A priori theoretical prediction of selectivity in asymmetric catalysis: Design of chiral catalysts by using quantum molecular interaction fields

Article abstract of DOI:10.1002/anie.200600329

Excellent forecast: The selectivities for a set of chiral catalysts were predicted by methods derived from quantum mechanical molecular interaction fields that were applied to ground-state structures rather than transition-state structures. The prediction

Full text of DOI:10.1002/anie.200600329

Communications
Computational Models
DOI: 10.1002/anie.200600329
A Priori Theoretical Prediction of Selectivity in
Asymmetric Catalysis: Design of Chiral Catalysts
by Using Quantum Molecular Interaction Fields**
James C. Ianni, Venkatachalam Annamalai, Puay-
Wah Phuan, Manoranjan Panda, and
Marisa C. Kozlowski*
Much effort has been devoted toward the development of
asymmetric catalysts for the synthesis of chiral compounds in
pure form.[1] Despite this body of work, the science of
asymmetric catalysis remains far from exact, and the process
of finding a new catalyst usually requires a large investment of
manpower and funding. Therefore, a major goal is the
development of more efficient methods for identifying
highly selective asymmetric catalysts. In terms of computa-
tional methods, the calculation of ground state and transition
state energies[2] for metal catalysts is promising.[3] Even so, a
detailed knowledge of the reaction mechanism is needed, and
many variables must be evaluated requiring extensive time,
effort, and resources. This communication illustrates the first
example of rapid quantitative structure–selectivity relation-
ship (QSSR) calculations being used to predict the enantio-
selectivities of new chiral catalysts. Subsequent synthesis and
assessment of the new chiral catalysts in the asymmetric
addition of Et2Zn to benzaldehyde (Scheme 1) revealed that
the QSSR calculations forecast the experimental results with
a high degree of fidelity.
In a prior report[4] we determined that grid-based QSSR
methods[5] derived from quantum mechanical molecular
interaction fields can be used to generate statistically valid
models that correlate enantioselectivities in the addition of
Et2Zn to PhCHO with catalyst geometries obtained from
transition structures. The true test of this method lies in
predicting the selectivity of catalysts with ligands that have
not been tested and for which the experimental results cannot
be easily anticipated based on previous results. To this end, we
designed P1, a cyclohexane with trans amino and hydroxy
[*] Dr. J. C. Ianni, V. Annamalai, Dr. P.-W. Phuan, Dr. M. Panda,
Prof. M. C. Kozlowski
Roy and Diana Vagelos Laboratories
Department of Chemistry
University of Pennsylvania
Philadelphia, PA 19104-6323 (USA)
Fax: (+1)215-573-7165
E-mail: marisa@sas.upenn.edu
[**] Financial support of this research was provided by the NIH
(GM59945). Computing resources from the NCSA (CHE020059)
and the NSF CRIF program (CHE0131132) are acknowledged. We
thank Steve Dixon, Giorgio Lauri, and Prof. Kenneth Merz for
assistance with the QM-QSAR program. We are grateful to Bill
Nugent for supplying the unpublished data for P3.
Supporting information for this article is available on the WWW
under http://www.angewandte.org or from the author.
5502
ꢀ 2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2006, 45, 5502 –5505

Angewandte
Chemie
critical as good QSSR models could not be obtained for the
monomers (trigonal at Zn). The dimers (tetrahedral at Zn)
likely reflect the interactions encountered during reaction
(tetrahedral at Zn). The syn zinc dimers were selected as they
have been identified crystallographically[9] and in solution[10]
for the chiral catalysts.[11] The geometry for the half of the
calculated dimer of T9 (PM3) was quite similar to the
reported crystallographic structure (rmsd = 0.23 Š) as well as
to other calculated dimers.[10,12] Similarly, the geometry of the
half dimer of T9 matches closely with that portion in
calculated transition states.[13]
Scheme 1. Generation of the catalyst and the b-aminoalkoxide-cata-
lyzed addition of Et2Zn to PhCHO.
