featuring the opposite absolute configuration of Phe (Figure 2d, 2e, 2f). Substrate S9 improved from a 0% ee with P2 to a remarkable 69% ee with P4 by the effect of this modification alone (+0.79 kcal/mol improvement in ΔΔG^‡_obs). The effect of the additional hydrogen bond donor was traced back to imparting a higher degree of preorganization to the interaction mode leading to the major enantiomer of the product, by virtue of an extra hydrogen bond of the catalyst backbone with itself. This modification not only resolved conformational landscape for the interaction of S9 with the catalyst (major enantiomer, from three to only one state within 2 kcal/mol from the most stable), but also tightened the internal network of hydrogen bonds between the catalyst and the free alcohol moiety of the substrate (Figure 2d).

P3 to P4 (S4): ^DPhe to ^LPhe.

Inversion of stereochemistry at the i+3 position proved pivotal at this stage, and afforded P4. The great success of P4 across the diols set arose from combining an end cap modification (NMe₂ to NHMe, as from P1 to P3) with the inversion of configuration of Phe (^DPhe to ^LPhe, P3 to P4). We assessed the influence of the phenylalanine configuration modeling substrate S4, which was desymmetrized in just 1% ee with P3 but improved drastically to 77% ee with P4 (+0.93 kcal/mol improvement in ΔΔG^‡_obs, Figure 3). Notably, while the interaction modes leading to the major enantiomer of the product feature identical hydrogen bond topology, P4 features a much tighter binding pocket compared to P3, which instead adopts a relatively flat conformation. In addition, the accessible conformational space is also significatively reduced (major enantiomer complex, 3 kcal/mol window: P3: 14 conformers, P4: 3 conformers).

Figure 3.

Three major catalyst improvements: from P3 to P4 (a), from P4 to P5 (b) and from P5 to P7 (c). Free energy of activation differences (ΔΔG^‡_obs) calculated from experimental selectivities (−40 °C). a) Overlay of the activation modes for the major diastereomer of P3 and P4 with S4. While P3 (^DPhe) features a relatively flat double β-turn, P4 (^LPhe) shows a more enclosed geometry, that interacts with the substrate more tightly. This is also reflected in the shorter hydrogen bond distances with the free alcohol of S4. b) Overlay of the activation modes for the major diastereomer of P4 and P5 with S10. While P4 (Pro) is more flexible and the favored conformation of the adduct gives up the enclosed, non-planar geometry of the double β-turn in exchange for substrate π-stacking, P5 (^LPip) features a stricter geometrical enforcement of the out-of-plane secondary structure and retains a conformation similar to the ones previously observed for other substrates. This is also reflected in the shorter hydrogen bond distances with the free alcohol of S10. b) Overlay of the activation modes for the major diastereomer of P5 and P7 with S12. The larger polarizable surface of Bip (P7) relative to Phe (P5) induces a tighter fit of the pedant substituent thanks to attractive dispersive interactions with the adamantly core of the Azc residue. This is also reflected in a shorter NH…OH hydrogen bond distances with the free alcohol of S10.

P4 to P5: Pro to Pip.

The next significative advancement in the generality of P4 was obtained replacing Pro with Pip (P5, Figure 3b). Analyzing the interactions of S10, which improved from 48% ee with P4 to 81% ee with P5 (+0.56 kcal/mol improvement in ΔΔG^‡_obs), we noted that the increased propensity of Pip to nucleate β-turns at the i+2 position was also reflected in an increased rigidity of the catalyst structure against substrate-induced deformations. This can also be observed in the flattening (and shortening) of the β-turn hydrogen bond. In the case of S10, we attribute the success of the point change to the additional hydrogen bond between the phenylalanine N–H bond and the nonreacting −OH group present in the complex leading to the major enantiomer of the product (Figure 3b). This is not present in any accessible state of the P4/S10 complex. We believe this is not present for P4 due to the increased flexibility of Pro relative to Pip that allows the “flattening” of the catalyst geometry in exchange for the π-stacking interaction with aromatic substrate S10.

