A complete map of potential pathogenicity markers of avian influenza virus subtype H5 predicted from 11 expressed proteins

Abstract

Background: Polybasic cleavage sites of the hemagglutinin (HA) proteins are considered to be the most important determinants indicating virulence of the avian influenza viruses (AIV). However, evidence is accumulating that these sites alone are not sufficient to establish high pathogenicity. There need to exist other sites located on the HA protein outside the cleavage site or on the other proteins expressed by AIV that contribute to the pathogenicity.

Results: We employed rule-based computational modeling to construct a map, with high statistical significance, of amino acid (AA) residues associated to pathogenicity in 11 proteins of the H5 type viruses. We found potential markers of pathogenicity in all of the 11 proteins expressed by the H5 type of AIV. AA mutations S-43(HA1)-D, D-83(HA1)-A in HA; S-269-D, E-41-H in NA; S-48-N, K-212-N in NS1; V-166-A in M1; G-14-E in M2; K-77-R, S-377-N in NP; and Q-48-P in PB1-F2 were identified as having a potential to shift the pathogenicity from low to high. Our results suggest that the low pathogenicity is common to most of the subtypes of the H5 AIV while the high pathogenicity is specific to each subtype. The models were developed using public data and validated on new, unseen sequences.

Conclusions: Our models explicitly define a viral genetic background required for the virus to be highly pathogenic and thus confirm the hypothesis of the presence of pathogenicity markers beyond the cleavage site.

Fig. 1

Schematic representation of the applied computational modeling methodology

Fig. 2

Accuracies of the cross-validation and the testing of the models on new, unseen data. a Quality measures for the rule-based models. Averaged Accuracy is the average of mean accuracy from the 10-fold cross-validation loop for the models created on 100 under-sampled subsets for each protein. Standard deviation from the 10-fold cross validation loop, averaged in a similar way as accuracy, is shown as error bars on the plot. b Re-classification of the training sequences of the H5N1 sequences. Accuracy is the percentage of correctly classified sequences. See also Additional file 4: Table S2. c Re-classification of the training sequences of the non-H5N1 sequences. Accuracy is the percentage of correctly classified sequences. See also Additional file 4: Table S3. d Accuracies of the classifiers when tested on the newly published unseen H5N1 sequences, i.e. sequences not included in the training of the models and with sequences identical to the training sequences removed. Accuracy is the percentage of correctly classified sequences. Classifiers consisted of the significant rules from all the rule-based models created for a given protein. See also Additional file 5: Table S4. e Accuracies of the classifiers when tested on the newly published unseen non-H5N1 sequences, i.e. sequences not included in the training of the models and with sequences identical to the training sequences removed. Accuracy is the percentage of correctly classified sequences. Classifiers consisted of the significant rules from all the rule-based models created for a given protein. See also Additional file 5: Table S5

Fig. 3

AA’s appearing in the most significant rules marked on the 3D structures of different proteins. AA residues appearing in the rules are shown as spheres. Positions from the high pathogenicity rules are shown in blue, positions from the low pathogenicity rules are in magenta and mutations associated with the shift of pathogenicity from low to high as defined by the rules are shown in red. a Mapping of amino acid positions associated with pathogenicity from the rules onto 3D structure of the HA protein of Influenza A virus (A/Hubei/1/2010 (H5N1)) (PDB: 4KTH). Chain A (HA1 residues) and chain B (HA2 residues) are presented in green, while the rest of the trimer is shown in gray. b A cartoon representation of chains A, B, C and D of the NA protein with AA positions from the rules (PDBID: 2HU4). Chain A, the one marked with rule positions, is shown in green and the others in gray. Residue R-371, shown as a sphere in orange, is a part of the catalytic site of the protein. Cyan spheres constitute Oseltamivir 2, a substrate bound to the protein. c A cartoon representation of the NP protein trimer (PDBID: 2IQH) with positions from the rules. Chain A is shown in green and the others are in gray. d AA’s from the rules marked on a cartoon representation of NS1 (PDBID: 3FST). e A cartoon representation of the PB2 protein cap-binding domain (PDBID: 4CB4) with AA’s from the rules