RESEARCH ARTICLE◥

ORGANIC CHEMISTRY

A data-intensive approach tomechanistic elucidation appliedto chiral anion catalysisAnat Milo,1* Andrew J. Neel,2* F. Dean Toste,2† Matthew S. Sigman1†

Knowledge of chemical reaction mechanisms can facilitate catalyst optimization, butextracting that knowledge from a complex system is often challenging. Here, we presenta data-intensive method for deriving and then predictively applying a mechanistic modelof an enantioselective organic reaction. As a validating case study, we selected anintramolecular dehydrogenative C-N coupling reaction, catalyzed by chiral phosphoricacid derivatives, in which catalyst-substrate association involves weak, noncovalentinteractions. Little was previously understood regarding the structural origin ofenantioselectivity in this system. Catalyst and substrate substituent effects were probedby means of systematic physical organic trend analysis. Plausible interactions betweenthe substrate and catalyst that govern enantioselectivity were identified and supportedexperimentally, indicating that such an approach can afford an efficient means ofleveraging mechanistic insight so as to optimize catalyst design.

Catalyst discovery and development oftenrely on empirical observations gainedthrough the laborious evaluation of mul-tiplepotential reactionvariables (1). Althoughhigh-throughput methods can streamline

this process (1–9), the ability to rationally designcatalysts that affect chemical reactions in a pre-dictable manner would be a transformative stepforward. In catalysis, the principal challenge liesin inferring how catalyst structural features affectthe mechanistic aspects of a given chemical re-action, including those that govern selectivitywhen multiple products are possible (10–12). Al-though mechanistic studies are able to guide therational design of catalytic systems, traditionalapproaches are not often suited to address thecomplexity of modern catalytic transformations.This limitation is especially apparent in cases inwhich selectivity is affected by subtle catalyst andsubstrate structural features (13), and/or theproduct-determining step of the reaction occursafter the rate-determining step. In order to addresssuch systems, we envisioned a strategy for mecha-nistic study involving the application of moderndata analysis techniques. This approach relies onthe generation of mathematical correlations bet-ween quantifiable properties describing the inter-acting reaction partners’ molecular structures(molecular descriptors) andameasurable outcomeof the reaction {for example, enantioselectivity,

which is represented as the energy differencebetween transition states leading to either enan-tiomer DDG‡ = –RTln[(S)/(R)], kilocalories/mole}(14). Combining appropriate experimental design,data organization, and trend analysis techniquesprovides the basis to distinguish causal relations,producing testable hypotheses regarding the struc-tural origin of the reaction outcome.New catalystscan be designed, and the ability of the models topredict new experimental outcomes can be usedas validation of themechanistic hypotheses. Here,we demonstrate that this approach enables in-depth mechanistic analysis of interactions thatgovern enantioselectivity, affording nonintuitiveinsight into the origin of asymmetric inductionand guiding rational catalyst design.

Choice of case study

To assess the applicability of this data-drivenapproach toward mechanistic analysis, a systemwas sought with a poorly understood mecha-nism that would be difficult to probe by usingstandard techniques. Accordingly, the field ofchiral anion catalysis was appealing becausediverse catalyst-substrate interactions contrib-ute to enantioselectivity, but their distinctiveeffects are difficult to deconvolute. Particularly,the oxoammonium salt (3)-mediated enantio-selective cross dehydrogenative coupling (CDC) cat-alyzed by chiral 1,1′-Bi-2-naphthol (BINOL)–basedphosphoric acids (PAs) (Fig. 1, 1) bearing triazolesat the 3 and 3′ positions reported in 2013 by Tosteand coworkers (15) was identified as a prototyp-ical example. This type of reaction could benefitfrom such an analysis because of the followingchallenges. First, high levels of enantioselectivitywere necessarily predicated on the oxidant’s(16–19) insolubility under the reaction condi-

tions, precluding rigorous kinetic analyses. Sec-ond, the enantioselectivity trends with respectto both catalyst and substrate were not obvious,with even modest structural modifications result-ing in substantial differences (Fig. 1C). Last, wehypothesized that enantioselectivity was governedby attractive noncovalent interactions (15). Thesesubtle interactions are ubiquitous in biologicaland catalytic systems (13) but are difficult tostudy or apply toward rational catalyst design,especially if several such interactions are involvedin determining reaction outcomes.The distinct mechanistic features of triazole-

PA catalysts are highlighted by the observationthat they result in opposite and enhanced en-antioselectivities (Fig. 1A) relative to more con-ventional PAs such as C8-TRIP (5) and TCYP (6),which are representative of BINOL-based PAsthat have seen the most extensive use (20–22).Additionally, electronically distinct pyrazolyl(pyr-1e) and imidazolyl (imid-1e) PAs affordproducts with significantly reduced enantio-selectivities relative to the parent triazolyl (Fig. 1A,1e), despite having nearly identical steric pro-files. Although these data allude to selectivitydetermination via attractive, noncovalent inter-actions between the catalyst and substrate, suchinteractions are difficult to further characterize.This limitation is not uncommon in enantio-selective catalysis. Thus, our goal was to developa general, data-driven technique for the evalu-ation of how subtle structural features affect se-lectivity, using this reaction as a challenging casestudy.