P5 to P7: Phe to Bip.

Notwithstanding the obtainment of a quite general catalyst for diol desymmetrization (P5, up to 95% ee, 74% median ee), further optimization was still undertaken. Notably, not only was further improvement achieved, but we were amazed to observe that some point changes still proved beneficial across the entirety of the test set. The last of such modifications was the replacement of Phe for Bip (phenylalanine to biphenylalanine, P5 to P7, Figure 3c). Substrate S12 was desymmetrized in 46% ee with P5 and in 79% ee with P7 (+0.53 kcal/mol improvement in ΔΔG^‡_obs). Analysis of the interactions leading to the major enantiomer of the product highlighted how Bip induces subtle but important geometrical alterations of the binding pocket. Chiefly, it does so by shielding a region of space in front of the Azc residue by virtue of the association of two dispersion energy-donating groups: the biphenyl and adamantly moieties. Moreover, numerous C–H bonds on the substrate are within 3.5 Å from carbon atoms of the Bip residue, engaging in additional weak interactions. This pocket deformation also causes a significative contraction and a slight elongation, respectively, of the two key hydrogen bonds involving the free alcohol moiety of substrate S12. It is noteworthy that while the interaction modes of both P5 and P7 with S12 show three conformations within 3 kcal/mol (G_233K), only one of these which is transition state-like (i.e. features the catalyst’s N⁺–O⁻ moiety next to an alcohol C-H bond).

Interactions of different diols with P7.

After observing the convergence of many substrates to a unified interaction topology across the different catalyst generations, we wanted to assess if multiple substrates that are effectively desymmetrized by the optimized catalyst P7 also favor the same interaction network, in addition to the already modeled S12. Interestingly, we observed different preferred interaction modes for different substrates (Figure 4). Certain constrained, cyclic diols like S4 and S15 bind with a slight alteration of this main mode, trading one of the hydrogen bonds (alcohol (O)-H to amide (C)=O) for a better overall geometrical fit within the pocket (Figure 4a, “incomplete” mode). This remarkable effect causes conformations of bound diols S4 and S15 that have one less hydrogen bond relative to the “main” mode to be five to six kcal/mol more stable than the latter. An “alternate” mode could also be observed as the preferred for the minor diastereomer of P7 with S10. Yet, most rigid substrates retain high levels of selectivity regardless of their preferred interaction network (96% ee for S4, 90% ee for S10). Evidently, this activation mode promiscuity is not necessarily tied to a degradation in selectivity, as long as the coexistence of multiple modes retains a large energetic splitting between the diastereomeric states (ΔG° = 7.71 kcal/mol for S4 and 5.98 kcal/mol for S10 between the most stable diastereomeric TS-like ground state conformers).

Figure 4.

Analysis of the preferred interaction modes of P7 with different substrates, from computational modeling. a) Schematic representation of the different modes. b) Most stable modes for each diastereomer of the reported substrates with P7 (“major” oxoammonium alkoxide diastereomer leading to the major observed enantiomer of the lactone product in the original work⁸ and vice versa).

On the contrary, P7 performs poorly on substrate S15 (26% ee). In the original report, a rationale was put forward discussing the low buried volume of this substrate inside the binding pocket.⁸ The atomistic modeling corroborates and complements this analysis by also unveiling how this substrate features the presence of a previously unobserved “unfolded” interaction mode, where the catalyst second β-turn hydrogen bond is lost (Figure 4). Substrate S15 also showed less energetically differentiated states (ΔG° = 2.36 kcal/mol between the most stable diastereomeric TS-like ground state conformers). It is noteworthy that while all the modeled complexes feature the same number of hydrogen bonds in the major and minor diastereomers, substrate S4 features one more hydrogen bond in the major. This is consistent with the substrate’s superior performance in the reaction (P7/S4, 96% ee, see Figure S7 for three-dimensional superposition of the interaction modes). While a quantitative energy decomposition analysis^39-41 would tease out the main components of ΔG° for different substrate-catalyst complexes, we limit this work to a more qualitative description of the non-covalent interaction networks. Moreover, our analysis emphasizes the likely contribution of multiple members of a transition state ensemble to overall selectivity, and the feature that different interaction modes within the ensemble may contribute to varying degrees for different substrates.

