Enantioselective Sulfonimidamide Acylation via a Cinchona Alkaloid-Catalyzed Desymmetrization: Scope, Data Science, and Mechanistic Investigation

Abstract

Methods to access chiral sulfur(VI) pharmacophores are of interest in medicinal and synthetic chemistry. We report the desymmetrization of unprotected sulfonimidamides via asymmetric acylation with a cinchona-phosphinate catalyst. The desired products are formed in excellent yield and enantioselectivity with no observed bis-acylation. A data-science-driven approach to substrate scope evaluation was coupled to high throughput experimentation (HTE) to facilitate statistical modeling in order to inform mechanistic studies. Reaction kinetics, catalyst structural studies, and density functional theory (DFT) transition state analysis elucidated the turnover-limiting step to be the collapse of the tetrahedral intermediate and provided key insights into the catalyst-substrate structure–activity relationships responsible for the origin of the enantioselectivity. This study offers a reliable method for accessing enantioenriched sulfonimidamides to propel their application as pharmacophores and serves as an example of the mechanistic insight that can be gleaned from integrating data science and traditional physical organic techniques.

Graphical Abstract

INTRODUCTION

Sulfur(VI) functional groups such as sulfonamides and sulfones have a rich history of impacting human life through medicinal and agrochemical applications. Their corresponding, largely untapped, chiral sulfur(VI) pharmacophores have recently emerged as targets of interest in medicinal chemistry.¹ As examples, sulfoximines,² sulfondiimines,³ and sulfonimidamides⁴ (Figure 1A) offer attractive properties for pharmaceutical applications: chirality, stability, solubility, desirable physicochemical characteristics, hydrogen bonding propensity, and multiple sites to incorporate structural diversity.⁵ Specifically, sulfonimidamides have been postulated to be a chiral bioisostere to carboxylic acids⁶ and sulfonamides,^7,8 common moieties in many drug candidates. However, this motif has yet to be widely deployed due to limited commercial availability and a dearth of practical asymmetric synthetic methods.

Figure 1.

(A) Medicinally relevant molecules containing sulfur(VI) functional groups. (B) Previously reported desymmetrization of sulfonimidamides. (C) This work.

Willis and co-workers have reported an enantioselective alkylation of protected sulfonimidamides using cinchona alkaloid-derived phase-transfer catalysts.⁹ Additionally, we have recently disclosed an enantioselective Pd-catalyzed aryl-carbonylation of sulfonimidamides with aryl and heteroaryl iodides (Figure 1B).¹⁰ This reaction leveraged the rapid tautomerization of the imido and amido nitrogen atoms on unprotected sulfonimidamide starting materials, which provided a unique opportunity to desymmetrize the prochiral nitrogen atoms via dynamic kinetic resolution. With the desire to expand the number of enantioselective transformations and gain convenient access to the chiral sulfonimidamide pharmacophore, we investigated direct desymmetrization through acylation with a labile electrophile and a cinchona alkaloid for nucleophilic catalysis. This resolution of the two unprotected nitrogen nucleophiles on a single sulfur chiral center is challenging due to the tendency toward oligomerization and the lack of obvious structural differences required for effective catalyst recognition. Additionally, the nitrogens in both the starting material and monoacylated product are nucleophilic, which can result in competitive uncatalyzed reactions and the ability to form bis-acylated products (as observed in reported aryl-carbonylation reactions).^10,11 Related enantioselective acylation of nitrogen nucleophiles are sparse but include both enzymatic and small molecule catalysis, highlighted by independent efforts of the De and Seidel,¹² Fu and co-workers,¹³ and Bode and co-workers¹⁴ teams.

By combining chemical intuition and high throughput experimentation (HTE), we have identified an enantioselective acylation of unprotected sulfonimidamides to access enantiomerically enriched N-trifluoroacetyl-sulfonimidamides via cinchona-phosphinate catalysis (Figure 1C). Applying this unique catalyst scaffold¹⁵ for an unreported reaction inspired a data science-informed evaluation of the substrate scope, which was leveraged to facilitate statistical modeling to drive traditional mechanistic studies. These investigations, which included reaction kinetics, catalyst structural studies, and density functional theory (DFT) transition state analysis, provided compelling evidence for the turnover-limiting collapse of the tetrahedral intermediate¹⁶ and key insights into the origin of enantioselectivity.

