Background: The majority of HIV-1 strains in brain use CCR5 for infection of macrophages and microglia, but dual-tropic (R5X4) strains have also been detected in brain and CSF of some patients with HIV-associated dementia (HAD). Bioinformatic algorithms based on viral genotype are a quick and easy method to predict co-receptor usage, but database sets are heavily dependent on CCR5-tropic sequences and may therefore underestimate CXCR4- or R5X4-tropism. Here, we cloned HIV-1 envelopes (Env) from an AIDS patient with severe HAD and determined co-receptor usage. We then added these sequences to a larger data set of R5X4 Env sequences and compared the ability of available algorithms to reliably predict R5X4-usage.

Methods: Env were cloned from brain and spleen tissue of an AIDS patient with severe HAD. 10 Env (5 brain and 5 spleen) were screened for co-receptor usage. Functional Env were sequenced and added to a database set of R5X4 Env for a total of 24 R5X4 Env sequences (8 brain and 16 blood/lymphoid) from 7 patients. We compared the ability of 5 bioinformatic algorithms to accurately predict co-receptor usage based on V3 sequence.

Results: Of 10 Env clones, 9 (n = 5 brain and 4 spleen) from the HAD patient were R5X4-tropic in a cell-cell fusion assay. HIV-1 isolates from these tissues were also R5X4-tropic and used CXCR4 efficiently for entry into primary macrophages and microglia. Comparisons of bioinformatic algorithms to predict co-receptor usage found that only one program (SVM_geno2pheno) correctly predicted the ability of Env to use CXCR4 with >90% accuracy (n = 23/24 predicted to use CXCR4). Confidence levels for these predictions were typically low (average p = 0.3), but were increased to >90% by inputting clinical data (CCR5-genotype and CD4⁺ counts). Two programs (PSSM-SINSI and SVM_genomiac) correctly predicted CXCR4-usage with >75% accuracy (19 of 24 and 18 of 24, respectively). The PSSM-X4R5 matrix correctly predicted CXCR4-usage for only 5 of 24 Env. Finally, the commonly used "11/25" rule predicted X4-usage by only 4 of 24 Env. Analysis of amino acid sequences identified 2 positions in the V3 loop (19 and 32) that contributed to incorrect coreceptor usage predictions.

Conclusions:  R5X4 HIV-1 is associated with brain infection in some HAD patients, but the frequency of these strains is underestimated by most commonly used predictive algorithms. SVM_geno2pheno is the most accurate predictor of CXCR4-usage by R5X4 HIV-1 in brain and other tissues.