54

This work addresses homogeneous asymmetrichydrogenation of C=O bonds, from ketones toalcohols, which has applications in the industrialproduction of pharmaceutical intermediates. Thecatalyst studied, [(S)-XylBINAP-RuH2-(S,S)-DPEN], is shown in Figure 1.

Each reaction mechanism may be understoodin terms of the energies of intermediates and theroles of ligands and additives, as determined bystate-of-the-art computational techniques. Theknowledge gained will then be exploited for thedesign and synthesis of ligands for improved cata-lyst systems. Advances in experimental techniqueswill allow rapid identification of lead ‘designed’catalysts by automated parallel screening. With

this scheme in mind, a consortium was assembledfrom leading experts (both industrial and academ-ic) in all areas of the workflow. The partners arethe Royal Institution of Great Britain, theUniversity of Liverpool, the University ofSouthampton, Johnson Matthey, AstraZeneca,GlaxoSmithKline and Pfizer. The aims of the pro-ject were to implement an evolutionaryimprovement in ligand/catalyst design strategies:

COMPUTATION ⇔ SYNTHESIS ⇔

ACCELERATED TESTING

This computation-guided approach for catalystdiscovery is expected to be more efficient, fasterdelivering and more revealing on the molecularaspects of a catalytic cycle than one-at-a-time syn-thesis or combinatory methodologies, whichusually screen catalysts at random (5); seeScheme I.

Since the project’s conception there has been astep change in the ability of industry to performhigh-throughput screening. This acceleration hasenormously reduced the time required to identifythe right catalyst for any desired transformationfrom a library of existing catalysts or ligands. Thepreparation of the library of ligands and catalystsremains, however, the bottleneck in this process.

Platinum Metals Rev., 2007, 51, (2), 54–62

DOI: 10.1595/147106707X180891

Fig. 1 The structure of [(S)-XylBINAP-RuH2-(S,S)-DPEN], asymmetric hydrogenation catalyst

Ru

PAr2 H H2

N H

Ru

P H N HAr2 H2

Modelling Reactions at the Active Sites ofChiral Ruthenium Catalysts Using DensityFunctional TheoryNEW APPROACH TO UNDERSTANDING OF CATALYTIC REACTIONS

By S. A. FrenchJohnson Matthey Technology Centre, Blounts Court, Sonning Common, Reading RG4 9NH, U.K.; E-mail: frencsa@matthey.com

Selectivity is a key success factor in the chiral catalyst technologies market. Understandingthe fundamental processes that occur when a reagent interacts with a homogeneous singlesite catalyst, both in its approach and at the active site, is therefore critical to the rationaldesign of new catalysts. Ruthenium-based asymmetric hydrogenation catalysts have beenconsidered as part of a collaborative research project. [(S)-XylBINAP-RuH2-(S,S)-DPEN],first developed by Noyori (1–3), is studied as the parent or prototype model of a series ofefficient hydrogenation catalysts, among them the catalysts based on the P-Phos, PhanePhosand ParaPhos ligand families (4).

For example, the selection of the correct sub-stituents at phosphorus in any family of bidentatephosphine ligands is largely a matter of trial anderror, with each component of the family requir-ing independent and often time-consumingsynthesis. The design and synthesis of new ligandbackbones is even more time-consuming, andthere is no certainty that the new ligands will beeffective in the desired transformation.Understanding the factors which govern the rela-tionship between the structure of the ligand andits efficacy in catalysis will accelerate the processof ligand design by evaluation through computersimulation (‘in silico’) of a large number of structur-al variations, among which only the mostpromising structures will actually be synthesised.

The project, supported financially by the U.K.Department of Trade and Industry’s‘Manufacturing Molecules Initiative’ (6), isfocused on two industrially important organicreactions: (a) the production of chiral alcohols viathe asymmetric reduction of ketones; and (b) C–Cbond forming reactions such as the Heck reaction.Molecular modelling has so far focused onReaction (a). Computer simulations have beenused, at the molecular-mechanical (7) and, asreported here, quantum-mechanical levels, toinvestigate the structures of proposed catalysts,and to probe the reaction mechanism.