RESULTS AND DISCUSSION

Discovery of the Active Catalyst and Optimization.

A preliminary exploration of an asymmetric N-acylation was performed on a model benzenesulfonimidamide substrate 1a with 2,2,2-trifluoroethyl trifluoroacetate 2a as the electrophile. Not only is 2a commercially available but it was initially selected to provide a labile group to allow facile further functionalization (vide infra). Cinchona-based organocatalysts emerged as a promising class for this transformation. Derivatives consisting of ester, amide, sulfonamide, urea, thiourea, squaramide, and phosphorus-based functional groups were evaluated via HTE (Figure 2A) (see the SI for HTE details). Several of the cinchona derivatives demonstrated encouraging results in catalyzing this enantioselective transformation. The lead catalyst, furnishing the desired product 4aa, was derived from a rarely utilized cinchona-phosphinate derivative.¹⁵ Notably, this catalyst displayed excellent chemoselectivity, with no detectable bis-acylated product 5aa observed. In contrast, bis-acylation was a competitive process for many of the other cinchona catalyst derivatives evaluated (up to 32% 5aa observed in the crude reactions). Laboratory scale validation of catalyst 3a confirmed the complete conversion of 1a with no detectable amount of bis-acylated product 5aa and high enantioselectivity of 4aa (95:5 er) (Figure 2B). We hypothesized that the relationship between the basic, nucleophilic sites of the phosphinate catalyst 3 quinuclidine fragment, 1a, and 2a is key to the observed chemoselectivity. While we have not specifically evaluated the pK_a of the cinchona-phosphinate catalyst experimentally, calculations indicate that the pK_a of the amino group in 4aa is lower by 1.6 units compared to starting sulfonimidamide 1a (7.6 vs 9.2, respectively), making the second acylation leading to 5aa more challenging.

Figure 2.

(A) HTE screening of cinchona-derived catalysts to identify leads. (B) Optimization of cinchona-phosphinate catalysts and reaction solvent. Standard reaction conditions were as follows: 1a (0.16 mmol, 1.0 equiv), 2a (0.48 mmol, 3.0 equiv), and catalyst 3 (0.032 mmol, 20 mol %) in THF (3.2 mL, 0.05 M) at −35 °C. ^aRelative product area % as determined by SFC analysis. ^bEnantiomeric ratio (er) of product in the crude reaction mixture, as determined by chiral SFC analysis.

We explored accessible phosphinate derivatives of the initial catalyst hit 3a in an attempt to further improve the catalytic asymmetric reaction (Figure 2B). Electron-poor (3b), electron-rich (3c), and sterically hindered (3d) aryl derivatives did not influence the catalyst selectivity. Conversely, the relative substituent size of alkyl derivatives significantly impacted enantioselectivity (3e-3g, 69:31 to 96:4 er), with the optimal catalyst being the dicyclohexylphosphinate derivative 3e (96:4 er). Modifications of the C6′-position on the quinoline fragment of 3a were also evaluated. The protio derivative 3h gave only moderate enantioselectivity (81:19 er) despite its distal location from the identified key phosphinate group. Installation of bulkier i-PrO or Ph substituents (3i and 3j, respectively) had little impact on enantioselectivity (95:5 to 94:6 er, respectively). Two other aryl substituents gave unpredictable results, with the electron-withdrawing 3,5-bis(trifluoromethyl)phenyl substituent on catalyst 3k adversely impacting selectivity (85:15 er), while little change to enantioselectivity was observed for electron-rich 4-methoxyphenyl derivative 3l (94:6 er). Notably, the diversity of results coupled with the challenges in understanding the origin of selectivity provided the basis for the mechanistic analysis presented below.

Additional optimization of the reaction parameters, including temperature, concentration, catalyst loading, and electrophile equivalents, was pursued for the top-performing phosphinate catalyst 3e (see the SI for details). Such changes to the standard reaction conditions reported in Figure 2B minimally affected enantioselectivity (93:7 to 97:3 er) or conversion (>98%). Conversely, the solvent was found to impact the reaction (Figure 2B), as only moderate selectivity (76:24 to 80:20 er) was observed in nonethereal media, and reduced conversion (7%) was found in DCM. THF was determined to be the optimal solvent.