Use of the QSSR method with these dimers (aligned by N,
Zn, O atoms; bold in Scheme 1) provided models that were
highly accurate. Notably, these ground state structures were
obtained much more readily and with less computational time
than the corresponding transition structures. A “leave-two-
out” analysis using all 153 combinations of 16 parameter-
ization- and 2 prediction-set elements from T1–T18 indicated
a high degree of cross-validation for these internal predictions
(r2cv = 0.85, see the Supporting Information). Figure 2 illus-
groups (Figure 1). This type of ligand, with a chiral quaternary
alcohol carbon, is not found in the parameterization set T1–
T18 used to calibrate the QSSR model.[6] Since syntheses of
Figure 2. a–c) Several of the models from the “leave-two-out” analyses.
d) Superimposed models from all 153 trials. Blue: ee value increases
with increasing grid-point energy; red: ee value decreases.
Figure 1. b-Amino alcohols used in parameterization (T1–T18) and
a priori predictions (P1–P7).
trates the grid points (sphere centers) correlated to selectivity
for three of the models from this analysis as well as an overlay
of all 153 models. The method is robust as the models are
relatively unaffected by the constitution of the parameter-
ization set. In addition, these points indicate which portions of
the catalyst are most important to the stereoselectivity. For
P2[7] and P3[8] had been reported, we wished to predict the
selectivities of P1–P6. When the QSSR model based on
transition state structures[4] was used to generate predictions,
the standard deviations were high and the predictions were
found to be dependent on the parameterization set. To solve example, the dimethylmethylene bridge of T9 and T11
these problems, further investigation of the QSSR protocol
was undertaken.
correlates to high selectivity, whereas the phenyl group
nearest the hydroxy group in T6–T8 correlates to moderate
selectivity.
With this model in hand, we generated predictions for
catalysts derived from ligands P1–P6. In the first analysis
(Table 1, single run) with the parameterization set comprising
T1–T18, predicted DG values were obtained and translated
into enantioselectivity values at 08C (273 K). In a second
It is highly desirable to be able to apply a QSSR protocol[5]
without knowing the reaction pathway (i.e., the transition
structures previously used in the parametrization).[4] We have
achieved this here using the catalyst geometries found in the
ground state dimers (aligned by N, Zn, O atoms; bold in
Scheme 1) of T1–T18.[6] The geometry around the zinc atom is
Angew. Chem. Int. Ed. 2006, 45, 5502 –5505
ꢀ 2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
5503

Communications
Table 1: QSSR a priori predictions for catalysts derived from P1–P7.
resulted in a dramatic improvement in
enantioselectivity largely independent of
the substituents on the nitrogen atom.
Also in agreement with the calculations,
the isopropenyl (P2) and cyclopropyl (P3)
substitution was found to have little effect
on the enantioselectivity (cf. P1d, P2d, P3).
To illustrate that the model provides
good predictions for poorly selective as well
as moderately and highly selective catalysts,
further catalysts were screened computa-
tionally. Of those, P7 was predicted to
provide low selectivity (50% ee) which
was borne out in subsequent experiments
(33% ee). Overall, the agreement between
the experimental values and the predictions
for a variety of ligands was very good as
illustrated in Figure 3 (r2pred: = 0.87). Based
upon these results, the mixture of P4c and
P5 obtained during the ligand synthesis
(Scheme 2) was also tested and provided
the product with very high selectivity
(97% ee) as predicted by the model. Thus,
Single run[a]
“Leave-two-out”[b]
Ligand
DG[c]
% ee[d]
Mean
SD
% ee[d]
CI
Expt.
DG[c]
DG[c]
% ee[e]
% ee[f]
P1a
P1b
P1c
P1d
P2a
P2b
P2c
P2d
P2e
P3
P4a
P4b
P4c
P4d
P5
1.74
0.81
1.16
0.80
1.01
0.92
1.14
1.00
0.74
1.11
2.88
2.10
3.61
3.67
2.71
0.89
0.47
92.3
63.5
78.9
63.0
73.1
69.3
78.5
72.7
59.2
77.4
99.0
96.0
99.7
99.8
98.7
67.6
41.2
1.68
0.91
1.45
1.01
1.05
1.03
1.19
1.13
0.82
1.26
2.68
2.01
3.77
3.78
2.69
0.93
0.55
0.22
0.33
0.38
0.28
0.26
0.35
0.31
0.24
0.43
0.29
0.31
0.38
0.77
0.75
0.16
0.21
0.24
91.5
68.8
87.2
73.2
74.9
73.9
80.2
78.0
64.1
82.3
98.6
95.2
99.8
99.8
98.6
69.8
50.1
3.2
15.9
8.2
11.8
10.3
14.6
10.0
8.7
23.1
8.4
0.8
3.2
0.3
–
–
89
83
67
70
87
81
48
84
–
–
97
94
98
59
33
0.2
0.4
9.8
8.0
P6
P7
[a] Compounds T1–T18 served as the parameterization set. [b] “Leave-two-out” cross-validating analysis
using 16 compounds from T1–T18 as the parameterization set (all 153 combinations). [c] Calculated DG
in kcalmolÀ1. [d] Generated from DG at 273 K; S enantiomer product. [e] 95% confidence interval.