Initially it was proposed to use activation ener-gies calculated by considering transition states(TS) between reactant and products to comparethe performance of catalysts. However, difficultiesin simulating the reactants and products correctlycaused the TS calculations to fail. Therefore analternative strategy was implemented; a geometricconstraint was applied and the reactant broughttowards the reactive centre, exploring the pathwayof the ketone molecule to the active site.Understanding of the correct relative positions of

reactants and products, and further understandingthe need to ‘lock’ conformations of ligands, hasled to the capability of performing TS calculationson ‘cut down’ (i.e. simplified) model catalysts, aswell as exploring the entry of reactants to a realcatalyst system.

MethodsThe processes of prime interest to us involve

the breaking and creation of bonds, which meansthat the electronic structure as well as the molecu-lar structure must be modelled. Traditionalmethods in electronic structure theory, in particu-lar Hartree-Fock theory and its descendants, arebased on the complicated many-electron wave-function. The main objective of density functionaltheory (DFT) is to replace the many-body elec-tronic wavefunction as the basic quantity by theelectronic density. Within this study we have usedthe DFT code DMol3 (8, 9) for both model TScalculations and constrained optimisations.

All the DFT investigations were performedusing the linear combination of atomic orbitalsapproximation, with a double numerical basis setaugmented by polarisation functions (with a 5.5 Åcut-off). The calculations employed the gradient-corrected Perdew-Becker-Ernzerhof (PBE)exchange-correlation functional. The fine accura-cy convergence criteria were used throughout forboth electronic structure and atomic optimisationcalculations. The criteria guarantee that the energyper bond, bond lengths and angles converge toapproximately 0.1 eV, 0.01 Å and 1º, respectively.

Constrained Optimisation CalculationsTo understand how the reactant molecule

approaches the metal centre and what restrictionsare placed on its passage, we have performed alarge number of simulations to compile a trajecto-ry of the path followed, and to compare the

Platinum Metals Rev., 2007, 51, (2) 55

Catalyst Modelling(Royal Institution)

Ligand Synthesis(Liverpool, Southampton

Universities)

Catalyst Screening(Johnson Matthey)

Commercial Application(Pharmaceutical

Companies)

Scheme I Roles of research consortium members

Platinum Metals Rev., 2007, 51, (2) 56

energy barriers that are encountered. At eachstage, the geometry of the system is optimisedwith respect to one constraint, namely the dis-tance between the hydride H on the rutheniumand the C of the carbonyl group in the ketone.(These species eventually become bonded to oneanother in the alcohol product.) The output fromone simulation is used to generate the initial con-figuration for the next. Initially we used the resultsto understand the interaction between reactantand catalyst, to identify the most relevant reac-tants and products for TS calculations, which aredescribed below. The application of the methodwas then extended to consider four ‘quadrants ofattack’ of the ketone to the active site, to probethe potential energy surfaces of the reaction. Thiswill provide vital information concerning theselectivity of the catalysts.

Transition State CalculationsThe TS calculations have required a workflow

applied to a cut down version of the catalyst(Figure 2), to arrive at a final model for the TS.The stages are as follows: (a) Relaxation of the reactant;(b) Construction of the product from the reac-

tant using ‘chemical intuition’ to moveatoms around;

(c) Constrained relaxation of the product, tak-ing care to avoid conformational changes inthe ring;

(d) Full relaxation of the product;(e) Linear synchronous transit (LST) method

to approximate reaction path and provideinput to full LST/quadratic synchronoustransit (QST) calculation with conjugategradient (CG) optimisation;

(f) Single-point calculation of TS with frequen-cy analysis;

(g) Animation of negative mode to checkwhich centres are involved;

(h) TS optimisation calculation where mode isfollowed;

(i) Single-point calculation of TS with frequen-cy analysis;

(j) Animation of negative mode.

Results and DiscussionAmong the models used to rationalise the

structure-activity relationship in asymmetrichomogeneous catalysis, the so-called ‘quadrantapproach’ is one of the simplest and most effec-tive.