Substrate Scope.

We recently reported using chemical space visualization to qualitatively sample a diverse range of sulfonimidamide substrates, which was also deployed here (Figure 3B).¹⁰ Examples were evaluated on bench-scale using the top-performing cinchona-phosphinate catalyst 3e (Figure 3A). Exploration of the scope began by testing various substituted aryl sulfonimidamides. Different halogens placed at the para- (1b), meta- (1c), and ortho- (1f) positions afforded the corresponding acylated products 4ba (98%, 96:4 er), 4ca (71%, 99:1 er), and 4fa (92%, 97:3 er) in high yields and excellent enantioselectivity. While 2-methylbenzenesulfo-nimidamide 1e (87%, 94:6 er) reacted similarly to 1a, increasing the electron density (1d, 1i) and/or size of the aryl substituents (1g, 1h) resulted in a slight decrease in the enantioselectivity (81:19 to 92:8 er). Next, alkyl sulfonimidamides were tested under the reaction conditions. Benzyl sulfonimidamide 1j reacted sluggishly with only moderate selectivity (55%, 67:33 er), whereas allyl 1k (99%, 79:21 er) and cyclopropyl 1l (98%, 91:9 er) substituted analogs afforded higher yields and improved enantioselectivity. The reaction was found to be compatible with various heteroaryl substrates such as pyridine 1m (87%, 99:1 er), pyrimidine 1n (89%, 99:1 er), furan 1o (81%, 99:1 er), 5-chlorothiophene 1p (95%, 99:1 er), and benzofuran 1q (65%, 95:5 er). However, 7-chlorothieno[3,2-b]pyridine 1r (14%, 96:4 er), benzothiazole 1s (18%, 94:6 er), and 1-methylimidazole 1t (7% conversion, 92:8 er) were formed in low yields, albeit with high enantioselectivity. Thus, these three substrates (1r–1t) were not included in subsequent HTE screening campaigns designed for statistical modeling to elucidate structure-enantioselectivity relationships (vide infra). Notably, no bis-acylated product was observed for any of the sulfonimidamide substrates tested. The absolute configuration of compounds 4aa, 4la, 4ma, and 4qa were determined to be (S) by single-crystal X-ray diffraction analysis.

Figure 3.

Exploration of the sulfonimidamide substrate scope. (A) Sulfonimidamide substrate scope. Reaction conditions: substrate 1 (1.00 mmol, 1.0 equiv), 2a (3.00 mmol, 3.0 equiv), and catalyst 3e (0.20 mmol, 20 mol %) in THF (20.0 mL, 0.05 M) at −35 °C. ^aIsolated yield after purification. ^ber of the isolated product, as determined by chiral SFC analysis. ^cAbsolute configuration was determined by X-ray crystallographic analysis. (B) PCA of synthetically feasible sulfonimidamides (43.5% of the total variance depicted with two principal components), selected substrates are labeled, and black crosses indicate substrates screened using HTE.

Electrophile Scope.

The electrophile was also investigated using the optimized reaction conditions for 1a with catalyst 3e (Figure 4). Variations of the leaving group (purple sphere) were investigated and compared to the model electrophile 2,2,2-trifluoroethyl trifluoroacetate 2a. Changing to the more electron-rich leaving group of ethyl trifluoroacetate 2b resulted in no conversion, which is presumably due to the poorer leaving propensity of this group. Consistent with this observation, the reactivity is re-established upon incorporating a better-leaving group, N-trifluoroacetoxy succinimide (2c); however, this reaction resulted in no selectivity. Interestingly, selectivity correlates with leaving group ability, wherein phenyl trifluoroacetate 2d and S-ethyl trifluoroethanethioate 2e demonstrate modest improvements in the enantioselectivity. Additional electrophiles were also evaluated to explore the range of acyl groups that can be incorporated using this method (green sphere). The use of 2,2,2-trifluoroethyl difluoroacetate 2f gave excellent selectivity (99:1 er) but with reduced conversion (48%) compared to 2a. More electron-rich electrophiles 2g, 2h, 2i, and 2j resulted in no conversion. Ultimately, 2,2,2-trifluoroethyl trifluoroacetate 2a was determined to be the optimal electrophile for this transformation.

Figure 4.

