Secondary Logo

Journal Logo


Current V3 genotyping algorithms are inadequate for predicting X4 co-receptor usage in clinical isolates

Low, Andrew Ja,b; Dong, Winniea; Chan, Dennisona; Sing, Tobiasc; Swanstrom, Ronaldd; Jensen, Marke; Pillai, Satishf; Good, Benjaming; Harrigan, P Richarda,b

Author Information
doi: 10.1097/QAD.0b013e3282ef81ea
  • Free



HIV gains entry into CD4-expressing cells by using the CXCR4 and/or CCR5 co-receptors [1–5]. The capacity to use CXCR4, measured by the ability of the virus to form syncytia in CXCR4 expressing MT-2 cells [6] has been associated with rapid CD4 decline, accelerated disease progression, and reduced survival time in untreated individuals [7–9]. Accurately determining co-receptor tropism is of current clinical concern, especially in the context of screening patients prior to CCR5 antagonist-based therapies, as patients with detectable CXCR4-using virus do not show a significant virological response when administered CCR5 antagonists [10–12].

HIV co-receptor use in clinical samples is most commonly determined using a recombinant phenotype assay [13,14]. Although highly accurate on clonal samples, when these assays [13,14] were compared using clinically derived samples they showed an 85.1% concordance with each other [15], with discordances likely due to low level minority species [15]. Bioinformatic predictors based on viral genotype may be able to predict co-receptor usage in a more cost-effective and timely manner. The 11/25 charge rule [16], which is based on the presence of positively charged amino acids at positions 11 and/or 25 of the third hypervariable loop (V3 loop) of the envelope glycoprotein gp120, is a simple genotypic predictor of syncytium-inducing (SI) HIV and has been shown to be associated with reduced CD4 response and decreased survival time following the initiation of HAART [17]. Although the 11/25 rule displays > 90% sensitivity and specificity for predicting the SI phenotype on clonal HIV sequences [18], this sensitivity is reduced to < 60% for predicting the X4/R5 phenotype on clonal data and is further reduced when tested on bulk (population-based) sequences from clinically derived samples [18–21]. Other published bioinformatic methods, such as support vector machines (SVM) [22,23], neural networks (NN) [21], and position specific scoring matrices (PSSM) [24] have also demonstrated relatively high sensitivities for predicting SI HIV (SVM = 75%; NN = 90%; PSSM = 93%), and with a modest decrease in sensitivity, CXCR4-using HIV (SVM = 67%; NN = 75%; PSSM = 62%), when tested on sequences derived from clonal samples [18].

The use of CCR5 antagonists requires screening for co-receptor usage prior to their use. The requirement for screening could result in publicly available genotype-based predictors developed for HIV clones being used for clinical screening purposes. However, the heterogeneous nature of the HIV viral population sampled in the peripheral blood of HIV-infected patients creates difficulty in determining co-receptor usage with both genotype and phenotype based approaches. Genotypic methods trained on clonal data may not be able to interpret sequence ambiguities present in bulk population-based sequence data accurately, and minority species may be lost during PCR amplification for both genotype and phenotype based approaches.

In this study, we compared the sensitivities and specificities of five publicly available co-receptor predictors that use V3 genotype, including the 11/25 rule [16], two PSSM (PSSMSI/NSI, and PSSMX4/R5[24], a NN, and two SVM (SVMgenomiac[22], and SVMgeno2pheno[23]) for predicting CXCR4 co-receptor phenotype (Monogram Trofile assay).


Study subjects: the British Columbia HOMER cohort

The present study represents a baseline cross-sectional analysis of the HOMER [25] cohort, where V3 genotypes and co-receptor phenotypes were determined prior to the initiation of HAART [17,19]. Plasma viral loads, CD4 cell counts as well as genotypic and phenotypic parameters represent the latest pretherapy measurements collected in the 180 days prior to HAART initiation [25].