Kinetic isotope effects

Before any mechanistic study focused on theorigins of selectivity, we sought to establish theenantioselectivity-imparting step (or steps) inthe catalytic cycle. Bearing in mind the aim ofmathematically relating structural features ofthe reacting components to enantioselectivity,this knowledge would reveal the elementary stepthat is being represented by the catalyst andsubstrate molecular descriptors (vide infra). Withrespect to the general mechanism in Fig. 1B, wesought to distinguish two scenarios: (i) Enan-tioselectivity is determined during oxidation ofsubstrate 2, or (ii) enantioselectivity is determinedduring the cyclization of an oxidized intermediate(Fig. 2A, A). With respect to the former scenario,it was conceivable that although the stereogeniccenter is formally set in the cyclization from theoxidized intermediate (A), the interactions bet-ween the substrate and catalyst during the oxi-dation event may preorganize the system foreffective enantioselection.To distinguish between these possibilities, a

set of kinetic isotope (KIE) experiments was per-formed by using 2a-d1 [90% D incorporation,74:26 enantiomeric ratio (er)]. We expected thatif the chiral phosphate were involved in substrateoxidation, different KIEs would be observed forthe formation of 4a-d1 when using enantiomericcatalysts. Indeed, (R)- and (S)-1e promoted thereaction with KIEs of 3.42 and 1.08 respectively,suggesting the involvement of the catalyst in the

RESEARCH

SCIENCE sciencemag.org 13 FEBRUARY 2015 • VOL 347 ISSUE 6223 737

1Department of Chemistry, University of Utah, 315 South1400 East, Salt Lake City, UT 84112, USA. 2ChemicalSciences Division, Lawrence Berkeley National Laboratory,and Department of Chemistry, University of California,Berkeley, CA 94720, USA.*These authors contributed equally to this work. †Correspondingauthor. E-mail: sigman@chem.utah.edu (M.S.S.); fdtoste@berkeley.edu (F.D.T.)

on Septem

ber 22, 2020

http://science.sciencemag.org/

Dow

nloaded from

738 13 FEBRUARY 2015 • VOL 347 ISSUE 6223 sciencemag.org SCIENCE

Fig. 1. System under study: asymmetric C-N dehydrogenative coupling. (A) BINOL-based phosphoric acid scaffold, enantioselective cross dehydrogenativecoupling reaction scheme, and nitrogen-deletion experiment. (B) Proposed mechanism involving a chiral phosphate-substrate ion pair. (C) Enantiomeric excess(ee) values obtained by using various substrate/catalyst combinations. (D) Library design and parameter identification strategy.

Fig. 2. Kinetic isotope effect studies and mechanistic implications. (A) Considerations regarding the origin of enantioselectivity. (B) KIEs of deuteratedenantiomerically enriched substrate 2a-d1 with (R) and (S) PA catalysts: adamantyl-substituted triazolyl (1e) and pyrazoyl (pyr-1e), and TCYP (6). (C) Revisedmechanism of enantioselectivity determination.

RESEARCH | RESEARCH ARTICLEon S

eptember 22, 2020

http://science.sciencem

ag.org/D

ownloaded from

oxidation (Fig. 2B). If enantioselectivity were alsoestablished in this step, different enantiomericexcess (ee) values would be expected when usingthe enantiomeric catalysts. The observation thatthe final products exhibited equal but oppositelevels of enantioselectivity is consistent withenantioselection occurring during the bond-forming event from an oxidized intermediatesuch asA. This result indicated that the two keysteps of this reaction (oxidation and cyclization)proceed under independent Curtin-Hammettcontrol (10, 11), with similar interactions presum-ably governing selectivity in both cases (Fig. 2C).The KIE values resulting from the less-selectivecatalysts pyr-1e and 6 (Figs. 1A and 2B) do notdisplay the same enantiomer-dependency as thosefrom 1e. This effect is consistent with the assertionthat similar catalyst-substrate interactions are in-volved throughout the mechanism, as well as thetriazole substituent’s superior ability to interactwith substrate 2. However, the specific natureof this interaction remained undefined. To thisend, a thorough analysis of an extensive data setcontaining structural perturbations to the cata-lyst and substrate could serve to illuminate theseenantioselectivity-directing interactions.