[f] From reactions performed at 273 K. S enantiomer product.
analysis (Table 1, “leave-two-out”) all combinations of 16
parameterization-set elements from T1–T18 were employed,
resulting in 153 different models and 153 sets of predicted DG
values for catalysts derived from P1–P6. From these data,
95% confidence intervals for the predicted mean ee values
were generated.
After these predictions were made, we discovered that
Singaram et al. had reported the selectivities for catalysts
derived from the limonene derivatives P2.[14] We were
delighted to find a high degree of correlation between the
calculated and reported selectivities. The correlation was
good even with the conformationally flexible n-butyl com-
pound P2e. For P3, Nugent provided us with unpublished
results (84% ee) which agreed very well with the QSSR
predictions (single 77% ee, mean 82% ee).[15] To complete
our study we synthesized representative examples of the
remaining compounds in pure form using the protocol
outlined in Scheme 2 for P4c and P5.[16] We were excited to
find that the experimental values for catalysts derived from
P1c, P1d, P4c, P4d, P5, and P6 (89, 83, 97, 94, 98, 59% ee)
were in good accord with our predictions (87, 73, 99, 99, 99,
69% ee, respectively)[17] and generally fell within the calcu-
lated 95% confidence intervals. Notably, in accord with the
calculations, a change from a methyl group (P1) to a phenyl
group (P4) at C1 immediately next to the hydroxy group
Figure 3. Plot of predicted vs. experimental enantiomeric excesses for
catalysts derived from P1–P7.
the quantum QSSR model guided our discovery of this highly
enantioselective catalyst system, which can be prepared in
two simple steps from commercial materials.
Since the initial pioneering theoretical analyses (MP2/6-
31G*/Zn//RHF/3-21G/Zn[12] and B3LYP/6-31G*/Zn//RHF/3-
21G/Zn[13]) by Yamakawa and Noyori of the asymmetric
alkylation of aldehydes, many groups have used various types
of transition state calculations employing parameterized force
fields (Q2MM[3c]), semiempirical approaches (AM1,[3a]
PM3[18]), mixed methods (RHF/LanL2DZ:UFF,[19] RHF/
[21]
LanL2DZ:MM3[20]), DFT, and post-HF methods (MP2[3b]
)
to understand or design selective chiral catalysts. Notably, the
complexity of the system, the number of relevant structures,
and the presence of the zinc atom require considerable
computational resources for full ab initio or DFTcalculations.
For a set of nine catalysts, PM3 performed very well[18] (rp2red:
=
0.85) but was less satisfactory with a more diverse set of
catalysts.[19a] Other cases with multiple catalysts that per-
Scheme 2. Synthesis of prediction-set b-amino alcohols P4c and P5.
DMM=dimethoxymethane.
5504
www.angewandte.org
ꢀ 2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2006, 45, 5502 –5505

Angewandte
Chemie
formed well utilized QM/MM methods[19] (rp2red: = 0.43) and
single-point DFTcalculations
S. M. Hooijschuur, G. M. Klaus, N. T. Luchters, P. Dani, G.
[21]
(r2pred: = 0.58). Overall, these
Verspui, A. A. Smith, E. W. P. Damen, B. McKay, M. Hoogen-
raad, QSAR Comb. Sci. 2005, 24, 94 – 98; f) S. Sciabola, A. Alex,
P. D. Higginson, J. C. Mitchell, M. J. Snowden, I. Morao, J. Org.
Chem. 2005, 70, 9025 – 9027.
calculations were very good in their relative ranking of
catalysts and for key transition structure features. The QSSR
method identified substituents on the nitrogen atom and on
the carbon atom adjacent to the hydroxy group as important
to enantioinduction, in agreement with these transition state
calculations. In addition, the QSSR method is able to identify
other locations at which substitution is advantageous, for
example, the distal bridging elements in T9–T13 and the
phenyl group in P5. Overall, our QSSR method performs as
well as or better (r2pred: = 0.87) than other methods, while
requiring fewer computational resources since ground states
are employed exclusively.
