StaRProtein, a web server for prediction of the stability of repeat proteins

Abstract

Repeat proteins have become increasingly important due to their capability to bind to almost any proteins and the potential as alternative therapy to monoclonal antibodies. In the past decade repeat proteins have been designed to mediate specific protein-protein interactions. The tetratricopeptide and ankyrin repeat proteins are two classes of helical repeat proteins that form different binding pockets to accommodate various partners. It is important to understand the factors that define folding and stability of repeat proteins in order to prioritize the most stable designed repeat proteins to further explore their potential binding affinities. Here we developed distance-dependant statistical potentials using two classes of alpha-helical repeat proteins, tetratricopeptide and ankyrin repeat proteins respectively, and evaluated their efficiency in predicting the stability of repeat proteins. We demonstrated that the repeat-specific statistical potentials based on these two classes of repeat proteins showed paramount accuracy compared with non-specific statistical potentials in: 1) discriminate correct vs. incorrect models 2) rank the stability of designed repeat proteins. In particular, the statistical scores correlate closely with the equilibrium unfolding free energies of repeat proteins and therefore would serve as a novel tool in quickly prioritizing the designed repeat proteins with high stability. StaRProtein web server was developed for predicting the stability of repeat proteins.

Fig 1. Evaluating the stability of general proteins using distance-dependant statistical potential based on general protein library.

RAPDF (general) represents the statistical RAPDF scores calculated using the general protein database [17]. RAPDF (Composite) represents the statistical RAPDF scores calculated using the composite protein database composed of α-, β- and α+β proteins (2566 proteins).

Fig 2. Evaluating the stability of α+β proteins using distance-dependant statistical potential based on α+β protein library (1271 proteins).

RAPDF (α+β) represents the statistical RAPDF scores calculated using the α- and β- databases. First 16 sets were single misfold decoy sets and the rest 4 sets were from multiple decoy sets with a representative decoy selected.

Fig 3. Evaluating the stability of α proteins using distance-dependant statistical potential based on α protein library (1007 proteins).

RAPDF (α) represents the statistical RAPDF scores calculated using the α- database. First 6 sets were single misfold decoy sets and the rest 14 sets were from multiple decoy sets with a representative decoy selected.

Fig 4. Evaluating the stability of β proteins using distance-dependant statistical potential based on β protein library (288 proteins).

RAPDF (β) represents the statistical RAPDF scores calculated using the β- database. First 4 sets were single misfold decoy sets and the rest 16 sets were from multiple decoy sets with a representative decoy selected.

Fig 5. PRIDE2 structure comparison of non-redundant repeat proteins (Drawtree).

The repeat proteins are divided into branches, which are shown as groups (A) AR (B) TPR.

Fig 6. Distance-dependant statistical potential based on ankyrin repeat protein library (33 proteins).

Homology models were used as decoys. RAPDF (Ankyrin) represents the statistical RAPDF scores calculated using the Ankyrin database.

Fig 7. Distance-dependant statistical potential based on TPR protein library (73 proteins).

Homology models were used as decoys. RAPDF (TPR) represents the statistical RAPDF scores calculated using the TPR database.

Fig 8. Predicted stability of designed repeat proteins using distance-dependant statistical potential based on TPR (light blue) or AR (blue) protein libraries.

Fig 9. Correlation between the RAPDF scores of CTPRan and the equilibrium unfolding free energies.

(A) RAPDF scores versus ΔG_D-N(kcal/mol), the thermal unfolding free energies (B) RAPDF scores versus ΔG_0-j(kcal/mol), the folding free energies calculated from fitting the Ising model.