London Dispersion Interactions Rather than Steric Hindrance Determine the Enantioselectivity of the Corey–Bakshi–Shibata Reduction

Abstract The well‐known Corey–Bakshi–Shibata (CBS) reduction is a powerful method for the asymmetric synthesis of alcohols from prochiral ketones, often featuring high yields and excellent selectivities. While steric repulsion has been regarded as the key director of the observed high enantioselectivity for many years, we show that London dispersion (LD) interactions are at least as important for enantiodiscrimination. We exemplify this through a combination of detailed computational and experimental studies for a series of modified CBS catalysts equipped with dispersion energy donors (DEDs) in the catalysts and the substrates. Our results demonstrate that attractive LD interactions between the catalyst and the substrate, rather than steric repulsion, determine the selectivity. As a key outcome of our study, we were able to improve the catalyst design for some challenging CBS reductions.


Introduction
Thed etailed understanding of reaction mechanisms and the origin of enantioselectivity is essential for successful catalyst design. Enantioselectivity imparted by chiral, smallmolecule catalysts is rationalized typically by preferential steric destabilization derived from the repulsive part of the van der Waals (vdW) potential because it can be readily understood and taught with hard-sphere classical mechanics models.I nc ontrast, London Dispersion (LD), the attractive part of the vdW potential, [1] is often neglected in mechanistic considerations and for catalyst design. [2] However,f or ad etailed understanding of ag iven catalytic system, all interactions must be considered, even though we are just learning how to conceptualize this for reaction planning. [3] Fortunately, modern computational techniques like dispersion corrected density functional theory (DFT) now allow adetailed analysis of all factors contributing to transition state stabilization for am uch better understanding of catalyst design. [4] Here we chose the Corey-Bakshi-Shibata (CBS) reduction, which is av ersatile method for the enantioselective reduction of prochiral ketones by oxazaborolidines (OXB), achieving high selectivities and yields [5] to demonstrated that all steric factors,a ttraction and repulsion, have to be taken into account to arrive at abalanced description of the mechanism and to design new,more selective catalysts.
Coreyswidely accepted mechanistic model bases stereoselection exclusively on steric repulsion between the boron substituent R on the catalyst and the large R L and small R S substituents of the ketone in as ix-membered boat-like transition state (Scheme 1). [5f, 6] With this model, one can qualitatively predict the enantiofacial discrimination of numerous substrates.H owever,t his model of steric destabilization does not offer as atisfying explanation for the selectivity and reactivity of some substrates.F or example, the reduction of trichloroacetophenone predominantly generates the (R)-enantiomer. [6a] This implies that the large phenyl group (R L )faces the boron substituent in the favored transition state,which is in contrast to Coreysstandard model depicted in Scheme 1. In the reduction of cyclopropyl isopropyl ketone (1-cyclopropyl-2-methylpropan-1-one) one would assume poor selectivity,b ecause both substituents are similar in steric size.N onetheless,t he reduction delivers the (R)-enantiomer with as electivity of 91 % ee. [5f] Similarly, ah igh ee (81 %) was also found for p-methoxy-p'-nitrobenzophenone with two groups of similar size. [5f] In these two cases,t he cyclopropyl substituent and the p-methoxyphenyl group act as R L ,r espectively,t hereby demonstrating that Scheme 1. CBS reduction of acetophenone and proposed transition structures for hydride transfer,f avoringt he (R)-product on the basis of minimizing the steric repulsion between "R"and "R L ". other factors must also play an important role in the transition state structure.F urthermore,d espite bearing bulky groups, there are substrates that do not deliver high selectivity,e .g., unbranched aliphatic ketones. [7] There are several reports on the stereoselection of the CBS reduction trying to shed light on the origin of its enantioselectivity.I n1 993 Liotta et al. used the MNDO semiempirical approach to suggest that the reduction is more likely to occur via achair-like transition state.Inaddition, the carbinol phenyl substituents of the catalyst are required to lie parallel to the R L substituent to minimize steric repulsion. [8] Meyer et al. investigated the role of steric repulsion in the transition structures of the reduction by determining kinetic isotope effects (KIE), as the C-D bond is effectively shorter than the C À Hbond, resulting in inverse 2 HKIEs for reactions in which steric repulsion increases in the transition structure. [9] They concluded that Coreyss teric reasoning is too simplistic, because in the reduction of acetophenone the chair-like transition state prevails,with the boron substituent only playing am inor role. [10] In ar ecent theoretical study, Lachtar et al. suggested that the origin of the enantioselectivity for the oxazaborolidine catalyzed reduction of ketimines can be traced back to noncovalent interactions in the preferred transition structure. [11] However,b yr eplacing the phenyl groups of the catalyst by hydrogens,t hey used ac omputationally reduced model that neglects major parts of these key noncovalent interactions.F urthermore,t heir B3LYP/6-31G(d,p) computations do not include dispersion corrections.
