Comparison of enzyme polarization of ligands and charge-transfer effects for dihydrofolate reductase using point-charge embeddedab initio quantum mechanical and linear-scaling semiempirical quantum mechanical methods

Abstract

Using quantum mechanical (QM) methods, we investigated the dependence of a number of factors on the polarization by the enzyme dihydrofolate reductase (DHFR) of its ligands—the substrates, folate and dihydrofolate, and the cofactor NADPH—and evaluated the implications for facilitation of the enzymic reductions. Two quite different levels of QM description of the biomolecular system were used. State-of-the-art ab initio QM calculations of the ligands were performed with the bulk DHFR environment modeled using atom-centered point charges. At the other extreme, semiempirical AM1 QM calculations using the linear-scaling Mozyme formalism incorporated in MOPAC2000 allowed for consistent treatment of the 3000-atom system of both enzyme and bound ligands. The study considered the effects of a number of factors on the polarization, including: (i) different levels of ab initio QM treatment (HF, MP2, DFT) and basis sets; (ii) different sets of molecular mechanics (MM) point charges in representing the bulk enzyme; (iii) inclusion of the bulk enzyme environment as either point charges in the ab initio calculations, or explicitly in the semiempirical calculations; (iv) ab initio QM calculations of substrate and ligand together (combined system) or separately (noncombined system); (v) degree of charge transfer between substrate and cofactor, and, for the semiempirical calculations, between bound ligands and enzyme; (vi) polarization of the enzyme in the semiempirical calculations; (vii) differences in the behavior of folate and dihydrofolate; and (viii) DHFRs from different species (E. coli and human) and different X-ray structure coordinate sets from the same species. Polarization was analyzed mainly by differences in point-charge distributions between gas-phase and bound ligands at the level of complete ligands, subcomponents of ligands (residues), and individual atoms in the pterin and nicotinamide rings involved in the DHFR reactions, but some electron density differences were also calculated. Consistent with our preliminary study (Greatbanks et al., Proteins 1999, 37, 157), and earlier work by Bajorath et al. (Proteins 1991, 9, 217; 11, 263) for noncombined ligand systems, the DFT calculations showed an unrealistically large dipolar character for individual ligands compared with HF and MP2 results and anomalously large charge transfer to folate from NADPH in combined calculations, which were not shown by the HF or AM1 results. The origin of this behavior is in the representation of the gas-phase anions (the substrates are dianionic and NADPH is tetraanioinic), with the point-charge enzyme-embedded calculations showing polarization similar to the HF results. The analysis highlights that successful modeling of the polarization properties depends on accurate representation of both the gas-phase and enzyme-bound electronic structures of the QM region. For both folate and dihydrofolate at the HF and MP2 levels, changes in density from enzyme binding in the region of the reducible bonds (N8C7 for folate, C6N5 for dihydrofolate) is small, with the bulk of the polarization taking place in the N2C2N3 region near the Asp27 (or Glu30) active-site group. Polarization at C7 for folate and C6 for dihydrofolate is negative (i.e., does not favor hydride-ion transfer), whereas the trend at N8 for folate, but not at N5 for dihydrofolate, favors protonation. For the Mozyme results, the substrate pterin-ring polarization trends are similar, and also with negligible charge transfer to NADPH, and only a very small charge transfer to the enzyme. For NADPH, the HF, MP2, and Mozyme results indicate charge polarization on binding to the enzyme at the active carbon (C4) of the nicotinamide ring is favorable for hydride-ion transfer (i.e., slightly more negative), with the active hydrogen (H4) also being more negative for the HF and MP2 results. For all methodologies, the nonactive hydrogen (H4′) becomes significantly more positive, which would reduce its potential for transfer. The Mozyme results show a net loss to the enzyme of ∼0.3 electrons, mostly from NADPH, which is strongly localized in the vicinity of the substrate glutamate and NADPH diphosphate and 3′-phosphate groups. © 2000 John Wiley & Sons, Inc. J Comput Chem 21: 788–811, 2000