Sequential Turnover of Human Immunodeficiency Virus Type 1 env throughout the Course of Infection

Abstract

We examined the rates of variant population turnover of the V1-V2 and V4-V5 hypervariable domains of the human immunodeficiency virus type 1 (HIV-1) gp120 molecule in longitudinal plasma samples from 14 men with chronic HIV-1 infection using heteroduplex tracking assays (HTA). Six men had high rates of CD4⁺ T-cell loss, and eight men had low rates of CD4⁺ T-cell loss over 2.5 to 8 years of infection. We found that V1-V2 and V4-V5 env populations changed dramatically over time in all 14 subjects; the changes in these regions were significantly correlated with each another over time. The subjects with rapid CD4 loss had significantly less change in their env populations than the subjects with slow CD4 loss. The two subjects with rapid CD4 loss and sustained low CD4 counts (<150/μl for at least 2 years) showed stabilization of their V1-V2 and V4-V5 populations as reflected by low levels of total change in HTA pattern and low HTA indices (a novel measure of the emergence of new bands and band distribution); this stabilization was not observed in other subjects. The stabilization of env variant populations at low CD4 counts following periods of rapid viral evolution suggests that selective pressure on env, likely from new immune responses, is minimal when CD4 counts drop dramatically and remain low for extended periods of time.

Human immunodeficiency virus type 1 (HIV-1) infection of an individual generates a population of related but distinct cocirculating viral variants (15). This high level of genetic diversity allows HIV-1 to adapt quickly to multiple selective pressures, including neutralizing antibodies, cytotoxic T lymphocytes, and antiretroviral drugs (17, 21, 32-34, 42). The highest level of diversity within the HIV genome is found in the env gene. The five hypervariable domains, named V1 through V5, form loops on the outside of the gp120 molecule (22). The V3 loop is involved in binding a chemokine receptor for viral entry after gp120 binds the CD4 receptor (40, 43). Changes in the V3 domain determine in part whether HIV utilizes CCR5 (R5 variants) or CXCR4 (X4 variants) chemokine coreceptors (6). The V1-V2 loop is thought to interact with the V3 loop within the gp120 trimer on the virion surface (5, 23). The V4 and V5 domains are on a different face of gp120 than V1-V2 and V3 are (22) and are not directly involved in receptor binding.

The V1-V2 and V4-V5 regions of gp120 are highly variable in sequence and length within an infected individual and between infected individuals (19, 37, 41). The domains are heavily glycosylated, and changes in the location and number of N- and O-linked glycosylation sites in the V1-V2 and V4-V5 regions are associated with escape from neutralization (reviewed in reference 29). Removal of some of the glycosylation sites from the V1 and V2 regions results in increased immunogenicity of the domains (31) and has been shown in one study to redirect the immune response toward V3 rather than V1-V2 (7). Sequence changes observed in V1-V2 and V3 to V5, including insertions, deletions, and point mutations, have been linked to escape from neutralizing antibodies. The accumulation of nonsynonymous substitutions was associated with escape in two separate studies (11, 12), while changes in the glycosylation sites of gp120 have been described in terms of an evolving glycan shield that leads to escape from neutralizing antibodies in a third study (42). It has also been shown that mutations in env lead to escape from neutralizing antibodies through conformational masking of epitopes (21). Mutations in env have also been shown to affect cytotoxic T-lymphocyte (CTL) epitopes (11, 17), indicating that both neutralizing antibodies and CTL apply selective pressure on HIV-1 in vivo. High levels of env sequence diversity have been linked with both slower disease progression and more-effective immune responses against the virus, both in simian immunodeficiency virus and HIV (3, 8, 14, 16, 26), thereby implying a link between strong immune selection and slower disease progression.

Heteroduplex assays are powerful methods for displaying the number and variety of variants within a viral population (reviewed in reference 4) without the potential selection bias inherent to sequence-based analyses. Heteroduplex mobility assays reveal the diversity within a sample through the visualization of heteroduplexes formed between different variants within the sample, while heteroduplex tracking assays (HTA) use a radiolabeled probe to display differences between the probe and the sample. We have previously used HTA to examine the diversity and changes in V1-V2 in monthly samples from subjects with low CD4 counts (18). In 12 of 21 of these subjects, at least one V1-V2 variant population was gained or lost over 5 to 9 months. Sequence analysis of the V1-V2 regions for several of the subjects revealed point mutations, recombination events, and deletions as the main mechanisms of sequence change. Delwart et al. analyzed the V3 to V5 variant populations in semiannual samples from subjects with different rates of progression, showing that subjects with faster CD4⁺ T-cell decline had slower diversification of the region from V3 to V5 and that progression of disease was correlated with reduced diversity (8). These latter results support the idea that the level of sequence diversity directly correlates with immune selection.