Herein, we aim at bringing together experimental and computational studies geared towards understanding ar eaction whose stereochemical outcome was classically interpreted as being derived solely on the basis of steric repulsion. We demonstrate that am ore detailed and hence more powerful mechanistic reasoning emerges when all interactions are taken into account and we gauge the role of attractive LD stabilization in this particular reaction.

Results and Discussion
This section is organized in three parts.F irst, ac omprehensive computational investigation of the various noncovalent interactions (NCI) in the transition structures provides acontemporary view of the origin of the enantioselectivity in the CBS reduction. Then we show how these insights help in the design of new catalysts to improve enantioselectivity, especially for some challenging substrates.Finally,weprovide an experimental validation of our improved understanding of catalyst design.

Reconsidering Steric Effects
None of the previously reported computational mechanistic studies include LD corrections, [8][9][10][11][12] which are needed to strike ap roper balance between repulsive and attractive noncovalent interactions.A st his is the very concept of an "equilibrium structure", we set out to determine the role LD plays in the CBS reduction. We first computed the reaction pathway for the reduction of acetophenone using ac omparison of B3LYP vs.B3LYP-D3(BJ) with appropriate basis sets and solvent inclusion (see Computational Details below);the difference should provide ag ood estimate of the role dispersion plays (Figure 1). Ad etailed potential energy surface (PES) of the complete reaction pathway and higherlevel single-point energy computations with DLPNO-CCSD-(T) [13] (domain-based local pair natural orbital CCSD(T)) for the key step are provided in the Supporting Information (Supporting Information, Figure S1).
In this simplified PES,w es tart with the catalyst, the reducing agent, and acetophenone as reference 1.The hydride transfer determining enantioselectivity occurs via two diastereomeric transition structures.IfLDisnottaken into account (color-coded in gray), the transition structures TS1' ' S and TS1' ' R are found to be very high in free energy exhibiting barriers of 29.8 kcal mol À1 and 31.7 kcal mol À1 ,r espectively. These energy barriers are too high for such afast reaction at 25 8 8C. After inclusion of LD (color-coded in black), the relative energies of the transition structures TS1 R and TS1 S are notably lower with barriers of only 13.7 kcal mol À1 and 15.7 kcal mol À1 ,r espectively,w hich is much more reasonable for ac atalyzed reaction that proceeds quickly at room temperature.T he calculated enantioselectivity of the reduction, which is expressed in the energy difference between the transition structures (DDG°)for hydride transfer,isÀ2.0 kcal mol À1 and thereby consistent with previously published experimental results (À2.2 kcal mol À1 ). [5b] We computed the transition structures in solvent within the limitations of an SCRF model. Theenergy difference of TS1 R and TS1 S in gas phase is 4.0 kcal mol À1 ,w hile it is 2.0 kcal mol À1 in THF.A s expected, the LD interactions are attenuated by the interaction with the solvent but, more importantly,t hey do not vanish. Catalyst regeneration and release of the boronate 7 is exergonic by À15.1 kcal mol À1 .For comparison, we also added DLPNO-CCSD(T)/cc-pVTZ single-point energies on the DFT-optimized geometries (and ZPVE corrections) of TS1 S and TS1 R ,w hich are 7.7 kcal mol À1 and 10.8 kcal mol À1 , respectively.T his indicates that the dispersion-corrected energy barrier is more reasonable and that more complete inclusion of electron correlation effects emphasize the importance of LD.