Determination of baseline HIV co-receptor phenotype

The Trofile assay, performed at Monogram Biosciences [13] has previously been used to assess co-receptor phenotype in the latest pretherapy plasma sample for each subject and the clinical predictors of the co-receptor phenotype were described [19]. Briefly, a reverse transcription (RT)-PCR product spanning the entire gp160 is digested, purified and ligated into an E. coli expression vector, and gene libraries are constructed. Recombinant viruses are harvested after 48 h and assessed for their ability to infect U87-CD4 cells expressing CCR5 or CXCR4, as determined by a luciferase read-out measured in relative light units (RLU) on each of the cell lines. The Trofile assay classifies samples as CCR5-using, CXCR4-using or DM (indicating dual and/or mixed-tropic virus) based on confirmation of decreased RLU values by 50% or more upon addition of specific co-receptor inhibitors. Phenotypic data from 977 isolates were available, with 799, 177 and one characterized as R5, R5/X4 (DM) and X4-only, respectively, with a higher prevalence of X4-capable virus being detected in individuals with lower CD4 cell counts [19]. Here, isolates phenotyped as DM or X4 were combined and designated as DM for the remainder of analyses; CCR5-only using virus is referred to as R5.

Determination of corresponding baseline envelope V3 sequence

Aliquots of the same baseline plasma samples were used to determine bulk population HIV V3 envelope sequence as described [17]. Matched baseline co-receptor phenotypes and V3 genotypes were available for 953 of 1188 (80.2%) HOMER subjects. Sequences with more than seven amino acid mixtures in the V3 were excluded due to combinatorial factors (alignment and submission to online algorithms was either not possible or was prohibitively time-consuming), leaving 920 samples with matched baseline co-receptor phenotype and genotype (GenBank accession EF637088-EF638007). V3 sequences with nucleotide mixtures were translated into all possible amino acid permutations, resulting in 5512 unique V3 amino acid sequences, as well as all possible nucleotide permutations, resulting in 11 447 unique V3 nucleotide sequences, referred to as the amino acid and nucleotide datasets, respectively. Sequence alignments were performed using MUSCLE [26], followed by visual inspection.

Bioinformatic predictions

After alignment, sequences with positively charged amino acids at codons 11 and/or 25 of the V3 loop, associated with an HIV SI phenotype, were classified as having an 11/25 genotype. Data from the NN [21] had been previously collected [17], with no changes to the algorithm made since. The PSSMX4/R5, which is the PSSM trained with X4 and R5 data, and PSSMSI/NSI, which is the PSSM trained on SI and non-syncitium inducing (NSI) data, available at: (February 2007) [24], use unaligned sequence data from the amino acid dataset as an input, and output categorical (R5 or X4) scores as well as a continuous output variable, which will be referred to as the PSSM score. The SVMgenomiac, available at: (February 2007) [22], outputs a categorical score (CCR5 or CXCR4) and uses the amino acid dataset aligned to the following standard, CTRPNNNT-RK*I*I–GPG*AFY*-TG*I-IGDIRQAHC, where (*) indicates any amino acid or gap. The SVMgeno2pheno available at: (February 2007) [27] also outputs a categorical score (R5 or X4) and uses the unaligned nucleotide dataset as an input. For this analysis, the default false-positive rate of 0.1 was chosen to yield a specificity of approximately 90%. Another more recent version of this SVM was also developed to include clinical data (CD4%, number of sequence ambiguities, host CCR5Δ32 heterozygosity, and presence of insertions/deletions) as input variables [28]. This model was compared to the previous SVM to determine the impact clinical markers have on co-receptor prediction. For all predictors, a sample was scored as being X4 if ≥ 25% of all expanded sequence permutations were classified as X4 [17].

Clonal analysis for low level detection of minority X4-capable variants

Cloning of amplified PCR product was performed using the Invitrogen TOPO TA cloning kit (catalog K4550-40, Burlington, Ontario, Canada), containing the PCR 2.1-TOPO vector with chemically competent TOP10F′ one shot cells, according to the manufacturer's instructions. Clones (N = 48) were picked for eight samples chosen to address whether bulk sequencing methods contained undetected, low level minority variants. Sequencing was performed on the population sample as well as all viable clones using standard automated sequencing techniques [17].