The space around the reactive centre is dividedinto four volumes, across which the substrate canbind to the metal centre in a number of differentconformations. The ligand will prevent access tosome quadrants by simple steric interaction,thereby forcing the substrate to bind to the metalin a preferred conformation that, upon transfer ofthe hydride from the metal to the substrate,becomes the precursor to the favoured enan-tiomer of the product. Such a model, althoughvery simplistic, allows straightforward rationalisa-tion of the sense of stereoinduction obtainablewith a number of well-known hydrogenation cat-alysts such as Ru-BINAP, Rh-DuPhos andRh-BisP* (10, 11).

Starting from this simplistic approach (seeFigure 3) we have developed a more sophisticatedmodel that takes into account the whole trajecto-ry of the substrate into the ‘reactive pocket’ of thecatalyst. It is well known that subtle modificationsof the substituents at phosphorus can producevery significant changes in the activity and selec-tivity of the catalyst (one example of this being theso-called ‘meta-effect’). We suggest that the rea-son for these effects may reside not only inchanges at the transition state, but also in thedocking of the substrate into the reactive pocket,well before the bond breaking/bond forminginteractions are established.

Constrained Optimisations: Initial QuadrantInitially we have considered the catalyst

[(S)-XylBINAP-RuH2-(S,S)-DPEN] as an exem-

PH2

PH2

Ru

NH2

NH2

H

H HH

HH

H

H

H

H

Fig. 2 Cut down model of the Noyori catalyst

Platinum Metals Rev., 2007, 51, (2) 57

plar of the class of asymmetric materials that weare interested in understanding. As mentionedabove, initially we focused on understanding thepathway of the reactant to the active site. Withthis in mind we chose to model the quadrant andorientation known to lead to the preferred prod-uct. Determining the conformational changesforced on the reactant or catalyst during approachwould provide a better understanding of wherereactants and products should be sited for TS cal-culations. The Platinum Metals Review websiteincludes an animation (12) showing the finalstructure from each of the constrained optimisa-tion calculations, starting with a constraint (Ru–H - - - C=O) of 8 Å and reducing the separa-tion between the reactant and catalyst in steps of0.5 Å. It is clear from the animation and subse-quent analysis of the potential energy surface forthe pathway (Figures 4 and 5) that we can beginto understand the complexity of these systems.There are two distinct energy barriers that the

reactant must overcome before arriving at theactive site of the catalyst. The reactant must firstpush into a pocket of the catalyst, before arrivingat its final alignment. The C=O of the ketone andthe Ru–N bonds lie parallel, thereby maximisingorbital overlap with the hydrogen atoms thattransfer to form the alcohol. The advantage ofcomputational models is that the changes in geo-metrical structure as the ketone approaches canbe observed ‘frame by frame’. It is then possibleto follow the trajectory of approach, analyse theposition of the barriers, and view the correspond-ing changes in atomic structure.

When the reactant enters the pocket of thecatalyst, which begins to occur when the con-straint (Ru–H - - - C=O) is between 5 Å and 4 Å,the ketone-catalyst system stabilises. This isshown by the total energy of the system decreas-ing, before it has to overcome the largest barrierbetween 3.5 Å and 2.75 Å. The barrier is due tothe interaction of the phenyl ring of the ketone

Ru

PP

N NORu

PP

N NORu

PP

N NO

Ru

PP

N NO

Q1(R)-alcohol

Q2(S)-alcohol

Q3(S)-alcohol

Q4(R)-alcohol

Ru

PP

N NORu

PP

N NORu

PP

N NO

Ru

PP

N NO

Q1(R)-alcohol

Q2(S)-alcohol

Q3(S)-alcohol

Q4(R)-alcohol

Q1(R)-alcohol

Q2(S)-alcohol

Q3(S)-alcohol

Q4(R)-alcohol

Q1Product:

(R)-alcohol

Q2Product:

(S)-alcohol

Q3Product:

(S)-alcohol

Q4Product:

(R)-alcohol

Fig. 3 Quadrants of inter-est showing the stericinteractions between thereactant and ligands of thecatalyst