Acloser look at the geometries of the transition structures TS1 R and TS1 S reveals chair-like conformations (Figure 2), which are 3.8 kcal mol À1 lower in energy,than the corresponding boat-like conformations suggested by Corey (Supporting Information, Figure S1). Thereby,t he catalyst binds to the ketone at the lone pair facing the small substituent (R S ) anti to the electron-rich substituent as it is also described in Coreys model. NCI plots indicate some differences of the noncovalent interactions between catalyst and substrate in the two transition structure conformations. [14] Contrary to Coreys model no steric destabilization (repulsion is color-coded red in the NCI plot) by hydrogen-hydrogen contacts can be found in less favored TS1 S .I nf act, the bond distances of around 2.5 in the preferred transition structure TS1 R lead to stabilizing s-p LD interactions [15] between acetophenone and the phenyl groups of the catalyst, as visualized by the green areas in the NCI plot. Additionally,the methyl substituent of the substrate interacts favorably with the boron substituent of the catalyst. In the less favored transition structure TS1 S the long distance between substituents of substrate and catalyst prevent optimal interactions.These computations suggest LD interactions to be important for enantiodiscrimination. We additionally employed LD potential maps developed by Pollice and Chen to visualize these LD interactions (Supporting Information, Figure S2). [16] These confirm the conclusions drawn from the qualitative NCI analysis.
To examine the general effect of LD interactions on the enantioselectivity,t hree literature examples with various substrates and catalysts were also studied ( Table 1). In the reduction of cyclohexyl methyl ketone (1-cyclohexylethanone,e ntry 2) the moderate enantioselectivity (85 % ee)i s likely due to decreased LD interactions of the cyclohexyl with the phenyl group of the catalyst. [5a] Similarly,r eplacing the phenyl substituents with as pirocyclopentyl group in the catalyst leads to diminished enantioselectivity of 67 % ee. [17] This implies that the LD interactions of the phenyl groups in the catalyst are crucial for enantioselectivity.N ote that the selectivities obtained from our LD corrected computations (B3LYP-D3(BJ)) are in better agreement with the experimental ee values (Table 1) than the uncorrected values. [5a,17] SAPT0 (Symmetry Adapted Perturbation Theory) was employed to analyze the different energetic contributions of the interactions between substrate and catalyst in the transition structures (Figure 3). [18] Thec omponents include electrostatics,e xchange,i nduction, and LD energies.T he electrostatic term arises from the large Coulomb interactions between the Lewis acid and Lewis base sites (carbonyl and boryl as well as amino and boryl groups). In the transition structures,e lectrostatics and induction dominate the interactions but they are counterbalanced by al arge exchange term (i.e., Pauli repulsion), indicating significant steric repulsion. However,t he larger exchange energy in favored TS1 R disagrees with Coreysm odel, in which the larger exchange term should favor TS1 S .Therefore,the selectivity is  not determined by steric repulsion (alone). Although LD is as mall part of the total interaction energy,i td ecisively contributes.T he LD energy preference for TS1 R is 5.9 kcal mol À1 ,w hich is in good agreement with the experimentally observed high selectivity. [5a,b] Thus,our computational results strongly suggest that LD interactions between catalyst and substrate also determine the enantioselectivity.

Improving Catalyst Design with Dispersion Energy Donors
Based on our new understanding of the origin of enantioselectivity in the CBS reduction, we hypothesized that higher enantioselectivity can be achieved by modifying the catalyst with dispersion energy donors (DEDs) [2c, 15b] that enable favorable substrate-catalyst interactions through increasing polarizability.T he catalyst modifications involve on the one hand the boron substituent and the carbinol substituents on the other.