V3 genotypes of known co-receptor tropism phenotype (N = 920) were submitted to six different V3 genotype-based algorithms. The initial sensitivity and specificity for predicting X4 capacity of the 11/25 rule (31% sensitivity/93% specificity), NN (44%/88%), PSSM(sinsi) (34%/96%), PSSM(x4/r5) (24%/97%), SVMgenomiac (22%/90%) and SVMgeno2pheno (50%/89%) demonstrates that although specificities for all genotype-based methods were high, the sensitivities for all methods, when tested on clinically derived samples, were relatively low (Table 1). The PSSM provide a raw score (PSSM score) in addition to the categorical predicted phenotype which, after averaging over all permutations, provided a single PSSM score for each sample. This PSSM score could be optimized with receiver-operating characteristic (ROC) curves by reducing the specificity to 90%, resulting in a sensitivity of 43.7% (cutoff = −5.93) for the PSSMSI/NSI and 43.7% (cutoff = −8.12) for the PSSMx4/r5. Adjusted sensitivities and specificities represented by a more aggressive approach for the 11/25 rule (31% sensitivity/92% specificity), PSSM(sinsi) (38%/95%), PSSM(x4r5) (27%/96%), SVMgenomiac (24%/88%) and SVMgeno2pheno (50%/89%) were calculated by categorizing a sample as X4 if any of its sequence permutations were scored as X4 (included in Table 1). This method could be applied to both SVM and PSSM methods, but not the NN. In addition, the PSSM methods were further optimized by assigning each sample the highest PSSM score (most likely to be X4) of all its permutations, instead of the average PSSM score. Using ROC curve analysis at a specificity of 90% on this data resulted in a sensitivity of 51.7% (cutoff = −4.81) for the PSSMSI/NSI and 49.0% (cutoff = −7.18) for the PSSMX4/R5. Overall, the total concordance of the bioinformatic methods with each other was 61.1% (35.8% in X4 samples only and 66.1% in R5 samples only). When limited to the methods with the greatest sensitivity (the PSSMs and SVMgeno2pheno) the concordance was 84.6% (X4 samples 78.1%; R5 samples = 85.8%).

Table 1
Table 1:
Sensitivity and specificity of the six V3 genotype-based algorithms.

In the phenotype assay, luciferase produced in CCR5 or CXCR4 expressing U87 cells is measured in RLU. To examine how sensitivity changed with Trofile Assay output, samples were grouped into 9 log10(CXCR4 RLU) strata (Fig. 1). All optimized predictive methods showed a significant increase in sensitivity with increasing CXCR4 RLU (P < <0.05).

Fig. 1
Fig. 1:
Sensitivity of genotype-based predictors stratified by CXCR4 RLU. Sensitivity, defined as the proportion of all positives (DM by Phenotype Assay) detected by the predictive method, is calculated for all predictors grouped within 11 CXCR4 RLU strata: <2.6 (n = 11), 2.6–3.0 (n = 22), 3.0–3.4 (n = 23), 3.4–3.8 (n = 13), 3.8–4.2 (n = 21), 4.2–4.6 (n = 26), 4.6–5.0 (n = 15), 5.0–5.4 (n = 10), ≥5.4 (n = 10). Linear regressions are derived from the sensitivity of each stratum, where each stratum is represented by incremental integer values. Symbols and linear regression coefficients for the methods used are: (a) 11/25 rule (▪ with a dashed, black trendline; R2 = 0.77), PSSMSI/NSI (♦ with a solid black trendline; R2 = 0.94), PSSMX4/R5 (
Table 1
with a solid grey trendline; R2 = 0.90); (b) Neural Network (▪ with a dashed, black trendline; R2 = 0.85), SVMgenomiac (♦ with a solid black trendline; R2 = 0.52), SVMgeno2pheno (
Table 1
with a solid grey trendline; R2 = 0.83).

