Determination of molecular descriptors from analysis of sequence and genome data

---

With the recent developments of new fast and more advanced techniques complete genomes have now been fully sequenced and even more is expected to appear in the near future. The requirement of computational tools to investigate sequences from these genomes without determination of experimental structures and time consuming biochemical characterizations has thus never been more prominent.

The project aims at building bioinformatics competence and tools for identification, exploration and development of macromolecules of industrial and biotechnological importance. This will be achieved by developing methods for identification of molecular descriptors for a limited number of protein functional and biophysical properties.

The project focuses on a series of homologous proteins that have been identified from available sequence databases. Using these proteinases as model systems, structural characteristics responsible for certain functional or biophysical properties (stability and specificity) will be studied in order to come up with molecular and structural descriptors.

Based on sequence alignments three different proteinases were selected to be further studied in terms of structural features responsible for their specificity and stability. The proteinases used are; subtilisin Carlsberg, proteinase K and thermitase.

The specificity of these enzymes has been studied by free energy calculations using molecular dynamics (MD) simulations in combination with the linear interaction energy (LIE) method. For subtilisin Carlsberg and thermitase crystal structures of complexes with the protein inhibitor eglin C, pdb entry 1CSE and 2TEC respectively, were used as starting points to study the effect of single point mutations on the association energies. In the case of proteinase K, no experimental structure was available and docking of eglin C were therefore done using subtilisin Carlseberg as a template due to the high sequence identity (~60 %). Steric clashes at the protein-protein interface were manually removed by adjusting the side chains. 20 different complexes were then generated for each of the proteinases where the only difference is the amino acid at the primary binding (P1) position. Experimental data is available for binding of several P1 variants of eglin C to subtilisin Carlsberg, and our LIE calculations agrees quite well with this data set. Furthermore, the LIE calculations clearly show that the specificity profiles of all three enzymes investigated are very broad, confirming that they can accommodate substrates with different amino acids at the P1 position.

Another important aspect has been to reduce the scaffold of the protein inhibitors to consist of only the residues of that constitute the proteinase binding loop. This will result in less CPU-demanding calculations as well as simply build of protein-protein models in cases where experimental structures are missing. Our calculations shows that the free energies of binding and the resulting specificity profiles for the reduced models are virtually identical to those obtained with the complete inhibitor scaffold.