Experimental design and analysis

The collection and organization of diverse datasets is at the foundation of data-driven analysisstrategies (23). Therefore, an effective experimen-tal library should include rational changes tovarious structural features that affect the reac-tion outcome of interest. To this end, substrates(2) were modified at positions hypothesized toinfluence enantioselectivity (at the 2-, 4-, and 6-positions of the benzyl ring and the distal arylring) (Fig. 1D), using substituents with variedelectronic and steric properties (according totheir Hammett spara and Sterimol B1 values, re-spectively; additional details are provided inthe supplementary materials, p4-8). Similarly,catalysts (1) were modified at the 2-, 4-, and 6-positions of the aryl ring attached to the triazole.Adamantyl-substituted catalysts 1e, pyr-1e, andimid-1e were also included to explore the effectof changes to the heterocyclic ring. In total, 12substrates and 11 triazolyl catalysts were selected(Fig. 1D). These libraries were then synthesized,and the enantioselectivity of each catalyst-substratecombination was obtained. Simultaneously, adiverse array of molecular descriptor valueswas collected to describe the structural featuresof each catalyst and substrate, including Sterimolparameters (24), length measurements fromgeometry optimized structures, and computa-tionally derived vibrational frequencies andintensities (details are provided in the supple-mentary materials, p4-8) (Fig. 1D) (25). Linearregression algorithms were then applied to var-ious subsets of the data to identify correlationsbetween molecular structure and the experi-mentally determined enantioselectivity. Subse-quently, analysis and refinement of the resultingmodels were used to produce explicit mechanis-tic hypotheses that were then tested and val-idated experimentally.

Modeling catalyst heterocyclic rings

Given the clear importance of the catalyst het-erocyclic ring in enantioselectivity determina-tion (vide supra), we initially sought to understandthe subset of results obtained by using catalysts1e, pyr-1e, and imid-1e. Accordingly, by usinglinear regression techniques the correlation de-picted in Fig. 3B was identified from a trainingset of 10 different substrate-catalyst combina-tions (Fig. 3, A and C, black squares). Of thelarge number of steric (26) and vibrational (25)terms initially investigated as molecular descrip-tors, four discrete vibrational parameters weresufficient to produce a correlation with enantio-selectivity: one catalyst descriptor (nY–N, a stretch-ing frequency on the heterocyclic ring), and twosubstrate descriptors (the stretching frequencyof the amide C=Ο bond, nC=Ο, and stretchingfrequency/intensity of the C–H bond undergoingoxidation, n/iC–H) (Fig. 3, B and D). A cross-term between the catalyst and substrate descrip-tors improves the overall quality of the model(nY–N x iC–H), suggesting a synergistic struc-tural effect. The model was evaluated by plott-ing measured versus predicted DDG‡ values(Fig. 3C), and as a validation of its robustness,the enantioselectivities of 10 catalyst-substratecombinations not included in the training setwere well-predicted (Fig. 3, A and C, red crosses).A slope approaching unity and intercept approach-ing zero over the training and validation setsindicate an accurate and predictive model, andthe R2 value of 0.90 demonstrates a high degreeof precision. The largest coefficient in this nor-malized model belongs to the heterocyclic ringvibrational frequency, signifying its substantialrole in the quantification of enantioselectivity.

This model is capable of predicting resultswhose origins are not obvious upon inspection.For example, comparison of the reaction out-comes using 1e and pyr-1e with substrate 2a(Fig. 3, entries 1 versus 13) may lead to the con-clusion that pyr-1e generally affords inferior se-lectivity. Indeed, experimental results for severaladditional substrates support this notion and areaccurately predicted by the model (for example,2-OMe benzyl substrate 2e, entries 5 versus 16).However, with substrate 2i (R1 = 2-OMe, R2 = Ph,entries 9 versus 17), the triazolyl and pyrazolylPAs afford the product with similar levels ofenantioselectivity. This counterintuitive result isprecisely predicted, indicating that the divergentenantioselectivity displayed by 1e, as comparedwithpyr-1e and imid-1e, is adequately addressedby the model.

Trend analysis

Although the model in Fig. 3B establishes thecapacity of the chosen parameters to describesubtle aspects of this system, the ultimate goalof this approach was to discern underlying mech-anistic phenomena. This objective could not beachieved by using merely the above correlationbecause it was produced by using a small sub-set of data, in which the catalyst heterocyclicrings bore the same substituent (adamantyl).We hypothesized that the complete data set con-tained invaluable information to this end be-cause it was produced by using strategicallymodified catalysts and substrates with substit-uents intentionally introduced to probe subtleeffects, resulting in 132 enantioselectivities bet-ween –54 and 94% ee, which corresponds to aDDG‡ range of 2.8 kcal/mol. In accordance with

SCIENCE sciencemag.org 13 FEBRUARY 2015 • VOL 347 ISSUE 6223 739

Fig. 3. Impact of heterocyclic catalyst substituent on enantioselectivity. (A) Predicted and measuredenantioselectivities for various substrates with adamantyl-substituted triazolyl (1e), pyrazoyl (pyr-1e),and imidazoyl (imid-1e) PAs. Values identified with an asterisk are external validations. (B) Mathematicalcorrelation of normalized catalyst and substrate vibrational parameters to enantioselectivity. (C)Predicted versus measured DDG‡ plot. (D) Illustration of the structural features implicated by theidentified parameters.