The dependence of predictor sensitivity on CXCR4 RLU suggests that population or bulk V3 genotype may be missing low level minority species. To test this hypothesis, eight patient samples, with plasma viral loads (pVL) > 4 log10 (to reduce the likelihood of PCR resampling) were chosen for cloning and sequencing. Two samples were chosen with matching R5 phenotype and genotype (11/25−) while three samples with matching X4-capable (DM) phenotype and genotype (11/25+) acted as controls. The final three samples were chosen with discordant DM phenotype and 11/25− genotype. A total of 48 clones were picked for each sample and after sequencing, each clone was categorized as R5 or X4 based on the 11/25 rule. All samples which were concordant between phenotype and genotype showed no clones with discordant genotypes from the bulk genotype (Table 2). Samples 1 and 2 were both R5 in the phenotype assay and bulk genotype (11/25 rule) and all clones were genotyped as being R5. Samples 3, 4 and 5 were DM in the phenotype assay and X4 in the bulk genotype. All clones for samples 3 and 4 were genotyped as being X4, while sample 5 was 23.5% X4 (26 R5 clones; 8 X4 clones). However, samples discordant between the genotype and phenotype showed varying levels of an X4 genotype population in the clonal samples when compared to the bulk sequence indicating a minority species which was undetectable in the bulk genotype. Samples 6, 7 and 8 were DM in the phenotype assay and R5 in the bulk genotype. Sample 6 was 22.2% X4 (35 R5 clones; 10 X4 clones), sample 7 was 21.4% X4 (33 R5 clones; 9 X4 clones) and sample 8 was 8.9% X4 (41 R5 clones; 4 X4 clones).

Table 2
Table 2:
Comparison of the ability of the six V3 genotype-based algorithms to detect minority genotypic variants.

The positive predictive value (PPV) of the different genotype-to-phenotype predictors increased with decreasing CD4 strata (P = 0.03–0.01) (Fig. 2). Sensitivity did not vary significantly across CD4 strata (data not shown). This dependence of PPV on CD4 indicates a potential benefit for including clinical information such as CD4 count or CD4 percentage (CD4%) into co-receptor predictors [27].

Fig. 2
Fig. 2:
Positive predictive value for X4-tropic virus stratified by CD4 count. Positive predictive value, defined as the proportion of all predicted positives which are true positives, is calculated for all predictors grouped within 7 CD4 strata: <25 (n = 69), 25–49 (n = 56), 50–99 (n = 75), 100–199 (n = 163), 200–349 (n = 238), 350–499 (n = 182), >500 (n = 137). Linear regressions are derived from the sensitivity of each stratum, where each stratum is represented by incremental integer values. Symbols and linear regression coefficients for the methods used are: (a) ‘11/25’ rule (▪ with a dashed, black trendline; R2 = 0.71), PSSMSI/NSI (♦ with a solid black trendline; R2 = 0.65), PSSMX4/R5 (
Table 1
with a solid grey trendline; R2 = 0.80); (b) Neural Network (▪ with a black trendline made a series of dots; R2 = 0.80), SVMgenomiac (♦ with a solid black trendline; R2 = 0.62), SVMgeno2pheno (
Table 1
with a solid grey trendline; R2 = 0.91).


Although implementations of HIV V3-loop based co-receptor predictors perform relatively well in clonal samples [18], their performance had not yet been tested on patient-derived samples. Our data suggests these methods, available in February 2007, are too insensitive to be implemented in a clinical setting, due in part to the presence of low level mixtures in standard bulk genotyping, and require improvement. Adjusting the cutoffs of the raw score output of the PSSM, [24] and using a more sensitive approach, where any X4 permutation results in an X4 sample to increase sensitivity are two effective examples which highlight the importance of optimizing these methods for testing on clinical samples.

The SVMgenomiac[22] had the lowest sensitivity of all the predictors, which is in contrast to the SVMgeno2pheno[23] (one the best performing predictors), although both predictors were based on the SVM machine-learning model. The SVMgenomiac was trained on a dataset comprised of multiple HIV clades, while all other methods were trained on datasets composed primarily of clade-B virus. The clinical test set from British Columbia is composed of 97.5% clade-B virus, thus biasing the results in favor of those methods trained on primarily clade-B sequences. Additionally, the PSSM and the SVMgeno2pheno were trained on the same set of clonal samples, and both methods performed nearly equally well at all CXCR4 RLU and CD4 strata, despite the different approaches used.

