Patterns of Selection Amongst Dengue Virus Serotypes and Efficacy of Computational Epitope Prediction Applications

Dengue virus (DENV) is a mosquito-borne virus of global concern that leads to infections with symptoms ranging from high fevers and headaches to death. Current estimates place almost half the world’s population living in areas that are at-risk for infection by dengue and that area continues to expand over time. There are four distinct serotypes of dengue (DENV1, DENV2, DENV3, and DENV4). Disease severity from dengue infection is significantly worse if there is a secondary infection by a heterotypic serotype. The problems of increased severity with secondary infection as well as host immune response evasion by the virus itself have made vaccine development especially difficult. Analysis of complete polyprotein sequences of all four DENV serotypes revealed that all serotypes are currently subject to ongoing purifying selection and have experienced past purifying selection. However, several noteworthy exceptions within some genes of varying serotypes suggest conflicting evolutionary pressures. Most notably, the envelope (E) gene of DENV4 showed nucleotide polymorphism patterns suggestive of positive selection or relaxed purifying selection in its epitope regions. The E protein interacts directly with the host cell and is necessary for viral entry suggesting that its polymorphism pattern is indicative of evasion from immune system recognition. Both the NS1 gene of DENV2 and NS3 gene of DENV3, which are critical in viral replication, exhibited significantly higher medians of nonsynonymous gene diversity in their epitope regions implying escape mutations in these proteins as well. Significant conflicting ratios of πN (mean nonsynonymous nucleotide diversity) to πS (mean synonymous nucleotide diversity) between pairs of serotypes indicate that each individual serotype is under conflicting evolutionary pressures and help to explain the severity of secondary heterotypic infections. A comparison of three epitope prediction applications using datasets of both DENV genomes coupled with known cytotoxic T lymphocytes (CTL) DENV epitopes revealed significant differences in RANK and sensitivity measures of the programs. As these programs are used to predict which epitopes should be further studied in the creation of subunit vaccines, it was apparent that applications containing a combination of several prediction methods in concert are much more efficient than programs involving only one approach