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5

Entry intoPocket

Constraint, ÅRel

ativ

e E

nerg

y, k

J m

ol–1

– N

orm

alis

ed a

t 5 Å

Fig. 4 The potential ener-gy surface of the pathwayfor the reactant toapproach the catalystactive site. The black cir-cles show the positions onthe trajectory that areillustrated in Figure 5

5

3

1

–1

–3

–5

–7

Platinum Metals Rev., 2007, 51, (2) 58

with ligands of the catalyst; this interactionincreases as the ketone is pulled down onto theactive site. At a constraint of 2.75 Å there is a con-formational change of the reactant, with thephenyl ring tilting so that all the carbons of thering are no longer in the plane of the other atom-ic centres of the ketone. The driving force for thischange of conformation is the formation of ahydrogen bond, which holds the ketone as itmoves closer to the active site. The hydrogenbond distance reduces further as the reactant ispulled further towards the catalyst, but the posi-tion of the oxygen atom does not change greatly;the major movement is that of H(NH) of the catalyst.

For the reactant to leave the pocket there isthen another barrier, which requires the alignmentof the C=O bond of the ketone to the underlyingRu–N bond of the catalyst. To this end, the car-bon moves further down, changing from sp2 to sp3

hybridisation, until it is in the same plane as oxy-gen. This results in the C=O bond lying parallel tothe Ru–N bond, maximising overlap with the twohydrogen atoms to be transferred from the catalystto the reactant. Simultaneously, the Ru–H bondelongates; this would contribute to the barrier at a(Ru–H - - - C=O) constraint of 2.25 Å.

Constrained Optimisations: Other QuadrantsHaving considered the approach of the reac-

tant along the favoured quadrant, we thenaddressed the question of whether computationalcalculations contain sufficient detail to predict theselectivity of a specific catalyst. A catalyst thatforms products with a high enantiomeric excess(ee) is highly desirable, as these reactions areimportant in a pharmaceutical context. Here thephysiological reaction to one enantiomer may dif-fer greatly from that to another. Using dockingcalculations, the aim is to investigate the variousarrangements, and thereby provide insight intohow catalysts may be optimised.

The difference in energy barriers between thedifferent quadrants of approach has the potentialto provide such discrimination. From Figures 2and 6 it is obvious that two of the possible orien-tations are sterically ‘favourable’ (quadrants Q1and Q3) and two sterically ‘unfavourable’ (Q2 andQ4) (1). However, this is not sufficient to under-stand the selectivity, as Q1 and Q3 lead to (R) and(S) products, respectively. In fact what this simpleanalysis demonstrates is that the channels thatwould be followed by reactants approaching alongQ2 and Q4 should be closed. However, of mostimportance is to understand why certain catalysts

3 Å 2.75 Å

2.25 Å 1.75 Å

Fig. 5 Molecular modelsshowing stages of thereactants’ pathway to thecatalyst active site

Platinum Metals Rev., 2007, 51, (2) 59

produce much higher ee than others. To this endwe must understand the difference between Q1and Q3.

Figure 7 shows clearly that for [(S)-XylBINAP-RuH2-(S,S)-DPEN], Q1 possesses a much lower

barrier to approach and would therefore beexpected to show high ee. This is confirmedexperimentally where, depending upon ex-perimental conditions, an ee of around 97%is achieved.

Q1 Q2

Q3 Q4

Fig. 6 Starting configura-tions with a constraint of 5 Å showing the quadrants

-40

-30

-20

-10

0

10

20

30

40

1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8

Constraint (Angs)

Rel

E -

8 A

ngs

- kJ/

mol

X-Q1X-Q2X-Q3X-Q4

Rel

ativ

e E

nerg

y, k

J m

ol–1

– N

orm

alis

ed a

t 8 Å

Constraint, Å

Fig. 7 The energy profiles for approach of the reactant to the catalyst along different quadrants. (Numbers are nor-malised to 8 Å, as here the reactant and catalyst do not interact)

With an understanding of the pathway of thereactant to the catalyst active site and of the stableposition of the ketone along this path, we couldreturn to considering the calculations of TS andtherefore the activation energies for the hydro-genation processes. We are currently consideringother catalysts and reactants to ascertain whetherwe can correlate the barrier for entry into the cat-alyst pocket with the selectivity of known systems.