In another study, predictions were performed using the SVMgeno2pheno model, but with additional clinical parameters included (log10[CD4%]; host Δ32 heterozygosity; number of ambiguous amino acid V3 positions; and a variable indicating the presence of insertions or deletions in the V3 sequence) [28]. Sensitivity of the predictor with clinical parameters was approximately 10% greater than the purely sequence based approach at a specificity of 90% [28]. Although it has been previously demonstrated that there is no significant association between CXCR4 RLU and CD4 cell count [29], the association of CD4 count with PPV and the increase in sensitivity, achieved by integrating clinical information into training and prediction, indicate the importance of integrating clinical parameters for co-receptor prediction. However, it is not yet clear whether low CD4 cell counts are indicative of a more favorable immune environment for X4-capable virus or if CD4 count is an independent predictor of X4-capable virus. The log-linear dependence of CXCR4 RLU sensitivity for all methods indicates that samples producing higher levels of CXCR4 RLU are more likely to be genotypically characterized as X4. Lower CXCR4 RLU signals may indicate a minority species at levels undetectable to bulk sequencing techniques and/or a genotypically unique X4-tropic viral species present in vivo, which predictors trained on clonal data cannot detect.

Results of the clonal analyses performed here indicate that bulk sequencing techniques were unable to reliably detect minority species present below approximately 22% which is in agreement with previous studies that estimate the limit of detection for minority species by direct sequencing of bulk PCR products to be 10–25% [30–33]. Unlike genotype-based HIV antiretroviral resistance testing, low level minority species of X4-using virus appear to be the rule rather than the exception. This is likely a reason for much of the discordance and lack of sensitivity observed, and also indicates that an area for improvement would be increasing the sensitivity of detection for minority variants. Techniques that may improve the detection of minority variants include adopting more aggressive base-calling techniques or examining the area under the curve of the chromatograms as a variable when nucleotide mixtures are observed. New technologies capable of more sensitive base-calling or detection of minority DNA species [34] may result in greatly increased sensitivities for genotype-based predictors. Alternatively, methods capable of separating genotypic variants, such as the heteroduplex tracking assays [35] combined with sequence analysis may provide another approach for the detection of minority variants.

In this large clinically based study, all V3 genotype and co-receptor phenotype results were derived on the same set of samples using well defined methodologies, and all predictive methods were tested on the same data. However, this study is limited by the fact that associations between HIV co-receptor usage and clinical parameters are based on cross-sectional data from a population of therapy-naive individuals initiating their first antiretroviral treatment during the period 1996–1999, and may not be representative of the HIV-infected population in general, nor of individuals treated with co-receptor antagonists. It should be emphasized that these results were limited to therapy naive individuals and therefore, cannot necessarily be extrapolated to therapy experienced individuals. The methods investigated in this study limited their training and test set to only the V3 loop of the gp120 surface glycoprotein. Although the V3-loop region contains the majority of mutations predictive of co-receptor usage, expanding the genotypic sequence sampled to regions outside the V3 loop may result in increased sensitivities. In addition, the Monogram Trofile assay, or any other co-receptor assay, cannot yet be considered a ‘gold-standard’. This was illustrated in a comparison of the Trofile assay to another well validated phenotype co-receptor assay, which yielded approximately 85% concordance [15]. Also, in clinical trials, maraviroc selected for preexisting X4 variants which were undetected by the Trofile assay [36]. Therefore, although this assay is in the unique position of being the sole test used in screening for co-receptor clinical trials, other phenotype assays are being developed [37,38], and there is still room for improvement.

In conclusion, the current sensitivities of genotype-based co-receptor predictors of 30–50% are inadequate for the prediction of CXCR4 co-receptor usage in a clinical setting. Although the exact level of desired sensitivity is not known, sensitivities upwards of 85% would be equivalent to the concordance of co-receptor phenotype assays [15] and should be sufficient. Although, even if a genotype-based method approached or exceeded the sensitivity of the phenotype assay, it would still need to be validated before it could be used as a widespread clinical tool. Importantly, the results obtained here indicate that the type of predictive model used is not nearly as important as the reliability and effectiveness of the parameters chosen to represent the phenotype. Genotype-based co-receptor predictors do have the potential to significantly decrease turn-around time and cost in comparison to current phenotype assay approaches, and although current implementations of genotype-based predictors are inadequate, there are identifiable methods to improve their sensitivity. In the interim, however, these approaches should not be used alone as screening tools for clinical use of co-receptor antagonists.


