Prediction of catalyst and substrate performance in the enantioselective propargylation of aliphatic ketones by a multidimensional model of steric effects

Article abstract of DOI:10.1021/ja4001807

The effectiveness of a new asymmetric catalytic methodology is often weighed by the number of diverse substrates that undergo reaction with high enantioselectivity. Here we report a study that correlates substrate and ligand steric effects to enantioselectivity for the propargylation of aliphatic ketones. The mathematical model is shown to be highly predictive when applied to substrate/catalyst combinations outside the training set.

Full text of DOI:10.1021/ja4001807

Communication
pubs.acs.org/JACS
Prediction of Catalyst and Substrate Performance in the
Enantioselective Propargylation of Aliphatic Ketones by a
Multidimensional Model of Steric Effects
Kaid C. Harper, Sarah C. Vilardi, and Matthew S. Sigman*
Department of Chemistry, University of Utah, 315 South 1400 East, Salt Lake City, Utah 84112, United States
S
* Supporting Information
these studies have focused exclusively on examining ligand
components. Encouraged by the correlative and predictive
models generated, we wanted to expand this approach to
incorporate other common reaction parameters. We selected to
probe what is perhaps the most impactful reaction consid-
eration in asymmetric catalysis, the substrate, as we believed
predicting and understanding substrate effects as a function of
catalyst structure is at the core of how one develops and applies
an asymmetric catalytic reaction.
ABSTRACT: The effectiveness of a new asymmetric
catalytic methodology is often weighed by the number of
diverse substrates that undergo reaction with high
enantioselectivity. Here we report a study that correlates
substrate and ligand steric effects to enantioselectivity for
the propargylation of aliphatic ketones. The mathematical
model is shown to be highly predictive when applied to
substrate/catalyst combinations outside the training set.
Parameter Selection. Sterimol parameters vary from many
common steric parameters because they describe a single
substituent with three parameters, B1, B5, and L (Figure 1),
lthough steric effects are ubiquitous in asymmetric
A
catalysis, they remain difficult to quantify and define for
precise application in catalyst design. Methods to measure
steric effects and other effects in asymmetric catalysis have
paralleled those developed for quantitative structure−activity
relationships (QSAR) in drug design.1 Typically, a QSAR study
interrogates a single protein with a library of inhibitors to
delineate the important substituent effect(s) responsible for the
desired outcome.2 The resultant mathematical models
generated using QSAR afford a testable hypotheses about the
key molecular interactions facilitating modern drug design.2e
In contrast to the specific protein−inhibitor interactions in
medicinal chemistry, asymmetric catalysis is complicated
because both the catalyst and the substrate are mutable and
can be probed to determine the important structural features of
each. Typically, catalysts have been examined independent of
substrate or vice versa.3 Indeed, the multifaceted interactions
between both substrate and catalyst are generally ignored in
examinations of asymmetric catalytic systems. Simultaneously
evaluating the relationship between these crucial reaction
variables is a goal of our ongoing program in understanding and
predicting asymmetric catalytic reactions.3g,4 Herein we present
the results of the simultaneous and systematic evaluation of
both catalyst and substrate steric effects on an enantioselective
Nozaki−Hiyama−Kishi (NHK) ketone propargylation mod-
eled with Sterimol parameters. Multidimensional modeling
allows for accurate prediction of enantioselectivity for 41
independent substrate and catalyst combinations for aliphatic
ketone propargylation.
Figure 1. Parameterization of an isopropyl group using Verloop’s
Sterimol parameters.
instead of a single parameter.5 The B1 parameter describes a
minimum width orthogonal to the primary bond, the B5
parameter describes the maximum width along the same axis,
and the L parameter is the length of the substituent along the
primary bond. These separate parameters can generate more
informative models than other simpler steric parameters but
also increase the dimensionality of the data three-fold. In order
to develop models using Sterimol parameters in combination
with linear regression techniques, a larger data set is required
than the nine-membered libraries previously utilized.4f−h
Because the application of experimental design is fundamental
to developing highly predictive models through multidimen-
sional regression techniques, the catalyst substituents and
ligand substituents need to be selected carefully using the
statistical principles known as the design of experiments
(DoE).6
Recently, we have reported several investigations of
asymmetric catalytic reactions using linear free energy relation-
ships where multiple steric and electronic components of the
ligand structure were examined simultaneously.4f−h This
multivariate examination of asymmetric catalytic reactions is
appealing because it examines how different catalyst compo-
nents might synergistically impact enantioselectivity. To date,
Substrate and Catalyst Design Matrix. Our previous
reports on the enantioselective propargylation of aliphatic
Received: January 7, 2013
Published: February 6, 2013
© 2013 American Chemical Society
2482
dx.doi.org/10.1021/ja4001807 | J. Am. Chem. Soc. 2013, 135, 2482−2485

Journal of the American Chemical Society
Communication
Figure 2. (a) Previous results showcasing sensitivity of both substrate and ligand steric effects. (b) Library design employing DoE principles. (c) Plot
of experimental vs predicted enantioselectivity using the model in eq 2. A total of 30 unique ligand/substrate experiments were performed.