First, we investigated the effect of the substituent at boron ( Table 2);D EDs including i Pr, t Bu, Cy,C H 2 Cy, c-C 5 H 9 were employed. In the reduction of cyclohexyl ketone LD interactions are the highest using CH 2 Cy with a DDG°of 3.5 kcal mol À1 (entry 1). As expected for large and highly polarizable groups,Cyand t Bu should also deliver significant enantiodiscrimination (entries 2and 4). For tert-butyl methyl ketone,only the t Bu and Me substituents show high selectivity (entries 8and 9). However,our experimental results demonstrate that the selectivity does not change much as compared to the original catalyst when using CH 2 Cy and Cy groups (Supporting Information, Figure S3). This implies that the interactions between substrate and the substituent at boron on the catalyst only have as ubtle effect on the enantioselectivity.
Thevariation of carbinol substituents of the catalyst leads to much larger changes of enantioselectivity.R eplacing the phenyl groups with aliphatic DEDs,e .g.,M e, i Pr, and n Bu show comparable or even reduced selectivity relative to the original catalyst with its unsubstituted phenyl groups. [6b] Thei ntroduction of DEDs in the meta-positions of the aryl groups of the catalyst results in comparably high or even slightly higher enantioselectivities relative to the original catalyst in the reduction of acetophenone ( Figure 4).
Thei ntermolecular stabilization by all-meta substitution in dispersion-driven systems has been demonstrated recently, e.g.,i nt he stabilization of molecular dimers [19] and in the catalytic hydroamination of olefins. [3d] Here we also find that DEDs in meta-aryl positions provide additional attractive interactions with, e.g.,the ethyl substituent of 2-butanone,as indicated in the NCI plots ( Figure 5). Again, more stabilizing interactions (À2.8 kcal mol À1 )b etween aryl groups on the catalyst and the ethyl group in the substrate are found in the preferred TS R .
To explore the general potential of aryl-substituted catalysts computationally,t he 3,5-t Bu 2 Ph catalyst was computed in the reduction of various substrates ( Figure 6). The modified catalyst shows comparable or improved enantioselectivity,e specially for substrates yielding low enantioselectivity with the original catalyst, e.g.,2 -butanone and cyclohexyl methyl ketone.F or 2-butanone, DDG°improves from 0.6 to 2.8 kcal mol À1 ,implying atheoretical change in ee from 47 %t o98%.

Experimental Validation
To examine our computational predictions,weperformed an experimental validation employing various catalysts and substrates.W estarted with comparing the effects of changing the substituents in the catalyst at the carbinol and boron positions.W ec ompared the original CBS catalyst to three modified versions in the reduction of three ketones bearing aromatic,b ranched or unbranched alkyl substituents.W e employed Coreyss tandard protocol using 10 mol %o f catalyst, 1.1 equivalents of reducing agent in THF at 50 8 8C for 1.5 h ( Figure 7). [6b] We chose slightly elevated temperatures,a sf or borane reductions there is an increase in   Table S2). [20] This resulted in nearly quantitative yields.A se xpected, by changing the carbinol substituent of the catalyst from hydrogen to phenyl, the selectivity increases. These initial findings support our proposal that the carbinol substituents are key for enantiofacial discrimination due to LD interactions with the substrate.Furthermore,when replacing the hydrogen at boron with ap henyl group,w e observe ad ecrease in enantioselectivity.T his disagrees with the steric repulsion hypothesis,w here (R)-selectivity should improve with increasing steric size of the substituent at boron. [6b] Consistent with our computations (Figures 2and 3), we relate this to stabilizing LD interactions between substrate and the phenyl group at boron in less favored TS S (Figure 2). This does not exclude the notion that catalysts bearing aphenyl group at boron are probably weaker Lewis acids and less effective,a nd the lower selectivity might also be aresult of amore prominent unselective background reaction.
