HIV-1 tropism determination using a phenotypic Env recombinant viral assay highlights overestimation of CXCR4-usage by genotypic prediction algorithms for CRF01_AE and CRF02_AG [corrected].

May 15, 2013 By:
  • Mulinge M
  • Lemaire M
  • Servais JY
  • Rybicki A
  • Struck D
  • da Silva ES
  • Verhofstede C
  • Lie Y
  • Seguin-Devaux C
  • Schmit JC
  • Bercoff DP.

BACKGROUND: Human Immunodeficiency virus type-1 (HIV) entry into target cells involves binding of the viral envelope (Env) to CD4 and a coreceptor, mainly CCR5 or CXCR4. The only currently licensed HIV entry inhibitor, maraviroc, targets CCR5, and the presence of CXCX4-using strains must be excluded prior to treatment. Co-receptor usage can be assessed by phenotypic assays or through genotypic prediction. Here we compared the performance of a phenotypic Env-Recombinant Viral Assay (RVA) to the two most widely used genotypic prediction algorithms, Geno2Pheno[coreceptor] and webPSSM. METHODS: Co-receptor tropism of samples from 73 subtype B and 219 non-B infections was measured phenotypically using a luciferase-tagged, NL4-3-based, RVA targeting Env. In parallel, tropism was inferred genotypically from the corresponding V3-loop sequences using Geno2Pheno[coreceptor] (5-20% FPR) and webPSSM-R5X4. For discordant samples, phenotypic outcome was retested using co-receptor antagonists or the validated Trofile(R) Enhanced-Sensitivity-Tropism-Assay. RESULTS: The lower detection limit of the RVA was 2.5% and 5% for X4 and R5 minority variants respectively. A phenotype/genotype result was obtained for 210 samples. Overall, concordance of phenotypic results with Geno2Pheno[coreceptor] was 85.2% and concordance with webPSSM was 79.5%. For subtype B, concordance with Geno2pheno[coreceptor] was 94.4% and concordance with webPSSM was 79.6%. High concordance of genotypic tools with phenotypic outcome was seen for subtype C (90% for both tools). Main discordances involved CRF01_AE and CRF02_AG for both algorithms (CRF01_AE: 35.9% discordances with Geno2Pheno[coreceptor] and 28.2% with webPSSM; CRF02_AG: 20.7% for both algorithms). Genotypic prediction overestimated CXCR4-usage for both CRFs. For webPSSM, 40% discordance was observed for subtype A. CONCLUSIONS: Phenotypic assays remain the most accurate for most non-B subtypes and new subtype-specific rules should be developed for non-B subtypes, as research studies more and more draw conclusions from genotypically-inferred tropism, and to avoid unnecessarily precluding patients with limited treatment options from receiving maraviroc or other entry inhibitors.

2013 May. PLoS One.8(5):e60566.
Other information