In conclusion, we have shown that grid-based QSSR
methods can be used to predict catalyst enantioselectivities
with a high degree of accuracy and precision. This method
requires knowledge of only the ground state catalyst struc-
ture. The empirical models are easily assembled from a small
set of ligands and their experimentally measured selectivities
such as might be determined in preliminary screening.
Relatively quick theoretical calculations provide models
that allow a researcher to readily distinguish poorly, moder-
ately, and highly selective catalysts. Within a catalyst series,
modifications that result in relatively small changes (i.e.,
P2a–P2e) can also be estimated with a good degree of
reliability. Studies are underway using this computational
method with other reactions, and preliminary results are
promising.[22] In addition, the readily prepared catalysts (two
steps) that we have discovered are being explored in other
transformations.
[6] Experimental selectivies for T1–T18: a) M. Kitamura, S. Suga,
K. Kawai, R. Noyori, J. Am. Chem. Soc. 1986, 108, 6071 – 6072;
b) L. Pu, H.-B. Yu, Chem. Rev. 2001, 101, 757, and references
therein.
[7] C. T. Goralski, B. Singaram, W. Chrisman, US Patent
6362373B1, 2002.
[8] a) G. S. Kauffman, G. D. Harris, R. L. Dorow, B. R. P. Stone,
R. L. Parsons, Jr., J. A. Pesti, N. A. Magnus, J. M. Fortunak, P. N.
Confalone, W. A. Nugent, Org. Lett. 2000, 2, 3119 – 3121;
b) R. L. Parsons, J. M. Fortunak, R. L. Dorow, G. D. Harris,
G. S. Kauffman, W. A. Nugent, M. D. Winemiller, T. F. Briggs, B.
Xiang, D. B. Collum, J. Am. Chem. Soc. 2001, 123, 9135 – 9143.
[9] M. Kitamura, S. Okada, S. Suga, R. Noyori, J. Am. Chem. Soc.
1989, 111, 4028 – 4036.
[10] M. Kitamura, M. Yamakawa, H. Oka, S. Suga, R. Noyori, Chem.
Eur. J. 1996, 2, 1173 – 1181.
[11] With homochiral catalysts, syn dimers have been observed
experimentally: a) E. Hecht, Z. Anorg. Allg. Chem. 2000, 626,
2223 – 2227; b) K. Nakano, K. Nozaki, T. Hiyama, J. Am. Chem.
Soc. 2003, 125, 5501 – 5510. With achiral catalysts, anti dimers
predominate: c) M. R. P. van Vliet, J. T. B. H. Jastrzebski, G.
van Koten, K. Vrieze, A. L. Spek, J. Organomet. Chem. 1983,
251, C17-C21; d) M. R. P. van Vliet, G. van Koten, P. Buysingh,
J. T. B. H. Jastrzebski, A. L. Spek, Organometallics 1987, 6, 537 –
546; e) J. T. B. H. Jastrzebski, J. Boersma, G. van Koten, W. J. J.
Smeets, A. L. Spek, Recl. Trav. Chim. Pays-Bas 1988, 107, 263 –
266; f) M. Kitamura, S. Suga, M. Niwa, R. Noyori, J. Am. Chem.
Soc. 1995, 117, 4832 – 4842.
[12] M. Yamakawa, R. Noyori, J. Am. Chem. Soc. 1995, 117, 6327 –
6335.
[13] M. Yamakawa, R. Noyori, Organometallics 1999, 18, 128 – 133.
[14] D. Steiner, S. G. Sethofer, C. T. Goralski, B. Singaram, Tetrahe-
dron: Asymmetry 2002, 13, 1477 – 1483.
