The utility of different bioinformatics algorithms for genotypic HIV-1 tropism testing in a large clinical cohort with multiple subtypes

Objectives: HIV-1 tropism needs to be determined before the use of CCR5 antagonist drugs such as maraviroc (MVC), which are ineffective against CXCR4-using HIV-1. This study assessed how different computational methods for predicting tropism from HIV sequence data performed in a large clinical cohort. The value of adding clinical data to these algorithms was also investigated. Design and methods: PCR amplification and sequence analysis of the HIV-1 gp120 V3 loop region was performed on triple replicates of plasma viral RNA or proviral DNA extracted from peripheral blood monocytes (PBMCs) in 242 patients. Coreceptor usage was predicted from V3 sequences using seven bioinformatics interpretation algorithms, combined with clinical data where appropriate. An intention-to-treat approach was employed for exploring outcomes and performance for different viral subtypes was examined. Results: The frequency of R5 predictions varied by 22.6%, with all seven algorithms agreeing for only 75.3% of tests. The identification of individuals likely to fail was poor for all algorithms. The addition of clinical data improved this, but at the expense of their ability to predict success. The clinical algorithms varied across subtypes, whereas other algorithms were more consistent. Furthermore, individuals with discordant clonal and clinical predictions were more likely to fail MVC treatment. Conclusion: Eligibility for MVC varied depending on the algorithm method used. The addition of clinical parameters alongside sequence data may help predict X4 emergence during treatment. It could be that V3 loop analysis in isolation may not be the best method for selecting individuals for MVC