A Multivariate Statistics Approach to Analyze the Transition from CCR5 to Dual and CXCR4 Tropism and to Predict the Coreceptor Usage of HIV-1.

Abstract

Background: The tropism of HIV-1 is determined by the use of either or both chemokine coreceptors CCR5 (R5) and CXCR4 (X4) for entry into the target cell. The coreceptor usage of HIV-1 is mainly determined by the third variable loop (V3) of the envelope glycoprotein gp120.

Materials and methods: We assembled a training dataset of 3071 V3 sequences with known tropism: 2428 R5-tropic, 358 dual R5X4-tropic, and 285 X4-tropic. We analyzed this dataset by using univariate statistics (t-Student, chi-square contingency table, and Cramer's V index), bivariate statistics (Pearson's correlation coefficient), and multivariate statistics (principal component analysis and Fisher's linear discriminant analysis).

Results: One third of the 35 amino acid positions in the V3 loop are key drivers of transition from R5 to R5X4 and X4 tropism. Principal component analysis (PCA) indicates that the switch between the three tropism states is a gradual process involving variation for several biochemical features. Linear discriminant analysis (LDA) is a good predictor of coreceptor usage. When applied to the training dataset, it showed a high sensitivity (92% of X4-tropic sequences predicted correctly) and a high specificity (90% of R5-tropic sequences predicted correctly).

Discussion: The finding that the sensitivity and specificity of LDA are similar, or better than that of four widely used prediction methods (Geno2pheno, WebPSSM, 11/24/25 charge rule, and net charge rule) indicates that LDA is a valuable complement to the existing models for prediction of coreceptor usage.

Conclusion: Parametric multivariate statistics is a useful tool to investigate the switch of HIV-1 tropism and to predict the CCR5 or CXCR4 coreceptor usage.