Transition StatesTo make the best use of available computer

resources, initial TS calculations were performedon a cut down model of a commercial catalyst(Figure 2). Acetone was initially considered as thereactant molecule. From our understanding of theapproach of the ketone to the catalyst, we havebeen able to determine a stable configuration forthe reactant. The constrained optimisation calcu-lations showed that with the ketone hydrogenbonded to the catalyst, (R2C=O - - - HNH) wasat a minimum energy position when the (Ru–H - - - C=O) was held at 2.75 Å, with thehydrogen bond holding the reactant in place. Wefurther optimised this structure to provide thestarting point for our TS calculations. The productstate has the alcohol physisorbed above the cata-lyst, and is stabilised by a hydrogen bond betweenCOH - - - NH.

The TS forms a six membered ring (Figure 8).The hydride bond is elongated from 1.7 Å to1.75 Å. It can be seen from Figure 9 that there isalso a change in the hybridisation of carbon,which moves from sp2 towards sp3. The TS has a (O)C - - - H(Ru) distance of 1.96 Å and (C)O - - - H(NH) of 1.86 Å. It is possible to visu-alise the imaginary mode that is associated withthe TS, and it is found that the two hydrogenatoms and carbon are the atomic centres thatmove the most. Figure 10 shows the energy pro-files for TS searches using the LST and QSTcalculation methods.

The activation energy that we have calculated is3 kcal mol–1, which is in agreement with previouscalculations on similar-sized models of the com-mercial catalyst, while the calculated reactionenergy is exothermic by 6 kcal mol–1. The hydro-

genation of acetone is therefore extremely facile,proceeding as follows: – Incipient bond formation is signalled by the

shortening of the O - - - H distance (from 2.11Å to 1.86 Å) in the N–H - - - O hydrogen bond,and of the Ru–H - - - C distance (from 3.02 Åto 1.95 Å);

– Small changes in the same direction areobserved in the other bond lengths (< 2%).The structure of the TS, therefore, resembles

much more that of the reactant complex[RuH2–acetone] than that of the product complex[RuH2–iPrOH]. This process is therefore a goodexample of the Hammond principle. This statesthat the structure of the transition state will resem-ble that of the product more closely than that ofthe reactant for endothermic processes, whereasthe opposite is true for exothermic reactions. A

Platinum Metals Rev., 2007, 51, (2) 60

Ru

NH2

NHH3P

H3P

H

H

H

C

O

H3C

H3C

Fig. 9 Molecular configuration for transition state inhydrogenation of acetone; EA = 3.07 kcal mol–1

Fig. 8 Valence bond representation of transition state inhydrogenation of acetone

previous computational study on a trans-dihydro(diamine)ruthenium(II) Noyori-typemodel catalyst has evaluated a reaction barrier forthe hydrogenation of acetone of 3.6 kcal mol–1 atB3LYP/6-31G** level (13). Our results are inapparent agreement with previous calculations onthe formaldehyde/methanol transformation bythe RuH(NH2CH2CH2NH)(η6-benzene) complexperformed at B3LYP (14) and generalised gradi-ent approximation (GGA) (15). This shows thatclassical reaction barriers computed with GGAfunctionals are smaller than those obtained withB3LYP by about 2 kcal mol–1.

We certainly anticipated that the methodologyused would impact on the activation energy (EA),and we are currently evaluating the effect ofchanging the density functional. Initial resultsshow that increasing electron localisation by mov-ing from GGA via hybrid to meta functionalsleads to a slight increase in EA. We are also con-sidering other ketones, and attempting to build upa larger model of the catalyst system, so that wecan make direct comparison with experimentaldata for industrially relevant systems.

