Correlation between genotypic predictions based on V3 sequences and phenotypic determination of HIV-1 tropism

Objective: Replacing phenotypic assays with simple genotypic predictions of HIV-1 coreceptor usage would make the clinical use of CCR5 antagonists easier.

Design: Paired genotypic and phenotypic determination of HIV-1 coreceptor usage was performed to assess several genotypic approaches for detecting CXCR4-using and CCR5-using viruses in a clinical setting.

Methods: HIV-1 coreceptor usage was prospectively assessed using plasma samples from 103 patients who were candidates for treatment with a CCR5 antagonist. Direct sequencing of the V3 region and a sensitive recombinant virus phenotypic entry assay were performed in parallel for each patient from the same bulk env PCR product.

Results: The 103 patients had a median CD4+ T lymphocyte count of 268 × 10⁶ cells/l and nadirs of 98 × 10⁶ cells/l. Paired genotypic and phenotypic data were obtained for 98 of the 103 patients. For detecting CXCR4-using viruses, the genotypic rule based on amino-acid residues at positions 11/25 and the overall net charge of V3 was 77% sensitive and 96% specific. The Geno2pheno bioinformatic tool was 88% sensitive and 87% specific. The WebPSSM tool prediction with the SI/NSI matrix was 77% sensitive and 94% specific. The global concordance between genotypic and phenotypic data was 91% with the rule combining the amino-acid residues at positions 11/25 and V3 net charge.

Conclusion: Genotypic predictions performed well in paired genotypic and phenotypic assessment of HIV-1 coreceptor usage. Multicenter studies analyzing the correlations between the genotypic determination of HIV-1 tropism and clinical response to CCR5 antagonists are needed to validate this approach in clinical practice.

Introduction

Human immunodeficiency virus type 1 (HIV-1) enters CD4-expressing cells using one or both of the chemokine receptors CCR5 and CXCR4. CCR5-using virus clones are classified as R5 variants, CXCR4-using virus clones are classified as X4 variants, and virus clones that use both coreceptors are classified as R5X4 dualtropic variants ^[1]. The use of both coreceptors by a virus population may be due to the presence of either R5X4 dualtropic clones or to a mixture of pure R5 and X4 virus clones (dual/mixed R5X4 phenotype) or both.

The determination of HIV-1 tropism is now of therapeutic interest because the coreceptors are new targets for drugs that block HIV-1 entry. CCR5 antagonists are presently used in clinical practice but they can only be used to treat patients harboring R5 viruses ^[2]. Thus HIV-1 tropism must be assessed in patients who are candidates for CCR5 antagonists. Recombinant virus phenotypic entry assays are now considered to be the gold-standard method for determining HIV-1 coreceptor usage ^[3]. However, their routine use is hampered by technical and cost limitations. Genotypic approaches have great potential, as they could be easier, faster, and cheaper than phenotypic assays for routinely assessing HIV-1 tropism. However, good genotype-phenotype correlations are required before they can be used.

Most genotypic determinants of HIV-1 coreceptor usage have been identified in the V3 region of env ^[4]. The presence of basic residues at V3 positions either 11 or 25 or both and an increased electrostatic net charge of V3 have been associated with CXCR4 usage ^[5,6]. Bioinformatic algorithms have been developed to improve the prediction of HIV-1 tropism from the V3 sequence, taking into account the key amino-acids at positions 11 and 25, plus other sites in V3 that differ between CCR5-using and CXCR4-using strains ^[7].

It has been reported that genotypic approaches lack sensitivity for predicting CXCR4 usage, and are thus not likely to replace the phenotypic assays ^[8]. However, we found excellent correlations between the V3 genotype and the phenotype in a set of 118 molecular clones ^[9]. We also showed that genotypes and phenotypes of bulk samples correlated well with the clonal composition of the virus quasispecies. These encouraging results for the genotypic prediction of HIV-1 tropism are in apparent contradiction with the reported lack of sensitivity of the genotype. There is thus an urgent need to compare the performances of genotypic and phenotypic methods for the routine assessment of HIV-1 tropism.

We have therefore assessed HIV-1 coreceptor usage both genotypically and phenotypically using plasma samples from a new cohort of 103 patients who were candidates for treatment with a CCR5 antagonist. We used the matched genotypic V3 data and phenotypic results to study the performance of genotypic approaches for predicting HIV-1 tropism in this clinical setting.

Methods

Study subjects and samples

We prospectively studied 103 HIV-1 infected patients for whom therapy had failed, as defined by a plasma HIV-1 RNA of at least 400 copies/ml and the presence of drug-resistance mutations for the three classes of NRTIs, NNRTIs, and protease inhibitors ^[10]. These patients were recruited at the Department of Infectious Diseases of Toulouse University Hospital, France, and were distinct from the patients previously described ^[9]. HIV-1 RNA was isolated and purified from 1 ml plasma samples using the QIAamp Viral RNA Mini Kit (Qiagen).