Ethical approval for this study was obtained from the institutional ethics board (Providence HealthCare/University of British Columbia).

We would like to thank James Goodrich and Howie Mayer from Pfizer Inc., as well as Brian Wynhoven, Theresa Mo and Conan Woods from the BC Centre for Excellence in HIV/AIDS for their contributions to this study. This study was supported through an R&D research grant from the Canadian Institutes for Health Research (CIHR) and by funding from Pfizer, Inc. Richard Harrigan has previously consulted for Monogram, Virco, and other companies in the pharmaceutical and diagnostic industries.


1. Deng H, Liu R, Ellmeier W, Choe S, Unutmaz D, Burkhart M, et al. Identification of a major co-receptor for primary isolates of HIV-1. Nature 1996; 381:661–666.
2. Dragic T, Litwin V, Allaway GP, Martin SR, Huang Y, Nagashima KA, et al. HIV-1 entry into CD4+ cells is mediated by the chemokine receptor CC-CKR-5. Nature 1996; 381:667–673.
3. Feng Y, Broder CC, Kennedy PE, Berger EA. HIV-1 entry cofactor: Functional cDNA cloning of a seven-transmembrane, G protein-coupled receptor. Science 1996; 272:872–877.
4. Moore JP, Kitchen SG, Pugach P, Zack JA. The CCR5 and CXCR4 coreceptors-central to understanding the transmission and pathogenesis of human immunodeficiency virus type 1 infection. AIDS Res Hum Retroviruses 2004; 20:111–126.
5. Zhang L, He T, Huang Y, Chen Z, Guo Y, Wu S, et al. Chemokine coreceptor usage by diverse primary isolates of human immunodeficiency virus type 1. J Virol 1998; 72:9307–9312.
6. Schuitemaker H, Kootstra NA, de Goede RE, de Wolf F, Miedema F, Tersmette M. Monocytotropic human immunodeficiency virus type 1 (HIV-1) variants detectable in all stages of HIV-1 infection lack T-cell line tropism and syncytium-inducing ability in primary T-cell culture. J Virol 1991; 65:356–363.
7. Schuitemaker H, Koot M, Kootstra NA, Dercksen MW, de Goede RE, van Steenwijk RP, et al. Biological phenotype of human immunodeficiency virus type 1 clones at different stages of infection: Progression of disease is associated with a shift from monocytotropic to T-cell-tropic virus population. J Virol 1992; 66:1354–1360.
8. Tersmette M, Gruters RA, de Wolf F, de Goede RE, Lange JM, Schellekens PT, et al. Evidence for a role of virulent human immunodeficiency virus (HIV) variants in the pathogenesis of acquired immunodeficiency syndrome: Studies on sequential HIV isolates. J Virol 1989; 63:2118–2125.
9. Koot M, Keet IP, Vos AH, de Goede RE, Roos MT, Coutinho RA, et al. Prognostic value of HIV-1 syncytium-inducing phenotype for rate of CD4+ cell depletion and progression to AIDS. Ann Intern Med 1993; 118:681–688.
10. Mayer H, Van der Ryst E, Saag M, Clotet G, Fätkenheuer N, Clumeck N, et al. Safety and efficacy of maraviroc, a novel CCR5 antagonist, when used in combination with optimized background therapy for the treatment of antiretroviral-experiences subjects infected with dual/mixed-tropic HIV-1: 24-week results of a phase 2b exploratory trial. XVI International AIDS Conference, August 13–18, Toronto, Ontario, Canada, 2006 [Abstract THLB0215].
11. Nelson M, Fätkenheuer N, Konourina I, Lazzarin A, Clumeck N, Horban A, et al. Efficacy and Safety of maraviroc plus optimized background therapy in viremic, ART-experienced patients infected with CCR5-tropic HIV-1 in Europe, Australia, and North America: 24-week results. 14th Conference on Retroviruses and Opportunistic Infections, February 25–28, Los Angeles, California, USA, 2007 [Abstract 104aLB].