Exploration of the electrophile scope. Reaction conditions: 1a (0.16 mmol, 1.0 equiv), electrophile 2 (0.48 mmol, 3.0 equiv), and catalyst 3e (0.032 mmol, 20 mol %) in THF (3.2 mL, 0.05 M) at −35 °C. ^aRelative product area % as determined by UPLC-MS/SFC analysis. ^ber of the product in the crude reaction mixture, as determined by chiral SFC analysis.

Applications of the Methodology.

To demonstrate the utility of this method, we prepared chiral sulfonimidamide analogs of the antitumor agent tasisulam^17–19 and the antiplatelet drug elinogrel²⁰ (Figure 5). Coupling 4pa with 2,4-dichlorobenzoic acid 6 using a standard 1-[bis-(dimethylamino)methylene]-1H-1,2,3-triazolo[4,5-b]-pyridinium 3-oxide hexafluorophosphate (HATU)/N,N-diisopropylethylamine (DIPEA) protocol followed by TFA deprotection using ammonia produced the tasisulam-sulfonimidamide analog 7 in 74% yield with no erosion in enantioselectivity, demonstrating the stability of these sulfonimidamide products to a highly enabling reaction. The elinogrel-sulfonimidamide analog 9 was synthesized in a three-step, one-pot procedure. Substrate 4pa was reacted with p-nitrophenyl chloroformate to produce the corresponding carbamate, which reacted with 8 to form unsymmetrical urea. Lastly, TFA deprotection gave the product in 42% yield and retained stereochemical integrity at 98:2 er.

Figure 5.

Application of this methodology to synthesize chiral sulfonimidamide analogs of sulfonamide-containing drug candidates.

Statistical Modeling for the Substrate–Catalyst Relationship.

Having evaluated the scope of the reaction using a singular catalyst, the nature of the structure-enantioselectivity relationship as it relates to both the substrate and catalyst was not obvious. We were especially interested in the role of the cinchona-phosphinate in catalyzing a highly enantioselective reaction for an unusual reactant. We deployed an arsenal of mechanistic and data-driven techniques to probe how the catalyst operates and what substrate features are required to achieve high enantioselectivity.

As the first stage, we designed an HTE screening campaign that assessed a combinatorial matrix of catalysts and substrates. The resultant quantity and diversity of data would allow for the construction of enantioselectivity correlations with substrate and catalyst structural features through statistical modeling.^21,22 By assessing a combinatorial matrix, we hypothesized that most general features required for effective asymmetric catalysis would be revealed and serve to seed further computational and mechanistic studies that are required to elucidate the mechanism of this reaction. Ten representative catalysts that sampled two points of modulation, the phosphinate and quinoline substituents, were selected as they showcased a diversity of responses in the initial optimization campaign. These catalysts were evaluated against the 17 sulfonimidamide substrates that provided reasonable yields in our initial scope evaluation (Figure 6A). The 170 reactions exhibited a well-distributed and wide range of observed selectivities ranging from 53:47 to 99:1 er, and 96% of the reactions achieved ≥85% conversion. The matrix approach highlights that each catalyst reacts uniquely with every substrate and vice versa, positioning this data set well for statistical modeling to decipher catalyst-substrate interactions that impact selectivity.

Figure 6.

Statistical modeling of the substrate catalyst relationship. (A) Employment of a combinatorial matrix approach for HTE screening of enantioselectivity for 17 sulfonimidamide substrates against 10 cinchona-phosphinate catalysts. (B) Data were curated for statistical model validation. (C) Multivariate linear regression (MLR) model for enantioselectivity built from Boltzmann averaged descriptors and depictions of the molecular descriptors included in the model. Computational method: M06-2X/def2-TZVP//B3LYP-D3BJ/6-31G(d,p).