Phenotypic characterization of HIV-1 coreceptor usage

PCR amplification of an HIV-1 env fragment encompassing the gp120 and the ectodomain of gp41

Reverse-transcription polymerase chain reaction (RT-PCR) and nested PCR were used to amplify the gp120 and the gp41 ectodomain. The RT-PCR was performed with the SuperScript III One-Step RT-PCR System (Invitrogen), with the following conditions: 60 min at 56°C; 2 min at 94°C; 30 s at 94°C, 30 s at 58°C, and 4 min at 68°C for 50 cycles. The nested PCR was performed with the Expand High Fidelity Plus PCR System (Roche), with the following conditions: 2 min at 94°C; 30 s at 94°C, 30 s at 60°C, and 3 min 30 s at 72°C for 35 cycles. The outer primer pair was 5′-GGCTTAGGCATCTCCTATGGCAGGAAGAAG-3′ and 5′-GGTCTTAAAGGTACCTGAGGTCTGACTGGA-3′. The inner primer pair was 5′-CCACCACTCTATTTTGTGCATCA-3′ and 5′-GGTGGTAGCTGAAGAGGCACAGG-3′ ^[11].

Vector construction

HIV-1 pNL4-3 (GenBank AF324493) was deleted between positions 6611 and 8257 to create the pNL43-Δenv vector. Briefly, pNL4-3 was deleted between the SalI (position 5785) and NotI (position 8796) restriction sites. A 5′ SalI-NheI fragment (positions 5785-6611) and a 3′ NheI-NotI fragment (positions 8257-8796) were obtained by PCR amplification and joined. The resulting SalI-NotI insert was cloned in the deleted pNL4-3 backbone to create the pNL43-Δenv vector. This vector contained a single NheI restriction site at position 6611 for linearization. The Firefly luciferase reporter gene was inserted into the nef gene of the pNL43-Δenv vector between the NotI (position 8796 of pNL4-3) and XhoI (position 8887 of pNL4-3) restriction sites to create the final pNL43-Δenv-Luc vector (Fig. 1a).

Fig. 1

Recombinant virus phenotypic assay

The phenotype of HIV-1 coreceptor usage was determined using a single-cycle recombinant virus entry assay. 293T cells were cotransfected using the calcium phosphate precipitation method with NheI-linearized pNL43-Δenv-Luc vector DNA and an env PCR product that encompassed the deleted region of the pNL43-Δenv-Luc vector, and carried short overlaps on each side that allow homologous recombination with the vector (Fig. 1b). The chimeric recombinant virus particles released into the supernatant were used to infect indicator cells bearing CD4 and CCR5 (U87.CD4.CCR5 cells) or CD4 and CXCR4 (U87.CD4.CXCR4 cells) (these reagents were obtained through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH, from Dr. HongKui Deng and Dr Dan R. Littman). The ability of chimeric recombinant viruses to enter and complete a single round of replication in target cells was assessed by measuring the luciferase activity in lyzed cells (as relative light units; RLU). The tested virus used either CCR5 or CXCR4 or both coreceptors if the luciferase activity was above a cutoff, defined as greater than 3 SD above the mean of the background of the indicator cell line. Recombinant viruses with known coreceptor usage (CXCR4-tropic HXB2 and CCR5-tropic BaL) were used as controls in all experiments. The specificity of low-level positive luminescent signals (defined as a RLU 1- to 10-fold the cut off value) was confirmed using coreceptor antagonists. JM-2987 (hydrobromide salt of AMD-3100) was used to block CXCR4-mediated entry and TAK-779 to block CCR5-mediated entry (these reagents were obtained through the NIH AIDS Research and Reference Reagent Program, Division of AIDS, NIAID, NIH) ^[12,13]. The inhibition studies used AMD3100 (0.1, 0.3, and 1 μmol/l) and TAK779 (0.3, 1, and 3 μmol/l). Inhibition by either drug was defined as a 10-fold decrease in RLU relative to the RLU measured in the absence of the drug. The ability of the assay to detect the presence of minor X4 variants in a virus population was assessed in three independent mixture experiments with X4/R5 DNA ratios of 0:100; 5:95; 10:90; 15:85; 20:80; 25:75; and 50:50.