Next, we experimentally probed the computationally predicted effects of the carbinol substituents shown in Figure 4. We employed catalysts with aryl groups bearing additional DEDs to check whether the LD interactions increase and thereby increase enantioselectivity (Figures 8  and 9). In all cases,at508 8Cafter 1.5 hthe reduction results in near quantitative yields.Inthe reduction of cyclohexyl methyl ketone all modified catalysts achieve higher selectivities due to additional LD stabilizations.A sL Dt hrough the metasubstituent seems to be maximized at methyl already,w e decided to test the 4-OMe-3,5-Me 2 Ph and 4-OMe-3,5-t Bu 2 Ph catalysts that should be even more polarizable due to electron donation from the methoxy group.I ndeed, the best selectiv-   ities were achieved with the 3,5-Me 2 Ph-and 4-OMe-3,5-Me 2 Ph catalysts.T he selectivities with 3,5-i Pr 2 Ph, 3,5-t Bu 2 Ph and 4-OMe-3,5-t Bu 2 Ph are slightly lower, as the substituents are getting too bulky for cyclohexyl ketone;t he results are similar for the reduction of 2-heptanone.W hile the overall selectivity is lower due to entropic penalty of the linear alkyl chain, [21] we observed the best selectivities with 3,5-Me 2 Ph and 4-OMe-3,5-Me 2 Ph. These experimental results fit the qualitative expectation from our improved model and confirm our computations of Figure 4, as the 3,5-Me 2 Ph catalyst was the also the best computed catalyst for cyclohexyl ketone.
Thetrend of increasing selectivities in the experiments is consistent with our computational results but the absolute selectivities differ.While computations suggest an increase of selectivity of up to 98 % ee by introducing DEDs,t he experimentally observed improvement is more moderate. There may be several reasons for this.First, the mechanism is more complex than accounted for in the computations. Second, the reduction features also an on-negligible background reaction with BH 3 ,w hich could have al arger impact on the modified catalysts,a st he activity of these is lower compared to the original catalyst, because the EDG in the carbinol position reduces the Lewis acidity at boron. [22] Third, the computations are not accurate enough as compared to highest level ab initio computations.This is certainly true but we are pleased to see trends with predictability leading to improved catalyst performance,i np articular, for the most challenging of substrates.
In order to also provide some counter examples,w e included catalysts with 3,5-(CF 3 ) 2 Ph and C 6 F 5 carbinol substituents and found that the fluorinated catalysts are much less selective (Figure 10), despite their high steric demand and significant activation of the boron Lewis acid. These results are in accord with the computed values (Supporting Information, Table S1) and suggest weakened LD interactions between the fluorinated aryl groups and the substrates,aswedecrease attractive s-p interactions through the strongly electron withdrawing fluorine substituents. [15c,23]   Moreover,t his is consistent with the reduction of special substrates like pentafluorobenzophenone and p-methoxy-pnitro-benzophenone (Table 3). In all cases,t he enantiomer maximizing the attractive s-p interactions in the TS is favored. Also,t he s-p interaction of aC 6 F 5 substituent to ap henyl ring is lower than the s-p of two phenyl groups (entry 1). [15c] For p-methoxy-p-nitro-benzophenone the electron-rich aryl group (4-OMe-C 6 H 4 )p rovides the stronger interaction with the catalyst (entries 2a nd 3). In the case of cyclopropyl isopropyl ketone (entry 4), the catalyst interacts with the more p-electron rich cyclopropyl substituent favoring the R enantiomer.W ith trichloroacetophenone (entries 5 and 6), the more attractive s-p interaction results in R selectivity,aschloromethanes strongly interact with aryl rings due to LD (Supporting Information, Table S4). [23a] Figure 11 summarizes the results for the reductions of av ariety of ketones with our best modified catalyst. These data also indicate that enantioselectivities increase with the computed polarizabilities per volume a/V,resulting in ahigher interaction energy of the substituent with the catalyst (Figure 12). [15a] These findings are confirmed by ac ompetitive rate analysis in the reduction of 2-pentanone and tert-butyl methyl ketone in the same reaction flask (Figure 13). After the given reaction times,w et ook as mall sample of the reaction mixture,q uenched it in citric acid, and analyzed the conversion after work up.W echose tert-butyl methyl ketone and 2-pentanone,a st hey are reduced with quite different selectivities ( Figure 11). While tert-butyl methyl ketone is   reduced in excellent selectivity,t he reaction should proceed slowly because the neopentyl position is traditionally viewed as highly sterically encumbered, thereby hampering the attack of nucleophiles. [25] Remarkably,t he consumption of tert-butyl methyl ketone occurs at ahigher reaction rate than that of 2-pentanone.Computations show that the complex of the catalyst with tert-butyl methyl ketone has asimilar energy as that with 2-pentanone (Supporting Information , Figure S6). This implies that electrostatic interactions with catalyst are similar, and electronic effects are not significant. We conclude that stabilizing LD interactions in the TS are at work because the rate of the sterically more demanding substrate is higher.

