Ž . Journal of Molecular Catalysis B: Enzymatic 7 1999 273–282 www.elsevier.comrlocatermolcatb Study of enzyme-catalyzed reactions in organic solvents using multiple linear regression David Nurok a, ) , Robert M. Kleyle b , Barry B. Muhoberac a , Megan C. Frost a , Paul Hajdu a , Daniel H. Robertson a , Sanjay V. Kamat c , Alan J. Russell c a Department of Chemistry, School of Science, Indiana UniÕersity–Purdue UniÕersity Indianapolis, 402 N. Blackford Street, Indianapolis, IN 46202-3274, USA b Department of Mathematical Sciences, School of Science, Indiana UniÕersity–Purdue UniÕersity Indianapolis, 402 N. Blackford Street, Indianapolis, IN 46202-3274, USA c Department of Chemical Engineering and Center for Biotechnology and Bioengineering, UniÕersity of Pittsburgh, Pittsburgh, PA 15261, USA Received 23 June 1998; received in revised form 23 October 1998; accepted 19 February 1999 Abstract We have used multiple linear regression to predict either initial rate, log initial rate or specificity for enzyme-catalyzed reactions performed in non-aqueous solvents. The Subtilisin Carlsberg catalyzed transesterification of N-acetyl-L-phenyl- alanine ethyl ester by methanol, 1-propanol, and 1-butanol was assayed in 30 non-aqueous solvents, and the lipase catalyzed transesterification of methyl methacrylate in 23 non-aqueous solvents. Both sets of reactions were performed at fixed thermodynamic water activity. The lipase catalyzed reactions were also performed in water saturated solvents and in dry solvents. The report illustrates that regression analysis may provide insight into how solvents can alter the activity and specificity of enzymes suspended therein. A regression model for the subtilisin catalyzed reaction suggests that solvents which have a flat hydrophobic region inhibit by competing with the substrate for an enzyme cleft. In the lipase catalyzed Ž . reaction, tetrachloroethylene is an outlier i.e., behaves differently to other solvents for all the regression models. This deviation, together with an element of structural similarity to the substrate, suggests that tetrachloroethylene acts as a competitive inhibitor. Log P is an important descriptor and it, or an expression containing log P, appears in all the regression equations. Log initial rate is predicted by a two-descriptor model for either enzyme system in solvents of high log P at fixed thermodynamic water activity. Regression models with the same two descriptors predict initial rate for the lipase system over the entire log P range for solvents maintained at fixed thermodynamic water activity and for dry solvents, but not for water saturated solvents. q 1999 Elsevier Science B.V. All rights reserved. Keywords: Enzyme-catalyzed reactions; Muliple linear regression; Solvents 1. Introduction For over a decade there has been a concerted attempt to correlate the properties of organic ) Corresponding author. Tel.: q1-317-274-6886; Fax: q1-317- 274-4701; E-mail: nurok@chem.iupui.edu solvents to the function of biocatalysts sus- w x pended therein 1–3 . This tour de force of research in non-aqueous enzymology has, how- ever, yet to yield a predictive equation which correlates the variation of activity or specificity of multiple enzyme–substrate pairs to solvent properties. Indeed, it is rare that a prediction 1381-1177r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved. Ž . PII: S1381-1177 99 00047-8