Genotypic prediction of HIV-1 coreceptor usage

The V3 region was directly sequenced from bulk env PCR products in both directions by the dideoxy chain termination method (BigDye Terminator v.3.1) on an ABI 3130 DNA sequencer (Applied Biosystems). Two primer pairs were used for V3 sequencing. The first primer pair was 5′-ACAATGYACACATGGAATTARGCCA-3′ and 5′-AGAAAAATTCYCCTCYACAATTAAA-3′, and the alternative primer pair was 5′-CTGTTAAATGGCAGTCTAGC-3′ and 5′-GTGATGTATTRCARTAGAAAAATTC-3′. The results were analyzed with Sequencher (Genecodes), blinded to the phenotype. Minority species were detected when the sequencer electrophoregram showed a second base peak.

We used a combination of criteria from the 11/25 and net charge rules to predict HIV-1 tropism from the V3 genotype ^[5,6,9]. One of the following criteria is required for predicting CXCR4 coreceptor usage: either 11R/K or 25K or both; 25R and a net charge of at least +5; a net charge of at least +6. The V3 net charge was calculated by subtracting the number of negatively charged amino acids (D and E) from the number of positively charged ones (K and R). All possible permutations were assessed when amino-acid mixtures were found at some codons of V3. The combination resulting in the highest net charge was used to predict the tropism. We then compared the performances of these combined criteria for detecting CXCR4-using viruses with that of the 11/25 rule alone, and that of the bioinformatic tools Geno2pheno (false positive rate 10%) and WebPSSM (with both the X4/R5 and the SI/NSI matrices), available at: http://coreceptor.bioinf.mpi-sb.mpg.de/cgi-bin/coreceptor.pl and: http://ubik.microbiol.washington.edu/computing/pssm/ (December 2007), respectively.

Results

Patient characteristics

The HIV-1 coreceptor usage of 103 patients with persistent viremia despite HAART who harbored viruses with triple-class drug resistance mutations, were studied, as these patients could be candidates for treatment with CCR5 inhibitors. Clinical characteristics of the patients and virological and immunological data are shown in Table 1. Eighty-five percent of the patients harbored subtype B HIV-1. The remaining 15% patients harbored A1, F1, G, J, CRF-01, CRF-02, and CRF-06 subtypes of HIV-1.

Table 1

Phenotypic characterization of HIV-1 coreceptor usage

The sensitivity of our phenotypic assay has been assessed in mixtures experiments with various X4/R5 proportions. Minor X4 variants were always detectable when present at a frequency of 10% or higher, and in two of three experiments when present at a frequency of 5% (data not shown). The env region of HIV-1 was successfully amplified from the plasma of all but one of the 103 patients studied. The phenotype of coreceptor-mediated entry was then successfully determined for 100 of these 102 patients; it revealed 74 virus populations with an R5 phenotype, 19 virus populations with a dual/mixed R5X4 phenotype, and seven virus populations with an X4 phenotype. Of these 100 samples, 89 gave high positive luminescent signals (RLU more than 10-fold the cut off value), and 11 gave low positive luminescent signals (RLU 1- to 10-fold the cut off value) on one or both indicator cell lines. The CCR5 and CXCR4 antagonists, TAK779, and AMD3100, confirmed the specificity of the low positive signals from these 11 samples.

Genotypic prediction of HIV-1 coreceptor usage based on combined criteria from the 11/25 and net charge rules

The V3 genotype was obtained by direct sequencing from bulk env PCR products from 100 of the 102 patients. The V3 genotype from the remaining two patients could not be analyzed because of overlapping peaks on the sequencer electrophoregrams. Both genotypic V3 data and phenotypic results were obtained for 98 patients and these were used to study the capacity of the genotype to predict HIV-1 tropism. The combined criteria from the 11/25 and net charge rules were used to classify 75 virus populations as R5 and 23 as X4 (Table 2). Because it is impossible to discriminate between pure X4 and R5X4 dualtropic viruses by genotypic approaches, they were both genotypically classified as CXCR4-using viruses. The combined criteria from the 11/25 and net charge rules detected CXCR4-using viruses with a sensitivity of 77% and a specificity of 96% (Table 3).

Table 2

Table 3

Genotypic prediction of HIV-1 coreceptor usage based on the 11/25 rule and the bioinformatic tools Geno2pheno and WebPSSM

The genotypic classification based on the 11/25 rule and the Geno2pheno and WebPSSM tools are compared with the observed phenotype in Table 2. The 11/25 rule detected CXCR4-using viruses with a sensitivity of 65% and a specificity of 94%. Geno2pheno was the most sensitive (88%), but was less specific (87%). PSSM_X4/R5 was less sensitive (69%) than PSSM_SI/NSI (77%), both were highly specific (97 and 94%, respectively) (Table 3).