ΔΔG⧧ = z0 + aB1L + bB5L + cLL + dB1S + fB5S + gLS
ketones using NHK conditions indicated that increased steric
bulk on one side of the ketone led to higher enantioselectivity
(Figure 2a).4g Coupling this sensitivity with the reported
carbamate substituents effects on the ligand provided the
foundation for the initial experimental design.4c,d,f,g Variation in
steric bulk at the two substituents could be quantified using
Sterimol parameters to develop a model capable of predicting
enantioselectivity.
+ hB1LB1S + iB1LB5S + jB1LLS + kB5LB1S
+ mB5LB5S + nB5LLS + oLLB1S + pLLB5S
+ qLLLS
(1)
(2)
ΔΔG⧧ = −2.19 + 1.47B1L + 0.94B1S − 0.27B1LB1S
− 0.09B1LB5S
Our choice of substituents for the substrate and ligand was
based on evenly distributing the Sterimol values for both
according to DoE principles (Figure 2b). The Sterimol
parameters present three steric dimensions to evenly span in
order to produce an effective experimental design. Upon
inspection of our previous efforts in enantioselective ketone
propargylation, both the substrate and ligand have an apparent
sensitivity to B1 and B5, allowing us to focus on an even
distribution of the these parameters.4g The L parameter was
also considered, but the collinear relationship between B5 and L
justified only ancillary attention.4h As we have previously
reported, a precise experimental design is difficult to achieve
using steric parameters because these parameters are not
continuous.4f Accordingly, ligands L1−L6 were chosen to have
adequate and reasonably even variation for both B1 and B5 (see
Supporting Information for details). Similarly, substrates S1−
S5 were selected.
Model Development. The target substrates are all
commercially available. To facilitate the library synthesis, we
developed a simplified methyl pyridine ligand as an attractive
alternative to the previously reported quinoline-based ligand
(Figure 1b).4g The advantages of using the methyl pyridine
core are the rapid (in as few as three steps), modular, and
scalable synthesis from commercial materials.
The dimensionality of eq 2 does not allow visualization by
graphical means, but a plot of experimentally measured values
for enantioselectivity, as well as those predicted by eq 2, is
depicted in Figure 2c and demonstrates a high correlation
between model and the experimentally observed enantio-
selectivity, with R2 = 0.88. The model also passes the f-test at a
95% confidence level. The training set employed 30 unique
ligand−substrate combinations, and the model is capable of
describing the variation of these data using only five terms. The
main characteristics of the model demonstrate the importance
of the B1 parameter on enantioselectivity. Large positive
coefficients for B1 are observed in the substrate and ligand
dimensions. The two cross-terms are interesting. The B1LB1S
cross-term has a negative coefficient, indicating a small but
negative effect on enantioselectivity. A possible explanation for
the negative B1LB1S steric effect is that the rate of reaction is
decreased significantly when larger groups are present in both
substrate and catalyst, which might magnify the effects of less
selective background reactions eroding the observed enantio-
selectivity. The final cross-term is the B1LB5S term, indicating
that enantioselectivity is negatively affected, albeit modestly, by
substrates with large B5 values.
Validation. To evaluate the predictive power of the model
described by eq 2, extensive independent validation was
performed. Two new ligands, L7 and L8, and four new
substrates, S6−S9, were selected to validate eq 2 (Figure 3a).