The data set was partitioned into three groups (Figure 6B). The first was an external validation set that was left out until the model building and selection were complete. This set included 17 data points (black boxes in Figure 6A, black stars in Figure 6C), representing each substrate once and each catalyst once or twice. Of the remaining 153 data points, a Kennard Stone algorithm was used to define 50% as the training set (solid gray circles, Figure 6C) and the other 50% as the test set (teal crosses, Figure 6C). Forward-stepwise linear regression was employed to build models for the observed selectivity (ΔΔG^‡) using DFT-derived molecular descriptors of both the catalyst and the substrate.²¹ Molecular mechanics conformational searches were used to generate conformer ensembles for both the substrates and the catalysts. After geometry optimization, each conformer’s steric and electronic properties were collected to determine ensemble-dependent descriptors capturing molecular flexibility (see the SI for details).²³ The resultant models built from these descriptors were parsed based on their cross-validation and test-set statistical measures.²¹ Averaged leave-one-substrate-out and leave-one-catalyst-out mean absolute error (MAE) values were used as an additional validation technique to ensure the model was not heavily biased to specific substrates or catalysts.²⁴ The presented model (Figure 6C) was selected to allow the descriptors and their coefficients to be interpreted to provide insight into the reaction mechanism; however, numerous models with similar statistics (presented in the SI) were determined. Averaged leave-one-substrate-out (0.224 kcal/mol) and leave-one-catalyst-out (0.223 kcal/mol) MAEs for this model indicate that in generalizing to the complete sampling of the substrate and catalyst members (training MAE = 0.185 kcal/mol), some precision is sacrificed. However, the external validation set was predicted similarly to the test set (test MAE = 0.177 kcal/mol, external validation MAE = 0.183 kcal/mol), demonstrating that the model can predict reactions of unseen substrate-catalyst combinations.

The four-parameter model consists of two terms for both the substrate and the catalyst. The natural bond orbital (NBO) partial charge of the substituent at the C6′-position on the quinoline classified the different substituents installed at this site (orange, Figure 6C). The more electron-rich (larger negative value) alkoxy-substituted catalysts give higher selectivities than the aryl or protio analogs. The buried volume collected within a 5 Å radius of C9 reads out the catalyst pocket established by the phosphinate substitution (red, Figure 6C). The larger buried volume provided by the cyclohexyl substituents (3e) restricts the pocket size, resulting in the most selective catalyst. The sulfonimidamide substrate is described by its molar volume (gray; Figure 6C). Although not independently correlated to the observed selectivity, we hypothesize that this parameter accounts for substrate-catalyst steric matching in the selectivity-determining transition state (vide infra). Lastly, the ¹³C NMR chemical shift of the α-carbon (green, Figure 6C) serves to classify the aryl from the pseudoaliphatic (i.e., benzyl, allyl, and cyclopropyl) substrates, which exhibit lower selectivities.

Experimental Mechanistic Studies.

Having established quantitative structure-enantioselectivity relationships, we began kinetic investigations of the mechanism of catalysis to inform subsequent computational models. We selected in situ infrared (IR) spectroscopy as our analytical technique of choice to monitor the reaction’s progress. Both the consumption of the starting electrophile 2a and the formation of product 4aa could be monitored via resolved peaks in the carbonyl region of the IR spectra (Figure 7A). From the outset of this study, we suspected that an uncatalyzed background acylation reaction might play an impactful role in the process and be competitive with cinchona-catalyzed acylation. To gain further insight into the rate of the uncatalyzed background reaction, the reaction was performed and monitored in the absence of a catalyst (black, Figure 7B). A 10-fold difference in the initial rates was observed between the 3e-catalyzed and uncatalyzed processes. Therefore, it is presumed that an aggregate rate of the cinchona-catalyzed reaction with a contribution from an uncatalyzed background reaction contributes to the reaction profile and observed selectivity. Importantly, precisely quantifying the exact background reaction contribution for each catalyst-substrate combination is challenging, but we estimate a ~5–30% background rate depending on the catalyst-substrate combination. This limits the selectivity ceiling to ~97:3 er for most examples.

Figure 7.

Experimental mechanistic investigation results. (A) Carbonyl region from a ReactIR waterfall plot for the reaction of 1a and 2a with catalyst 3e under the standard catalytic conditions, showing the conversion of 2a and the formation of 4aa. (B) Reaction profiles for a series of experiments using varying initial concentrations of catalyst 3e, the quinoline N-oxide catalyst derivative 3o, and the uncatalyzed background reaction. (C) Catalyst derivatives used to elucidate active site(s). Reaction conditions: 1a (0.16 mmol, 1.0 equiv), 2a (0.48 mmol, 3.0 equiv), and catalyst 3 (0.032 mmol, 20 mol %) in THF (3.2 mL, 0.05 M) at −35 °C. ^aRelative product area % determined by SFC analysis. ^ber of the product in the crude reaction mixture was determined by chiral SFC analysis.