Discussion

CCR5 antagonists are new drugs for treating HIV-infected patients, but they can only be used to treat patients harboring R5 viruses. Hence, efficient, affordable methods for routinely characterizing and monitoring HIV-1 tropism are needed. Replacing the complicated phenotypic assays with simple genotypic predictions of HIV-1 tropism could therefore be most important for clinical practice. However, genotypic assays must be based on strong, clear correlations between the genotype and the phenotype.

The study of the genotype-phenotype correlations in a population is complicated by the presence of minor species. Our recent work on clones showed excellent correlations between genotypes and phenotypes in a set of 118 V1-V3 molecular clones. The sensitivity of the V3 genotype for detecting CXCR4-using viruses at a clonal level was 94%, much better than previously reported. The study of bulk samples from 26 patient shows that population-based genotypic and phenotypic approaches correlated well with clonal analyses ^[9].

The results of the present study on an independent cohort of 103 patients confirm our previous findings. We used an improved single-cycle recombinant virus phenotypic assay that combines the simplicity of homologous recombination between an env PCR product and the pNL43-Δenv vector and the enhanced sensitivity of bioluminescent detection. Our phenotypic assay can detect minor X4 variants when present at a frequency of 5-10%, a sensitivity similar to that of the 'Trofile' assay ^[3]. Phenotypic results were obtained for 100 (97%) of the 103 patients tested. This success rate is better than that reported for the 'Trofile' assay (84.9%), and similar to that of the 'Tropism Recombinant Test (TRT)' assay (96.8%) ^[14].

Only 25% of the patients studied here harbored R5X4 or X4 viruses, compared with a frequency of 40-50% in previous studies of treatment-experienced patients ^[15,16]. This lower frequency of CXCR4-using viruses in our study is probably due to the higher CD4+ T-cell count and nadir of the patients studied. We also found R5X4/X4 viruses in 60% of patients with low nadir of CD4+ T cells and poor immunological response to HAART ^[17].

The genotypic approaches performed well in paired genotypic and phenotypic evaluation of HIV-1 coreceptor usage in 98 patients, with a global concordance of 88-91% between the genotype and the phenotype. The CXCR4-using viruses were detected by the combined criteria from the 11/25 and net charge rules with a sensitivity of 77% and a specificity of 96%. The bioinformatic tools Geno2pheno and WebPSSM also performed well. WebPSSM can use an X4/R5 or an SI/NSI matrix. Both are contributory because there is no absolute concordance between the X4 and SI phenotypes. The performances of the SI/NSI matrix seem to be slightly better than those of the X4/R5 matrix for predicting HIV-1 tropism. The genotypic algorithms have mostly been developed for viruses of clade B. Only 15% of our patients harbored viruses of non-B subtypes, and we found the same genotype-phenotype concordance among viruses of clade B (76/84) or non-B (13/14). A recent study on a data set of 115 non-B subtypes and 35 B subtypes viruses found that the bioinformatics tools were less sensitive for detecting CXCR4-using viruses in non-B than in B viruses ^[18]. There were not enough patients with non-B viruses in our study for us to address this issue. Studies analyzing genotype-phenotype correlations for each non-B subtype are needed to assess and improve the genotypic predictions.

Sensitivities of only 22-50% for predicting CXCR4 usage have previously been reported for both the 11/25 rule and for bioinformatic tools like Geno2pheno and WebPSSM ^[8]. The poor genotype-phenotype correlations could be due to a certain degree of heterogeneity in the paired genotypic and phenotypic data used. Differences between the assays or PCR amplification bias between the products used for the phenotypic assay and the V3 genotype could explain some of the discrepancies.

The high genotype-phenotype concordance found in our 98 patients could be due to the homogeneous assessment of HIV-1 tropism using single phenotypic and genotypic assays, performed in parallel for each patient from the same bulk env PCR products. This prevents sampling bias of the virus population between the genotypic and phenotypic assays. Degenerate primers were used for V3 sequencing to take into account the diversity of the virus quasispecies. Careful attention should be paid to minor species when analyzing the bulk V3 genotypes, as previously shown by comparison with clonal analyses ^[9]. Two recent studies also found that different bioinformatic algorithms based on the V3 genotype provided 70-86% concordance with the results of the 'Trofile' and 'TRT' phenotypic assays, in agreement with our results ^[14,19].

Thus, direct V3 sequencing could represent an alternative to phenotypic assays for assessing HIV-1 tropism. Multicenter studies analyzing the correlations between the genotypic determination of HIV-1 tropism and clinical response to CCR5 antagonists are needed to validate this approach in clinical practice.

Acknowledgement

Author contributions: S.R., P.D., and J.I. designed the study, analyzed the data, and wrote the manuscript. M.C. and C.S. performed the experiments. M.M., L.C., B.M., and P.M. contributed substantially to study conception and provided clinical samples. K.S.-S. performed statistical analysis.

Financial support: INSERM U563.