Background:  Both HIV dual infection (DI) and infection with X4-tropic virus are associated with accelerated disease progression and increased viral load, but viral tropism during DI remains unclear. Using ultra-deep sequencing (UDS) to screen for DI in a large cohort, and computational tools to predict tropism, we estimated the frequency of X4-tropism and determined whether tropism changes with DI.

Methods:  A subset of ART-naïve patients from the San Diego Primary Infection Cohort was screened for intraclade B DI by UDS of 3 coding regions on a 454 sequencer. Samples were defined as DI based on the presence of 2 phylogenetically distinct HIV populations in a plasma sample. Superinfection (SI) was defined when the baseline viral population was augmented or replaced by a divergent population. Geno2pheno was used to predict the proportion (D) of X4-tropic reads in each env sample with a complete V3 loop, N = 109 (false positive rate ≤10%) with D >1% classified as X4. A subset underwent phenotypic testing (0.3% limit of detection, Trofile ES Assay, Monogram Biosciences). We repeated UDS on a subset of samples to evaluate reproducibility.

Results:  Fifty-one of one hundred and nine (47%) patients had a subpopulation of X4-tropic virus (mean D = 25%, IQR 35.5%). We found no significant association between X4 tropism, estimated duration of infection (mean 1.3 years, range 0.08 to 6.3), or DI at time of sampling. In a longitudinal subset (N = 20), 7 patients (35%) had X4 at baseline, 4 (20%) remained X4 throughout over 9.4 person-years of follow up, 3 SI patients (15%) lost X4 over 9.3 person-years of follow up, and 4 (20%) developed X4 over 31 person-years, 3 of which were SI. In 11 subjects with phenotypic and genotypic data, results were concordant in 9 (82%) cases. In 36 patients with replicated UDS, X4-tropism determination was consistent in 33/36 (92%) cases.

Conclusions:  In the San Diego Primary Infection Cohort, we detected high (47%) prevalence of X4 tropism. Tropism change over time was rare and was associated with SI. Computational predictions from UDS data were reproducible and in good agreement with phenotypic measurements.