Reaction progress kinetic analysis (RPKA)^25,26 and variable time normalization analysis (VTNA)²⁷ were performed on the reaction using catalyst 3e and revealed a positive, near first-order dependence on the concentration of sulfonimidamide substrate 1a, electrophile 2a, and catalyst 3e, allowing us to approximate the rate law as eq 1. The rate law is consistent with the formation of a ternary complex in the transition state.²⁸ However, due to the possible engagement of three different functional groups of the cinchona-phosphinate catalyst (i.e., phosphinate, quinuclidine, and/or quinoline groups) in contacts, we designed a set of experiments to elucidate the precise roles these fragments play in catalysis (Figure 7C).

$$\text{rate}\cong k{\left[\text{sulfonimidamide}\right]}^1{\left[\text{electrophile}\right]}^1{\left[\text{catalyst}\right]}^1$$ (1)

Modifications to the phosphinate group of the top-performing cinchona-phosphinate catalyst 3e demonstrated that substituting cyclohexyl with methyl substituents (3g) resulted in decreased enantioselectivity. This structural analog allowed us to compare with the thiophosphinate derivative 3m, which resulted in a racemic product. This provides evidence that the phosphinate group is critical for selective catalysis. The next step was to differentiate the importance of basic nitrogens on the quinoline and quinuclidine sites. This was accomplished by selectively synthesizing the N-oxides at each nitrogen and sequentially determining how each impacts asymmetric catalysis. Excess m-CPBA in CHCl₃ was used to oxidize both nitrogen sites (3p), and compound 3o was provided by subsequent selective quinuclidine N-oxide reduction (NaHSO₃, HCl/acetone), while oxidation using urea hydrogen peroxide in 2,2,2-trifluoroethanol afforded 3n selectively. The quinuclidine N-oxide (3n) significantly impacted both the conversion and the enantioselectivity of the transformation; this was further confirmed by a similar result observed for the doubly blocked catalyst (3p). In contrast, the quinoline N-oxide (3o) did not impact the reaction outcome compared to the parent catalyst 3e. This result was corroborated by kinetic experiments, where we observed a perfect overlay between the quinoline N-oxide (3o) and parent catalyst (3e) reaction profiles under standard reaction conditions (orange and teal, Figure 7B). These studies provided crucial insights into developing a computational model for asymmetric catalysis.

Computational Mechanistic Studies.

Once the quinuclidine and phosphinate fragments were determined as crucial for the success of the enantioselective reaction, catalyst-substrate complex structures were evaluated by docking the sulfonimidamide substrate in catalyst pockets proximate to the two groups. Choosing catalyst 3e as a model catalyst, we found that the quinuclidine nitrogen and the phosphinate oxygen atoms can act as hydrogen-bond acceptors (Figure 8A). The cinchona-phosphinate catalyst efficiently forms doubly hydrogen-bound structures, anchoring the sulfonimidamide substrate (Figure 8B). During this binding event, we note that there does not seem to be any differentiation between imido versus amido nitrogen binding or preferential binding of one of the rapidly tautomerizing 1a enantiomers.²⁹ We did, however, proceed with these structures as a starting point to evaluate the addition of 2,2,2-trifluoroethyl trifluoroacetate 2a.

Figure 8.

Substrate-catalyst hydrogen-bonding interactions. (A) Hydrogen-bond acceptor sites on cinchona-phosphinate catalyst 3e. (B) The schematic model and illustrative optimized structure showing the catalyst-substrate binding mode with two hydrogen-bonds.

In light of our kinetic results, we extensively evaluated the direct addition of 2,2,2-trifluoroethyl trifluoroacetate 2a to benzenesulfonimidamide 1a as well as the subsequent collapse of tetrahedral intermediate 10aa as the possible turnover limiting steps. Both scenarios fit the experimental rate law and agree with catalyst active site control experiments; thus, it is challenging to rule out one scenario conclusively over the other based purely on empirical data.