In the present study, we have used HTA analysis to examine the V1-V2 and V4-V5 regions of HIV-1 env in semiannual plasma samples from subjects in the San Francisco Men's Health Study. Subjects with different rates of progression, as defined by rate of CD4 cell loss, were compared for rates of change in the env viral population using a newly developed HTA index algorithm that highlights periods of increasing env diversity.

MATERIALS AND METHODS

Subject population.

Subjects were chosen from the San Francisco Men's Health Study (38). Men in San Francisco, California, were enrolled in the study in 1984, and blood samples were collected every 6 months. Of the men who were HIV seropositive at enrollment, CD4 profiles were examined and 14 men were randomly selected, 6 with rapid CD4 decline (at least 125 cells/μl lost per year overall) and 8 with slow CD4 decline (less than 100 cells/μl lost per year overall). Subjects with wildly fluctuating CD4 counts were excluded. Plasma viral loads were determined using the Amplicor HIV Monitor kit (Roche) after pretreatment with heparinase I (Sigma). Analysis of these samples was exempt from Institutional Review Board oversight, because they were existing samples and were coded to ensure the privacy of the subjects.

Viral RNA isolation and RT-PCR.

Viral RNA was isolated from 140 μl of plasma using the QIAamp viral RNA mini kit (QIAGEN) following treatment with heparinase I (Sigma) or from 200 μl plasma using the QIAamp UltraSens virus kit (QIAGEN). In both cases, RNA was eluted with 60 μl buffer. For samples with less than 20,000 copies/ml, 0.5 ml of plasma was pelleted prior to RNA isolation. V1-V2 reverse transcription-PCR (RT-PCR) conditions were modified from those of Kitrinos et al. (18) as follows. Primers for RT-PCR were V1new (HXB numbering 6548 to 6577) (5′-AATCAGTTTATGGGATCAAAGCCTAAAGCC-3′) and V2 (HXB numbering 6951 to 6980) (5′-CTTAATTCCATGTGTACATTGTACTGTGCT-3′) or V4 (HXB numbering 7349 to 7378) (5′-TTTTAATTGTGGAGGGGAATTTTTCTACTG-3′) and V5 (HXB numbering 7647 to 7676) (5′-ATATAATTCACTTCTCCAATTGTCCCTCAT-3′) for V1-V2 and V4-V5 amplification, respectively. The V4-V5 primers were anchored on invariant positions in the subtype B consensus sequence from the HIV Sequence Database (http://hiv-web.lanl.gov) and bounded the V4 and V5 coding regions. RT reactions were generated via a two-step method as follows. (i) Five microliters of viral RNA (at least 200 copies of RNA), 2 mM (each) deoxynucleoside triphosphates, and 15 pmol primer V2 or V5 were mixed, heated to 85°C for 5 min, and then cooled on ice. (ii) Eight microliters of an RT mix (1× RT buffer, 2.5 mM MgCl₂, 6.5 mM dithiothreitol, 20 U RNase OUT, and 100 U Superscript III reverse transcriptase [Invitrogen]) was added to each tube. RT reactions were incubated first at 55°C for 1 h and then at 70°C for 15 min to inactivate the enzyme. After a brief cooling of the reaction mixtures, 2 U of Escherichia coli RNase H (Invitrogen) was added to each reaction mixture and incubated at 37°C for 20 min to degrade residual RNA. The reaction mixtures were cooled, and 35 μl of a PCR mix (1× High Fidelity buffer [Invitrogen], 1.25 mM MgSO₄, 2 mM deoxynucleoside triphosphate mix, 7.5 pmol primer V2 or V5, 15 pmol primer V1new or V4, and 2.5U High Fidelity Platinum Taq DNA polymerase [Invitrogen]) were added. The PCR mixtures were amplified in a Stratagene Robocycler 40 with the following program: 1 cycle of 95°C for 3 min, 55°C for 1 min, and 72°C for 2 min; 9 cycles of 95°C for 1 min, 55°C for 1 min, and 72°C for 2 min; 10 cycles of 95°C for 1 min, 55°C for 1 min, and 72°C for 3 min; 10 cycles of 95°C for 1 min, 55°C for 1 min, and 72°C for 4 min; 4 cycles of 95°C for 1 min, 55°C for 1 min, and 72°C for 5 min; and a final cycle of 95°C for 1 min, 55°C for 1 min, and 72°C for 10 min. Expected product lengths were approximately 420 bp for V1-V2 and approximately 350 bp for V4-V5. Each RT-PCR was performed at least twice.