ConclusionsThe two complementary DFT simulation

methodologies of transition state searches andconstrained geometry optimisations are now yield-ing results that are of considerable importance to

understanding catalyst behaviour, potentially lead-ing to the prediction and design of new catalystsfor the ketone hydrogenation reaction.

AcknowledgementsThe author would like to thank R. Catlow, E.

Palin and D. Di Tommaso (Royal Institution); J.Xiao, Z. Chen, X. Wu and J. Ruan (University ofLiverpool); A. Danopoulos and N. Stylianides(University of Southampton); F. King, F. Hancockand A. Zanotti-Gerosa (Johnson Matthey); P.Hogan and M. Purdie (AstraZeneca); P.Ravenscroft (GlaxoSmithKline); and P. Levett andA. Pettman (Pfizer).

References1 R. Noyori, Angew. Chem. Int. Ed., 2002, 41, (12),

2008 – Nobel Lecture 2 R. Noyori and T. Ohkuma, Angew. Chem. Int. Ed.,

2001, 40, (1), 40 3 T. Ohkuma, M. Koizumi, K. Muñiz, G. Hilt, C.

Kabuto and R. Noyori, J. Am. Chem. Soc., 2002, 124,(23), 6508

4 A. Zanotti-Gerosa, W. Hems, M. Groarke and F.Hancock, Platinum Metals Rev., 2005, 49, (4), 158

5 A. Hagemeyer, B. Jandeleit, Y. Liu, D. M. Poojary,H. W. Turner, A. F. Volpe and W. H. Weinberg,Appl. Catal. A: Gen., 2001, 221, (1–2), 23

6 ‘Manufacturing Molecules Initiative (MMI): ASource of Funding’, URN 02/527, U.K.Department of Trade and Industry, London, 2002

7 E. J. Palin, G. A. Grasa and C. R. A. Catlow, Mol.Simulat., 2006, 32, (10–11), 901

Platinum Metals Rev., 2007, 51, (2) 61

–1349.93

–1349.94

–1349.95

–1349.96

–1349.97

–1349.98

–1349.99

–1350.00

–1350.01

–1350.02

Ene

rgy,

Ha

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0Path Coordinate

Energy vs. LST path 1 Energy vs. CG path 1 Energy vs. QST path 2Energy vs. CG path 2 Energy vs. QST path 3 Energy vs. CG path 3Transition state

Fig. 10 Energy profilesof TS searches in hydro-genation of acetone

Platinum Metals Rev., 2007, 51, (2) 62

8 B. Delley, J. Chem. Phys., 1990, 92, (1), 508 9 B. Delley, J. Chem. Phys., 2000, 113, (18), 7756

10 I. D. Gridnev and T. Imamoto, Acc. Chem. Res.,2004, 37, (9), 633

11 J. M. Brown, ‘Hydrogenation of FunctionalizedCarbon-Carbon Double Bonds’, in“Comprehensive Asymmetric Catalysis”, eds. E. N.Jacobsen, A. Pfaltz and H. Yamamoto, Springer,1999, Vol. 1, p. 121

12 http://www.platinummetalsreview.com/images/Q1_new.gif13 K. Abdur-Rashid, S. E. Clapham, A. Hadzovic, J. N.

Harvey, A. J. Lough and R. H. Morris, J. Am. Chem.Soc., 2002, 124, (50), 15104

14 M. Yamakawa, H. Ito and R. Noyori, J. Am. Chem.Soc., 2000, 122, (7), 1466

15 J.-W. Handgraaf, J. N. H. Reek and E. J. Meijer,Organometallics, 2003, 22, (15), 3150

The Author

Sam French holds a degree in Chemistry withMedicinal Chemistry from the University ofWarwick, U.K., as well as a Ph.D. from the RoyalInstitution of Great Britain (RI). At the RI, he workedunder the supervision of Prof. Richard Catlow FRS.He held three subsequent postdoctoral researchassistant positions at the RI. All involved largecollaborations with industrial input, specialising inQM/MM techniques and grid technology in

chemistry and catalysis. He joined Johnson Matthey as a SeniorScientist in November 2004, to initiate and lead the ComputationalChemistry Group.