Each new ligand and substrate was evaluated in combination
with all other substrates and ligands, respectively. The results
are 41 measured enantioselectivities, which were compared
with the values predicted by eq 2. The comparison between the
predicted enantioselectivity and those observed experimentally
for each new ligand−substrate combination is depicted in
Figure 3b. A perfect model would possess a slope of 1.00, where
deviation from unity represents decreasing predictive power for
a model. The slope of 0.98 indicates that eq 2 is highly
predictive. By comparison to the QSAR literature, similar
models are considered predictive when slope values are
Propargylation of S1−S5 using ligands L1−L6 led to 30
unique observed enantioselectivities, each of which was
replicated and averaged. Equation 1 is the base equation from
which all models were derived. It contains the Sterimol
parameters for the ligand substituent (B1L, B5L, and LL) as well
as the Sterimol parameters for the substrate substituent (B1S,
B5S, and LS). The base equation also contains all potential cross-
terms between the Sterimol parameters for both the substrate
and ligand to examine potential synergistic relationships. A
backward stepwise regression was performed, removing terms
and optimizing the model based on f-tests of statistical
significance for the model and p-tests for the individual
coefficients.6 The result of this regression is a simplified model
shown as eq 2.
2483
dx.doi.org/10.1021/ja4001807 | J. Am. Chem. Soc. 2013, 135, 2482−2485

Journal of the American Chemical Society
Communication
Figure 4. Evaluation of 3-methyl-2-pentanone (S9).
we evaluated two cyclic ketones, S10 and S11 (Figure 5). To
predict the enantioselectivities for these cyclic ketones, the
Figure 3. (a) Validation ligands and substrates. (b) Plot of
experimental vs predicted enantioselectivity using the model defined
in eq 2. (c) Table of new substrate and ligand combinations comparing
measured vs predicted enantioselectivity.
between 0.6 and 1.4. The validation set includes eight pairings
of a new ligand and a new substrate. Direct comparison of these
completely interpolative values is shown in Figure 3c and
highlights the efficacy of the model.
The validation substrate set includes racemic 3-methyl-2-
pentanone, which contains a chiral center. Often in asymmetric
catalysis, enantiomers of the substrate interact distinctively with
a chiral catalyst, leading to increased rate for matched
diastereomeric catalyst/substrate pairs and decreased rate for
mismatched pairs. Subjecting S9 to the ketone propargylation
protocol led to equal generation of both diastereomers for all
eight ligands evaluated. This general lack of kinetic resolution
between catalyst and substrate alone is not compelling in a
synthetic setting. However, chiral separation of the product
diastereomers revealed that both were formed with the same
enantiomeric ratio for each of the eight ligands evaluated
(Figure 4). This suggests that the catalyst is not differentiating
the substrate according to the adjacent carbons’ configuration.
Juxtaposed to the lack of diastereomeric resolution is the
predictive power of eq 2, which predicts with reasonable
accuracy the enantioselectivity for this substrate (Figure 4). The
fact that eq 2 is heavily dependent on B1 or proximal steric
effects implies that the facial selectivity imparted by the catalyst
is governed by this consequence independent of substrate
chirality, although the steric difference between an ethyl and a
methyl group is modest at best. Mechanistically, these results
may suggest an open transition state, significantly reducing the
number of hypothetical mechanisms of asymmetric induction.
Equation 2 was generated with a training set that included
only methyl ketones, and the resultant model suggests that the
substrate is orienting itself in such a manner to minimize steric
interactions among its larger group and the carbamate
substituent of the ligand. To expand the utility of the model,
Figure 5. Evaluation of cyclic ketones.
existing model was revised to include the calculated Sterimol
parameters for each side of the ketone, and the smaller
substituent parameters were subtracted from the corresponding
larger parameters. For the training set, this equates to the
subtraction of the B1, B5, and L values for the methyl group. A
new model was derived using the difference between Sterimol
values of the substrate, and the only effect is a change to the
coefficient values shown in eq 3.
ΔΔG⧧ = −0.58 + 0.81B1L + 0.94B1S − 0.27B1LB1S
− 0.087B1LB5S
(3)
The Sterimol values of the cyclic ketones could then be
calculated and the difference taken and applied to eq 3. Figure 5
depicts the plots of predicted enantioselectivities from eq 3 for
S10 and the experimentally observed values. The results show
that eq 3 is capable of predicting enantioselectivity with
reasonable accuracy, validating the subtractive model. However,
the predicted enantioselectivities for S11 were much higher
2484
dx.doi.org/10.1021/ja4001807 | J. Am. Chem. Soc. 2013, 135, 2482−2485

Journal of the American Chemical Society
Communication
Chirality 2003, 15, 360. (e) Kozlowski, M. C.; Panda, M. J. Org. Chem.
2003, 68, 2061. (f) Jiang, C.; Li, Y.; Tian, Q.; You, T. J. Chem. Inf.