Heteroduplex tracking assays.

The HTA probes were generated using the same RT-PCR primers as described above. For each region, two probes from molecular HIV-1 clones were generated. The Ba-L (35) and JR-FL (20) sequences were selected for V1-V2 analysis, while the NL4-3 (2) and YU-2 (25) sequences were selected for V4-V5 analysis because of the availability of the sequences and their length differences in these regions. Initial RT-PCR products from a subject were analyzed with both probes to determine which probe resulted in the best separation of the bands; this probe was used for subsequent HTA analysis for that subject. The V1new and V2 primers were used to amplify the V1-V2 region from the Ba-L and JR-FL envelope clones (from Nathaniel Landau and Irvin Chen, respectively). The V4 and V5 primers were used to amplify the V4-V5 region from the NL4-3 and YU-2 envelope clones (from Malcolm Martin and the NIH AIDS Research and Reference Reagent Program). All of the PCR products were cloned using the Perfectly Blunt cloning kit (pT7Blue-3 for the V1-V2 probes and pT7Blue for the V4-V5 probes, both vectors from Novagen). The downstream EcoRI site was filled in using Klenow fragment of DNA polymerase I for each V1-V2 probe. Probes were end labeled essentially as described previously (28) by digesting the plasmids with EcoRI, adding [α-³⁵S]dATP and Klenow fragment to radioactively label one end of the probe sequence, and digesting the plasmids with PstI (V1-V2) or NheI (V4-V5) to release the probe. Radiolabeled probes were purified using ProbeQuant columns (Amersham).

Heteroduplex reactions contained 8 μl RT-PCR product, 1× annealing buffer (0.1 M NaCl, 10 mM Tris-HCl [pH 7.5], 2 mM EDTA), 0.5 to 1 μl labeled probe, and 0.1 μM V1 or V4 primer. Reaction products were denatured at 95°C for 2 min and then cooled at room temperature for 5 min before loading onto 6% (V1-V2) or 7% (V4-V5) polyacrylamide gels. Gels were run at 20 mA, vacuum dried, and exposed to both X-ray film and a PhosphorImager plate (Molecular Dynamics).

Duplicate RT-PCR products were analyzed separately by HTA to compare the patterns. If the two products from the same RNA sample did not have very similar patterns, a third RT-PCR was run to determine whether one of the first products should be excluded. Two products with very similar patterns were obtained for all RNA samples from which RT-PCR was successful. Missing lanes in the HTA are due to a lack of amplification from that sample.

Cloning of V1-V2 RT-PCR products.

RT-PCR products were cloned into the pT7Blue-3 vector according to the manufacturer's instructions (Perfectly Blunt cloning kit; Novagen). Clones were screened by colony PCR and HTA analysis as described previously (18). Clones of interest were prepared and sequenced by the University of Cincinnati DNA Core Facility.

Data analyses.

The phosphorimaging data were used to quantify the relative abundance of each band within a sample using FragmeNT Analysis software (Molecular Dynamics). Individual bands/variants were identified as previously described (18). Bands were compared for migration distance using plots generated using ImageQuant software (Molecular Dynamics). Plots were aligned using the single-stranded probe band. The relative migration distance and relative abundance of the bands were used to generate matrices representing the HTA patterns. The difference between lanes representing 6-month intervals was calculated as total change using the formula:

where A and B represent two different lanes of an HTA gel containing a total of n distinct bands in the two lanes (27). A_i and B_i are the frequencies of each band i in each lane, where ΣA_i = ΣB_i = 1. A band that is present in only one lane is assigned a frequency of zero in the lane from which it is absent. The absolute value of each difference was taken to allow for both increases and decreases in relative abundance. The sum of all differences is divided by two to reflect the fact that each increase in relative abundance results in a concomitant decrease. Thus, two lanes with no bands in common (reflecting a complete turnover in the virus population) are assigned a total change of 100%. Shannon entropy (S) was calculated for each lane using the following equation:

where P_i is the frequency of each band i in lane P (8). The entropy is normalized as S/ln(n) because maximum entropy is ln(n) and the number of band positions is different among patients. Normalized entropy ranges from zero when there is only one band to one when there is even distribution among all band positions.

A major limitation of the total change and entropy measures previously described is that they do not highlight the emergence of new bands in the context of other band pattern changes and do not account for differences in time intervals between samples, limiting the intervals that could be compared. Thus, we developed a new mathematical algorithm that emphasizes the emergence of new bands, termed the HTA index, or λ, for each lane B in a gel as a comparison of lanes A and B. It is described by the following equation:

and includes terms for the number of bands in lane B that were not present in the preceding lane A (β), the positive difference in band abundance between the two lanes (δ), the amount of variability in lane B (ω), and the difference in the time interval from the standard 6 months (τ). The HTA index and its derivation are described in more detail in the supplemental material. The HTA index algorithm can be downloaded at http://www.hepato-site.net/evoindex/.

Statistical analyses.

To confirm the extent to which each predictor variable affected the overall index for each patient, simulations were generated by varying each predictor by a predetermined amount within reasonable ranges given the observed data. All components of the HTA index were evaluated in this systematic manner. For regression of HTA index simulation results on predictor variables, dummy variables were created for changes in viral load (stable versus changing, increasing versus not increasing, decreasing versus not decreasing) and new band count. Predictor variable contribution to index variability was assessed by the adjusted R². Goodness of fit was assessed by assessing normality and homoskedasticity of residuals. Due to observed heteroskedasticity, robust standard errors were generated as suggested by Davidson and MacKinnon (7a). Manual backwards stepwise multiple linear regression was performed on all predictor variables examined in univariate analysis, first fitting the full model and then eliminating factors that were insignificant or colinear. All analyses were performed in Stata version 8.0 (Stata Corp., College Station, TX). In all analyses, a two-tailed alpha of ≤0.05 was considered statistically significant.

Total change, entropy, and HTA indices were transformed and compared for each region (V1-V2 and V4-V5) in Stata (version 8.0; Stata Corp., College Station, TX) using Student's t test with the Satterthwaite correction for unequal variance. Because total change values (expressed as a decimal) fall between 0 and 1, the logit transformation was used. The HTA index was normalized via log₁₀ transformation.

The outcomes of percent change over time and entropy were assessed for predictor variables, including rapid/slow CD4 loss, CD4 count (continuous) and category (above/below 150 cells/μl), and HIV viral load, via a generalized estimating equation (GEE) model in SAS (version 8.0; SAS Institute, Cary, NC) with an inverse Gaussian distribution and identity link, and exchangeable correlation structure. The GEE method was used because it is highly robust to deviations from normality and correlation structure can be specified. This model is able to parse within-subject variability from between-subject variability with great flexibility, meaning that the model accounts for differences in sampling time or number and differences in magnitude of within-subject correlation (10). Model results are reported as increase/decrease in total change compared to referent group, P value for the difference, and 95% confidence interval (95% CI). Model fit was assessed by comparing naïve and robust standard errors, which did not differ by more than 10%.

Pairwise correlations were used to assess correlations between (i) V1-V2 and V4-V5, (ii) total change and entropy, (iii) HTA index and entropy, and (iv) HTA index and total change.

Nucleotide sequence accession numbers.

The V1-V2 sequences from subjects 1048 and 778 have been deposited in GenBank under the accession numbers DQ885201 to DQ885217.

RESULTS

V1-V2 and V4-V5 undergo frequent changes except during prolonged periods of low CD4 counts.