12. Lalezari J, Goodrich J, DeJesus E, Lampiris H, Gulick R, Saag M, et al. Efficacy and safety of maraviroc plus optimized background therapy in viremic ART-experienced patients infected with CCR5-tropic HIV-1: 24-week results of a phase 2b/3 Study in the US and Canada. 14th Conference on Retroviruses and Opportunistic Infections, February 25–28, Los Angeles, California, USA, 2007 [Abstract 104bLB].
13. Whitcomb JM, Huang W, Fransen S, Limoli K, Toma J, Wrin T, et al. Development and characterization of a novel single-cycle recombinant-virus assay to determine human immunodeficiency virus type 1 coreceptor tropism. Antimicrob Agents Chemother 2007; 51:566–575.
14. Trouplin V, Salvatori F, Cappello F, Obry V, Brelot A, Heveker N, et al. Determination of coreceptor usage of human immunodeficiency virus type 1 from patient plasma samples by using a recombinant phenotypic assay. J Virol 2001; 75:251–259.
15. Skrabal K, Low AJ, Dong W, Sing T, Cheung PK, Mammano F, et al. Determining human immunodeficiency virus coreceptor use in a clinical setting: Degree of correlation between two phenotypic assays and a bioinformatic model. J Clin Microbiol 2007; 45:279–284.
16. De Jong JJ, De Ronde A, Keulen W, Tersmette M, Goudsmit J. Minimal requirements for the human immunodeficiency virus type 1 V3 domain to support the syncytium-inducing phenotype: Analysis by single amino acid substitution. J Virol 1992; 66:6777–6780.
17. Brumme ZL, Dong WW, Yip B, Wynhoven B, Hoffman NG, Swanstrom R, et al. Clinical and immunological impact of HIV envelope V3 sequence variation after starting initial triple antiretroviral therapy. AIDS 2004; 18:F1–F9.
18. Jensen MA, van 't Wout AB. Predicting HIV-1 coreceptor usage with sequence analysis. AIDS Rev 2003; 5:104–112.
19. Brumme ZL, Goodrich J, Mayer HB, Brumme CJ, Henrick BM, Wynhoven B, et al. Molecular and clinical epidemiology of CXCR4-using HIV-1 in a large population of antiretroviral-naive individuals. J Infect Dis 2005; 192:466–474.
20. Fouchier RA, Brouwer M, Broersen SM, Schuitemaker H. Simple determination of human immunodeficiency virus type 1 syncytium-inducing V3 genotype by PCR. J Clin Microbiol 1995; 33:906–911.
21. Resch W, Hoffman N, Swanstrom R. Improved success of phenotype prediction of the human immunodeficiency virus type 1 from envelope variable loop 3 sequence using neural networks. Virology 2001; 288:51–62.
22. Pillai S, Good B, Richman D, Corbeil J. A new perspective on V3 phenotype prediction. AIDS Res Hum Retroviruses 2003; 19:145–149.
23. Sing T, Lengauer T, Beerenwinkel N, Kaiser R, Daumer M, Low A, et al. Immunological and virological parameters improve genotypic prediction of HIV-1 coreceptor usage. BioSapiens-viRgil Workshop on Bioinformatics for Infectious Diseases, September 21–23, Bonn, Germany, 2005[ Abstract 22].
24. Jensen MA, Li FS, van 't Wout AB, Nickle DC, Shriner D, He HX, et al. Improved coreceptor usage prediction and genotypic monitoring of R5-to-X4 transition by motif analysis of human immunodeficiency virus type 1 env V3 loop sequences. J Virol 2003; 77:13376–13388.
25. Hogg RS, Yip B, Chan KJ, Wood E, Craib KJ, O'Shaughnessy MV, et al. Rates of disease progression by baseline CD4 cell count and viral load after initiating triple-drug therapy. JAMA 2001; 286:2568–2577.
26. Edgar RC. MUSCLE: Multiple sequence alignment with high accuracy and high throughput. Nucleic Acids Res 2004; 32:1792–1797.