Comput. Sci. 2003, 43, 1876. (g) Melville, J. L.; Andrews, B. I.; Lygo,
B.; Hirst, J. D. Chem. Commun. 2004, 1410. (h) Melville, J. L.;
Lovelock, K. R. J.; Wilson, C.; Allbutt, B.; Burke, E. K.; Lygo, B.; Hirst,
J. D. J. Chem. Inf. Model. 2005, 45, 971. (i) Ianni, J. C.; Annamalai, V.;
Phuan, P.-W.; Panda, M.; Kozlowski, M. C. Angew. Chem., Int. Ed.
2006, 45, 5502. (j) Chen, J.; Wen, J.; Li, M.; You, T. J. Mol. Catal. A:
Chem. 2006, 258, 191. (k) Urbano-Cuadrado, M.; Carbo, J. J.;
Maldonado, A. G.; Bo, C. J. Chem. Inf. Model. 2007, 47, 2228. (l) Jiang,
C.; Li, D.; Wen, J.; You, T. J. Mol. Model. 2007, 13, 91. (m) Donoghue,
P. J.; Helquist, P.; Norrby, P.-O.; Wiest, O. J. Chem. Theory Comput.
2008, 4, 1313. (n) O. Nilsson Lill, S.; Forbes, A.; Donoghue, P.;
Verdolino, V.; Wiest, O.; Rydberg, P.; Norrby, P.-O. Curr. Org. Chem.
2010, 14, 1629. (o) Denmark, S. E.; Gould, N. D.; Wolf, L. M. J. Org.
Chem. 2011, 76, 4337. (p) Denmark, S. E.; Gould, N. D.; Wolf, L. M. J.
Org. Chem. 2011, 76, 4260. (q) Weill, N.; Corbeil, C. R.; De Schutter,
J. W.; Moitessier, N. J. Comput. Chem. 2011, 32, 2878.
(2) (a) Tute, M. S. In Comprehensive Medicinal Chemistry; Hansch,
C., Sammes, P. G., Taylor, J. B., Eds.; Pergamon Press: Oxford, 1990;
Vol. 4, p 1. (b) Hansch, C.; Leo, A. Exploring QSAR: Fundamentals and
Applications in Chemistry and Biology; American Chemical Society:
Washington, DC, 1995. (c) Kubinyi, H. Drug Discovery Today 1997, 2,
457. (d) Kubinyi, H. Drug Discovery Today 1997, 2, 538. (e) Gonzalez,
M. P.; Teran, C.; Saiz-Urra, L.; Teijeira, M. Curr. Top. Med. Chem.
2008, 8, 1606. (f) Verma, J.; Khedkar, V. M.; Coutinho, E. C. Curr.
Top. Med. Chem. 2010, 10, 95.
than observed (Figure 5). Evaluation of the predicted versus
measured plot reveals good correlation of the predicted values
but poor accuracy, suggesting that the source of this error is
systematic, with each enantioselectivity being overestimated by
∼1 kcal/mol. Because S10 and S11 present similar steric
environments, especially in regard to the B1 parameter, we
explored a potential source of this systemic error. The purpose
of this study was to evaluate the steric effects of ligand and
catalyst simultaneously; to simplify the problem, we did not
incorporate an electronic parameter in either our design matrix
or our analysis, even though previous efforts suggest that
electronic effects may be pertinent.4g To examine the potential
electronic difference between these ketones, the carbonyl
stretching frequencies were measured, and a significant
disparity is observed (1733 cm−1 for S10 and 1753 cm−1 for
S11). The electronic difference between the ketones could be
the source of this large systematic error, and ongoing studies
are focused on the complex analysis associated with these
combined effects.
In conclusion, through the application of Sterimol parame-
ters and multivariate linear regression models, catalyst and
substrate steric effects have been correlated. Extensive
validation of these models resulted in excellent predictive
power. Although this analysis was limited to steric interactions,
the potential for applying a multivariate parametrization
approach to include electronic effects could greatly enhance
the application and breadth of the predictive power. Ultimately,
a substrate scope could be designed using DoE principles to
explore known reaction sensitivities, followed by regression
analysis to generate models capable of predicting the
performance of a wide range of substrates. A substrate scope
defined by a model, rather than the simple or available
substrates often reported, simplifies a key challenge in applying
asymmetric catalysis: knowing how to extrapolate catalyst
performance to substrate types not included in the original
scope evaluation. Not only would this impact the synthetic user
of the method, but the results would likely expose the key
features in the origin of asymmetric induction. These are goals
of our ongoing program.