Using DFT calculations, we identified tetrahedral adduct transition state structures (S)-11aa and (R)-11aa and tetrahedral intermediate collapse transition state structures (S)-12aa and (R)-12aa that give access to enantiomers (S)-4aa and (R)-4aa (Figure 9).³⁰ We note that these structures were obtained after thorough consideration and a comprehensive search of all accessible conformers and rotamers for the aforementioned structures. The calculated addition and elimination transition state barriers (ΔG^‡_(add) = 25.0 kcal/mol versus ΔG^‡_(col) = 26.6 kcal/mol, respectively) should be surmountable under the reaction conditions. While both addition (ΔΔG^‡_(S)vs(R) = 1.4 kcal/mol) and elimination (ΔΔG^‡_(S)vs(R) = 1.4 kcal/mol) show a slight preference for the observed (S)-enantiomer, the energy barrier for the tetrahedral collapse is higher by ΔΔG^‡_(add)vs(col) = 1.6 kcal/mol. At this point, we address that upon the addition of the electrophile, a second stereocenter is generated in the tetrahedral adduct. For instance, starting material (S)-1a reacting with electrophile 2a could form two tetrahedral adducts 10aa with (S, R) and (S, S) configurations for the sulfur and carbon centers, respectively. However, only the (S, R)-adduct is accommodated in the tetrahedral collapse transition state (S)-12aa due to formation of a 6-membered ring intermediate and proper alignment of the trifluoroethoxy group for elimination by proton transfer from quinuclidine. We suspect that the tetrahedral adduct formation is reversible, and thus the other (S, S)-adduct can revert to starting materials that eventually proceed through the collapse transition state (S)-12aa. Furthermore, the necessity to have a good leaving group seems to be in accordance with our electrophile scope evaluation. Thus, we conclude that the collapse of the tetrahedral intermediate is most likely the rate- and selectivity-determining step.

Figure 9.

Calculated transition state structures and energy barriers for tetrahedral intermediate formation and tetrahedral intermediate collapse. Computational method: ωB97XD/def2TZVPP (SMD(THF))//ωB97XD/def2SVP.

Stereochemical Model.

Having used statistical modeling, reaction kinetics, catalyst structural studies, and DFT calculations as complementary tools to interrogate this transformation, a stereochemical model can be proposed. According to our mechanistic model, the sulfonimidamide substituent is placed pseudoequatorially in the 6-membered ring transition state leading to the preferred (S)-enantiomer. Smaller substrates in the transition state will have a lower preference for the sulfonimidamide substituent being placed pseudoequatorially versus pseudoaxially ((S)-12, Figure 10A), leading to selectivity erosion. However, larger substrates in the preferred pseudoequatorial position will lead to unfavorable steric interactions with the phosphinate substituents ((R)-12, Figure 10A). To identify substrates for which it is difficult for the catalyst to impart selectivity, we further investigated the steric terms in the original statistical model (Figure 6C). We plotted the substrate volume against the catalyst buried volume and overlaid the measured ΔΔG^‡ as a heatmap (Figure 10B). It confirms that the catalyst pocket is not amenable to small (<1000 Bohr radius³/mol; 1k and 1l) nor large (>1490 Bohr radius³/mol; 1i, 1g, 1h, 1q, 1j) substrates.

Figure 10.

(A) Tetrahedral intermediate collapse transition state structures of catalyst 3e leading to the (S)-enantiomer and the (R)-enantiomer. (B) Substrate volume plotted against catalyst buried volume with the measured ΔΔG^‡ overlaid as a heatmap. The gray-shaded region indicates a high enantioselectivity. ^aData points for catalysts 3k and 3h overlap in the plot because they have nearly the same %V_bur. (C) Transition states for the tetrahedral intermediate collapse of catalyst 3g. (D) Impact of quinoline rotamers of catalyst 3k on the overall dipole moment and ground state energies. (E) Reaction profiles for various catalysts and the uncatalyzed background reaction. (F) MLR model for enantioselectivity trained without catalysts 3g, 3h, and 3k and depictions of the molecular descriptors from the lowest energy conformer included in the model (12 reactions were used as external validation, and a Kennard Stone algorithm was used to split the remaining 107 reactions 50:50 into training:test sets).