Fourteen subjects were selected from the San Francisco Men's Health Study on the basis of their CD4 cell profiles over time (Table 1). Six of the subjects had rapid loss of CD4 cells (at least 125 cells/μl/year), and the other eight subjects had slow loss of CD4 cells (less than 100 cells/μl/year). The dates of plasma sampling, AIDS presentation, and death are given in Table 1. The CD4 counts for the individual time points are shown in Fig. 1 (viral loads are shown in Fig. S1 in the supplemental material). Viral RNA was isolated from the sequential plasma samples from each subject, and the V1-V2 and V4-V5 variants were amplified from each RNA sample for analysis by HTA. Both the V1-V2 and V4-V5 regions are highly variable within and between subjects, including multiple point mutations, insertions, and deletions. This high level of diversity, especially in length polymorphisms, allows efficient separation of the variant populations by HTA (9, 18). To quantify the amount of change in bands occurring in each interval, we used phosphorimaging data to determine the relative abundance of each band within a lane. The relative abundance and band comparison data were used to generate matrices representing the HTA patterns, which were used to calculate the total amount of change (as a percentage) over 6-month intervals (shown for all subjects in Fig. 1). Only 6-month intervals (range, 0.32 to 0.78 years) were considered for the total change analysis to standardize for time. Shannon entropy was also calculated for each time point (see Fig. S2 in the supplemental material). The HTA gels used to calculate the total change values and entropy are shown in Fig. S1 in the supplemental material. For all of the subjects (subjects with rapid CD4 loss in Fig. 1A, subjects with slow CD4 loss in Fig. 1B), the total change was often 100% over 6 months, indicating that none of the bands in the first sample was present 6 months later.

TABLE 1.

Summary of CD4 loss, sampling, and endpoint data

Subject	Rate of CD4 cell loss (no./μl/yr)a	No. of plasma samplesb	Date (mo/yr)
			Sample collection	AIDS diagnosis	Death
Rapid decline     in CD4 cells
1048	125	11	3/85-9/92	1/90	10/92
778	182	10	10/84-10/89	4/88	5/90
703	299	5	10/84-5/87	4/87	11/87
641	265	7	9/84-2/88	NAc	9/88
582	182	5	9/84-10/89	6/89	3/90
446	157	4	3/85-2/90	8/88	6/90
Slow decline     in CD4 cells
953	65	12	12/84-7/91	8/93	10/94
822	97	15	10/84-9/92	NA	NA
815	53	11	10/84-2/92	7/94	8/94
576	30	12	4/85-4/92	1/96	NA
150	51	10	2/86-7/92	7/94	2/95
569	47	13	9/84-6/92	2/88	8/96
179	82	14	8/84-2/92	NA	NA
870	1	9	12/84-8/92	6/94	NA

^aOverall rate of CD4⁺ T-cell decline over the study period for each subject.

^bNumber of samples available for analysis.

^cNA, not applicable.

FIG.1.

CD4 counts, total change in HTA pattern, and HTA indices for each of the 14 subjects. (A) Data for subjects with rapid CD4 loss; (B) data for subjects with slow CD4 loss. The top graphs show the CD4 counts for each subject. The middle graphs show the total change values for 6-month intervals in V1-V2 (▵) and V4-V5 (▪). The bottom graphs show the HTA indices for V1-V2 (○) and V4-V5 (⧫). For the total change and HTA index graphs, each point is plotted at the time of the later sample of the interval.

The variant populations in subjects 778 and 1048 showed the lowest levels of total change of all subjects (Fig. 1A). Each subject had at least two consecutive intervals with less than 50% total change in V1-V2 and V4-V5 that started after the CD4 counts for subjects 778 and 1048 had been below 150/μl for 6 and 18 months, respectively. Interestingly, these sustained low CD4 counts coincided with maintenance of both V1-V2 and V4-V5 HTA bands (see Fig. S1A in the supplemental material). Identical nucleotide sequences were found for one of the bands from samples from subject 1048 (zero to four differences among three sequences from each sample), and nearly identical nucleotide sequences were found for one of the bands from samples from subject 778 (one to three nucleotide differences among three and two sequences, respectively) (see Fig. S3 in the supplemental material).