RESEARCH | RESEARCH ARTICLEon S

eptember 22, 2020

http://science.sciencem

ag.org/D

ownloaded from

a data-intensive strategy, none of these mea-sured values was discarded because even lowenantioselectivity carries information regardingthe catalyst-substrate interactions at the origin ofasymmetric induction.Yet before producing a global, predictive mod-

el, we considered that a series of focused corre-lations, coupled with an evaluation of overalltrends, might serve to reveal fundamental fea-tures of the system. To this end, for each individ-ual substrate a correlation was produced relatingits observed enantioselectivity values for the en-tire set of catalysts, with parameters describingthe catalyst structure (2a-2l, 12 models in total)(fig. S3). The same strategy was applied to allaryl-substituted triazole catalysts by using param-eters describing the substrate structure (1b-1d,1f-1k, 9 models total) (table S7). This organiza-tional schemewas viewed as ameans to facilitatethe identification of catalyst features that affectparticular substrate types (and vice versa). Sub-strates or catalysts with similar structures behaveanalogously not only in a qualitative manner, butalso in terms of the molecular descriptors thateffectively serve to predict their enantioselectiv-ities (individual substrate and catalyst measuredversus predicted DDG‡ plots and equations areavailable in figs. S2, S3, and S4 and table S7).These quantitative correlations, together with sys-tematically organized trends of experimentaloutcomes, can guide the development of testablemechanistic hypotheses.To simultaneously inspect multiple aspects of

large and intricate data sets, a communicative

visualization of data is crucial (27). Thus, weelected to present information gained from thesefocusedmathematicalmodels, alongsidemultipleobserved enantioselectivity results, organized ac-cording to trends in catalyst or substrate struc-tural features. Demonstrating this visualizationtechnique, the enantioselectivity trend for eachcatalyst (in Fig. 4, each line represents a catalyst)was plotted as a function of the substrates (inFig. 4, each x axis tick-line represents a substrate),and vice versa (Fig. 5). To afford a quantitativetrend analysis, the plots were arranged accordingto which positions were modified on the catalystor substrate structures, and the corresponding R1

or R2 substituent’s Sterimol B1 value (additionalvisualizations are provided in figs. S5 and S6)(24). For example, in the purple frame in Fig. 4, 2-substituted benzyl substrates are displayed fromthe largest to smallest substituent. The catalystmolecular descriptors required for correlatingeach subset of substrates are also depicted (Fig. 4),along with the substrate descriptors for each sub-set of catalysts (Fig. 5). For example, the catalystmolecular descriptors used as parameters for thecorrelation of enantioselectivity obtained for 2-substituted benzyl substrates are presented belowthe blue frame in Fig. 4 (nN–N, ∠tor).Analysis of this systematic data arrangement

reveals that in general, catalyst performance cor-relates with the aryl substitution pattern in theorder 2,4,6 > 2,6 > 4 (Fig. 5, gray, orange, andblue frames, respectively). Additionally, by juxta-posing Figs. 4 and 5, it appears that the reactionis mainly under catalyst control because catalyst

features affect enantioselectivity in a more con-siderable and systematicmanner (Fig. 5), whereasfor each substrate, the spread of observed enan-tioselectivity is broader (Fig. 4). All substratesbearing a 2-substituted benzyl group—even thosewith substitution at R2 (Fig. 4, purple and blueframes, respectively)—can be modeled by usingthe torsion angle between the triazole and itssubstituent, and a triazole vibration frequency(∠tor, nN–N) (individual models are provided intable S7). The torsion angle represents a stericeffect yet also contains information concerningthe conjugation between the triazole ring andits substituent. The vibration frequency canserve as a correction to both of these effectsbecause it takes into account nonadditive fea-tures of the substituents’ charge and mass dis-tribution (25).The models for substrates with 4-benzyl sub-

stitution (Fig. 4, green frame) contain the sametwo terms (∠tor, nN–N) and an additional stericdescriptor (the catalyst aryl ring minimal width,B1). Similar interactions with the triazole ringshould still be present for these 4-benzyl sub-strates, but the presence of a B1 term suggests anadditional steric interactionbetween the substrateand catalyst substituents, which is avoided in thecase of hydrogen at the 4-benzyl position. Thisclaim is supported by the lower enantioselectiv-ities observed for substrates with larger 4-benzylsubstituents, especially when using catalysts withlarger 2,6-groups. Thus, the lack of the catalystB1 term in the models describing 2- relative to4-benzyl substrates, and their overall higher

740 13 FEBRUARY 2015 • VOL 347 ISSUE 6223 sciencemag.org SCIENCE

Fig. 4. Graphical representation of catalyst structure-selectivity trends as a function of substrate. Enantioselectivity trends for every catalyst against allsubstrates. Each trend line represents a catalyst, and each x axis tick-line represents a substrate.

