Improving the quality of protein crystals using rational design of crystal contacts
Final Report Abstract
In the course of the project we created a publicly available crystal contacts database holding information relevant structure parameters, such as solvent accessibility, secondary structure as well as biological and crystallographic interfaces. For each amino acid residue its involvement in crystal contacts was calculated. This information was utilized to develop a machine learning procedure, which allows to distinguish good quality from poor quality crystals. In a next step we proposed a novel algorithm to predict the location of crystal contacts from primary sequence of proteins and to suggest mutations that improve crystal quality and hence facilitate structure determination. In parallel, we also developed a novel and very competitive method for predicting solvent accessibility of proteins from sequence alone. Finally, amino acid sequence substitutions were tested experimentally for their impact on protein crystal quality. Altogether, nine protein-encoding genes were cloned in a bacterial system and expressed end subjected to mutagenesis. These genes encoded the following proteins: DpnI restriction endonuclease from “Streptococcus pneumonia”, NlaIV restriction endonuclease from “Neisseria lactamica”, NgoPII restriction endonuclease from “Neisseria gonorrhoeae”, Mini-III RNase from “Fusobacterium nucleatum”, Mini-III RNase from “Bacillus subtilis”, XseAB DNA repair enzyme from “Escherichia coli”, HIV virus reverse transcriptase, RuvC from “Thermus thermophiles”, and human RNase H2. Six of these selected target proteins were crystallized as a wild-type. 41 protein variants were tested for potentially improved properties of the crystallization. Unfortunately no significant improvement in crystallizability or crystal quality was observed.