27. Sing T, Harrigan R, Beerenwinkel N, Daumer M, Kaiser R, Low A, et al. Immunologic markers improve genotypic prediction of HIV-1 co-receptor usage. Int Workshop on Targeting HIV Entry, December 2–3, Bethesda, Maryland, USA 2005 [Abstract 30].
28. Sing T, Harrigan R, Beerenwinkel N, Daumer M, Kaiser R, Low A, et al. Immunologic markers improve genotypic prediction of HIV-1 coreceptor usage. Keystone symposium on genome sequence variation and the inherited basis of common disease and complex traits, January 8–13, Big Sky, Montana, USA, 2006 [Abstract 30].
29. Low AJ, Brumme CJ, Brumme ZL, Hogg RS, Montaner JS, Harrigan PR, et al. Immunologic, virologic, and genotypic correlates of phenotypic CCR5/CXCR4 co-receptor usage assay readouts in a large panel of clinically derived isolates. 14th Conference on Retroviruses and Opportunistic Infections, February 25–28, Los Angeles, California, USA, 2007 [Abstract 177].
30. Larder BA, Kohli A, Kellam P, Kemp SD, Kronick M, Henfrey RD. Quantitative detection of HIV-1 drug resistance mutations by automated DNA sequencing. Nature 1993; 365:671–673.
31. Leitner T, Halapi E, Scarlatti G, Rossi P, Albert J, Fenyo EM, et al. Analysis of heterogeneous viral populations by direct DNA sequencing. BioTechniques 1993; 15:120–127.
32. Schuurman R, Demeter L, Reichelderfer P, Tijnagel J, de Groot T, Boucher C. Worldwide evaluation of DNA sequencing approaches for identification of drug resistance mutations in the human immunodeficiency virus type 1 reverse transcriptase. J Clin Microbiol 1999; 37:2291–2296.
33. Van Laethem K, Van Vaerenbergh K, Schmit JC, Sprecher S, Hermans P, De Vroey V, et al. Phenotypic assays and sequencing are less sensitive than point mutation assays for detection of resistance in mixed HIV-1 genotypic populations. J Acquir Immune Defic Syndr 1999; 22:107–118.
34. Lewis M, James I, Braverman M, Desany B, Jarvie T, Penny M, et al. Evaluation of an ultra-deep sequencing method to identify minority sequence variants in the HIV-1envgene from clinical samples. 14th Conference on Retroviruses and Opportunistic Infections, February 25–28, Los Angeles, California, USA, 2007 [Abstract 680].
35. Nelson JA, Baribaud F, Edwards T, Swanstrom R. Patterns of changes in human immunodeficiency virus type 1 V3 sequence populations late in infection. J Virol 2000; 74:8494–8501.
36. Lewis M, Simpson P, Francen S, Huang W, Whitcomb J, Mosley M, et al. CXCR4-using virus detected in patients receiving maraviroc in the Phase III studies MOTIVATE 1 and 2 originates from a preexisting minority of CXCR4-using virus. 16th International HIV Drug Resistance Workshop, June 12–16, Barbados, West Indies, 2007 [Abstract 56].
37. Roulet V, Faudon J-L, Rochas S, Labernardiere J-L, Mammano F, Level-Binay S, et al. A sensitive assay for the determination of viral tropism suitable for clinical management of HIV subtype B and non-B infected individuals treated with coreceptor antagonists: PHENOSCRIPT®ENV. 16th International HIV Drug Resistance Workshop, June 12–16, Barbados, West Indies, 2007 [Abstract 147].
38. Vandenbroucke I, Van Eygen V, Rondelez E, Van Baelen K, Styver LJ. Detection limit of X4 minority virus in abundant R5-tropic virus populations in standard genotypic and phenotypic assays. 16th International HIV Drug Resistance Workshop, June 12–16, Barbados, West Indies, 2007 [Abstract 139].

CCR5; co-receptor; CXCR4; HIV; R5; V3; viral phenotype; X4

© 2007 Lippincott Williams & Wilkins, Inc.