RESEARCH | RESEARCH ARTICLEon S

eptember 22, 2020

http://science.sciencem

ag.org/D

ownloaded from

enantioselectivities, are thought to arise from abetter accommodation of the former substrates’shapes in the catalyst active site.The individual catalyst trends and models car-

ry complementary information, revealing the sub-strate descriptors relevant to each catalyst subset(Fig. 5). The parameters that correlate with 4-substituted aryl catalysts’ enantioselectivity (Fig. 5,blue frame) include thesubstrate carbonyl-stretchingvibrational frequency and intensity (n/iC=O), theamide N–H vibrational frequency (nN–H), and across-term between the two (nC=O x nN–H).These values vary in response to substitution onthe substrate benzyl ring or the distal aryl ring,with the former having a greater effect (table S2).These same parameters effectively correlate withtheenantioselectivity observed for 2,6-disubstitutedaryl catalysts (Fig. 5, orange frame), along withan additional vibrational term (nBn) describing abenzyl ring stretch. The enantioselectivity spreadof catalysts with larger substituents at the 2,4,6-position [for example, 38 to 93% ee for 2,4,6-(Cy)3-Ph catalyst 1d (Fig. 5, gray frame)] can bedescribed by using two terms associatedwith thebenzyl ring (nBn, iN–H), stressing this ring’s rolein determining enantioselectivity.

Trend interpretation

Collectively, these results suggest that a p in-teraction is established between the triazole ringand substrate during the enantioselectivity deter-mining step, the strength of which is modulatedby local steric and electronic structural featuresof both interacting partners (13, 28–34). Fur-thermore, p interactions are often strengthened

by heteroatoms (35–37), which could explain thereduced enantioselectivity values obtained byusing imid-1e, pyr-1e, and TCYP catalysts com-pared with 1e (Fig. 1A), as well as the similarKIE values obtained when using both the (R)-and (S)-enantiomers of pyr-1e and TCYP ascatalysts, compared with the divergent ones dis-played by (R)- and (S)-1e (Fig. 2B). In relation tothe substrate, participation of both the benzylgroup and the distal aryl group in putative pinteractions are supported by the presence ofmolecular descriptors that are sensitive to sub-stitution on these rings in every catalyst mod-el (nC=O, iC=O, nN–H, and nBn) (Fig. 5 andtable S2).The energy stabilization gained from p inter-

actions is affected by the distance and geometryof the rings involved (13, 30–37). If a p interac-tion between the substrate and triazole is at theorigin of enantioselectivity determination, thedirectionality of the triazole—represented bythe torsion angle between the triazole and itssubstituent—is expected to directly affect enan-tioselectivity. In agreement with this hypothe-sis, catalysts with more pronounced torsionaleffects lead to higher enantioselectivity valuesfor substrates with relatively small substituentsat the benzyl 4-position. The torsion angle ap-proaches perpendicularity (90°) owing to larger2,6-substituents on the catalyst aryl ring con-nected to the triazole (Fig. 4, purple and blueframes, blue lines). Moreover, large catalyst 2,6-aryl substituents are presumed to serve as asteric barrier, docking the substrate in place foran improved overlap with the catalyst triazole

ring. Correspondingly, substrates with elongated4-substituents (R = Me, OMe) lead to lower en-antioselectivities by using catalysts with largesubstituents at the 2,6-position (Fig. 4, greenframe, blue lines). For these substrates, the stericrepulsion exerted by the large 2,6-substituentsaffords a weaker or less directing p interactionand, subsequently, lower enantioselectivity. Thus,the importance of the torsion angle and vibrationparameters for correlating enantioselectivity inthe individual models and overall trends (Fig. 4,fig. S2, and table S7) is proposed to reflect theangle at which the triazole engages the substrateand the steric role of the catalyst aryl group.Lending further credence to this proposal, cat-alysts with reduced torsion angles (such as cat-alysts with triazole R substituents: Ph, 4-NO2Ph,4-OMePh, or 4-SO2MePh) that do not introducethe proposed directional and steric effects lead todiminished enantioselectivities overall (Fig. 5).

Comprehensive model and probesof mechanistic hypotheses

On the basis of these hypotheses, we set out todesign a series of new catalysts to specificallyprobe putative interactions. To facilitate catalystdevelopment, a predictive model (Fig. 6) wasgenerated for the entire substrate set with thearyl-substituted catalysts (1b-1d and 1f-1k). Thismodel contains 108 combinations (9 catalyststimes 12 substrates) from the initial library ofexperiments, where half were used as a trainingset (Fig. 6B, black squares) and the other half asexternal validations (Fig. 6B, red crosses). Newcatalysts were proposed to address hypotheses

SCIENCE sciencemag.org 13 FEBRUARY 2015 • VOL 347 ISSUE 6223 741

Fig. 5. Graphical representation of substrate structure-selectivity trends as a function of catalyst. Enantioselectivity trends for every substrate againstall catalysts. Each trend line represents a substrate, and each x axis tick-line represents a catalyst.

