Hybrid Quantum Mechanical and Molecular Mechanical Studies of the Reaction Mechanism of Dihydrofolate Reductase

             
                             

The enzyme Dihydrofolate Reductase (DHFR)
catalyzes the reduction of dihydrofolate to
tetrahydrofolate in the presence of NADPH co-factor. Tetrahydrofolate is important in the formation of DNA and thus DHFR is an attractive target for developing cytotoxic drugs to treat proliferative diseases such as cancer.

As a first step, prior to a full QM/MM study of details of the mechanism, we have applied ab initio electronic structure techniques to determine the importance of the bulk enzyme environment in polarisation of the substrate and co-factor.

These calculations have used very large QM models and point-charge representations of the bulk enzyme. The effect of a number of different aspects of the model have been considered, including different theoretical methods, basis set size, different charge representations of the enzyme and different enzyme structures.

               
                 
                 
                 

 

Principal Investigator

Jill Gready

John Curtin School of Medical Research

Co-Investigators

Stephen Greatbanks

Peter Cummins

John Curtin School of Medical Research

Alistair Rendell

ANU Supercomputer Facility

Projects

w05 - VPP, PC

                 
                         

                 
                         

What are the results to date and the future of the work?

The calculations performed so far have used Hartree-Fock, DFT, MP2 and linear-scaling semiempirical methods, as implemented in the Gaussian and Mopac98 codes. The Hartree-Fock and MP2 methods, where the enzyme is represented as point charges, have shown that the enzyme environment provides only a small degree of polarisation to the substrate and co-factor (measured in terms of changes in atom-centered charges and electron density). This behaviour is very different from the polarization calculated using DFT.

All the DFT methods considered result in a substantial over-polarization of the ligands. Both dihydrofolate (and the alternative substrate, folate) and the NADPH cofactor are substantially polarized on binding to the enzyme. This anomalous polarization arises as a result of the inability of DFT methods to correctly model the gas-phase structure of the anionic ligands, a finding which is supported by advances reported recently in the literature.

The linear-scaling semiempirical calculations, among the first of their type to include the entire enzyme system and ligands explicitly, show results similar to the Hartree-Fock calculations.

We intend to use QM/MM methods to investigate aspects of the mechanism of the reaction, and particularly the hydride-ion transfer step. We intend to use DHFRs from different organisms to determine the features which allow different DHFRs to perform the same biological function.

           
                 
                             
Appendix A -

 

 

 

 

 

 

 

 

Appendix A -
             

       

What computational techniques are used?

The project has made extensive use of the Gaussian 94 program suite, requiring modifications to the code to increase the degree of vectorisation in link 502 (scf energy evaluation) to allow for calculations on these very large systems, near the limit of traditional Hartree-Fock techniques. The largest calculations are of ~1850 basis functions. Using the modified link 502, speedups for the systems considered are more than a factor of 3. These modifications have subsequently been incorporated by Gaussian inc. into Gaussian98.

Other modifications have been made to link 502 to allow for "variable level-shifting" of the virtual orbitals. This was to overcome extreme oscillatory behaviour in the convergence of the DFT calculations. Modifications were also made to link 9999 to provide a selective checkpointing facility to reduce substantially the size of the restart files. The calculations have required both large amounts of memory and substantial disk space. The modified versions of the code described above achieve high ~80% vectorization.

       
- Appendix A