Application to regioselective diol oxidations.

Recently, the Lin and Miller groups⁴² reported the use of an oxo-analog of P7 (OAzcP7, Figure 5) to enable the regiocontrolled oxidation of unsymmetrical diols. In this reaction, the highly preorganized peptidyl template enabled the selective oxidation of the more hindered alcohol group over the more accessible one, overriding intrinsic steric biases. The authors rationalized this selectivity with the ability of OAzcP7 to selectively sequester the more accessible primary alcohol within the sterically demanding hydrogen bond peptidyl framework, with the more hindered alcohol available to form a more stable oxoammonium alkoxide adduct (Figure 5a, blue). This heuristic is in alignment with ground state computational modeling we conducted on representative substrate D2 (Figure 5b), in analogy with the rest of the catalyst-substrate pairings of this work. The energetic difference between the best conformer of each interaction mode is 1.86 kcal/mol, which is in qualitative agreement with the experimental selectivity obtained (ΔG^‡_{calc., 228 K} = 1.86 kcal/mol, ΔΔG^‡_{obs., 228 K} ~ 2.0 kcal/mol). The favored interaction mode also features a hydrogen bond network involving the unbound alcohol group of the substrate that is strengthened relative to the unfavored regioisomeric complex (Figure 5a, blue structure featuring a tighter hydrogen bond network relative to orange).

Figure 5.

Application of the rigid, preorganized peptidic template of P7 to the regioselective oxidation of primary alcohols. The secondary structure of the catalyst (OAzcP7) is highly sensitive to steric effects, and the preferential non-covalent interaction of the least hindered alcohol with the peptide backbone results in the formation of the covalent oxoammonium adduct of the most hindered alcohol, resulting in contrasteric oxidation. a) Overlay of the best interaction modes for the different regioisomers of diol adducts. b) Substrate D2 used in the modeling. c) Structural differences of the regioisomeric adducts, and effect on the regioselectivity of the oxidation. Energetic values are computational relative energies of regioisomeric oxoammonium alkoxide ground states (ΔG°, −40 °C).

Conclusions

In conclusion, we described our approach to the experimental optimization of an asymmetric meso-diol oxidation catalyst towards high generality. We then performed a large scale DFT modeling of many catalyst-substrate pairs (~6000 structures) to unearth the underpinnings of each generational catalyst improvement, as well as the drivers behind the selectivity of the optimized catalyst towards different substrates. Highlights of the modeling include clarification of the conformational evolution throughout the catalyst generations; divergence of catalytically relevant conformations from those observed in several catalysts’ crystal structures and unexpected diversity of catalyst-substrate interactions that might not have been expected with a “general” catalyst. Our results show that catalyst flexibility and the coexistence of different catalyst-substrate interaction modes for different substrates are not at odds with the goals of enantioselectivity and substrate generality. In fact, an alternative perspective on the system may relate to the concept of “confinement.”⁴³ In fact, enclosed environments seem to facilitate the energetic splitting between multiple diastereomeric or topologically different states regardless of their exact nature, which can in fact vary across substrates, while still contributing to high selectivity.

Supplementary Material

A Supporting Information document is available free of charge on the ACS Publications website containing experimenta details, computational methodologies and energies for final DFT ensembles (PDF format). The .xyz files containing the computational coordinates for these ensembles can be found on the Github repository for this work.

ABBREVIATIONS

Acpc

Aminocyclopropane carboxylic acid

Aib

Aminoisobutyric acid

Bip

Biphenylalanine

Pip

Pipecolic acid

Phe

Phenylalanine

Pro

Proline

RMSD

root-mean-squared deviation (of atomic positions, in Å)

Tic

Tetrahydroquinoline carboxylic acid.