RESEARCH | RESEARCH ARTICLEon S

eptember 22, 2020

http://science.sciencem

ag.org/D

ownloaded from

raised by the focused models and trend analysis,and their enantioselectivity was predicted beforesynthesis by the comprehensive model (Fig. 6).First, to probe whether the aryl substituent on

the triazole ring plays a primarily steric role,rather than directly engaging the substrate in a pinteraction, perfluorophenyl catalyst 1l was de-signed and evaluated. Substituent local sterics andcharge distribution have been shown to stronglyaffect noncovalent p interactions (28–30, 32–39).Therefore, we expected that if enantioselectivitywere predominantly dependent on the aryl sub-stituent directly engaging as a partner in a p in-teraction (as opposed to taking an ancillary role inp interactions involving the triazole), perfluoro-phenyl catalyst 1lwoulddeviate significantly fromits Ad (1e), 2,6-(F)2-Ph (1k) or 2,6-(MeO)2-Ph (1j)counterparts. However, all four catalysts behavesimilarly with respect to the magnitude and signof enantioselectivity (Fig. 6C, entries 1 to 9 and 24to 26). This result is well predicted by the model(Fig. 6B) and is consistent with the hypothesisthat the main function of the aryl substituent issteric.Next, catalyst 1m, bearing a single isopropyl

group at the 2-position of the triazole aryl sub-stituent, was prepared to probe the hypothesisthat steric repulsion exists between larger cata-lyst 2,6-substituents and elongated substrate 4-substituents. We anticipated that an isopropylgroup would induce the torsion necessary toenforce the proposed benzyl-triazole p interac-

tion, while avoiding a direct steric interactionbetween the substrate benzyl 4-position and thecatalyst aryl 2,6-substituents. Indeed, for all 4-substituted substrates tested (Fig. 6C, entries 10to 12), 1m provided the corresponding product inhigher enantioselectivity than that of 1c, whichbears isopropyl groups at both ortho positions ofthe triazole aryl substituent (Fig. 6C, entries 13 to15). For 4-NO2-Bn substrate 2d, the 2-iPr catalyst1m resulted in the highest enantioselectivity ob-served to date (Fig. 6, B and C, entry 12), aspredicted by the model.Last, in order to evaluate the capacity to obtain

improved enantioselectivity as a result of a data-intensive approach, and the hypothesis that tor-sion leads to enhanced enantioselectivity for the2-substituted substrates, several proposed cata-lysts were evaluated by using the model in Fig. 6.Catalyst 1n was selected because it is syntheti-cally feasible, accommodates a torsion angle closeto 90°, and was predicted to give improved enan-tioselectivity for all substrates bearing hydrogenat the 4-benzyl position. This prediction was ver-ified in practice for all eight substrates evaluated,affording thehighest enantioselectivities observedto date (Fig. 6C, entries 16 to 23). These resultsconfirm that a perpendicular geometry of thetriazole and the aryl ring can indeed lead tohigher enantioselectivities, supporting the prem-ise that the orientation of the triazole ring cou-pled with its R group’s steric constraints controltriazole p interactions.

The overall analysis of the triazole-PA casestudy demonstrates the complementary mannerin which classical physical organic techniquesandmoderndata analysis strategies canbemergedtoward amore completemechanistic assessment(40). This approach is based on the use of em-pirical data, which is often a prerequisite for arational reaction optimization process, to con-comitantly conduct a mechanistic investigation.Information of this sort that could be used for anin-depth analysis is often omitted from reports inthe field of catalysis because only results leadingto the desired outcomes are generally presentedandpursued. Yet because high-throughput (2,4,8),automated (7)methods for reaction developmentare now common, data analysis strategies couldbe applied in parallel to optimization procedures,allowing for simultaneousmechanistic and struc-tural analysis.Creatively collecting and organizing data to

examine proposed hypotheses affords improvedgeneralizations, particularly as data sets becomelarger and more complex (23). This idea holdstrue for the analysis of reaction trends by param-eters that reflect structural modification. Indeed,the focused catalyst and substrate models—andtheir organization according to fundamental,quantitative, physical-organic trends—providednonintuitive insights regarding interactions in-volved in enantioselectivity determination.Althoughthis approach is general for the prediction andstudy of chemical reaction outcomes, this case

742 13 FEBRUARY 2015 • VOL 347 ISSUE 6223 sciencemag.org SCIENCE

Fig. 6. Validation of mechanistic hypotheses through directed catalyst design. (A) Normalized equation for the prediction of enantioselectivity by usingparameters that describe catalyst and substrate structural features. (B) Predicted versus measured DDG‡ plot. (C) Hypothesis-driven external predictions andtheir comparison with other relevant catalysts.

RESEARCH | RESEARCH ARTICLEon S

eptember 22, 2020

http://science.sciencem

ag.org/D

ownloaded from

study was chosen as a stringent benchmark be-cause it uses weak, noncovalent interactions forasymmetric induction. These interactions are typ-ically in the energy range required to distinguish ahighly enantioselective reaction from its racemateforming counterpart (2 to 3 kcal/mol) (12, 13, 32, 37),providing seemingly endless approaches to rationalcatalyst design. However, controlling noncova-lent interactions represents a notable challengein the design of catalytic systems because of mul-tiple energetically accessibleorientations (13). Com-plemented with rigorous experimental analysis,the disclosed data-intensive approach is suited toaddressing such intricacies and holds potentialfor the analysis of increasingly complicated cat-alytic systems streamlining both reaction andcatalyst development.

