Discordance in HIV-1 co-receptor use prediction by different genotypic algorithms and phenotype assay: intermediate profile in relation to concordant predictions.

Concordant and discordant genotypic predictions of HIV-1 co-receptor tropism were analyzed. V3 region was sequenced from plasma samples of patients screened for R5 tropism by the Trofile® assay, before CCR5 antagonist prescription. Ten tools including geno2pheno, PSSM, an "11/25" and "net charge" rule, and other published algorithms were used. Patients were grouped according to concordance or discordance between tools and Trofile® result. Trofile® tropism reports from 50 patient samples were R5 in 38 and Dual/Mixed (DM) in 12. Prediction with the genotypic tools were concordant for 23 R5 samples, and discordant for the 15 other ones. From Trofile® DM strains were concordant in 6 and discordant in 6. V3 sequences were not clearly distinct between R5 and DM strains, except a greater diversity in the later. Discordances were found with any tool or combination of them, so that no one can be proposed as better than the others. Predictive values of each algorithm were similar and rather good (efficacy ranged from 74% to 84%), but the rate of non-confirmed prediction is greater when compelling the results of all tools with each individual sample. The mean of quantitative values obtained with one tool when another tool give the opposite prediction were different from those obtained when all tools agree with that prediction. The two discordant groups were often not distinguishable from each other. These results suggest that viruses giving discordant prediction with bioinformatic tools could be functionally distinct and/or in a different evolutionary state compared to those with concordant prediction.