Conclusion
We present ac ombined computational and experimental exploration of the origin of the enantioselectivities in CBS reductions.C ontrary to the current hypothesis that makes steric repulsion solely responsible for enantioselection, our computations reveal the presence of stabilizing noncovalent interactions in the hydride transfer transition structure.N CI plots qualitatively aided in visualizing these intermolecular interactions particularly between the substrate and the phenyl substituents of the catalyst. Aq uantitative SAPT analysis suggests that LD interactions tip the balance in favor of attractive noncovalent steric interactions to achieve high enantioselectivity.
Catalysts bearing DEDs in the meta-positions of the aryl groups increase the enantioselectivity for different substrates, as confirmed computationally and experimentally.M ore polarizable substrates lead to stronger LD interactions with the catalyst and therefore to higher enantioselectivities as well as faster reaction rates.Ifsteric repulsion were the chief selector,t he rates would diminish with increasing selectiv- Figure 11. Reductionsofvarious substrates employing our new modified CBS catalyst. Figure 12. Increasing polarizability per volume a/V of the substrates typically leads to higher enantioselectivites in the reduction with agiven catalyst (4-OMe-3,5-Me 2 Ph). Computedv alues of polarizability (a)and volume (V) of the corresponding substituent. Level of theory:PBE0/augcc-pVDZ//B3LYP-D3(BJ)/6-311G(d,p). [24] Figure 13. Ketone-to-alcohol ratios in the competitive reduction of tertbutyl ketone and 2-pentanoneafter the given reaction times.
ity-the opposite is the case.Even though the overall positive effect on enantioselectivity through the addition of DEDs is moderate,i tp rovides strong evidence that the success of the CBS reduction is due to an excellent balance of attractive and repulsive steric interactions,w ith LD interactions being key to rationalizing the experimental findings.O ur study therefore emphasizes that attractive LD interactions can and should be used as am odern catalyst design principle.

Computational Methods
All computations were performed with the Gaussian16 or ORCA [26] program suite.G eometries were optimized with dispersion corrections [DFT-D3 [4a] (BJ) [4b] ]a nd without dispersion corrections in conjunction with the B3LYP functional combining a6 -311G(d,p) basis set. Vibrational frequencies were computed for each optimized structure to verify the stationary structures as minima or saddle points.S olvent effects were included by single-point energy computations with the SMD model [27] at the same level as for the optimized geometry.H igher level single-point energies were computed by the domain-based local pair natural orbital CCSD(T) (labeled DLPNO-CCSD(T)) method with ac c-pVTZ basis set. TheS APT analysis was performed at the SAPT0/jun-ccpvdz level of theory on the optimized geometries [28] utilizing the PSI4 code. [29] Conformational analyses were performed using xtb (version 5.8) employing GFN2-xTB by simulated annealing molecular dynamics (MD) simulations in the gas phase. [30] All energies discussed are Gibbs free relative energies at 298.15 Ka nd 1atm in kcal mol À1 unless noted otherwise.E ffects of zero-point vibrational energy (ZPVE) corrections are included.