Analysis of V3 loop sequences using various bioinformatic tools designed for genotypic HIV-1 tropism testing

Abstract

Interest in genotypic tropism prediction using bioinformatic tools trained on phenotype-genotype data has recently been rekindled by introduction of coreceptor inhibitors into clinical practice. In this study it was aimed to compare predictions obtained by six algorithms and investigate concordance of prediction. Fifty V3 loop sequences were analysed with geno2pheno (g2p) used as reference and PSSM_sinsi, SVM, charge rule, PART and C4.5 tools. For every patient sample Δ32 CCR5 genotyping and subtyping basing on partial pol sequence was also performed. Phylogenetic tree of V3 sequences was reconstructed by maximum likelihood with example 3D protein structure visualized. Subtype B was noted in 72.0% while CCR5 Δ32 allele in 25.0% of cases. X4 tropism was found in 30.0% of patients by g2p with 10% false positivity rate (FPR), 8% – PSSM_sinsi, 32% – SVM, 12% – C4.5, 18% – PART and 20% by charge rule. Mean 76.2% concordance among the tools for X4 tropism prediction was noted with g2p FPR < 5%, while for R5 prediction 92,6% concordance was obtained with FPR > 20%. In phylogeny no clustering between the X4 and R5 sequences was observed except for one taxonomical unit. Genotype V3 predictions are concordant for geno2pheno 0%-5% and 20-100% FPR while g2p based interpretation of sequences with false positive rate of 5-20% or derived from non-B subtype should be cautious. In data interpreted with g2p 10% FPR neither X4 nor R5 tropism was associated with presence of single Δ32 CCR5 allele or HIV-1 non-B variant and no significant clustering of sequences with the same tropism was noted.