Received: January 25, 2006
[15] a) W. A. Nugent, personal communication; b) Subsequent to our
calculations a report of 81% ee for P3 in this same reaction
appeared: S. N. Joshi, S. V. Malhotra, Tetrahedron: Asymmetry
2003, 14, 1763 – 1766.
[16] See the Supporting Information for further details and full
characterization. Epoxidation was accomplished with Shiꢀs
protocol (Z. Wang, Y. Tu, M. Frohn, J. Zhang, Y. Shi, J. Am.
Chem. Soc. 1997, 119, 11224 – 11235) and epoxide opening with
Singaramꢀs conditions (see reference [14]).
Keywords: asymmetric addition · benzaldehyde ·
enantioselectivity · structure–activity relationships · zinc
.
[1] Comprehensive Asymmetric Catalysis (Eds.: E. N. Jacobsen, A.
Pfaltz, H. Yamamoto), Springer, New York, 1999.
[2] a) ACS Symposium Series 721: Transition State Modeling for
Catalysis (Eds.: D. G. Truhlar, K. Morokuma), American
Chemical Society, Washington, DC, 1999; b) S. Niu, M. B. Hall,
Chem. Rev. 2000, 100, 353 – 406.
[17] The enantiomers of the predicted catalysts P1, P4, and P5 were
synthesized and yielded the enantiomer (R) opposite to that
predicted (S).
[3] For the reaction in Scheme 1: designing catalysts using transition
state (TS) calculations: a) frozen TS core: A. Vidal-Ferran, A.
Moyano, M. A. Pericµs, A. Riera, Tetrahedron Lett. 1997, 38,
8773 – 8776; b) MP2/LanL2DZ//HF/LanL2DZ: M. C. Kozlow-
ski, M. Panda, J. Org. Chem. 2003, 68, 2061 – 2076. For comments
on the complexity of the transition state calculations: c) T.
Rasmussen, P.-O. Norrby, J. Am. Chem. Soc. 2003, 125, 5130 –
5138.
[18] B. Goldfuss, K. N. Houk, J. Org. Chem. 1998, 63, 8998 – 9006; C.
Jimeno, M. Pastꢁ, A. Riera, M. A. Pericµs, J. Org. Chem. 2003,
68, 3130 – 3138.
[19] a) B. Goldfuss, M. Steigelmann, S. I. Khan, K. N. Houk, J. Org.
Chem. 2000, 65, 77 – 82; b) B. Goldfuss, M. Steigelmann, F.
Rominger, Eur. J. Org. Chem. 2000, 1785 – 1792.
[20] J. Vµzquez, M. A. Pericµs, F. Maseras, A. Lledꢁs, J. Org. Chem.
2000, 65, 7303 – 7309.
[21] DFTBP86/DN**//PM3 M. Sosa-Rivadeneyra, A. Muꢂoz-
Muꢂiz, C. Anaya de Parrodi, L. Quintero, E. Juaristi, J. Org.
Chem. 2003, 68, 2369 – 2375.
[4] M. C. Kozlowski, S. Dixon, M. Panda, G. Lauri, J. Am. Chem.
Soc. 2003, 125, 6614 – 6615.
[5] For other QSAR applications in enantioselective catalysis see:
a) J. D. Oslob, B. Škermark, P. Helquist, P. O. Norrby, Organo-
metallics 1997, 16, 3015 – 3021; b) K. B. Lipkowitz, M. J. Prad-
han, J. Org. Chem. 2003, 68, 4648 – 4656; c) S. Alvarez, S.
Schefzick, K. Lipkowitz, D. Avnir, Chem. Eur. J. 2003, 9, 5832 –
5837; d) M. Hoogenraad, G. M. Klaus, N. Elders, S. M. Hooij-
schuur, B. McKay, A. A. Smith, E. W. Damen, Tetrahedron:
Asymmetry 2004, 15, 519 – 523; e) J. B. van der Linden, E. J. Ras,
[22] P.-W. Phuan, J. C. Ianni, M. C. Kozlowski, J. Am. Chem. Soc.
2004, 126, 15473 – 15479.
Angew. Chem. Int. Ed. 2006, 45, 5502 –5505
ꢀ 2006 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
5505