215    

Background:  The use of both CCR5 and CXCR4 as co-receptors during host cell entry in HIV-1 subtypes A, B, and D is well known. In these subtypes, there is a strong association between CXCR4 usage and accelerated disease progression. However, subtype C, responsible for 42% of global infections, rarely uses CXCR4. The ability to screen large C-infected cohorts for X4 viruses, and relate their presence to disease status, is vital to understanding this important difference. However, in vitro determination of co-receptor usage is not always feasible, especially in those developing countries where C predominates. A reliable phenotype prediction method, based on sequence, could provide for rapid and less expensive screening. Existing methods using B V3 loop data do not perform well when applied to other subtypes. We hypothesized that we could develop a reliable C-specific phenotype predictor using V3 sequences of C isolates of known phenotype.

Methods:  We derived predictors from position-specific scoring matrices (PSSM) based on a subtype C training set of 229 R5 V3 sequences (from 200 subjects) and 51 X4 V3 sequences (20 subjects). Specificity and sensitivity distributions were estimated by combining dataset bootstrapping with leave-one-out cross-validation. Non-independence of sequences was attenuated by randomly sampling single sequences from individuals on each bootstrap iteration. We also performed V3-based heteroduplex tracking assays (V3-HTA) using a known R5 as probe and a subset (n = 21) of the training set as targets. We compared mobilities to PSSM scores using linear regression.

Results:  The C-specific predictor (C-PSSM) had specificity 91% [95% CI 89% to 93%] and sensitivity 84% [CI 79% to 89%], based on leave-one-out bootstrap using all sequences. Analysis based on single sequences per individual gave comparable specificity (94% [92% to 96%]) and somewhat lower sensitivity (75% [68% to 82%]). These are similar to B-PSSM results (median specificity 93%; sensitivity 88%). The C-PSSM is significantly more sensitive than the B-PSSM applied to the C sequences (sensitivity 37%). V3-HTA mobilities correlated with C-PSSM score (p < 0.0001; r²: 0.56).

Conclusions:  We derived a C-specific phenotype predictor that performs nearly as well on C V3 loops as do existing B-specific methods on B V3 loops, and can thus be applied to characterize C V3 sequence of unknown co-receptor use. V3-HTA could provide an inexpensive method for scoring large numbers of subtype C V3 samples.

Keywords: Coreceptor usage; Subtype C; PSSM