REFERENCES AND NOTES

1. K. D. Collins, T. Gensch, F. Glorius, Nat. Chem. 6, 859–871(2014).

2. M. T. Reetz, Angew. Chem. Int. Ed. Engl. 41, 1335–1338(2002).

3. D. W. Robbins, J. F. Hartwig, Science 333, 1423–1427 (2011).4. A. McNally, C. K. Prier, D. W. C. MacMillan, Science 334,

1114–1117 (2011).5. J. R. Schmink, A. Bellomo, S. Berritt, Aldrichim Acta 46,

71–80 (2013).6. M. R. Friedfeld et al., Science 342, 1076–1080 (2013).7. P. L. Heider et al., Org. Process Res. Dev. 18, 402–409 (2014).8. P. Metola, S. M. Nichols, B. Kahr, E. V. Anslyn, Chem. Sci.

5, 4278–4282 (2014).9. A. Buitrago Santanilla et al., Science 347, 49–53

(2015).10. D. Y. Curtin, Rec. Chem. Prog. 15, 110–128 (1954).11. J. Halpern, Science 217, 401–407 (1982).12. E. V. Anslyn, D. A. Dougherty, Modern Physical Organic

Chemistry (University-Science Books, Herndon, VA, 2006).13. R. R. Knowles, E. N. Jacobsen, Proc. Natl. Acad. Sci. U.S.A.

107, 20678–20685 (2010).14. K. C. Harper, M. S. Sigman, Proc. Natl. Acad. Sci. U.S.A. 108,

2179–2183 (2011).15. A. J. Neel, J. P. Hehn, P. F. Tripet, F. D. Toste, J. Am. Chem.

Soc. 135, 14044–14047 (2013).16. J. M. Bobbitt, C. Brückner, N. Merbouh, in Organic Reactions

(John Wiley & Sons, New York, 2004).17. C.-J. Li, Acc. Chem. Res. 42, 335–344 (2009).18. C. S. Yeung, V. M. Dong, Chem. Rev. 111, 1215–1292

(2011).19. O. García Mancheño, T. Stopka, Synthesis 45, 1602–1611

(2013).20. T. Akiyama, Chem. Rev. 107, 5744–5758 (2007).21. M. Terada, Synthesis 2010, 1929–1982 (2010).22. D. Parmar, E. Sugiono, S. Raja, M. Rueping, Chem. Rev. 114,

9047–9153 (2014).23. N. Silver, The Signal and the Noise: Why so Many Predictions

Fail–But Some Don't (Penguin Press, New York, 2012).24. A. Verloop, J. Tipker, Pharmacochem. Libr. 2, 63–81

(1977).25. A. Milo, E. N. Bess, M. S. Sigman, Nature 507, 210–214

(2014).26. K. C. Harper, E. N. Bess, M. S. Sigman, Nat. Chem. 4, 366–374

(2012).27. M. Lima, Visual Complexity: Mapping Patterns of Information

(Princeton Architectural Press, New York, 2013).28. C. A. Hunter, J. K. M. Sanders, J. Am. Chem. Soc. 112,

5525–5534 (1990).29. F. Cozzi, M. Cinquini, R. Annunziata, T. Dwyer, J. S. Siegel,

J. Am. Chem. Soc. 114, 5729–5733 (1992).30. C. R. Martinez, B. L. Iverson, Chem. Sci. 3, 2191 (2012).31. E. H. Krenske, K. N. Houk, Acc. Chem. Res. 46, 979–989

(2013).32. K. E. Riley, P. Hobza, Acc. Chem. Res. 46, 927–936

(2013).33. C. D. Sherrill, Acc. Chem. Res. 46, 1020–1028 (2013).34. R. M. Parrish, C. D. Sherrill, J. Am. Chem. Soc. 136,

17386–17389 (2014).35. E. G. Hohenstein, C. D. Sherrill, J. Phys. Chem. A 113, 878–886

(2009).

36. S. E. Wheeler, Acc. Chem. Res. 46, 1029–1038 (2013).37. B. F. Shchegolev, M. L. McKee, A. V. Zhuravlev,

E. V. Savvateeva-Popova, Biofizika 58, 461–467 (2013).38. S. E. Wheeler, K. N. Houk, J. Am. Chem. Soc. 130,

10854–10855 (2008).39. D. O’Hagan, Chem. Soc. Rev. 37, 308–319 (2008).40. T. T. Tidwell, Z. Rappoport, C. L. Perrin, Eds., Pure Appl. Chem.

69, 1–76 (1997).

ACKNOWLEDGMENTS

We thank the NSF (CHE-0749506 and CHE-1361296) and theNational Institute of General Medical Sciences (R01 GM104534) forpartial support of this work. The support and resources from theCenter for High Performance Computing at the University of Utah

are gratefully acknowledged. A.J.N. gratefully acknowledges anAmgen Fellowship in Organic Chemistry for funding and Jörg Hehnfor early contributions to this work.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/347/6223/737/suppl/DC1Materials and MethodsSupplementaryTextFigs. S1 to S7Tables S1 to S6References (41–53)

9 September 2014; accepted 13 January 201510.1126/science.1261043

REPORTS◥

SUPERCONDUCTIVITY

Light-induced superconductivityusing a photoactive electricdouble layerMasayuki Suda,1,2* Reizo Kato,2 Hiroshi M. Yamamoto1,2*

Electric double layers (EDLs) of ionic liquids have been used in superconductingfield-effect transistors as nanogap capacitors. Because of the freezing of the ionicmotion below ~200 kelvin, modulations of the carrier density have been limited to thehigh-temperature regime. Here we observe carrier-doping–induced superconductivity in anorganic Mott insulator with a photoinduced EDL based on a photochromic spiropyranmonolayer. Because the spiropyran can isomerize reversibly between nonionic andzwitterionic isomers through photochemical processes, two distinct built-in electricfields can modulate the carrier density even at cryogenic conditions.