The initial statistical model was trained on the observed selectivity (i.e., aggregate catalyst and background selectivity); however, subsequent kinetic studies highlighted the non-innocence of the background uncatalyzed reaction. We thus hypothesized that our initial model (Figure 6C) for the observed selectivity may have suffered from competing processes. To better understand the interplay of the catalyst and substrate steric matching required for enantioinduction and determine which catalysts and/or substrates may be most impacted by higher background reactivity, we further analyzed the steric heatmap (Figure 10B). There is a clear optimal region (gray-shaded area). The catalyst active region excludes catalyst 3g, which contains the methyl-substituted phosphinate substituent. This observation can be explained by the fact that catalyst 3g exhibits higher conformational flexibility seen in cinchona-derived catalysts^31–34 and is able to accommodate the substrate in various orientations (Figure 10C). The small methyl substituents are sterically less encumbered and allow for facile quinoline rotation in the catalyst, which leads to a faster reaction rate (green; Figure 10E). Additionally, the collapse of (R)-enantiomer tetrahedral intermediates having the sulfonimidamide substituents pseudoequatorially can be more readily accessed by catalyst 3g ((R)-14, Figure 10C), which leads to degraded selectivity. The behavior of catalysts 3k and 3h (light purple shaded area, Figure 10B) is similar to those in the optimal region. However, the significant perturbation of the quinoline group (3,5-(CF₃)₂C₆H₃) for catalyst 3k impacts the lowest ground state conformation of the catalyst (Figure 10D). To minimize the molecule’s overall dipole, catalyst 3k prefers to have the quinoline positioned orthogonally to the catalysts in the optimal region, thus lowering the ground state by ~1.0 kcal/mol. As a result, the overall activation barrier to the tetrahedral adduct collapse is increased (transition state analysis for catalyst 3h, as well as for 3g and 3k, is provided in the SI), and the uncatalyzed background reaction becomes more competitive (purple, Figure 10E). Catalyst 3h lies in some degree between catalysts 3g and 3k, as the erosion of the er is observed due to a combination of conformational flexibility, lower ground state energy, and slower rate (orange, Figure 10E).

With evidence that catalysts 3g, 3h, and 3k proceed via alternative modes of action, we hypothesized a more robust statistical model could be constructed that better describes the primary mode of enantioinduction if these three catalysts were removed from the data set. It is probable that removing catalysts that are slower and have a more competitive uncatalyzed rate can serve to reduce the noise in the modeled selectivity data (i.e., observed selectivity would be more indicative of innate catalyst selectivity). The best four-parameter model consisted of one catalyst term and three substrate terms (Figure 10F). The model is substrate-dominated, with a notable catalyst term (average Sterimol L, teal, Figure 10F) that reads out substitution of both the phosphinate and quinoline C6′-position. The substrate molar volume (gray, Figure 10F) remains a model parameter, in addition to the Sterimol B₁ value (purple, Figure 10F) and buried volume within a 2.5 Å radius of the α-carbon (green, Figure 10F). The sulfonimidamide substrate parameters serve to describe the nature of the substrate in the catalyst pocket, for which the substrate must provide enough steric bulk to secure its conformation for enantioinduction. Although all the statistical measures of this model improve compared to the model trained on the full data set (Figure 6C), of significance is the increased precision (~0.1 kcal/mol) of the model in leave-one-substrate-out (MAE = 0.125 kcal/mol) and leave-one-catalyst-out (MAE = 0.132 kcal/mol) analysis.

CONCLUSIONS

An extensive experimental and computational analysis of this cinchona-phosphinate-catalyzed chemospecific acylation of sulfonimidamides was conducted. A data science-informed substrate and catalyst scope evaluation allowed for statistical modeling to inform traditional physical organic mechanistic studies, including reaction kinetics, catalyst structural studies, and DFT transition state analysis. The catalyst “active site” was mapped by deconstructing the important model parameters in conjunction with insight from these traditional physical organic techniques. We have developed a comprehensive understanding of the catalyst-substrate structure–activity relationship that accounts for the selectivity of this reaction and conclude the dicyclohexylphosphinate-cinchona alkaloid catalyst 3e was robust for most sulfonimidamide substrates. Ultimately, the enantioselectivity determining event was proposed to be the collapse of the tetrahedral intermediate, which is often hypothesized for enzymatic processes,³⁵ not small molecule catalysis. This should provide inspiration for the development of related reactions involving the use of chiral sulfur(VI) pharmacophores. Additionally, this study highlights how data science tools can be merged with traditional physical organic methods in the pursuit of mechanistic analysis.