(3) (a) Jacobsen, E. N.; Zhang, W.; Guler, M. L. J. Am. Chem. Soc.
1991, 113, 6703. (b) Lewis, C. A.; Gustafson, J. L.; Chiu, A.; Balsells,
J.; Pollard, D.; Murry, J.; Reamer, R. A.; Hansen, K. B.; Miller, S. J. J.
Am. Chem. Soc. 2008, 130, 16358. (c) Rodríguez-Escrich, S.; Reddy, K.
S.; Jimeno, C.; Colet, G.; Rodríguez-Escrich, C.; Sola,
̀
L. s.; Vidal-
Ferran, A.; Pericas, M. A. J. Org. Chem. 2008, 73, 5340. (d) Zuend, S.
̀
J.; Jacobsen, E. N. J. Am. Chem. Soc. 2009, 131, 15358. (e) Knowles, R.
R.; Jacobsen, E. N. Proc. Natl. Acad. Sci. U.S.A. 2010, 107, 20678.
(f) Knowles, R. R.; Lin, S.; Jacobsen, E. N. J. Am. Chem. Soc. 2010, 132,
5030. (g) Gustafson, J. L.; Sigman, M. S.; Miller, S. J. Org. Lett. 2010,
12, 2794. (h) Uyeda, C.; Jacobsen, E. N. J. Am. Chem. Soc. 2011, 133,
5062.
(4) (a) Jensen, K. H.; Sigman, M. S. Angew. Chem., Int. Ed. 2007, 46,
4748. (b) Miller, J. J.; Sigman, M. S. Angew. Chem., Int. Ed. 2008, 47,
771. (c) Sigman, M. S.; Miller, J. J. J. Org. Chem. 2009, 74, 7633.
(d) Jensen, K. H.; Sigman, M. S. J. Org. Chem. 2010, 75, 7194.
(e) Jensen, K. H.; Webb, J. D.; Sigman, M. S. J. Am. Chem. Soc. 2010,
132, 17471. (f) Harper, K. C.; Sigman, M. S. Proc. Natl. Acad. Sci.
U.S.A. 2011, 108, 2179. (g) Harper, K. C.; Sigman, M. S. Science 2011,
333, 1875. (h) Harper, K. C.; Bess, E. N.; Sigman, M. S. Nat. Chem.
2012, 4, 366.
(5) (a) Verloop, A. In Drug Design; Ariens, E. J., Ed.; Academic Press:
New York, 1976; Vol. III, p 133. (b) Verloop, A., Tipker, J. In
Biological Activity and Chemical Structure; Buisman, J. A., Ed.; Elsevier:
Amsterdam, 1977; p 63. (c) Verloop, A., Tipker, J. In QSAR in Drug
Design and Toxicology; Hadzi, D., Jerman-Blazic, B., Eds.; Elsevier:
Amsterdam, 1987; p 97. (d) Verloop, A. In IUPAC Pesticide Chemistry;
Miyamoto, J., Ed.; Pergamon: Oxford, 1983; Vol. 1, p 339.
(6) Deming, S. N.; Morgan, S. L. Experimental Design: A Chemometric
Approach, 2nd revised and expanded edition.; Elsevier: Amsterdam,
1993.
ASSOCIATED CONTENT
* Supporting Information
Experimental procedures, model development, and character-
ization data for new substances. This material is available free of
charge via the Internet at http://pubs.acs.org.
■
S
AUTHOR INFORMATION
Corresponding Author
sigman@chem.utah.edu
■
Notes
The authors declare no competing financial interest.
ACKNOWLEDGMENTS
■
K.C.H. thanks the ACS Division of Organic Chemistry and the
University of Utah Graduate School for fellowships. This work
was supported by the National Science Foundation (CHE-
1110599).
REFERENCES
■
(1) (a) Kozlowski, M. C.; Waters, S. P.; Skudlarek, J. W.; Evans, C. A.
Org. Lett. 2002, 4, 4391. (b) Kozlowski, M. C.; Panda, M. J. Mol.
Graphics Modell. 2002, 20, 399. (c) Lipkowitz, K. B.; Kozlowski, M. C.
Synlett 2003, 1547. (d) Fristrup, P.; Tanner, D.; Norrby, P.-O.
2485
dx.doi.org/10.1021/ja4001807 | J. Am. Chem. Soc. 2013, 135, 2482−2485