The electric potential difference that usuallyexists across a phase boundary creates twolayers of space charges of different signs,the so-called electric double layer (EDL) (1,2).In recent years, EDL transistors with ionic

liquids or electrolytes as gate dielectrics havebeen used increasingly (3, 4). Under an appliedgate voltage, ions are driven to the surface of thechannel materials, forming an EDL that acts as ananogap capacitor. Using this method to accu-mulate a large numbers of carriers (~1014 cm−2)(5–9), field-induced superconductivity has beenrealized in various materials. However, modu-lations of the carrier density in EDL transistorshave been limited to the high-temperature regimebecause the ionic motions of ionic liquids or elec-trolytes are frozen below ~200 K.Hereweobserve field-induced superconductivity

even at low temperatures by using a photoactiveEDL system, a spiropyranmonolayer. Spiropyranis aphotochromicmolecule that canphotoisomerize

reversibly between the nonionic spiropyran (SP)form and the zwitterionic merocyanine (MC)form upon ultraviolet (UV) (forward reaction) orvisible light irradiation (reverse reaction) (fig. S1)(10, 11). Therefore, an EDL-like large electric fieldacross the film can be induced or be eliminatedby UV light or visible light irradiation, respec-tively. The advantage of the photochromic EDLsystemover the conventional one is that it retainsthe modulation capability even at low temper-atures, because these photochromic reactionsproceed with photon-energy dissipation undernonequilibriumconditions.Hence, ahighly alignedSP monolayer is a promising candidate for aphotoinducedEDL system that candirectly switchsuperconductivity only by photoirradiation. Al-though there have been attempts to tune theconductivity and even superconductivity usingphotochromic layers (12–17), photoinduced phasetransitions (including superconducting transi-tions) have not been realized yet.We fabricated a photoactive superconducting

device by laminating a thin single crystal ofk-(BEDT-TTF)2Cu[N(CN)2]Br (k-Br) [BEDT-TTF:bis(ethylenedithio)tetrathiafulvalene] on top ofan Al2O3(or HfO2)/Nb:SrTiO3 substrate coveredwith a self-assembled monolayer of spyropiran

SCIENCE sciencemag.org 13 FEBRUARY 2015 • VOL 347 ISSUE 6223 743

1Research Center of Integrative Molecular Systems (CIMoS),Institute for Molecular Science, Okazaki, Aichi 444-8585Japan. 2RIKEN, Wako, Saitama 351-0198 Japan.*Corresponding author. E-mail: msuda@ims.ac.jp (M.S.);yhiroshi@ims.ac.jp (H.M.Y.)

RESEARCHon S

eptember 22, 2020

http://science.sciencem

ag.org/D

ownloaded from

A data-intensive approach to mechanistic elucidation applied to chiral anion catalysisAnat Milo, Andrew J. Neel, F. Dean Toste and Matthew S. Sigman

DOI: 10.1126/science.1261043 (6223), 737-743.347Science 

, this issue p. 737; see also p. 719Sciencebond-forming reaction (see the Perspective by Lu). They then apply this model to improve the catalyst globally.a predictive model of how selectivity depends on multiple characteristics of the catalyst and substrate in a C-N

use multidimensional analysis techniques to generateet al.tradeoffs that in combination offer the best performance. Milo descent. So it is in chemistry, where optimizing each structural feature of a catalyst consecutively might gloss over subtlehill, and you only take steps toward higher ground, you might never find a peak on a route that requires a preliminary

Optimization strategies are often likened to hikes in a hilly landscape. If your goal is to get to the top of the highestOptimizing a catalyst many ways at once

ARTICLE TOOLS http://science.sciencemag.org/content/347/6223/737

MATERIALSSUPPLEMENTARY http://science.sciencemag.org/content/suppl/2015/02/11/347.6223.737.DC1

CONTENTRELATED http://science.sciencemag.org/content/sci/347/6223/719.full

REFERENCES

http://science.sciencemag.org/content/347/6223/737#BIBLThis article cites 46 articles, 8 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

is a registered trademark of AAAS.ScienceScience, 1200 New York Avenue NW, Washington, DC 20005. The title (print ISSN 0036-8075; online ISSN 1095-9203) is published by the American Association for the Advancement ofScience

Copyright © 2015, American Association for the Advancement of Science

on Septem

ber 22, 2020

http://science.sciencemag.org/

Dow

nloaded from