Appraising the performance of genotyping tools in the prediction of coreceptor tropism in HIV-1 subtype C viruses

Abstract

Background: In human immunodeficiency virus type 1 (HIV-1) infection, transmitted viruses generally use the CCR5 chemokine receptor as a coreceptor for host cell entry. In more than 50% of subtype B infections, a switch in coreceptor tropism from CCR5- to CXCR4-use occurs during disease progression. Phenotypic or genotypic approaches can be used to test for the presence of CXCR4-using viral variants in an individual's viral population that would result in resistance to treatment with CCR5-antagonists. While genotyping approaches for coreceptor-tropism prediction in subtype B are well established and verified, they are less so for subtype C.

Methods: Here, using a dataset comprising V3 loop sequences from 349 CCR5-using and 56 CXCR4-using HIV-1 subtype C viruses we perform a comparative analysis of the predictive ability of 11 genotypic algorithms in their prediction of coreceptor tropism in subtype C. We calculate the sensitivity and specificity of each of the approaches as well as determining their overall accuracy. By separating the CXCR4-using viruses into CXCR4-exclusive (25 sequences) and dual-tropic (31 sequences) we evaluate the effect of the possible conflicting signal from dual-tropic viruses on the ability of a of the approaches to correctly predict coreceptor phenotype.

Results: We determined that geno2pheno with a false positive rate of 5% is the best approach for predicting CXCR4-usage in subtype C sequences with an accuracy of 94% (89% sensitivity and 99% specificity). Contrary to what has been reported for subtype B, the optimal approaches for prediction of CXCR4-usage in sequence from viruses that use CXCR4 exclusively, also perform best at predicting CXCR4-use in dual-tropic viral variants.

Conclusions: The accuracy of genotyping approaches at correctly predicting the coreceptor usage of V3 sequences from subtype C viruses is very high. We suggest that genotyping approaches can be used to test for coreceptor tropism in HIV-1 group M subtype C with a high degree of confidence that they will identify CXCR4-usage in both CXCR4-exclusive and dual tropic variants.

Figure 1

Performance of each of the genotyping algorithms in predicting CXCR4-usage. Sensitivity for both the CXCR4-using and CXCR4-exclusive datasets was calculated as the number of viral sequences predicted as CXCR4-using divided by the total number of CXCR4-using or CXCR4-exclusive sequences tested. Specificity corresponds to the number of CCR5-using viruses predicted as R5 divided by the total number of CCR5-using viral sequences evaluated.

Figure 2

Ability of each approach at predicting CXCR4-usage in dual-tropic viral sequences. The percentage of dual-tropic sequences predicted as CCR5-using and CXCR4-using is shown with dark and light shaded areas of each bar corresponding to the percentage of sequences predicted as CCR5-using and CXCR4-using respectively.