The accurate prediction of enzyme kinetics from first principles is one of the central goals of theoretical biochemistry. Currently there is considerable debate [1][2][3] about the applicability of transition state theory (TST) to compute rate constants of enzyme-catalyzed reactions. Classical TST is known to be insufficient in some cases, but corrections for dynamical recrossing and quantum mechanical tunneling can be included. [1,2,4] Many effects that go beyond the framework of TST have been proposed, particularly focusing on the possible role of protein dynamics and conformational effects on the enzyme activity. Unfortunately, the overall importance of these effects for the effective reaction rate is difficult (if not impossible) to determine experimentally. However, if one could compute the quasi-thermodynamical free energy of activation with chemical accuracy (i.e. 1 kcal mol À1 ), comparison with the effective measured free energy of activation would directly show the importance of other effects.Combined quantum mechanical/molecular mechanical (QM/MM) methods have become an important tool for computational modeling of enzyme-catalyzed reactions. Only the substrate(s) and relevant residues in the active site are treated quantum mechanically, the rest of the protein is described at the empirical MM level. This lowers the computational expense, making it possible to treat large enzyme systems and the surrounding solvent, and to sample phase space. Nevertheless, the number of QM atoms is still relatively large, and until now only low levels of QM theory, such as semiempirical methods or density functional theory (DFT), have been feasible. Semiempirical methods, though applicable to large systems, are generally not accurate enough because computed free energies of activation may have an error of ten or more kcal mol À1 . DFT (especially with the B3LYP functional [5] ) offers improved accuracy but still lacks key physical interactions (e.g., dispersion). Often, DFT underestimates barrier heights by several kcal mol À1 , which cannot be systematically improved. Thus, when theoretical barriers do not agree with those from experiment, it is not clear whether the discrepancy arises from deficiencies in the electronic structure theory and the sampling, in the experimental observations, or in the underlying theoretical framework of QM/MM and TST.Consequently, there is a need for high-level electronic structure calculations for reliable predictions of enzyme reactivity. The ab initio electron correlation methods MP2 (Møller-Plesset second-order perturbation theory), CCSD (coupled-cluster theory with single and double excitations), and CCSD(T) (CCSD with a perturbative treatment of triple excitations) provide a well-established hierarchy that converge reliably to give high accuracy, and rate constants of gasphase reactions involving only a few atoms can be predicted with error bars comparable to those found experimentally. [6,7] However, the computational expense of these methods increases very rapidly with the number o...