Because of the variable number of intervals per subject, we examined the relationship between the CD4 count to the degree of band changes in V1-V2 and V4-V5 over time using GEE models including total change values for all 6-month intervals and entropy values for all samples. The models indicated that the CD4 count was predictive of total change, with each increase in CD4 count of 10 cells/μl predicting 0.4% increases in both total V1-V2 and total V4-V5 changes (for V1-V2, 95% CI of 0.1 to 0.7, P = 0.006; for V4-V5, 95% CI of 0.2 to 0.6, P < 0.0001). However, when intervals with CD4 counts less than 150/μl were excluded from the data set, CD4 count was no longer predictive of total change (P = 0.80 for V1-V2 and P = 0.21 for V4-V5). Other variables including viral load and presence of X4 variants (J. A. E. Nelson, unpublished data) did not show any significant correlation with total change for either region. GEE analysis did not demonstrate a significant association between CD4 count and changes in entropy over time for either V1-V2 or V4-V5 regions (P = 0.30 and 0.23, respectively). Therefore, the low CD4 count intervals (mostly from subjects 778 and 1048) were responsible for the predictive value of CD4 count on total change; there was no difference in the amount of change over the course of infection in the other subjects.

A novel index of viral evolution reveals the emergence of new variants.

A limitation of the calculation of total change in the variant populations using the data from this cohort was that the length of time between samples was not considered. Although the San Francisco Men's Health Study involved 6-month visits, in many cases samples were not available such that there were time intervals significantly longer than 6 months. In addition, neither total change in HTA pattern or entropy highlighted the emergence of new bands as an indicator of env evolution. Therefore, we developed a new algorithm, the HTA index, that combines quantitative aspects of the total change calculation and the band distribution of entropy. Terms for the emergence of new bands and a time interval correction were also added (see supplemental material for a full discussion). Importantly, the HTA index is not dependent on the total number of samples per subject, and thus, the index is valid whenever there are two chronologically adjacent lanes. Simulations of the HTA index in which single variables were changed indicated that the baseline band count (in the first lane of adjacent lanes) and the number of new bands in the second lane had significant positive effects on the HTA index, while the distribution of the bands within the second lane and the length of the time interval had significant negative effects on the HTA index (Table 2). Other variables including viral load, lane differences, and differences in band count between adjacent lanes were not significant or were colinear with either baseline band count or number of new bands. This analysis demonstrated that high numbers of new bands led to high HTA indices and that long time intervals and poor band distribution decreased the HTA index despite high numbers of new bands. A high baseline band number also had a small positive effect on the HTA index.

TABLE 2.

Multiple regression model of HTA index simulation results^a

Factor	Beta coefficient	Robust SE	Robust 95% CIb		P value
			LL	UL
No. of baseline bands	1.55	0.13	1.29	1.81	<0.0001
Band distribution	−97.16	4.00	−105.00	−89.32	<0.0001
Length of time interval	−138.62	5.42	−149.25	−127.99	<0.0001
No. of new bands	12.41	0.30	11.81	13.01	<0.0001

^aAdjusted R² = 0.49.

^bThe lower limit (LL) and upper limit (UL) of the 95% confidence limits are given.

The HTA indices are shown in Fig. 1 for both V1-V2 and V4-V5 over time for all of the subjects. Higher HTA indices indicate the appearance of many new bands (>10) over a shorter time interval (<1 year), while lower HTA indices typically reflect few new bands or a time interval of more than 1 year. Overall, there was much more fluctuation in the HTA index than was seen with total change, because high levels of total change do not require the emergence of new bands and the fluctuations in the HTA index coincided with fluctuations in total change at times (subjects 815 and 150, for example). The two subjects with V1-V2 and V4-V5 stabilization, subjects 1048 and 778, had low HTA indices when their CD4 counts were low, and they had low HTA indices overall (all below 20). Subjects 446 and 870 also had low HTA indices, but each of these subjects had only one 6-month interval; their longer intervals had low HTA indices because of the length of time. Subjects 953, 576, 569, and 150 had the highest HTA indices (>50), indicating the highest levels of emergence of new bands (12 to 17 new bands) and largest numbers of cocirculating variant populations (12 to 17 bands). The highest peaks in HTA index were not associated with either high or low CD4 counts but were generally flanked by low HTA indices (Fig. 1B). These elevated HTA indices reflected high numbers of new bands emerging over 6-month intervals. In some of the subjects, there was a correlation in the fluctuation of the HTA indices for V1-V2 and V4-V5, such as in subjects 1048 and 778 with rapid CD4 decline (Fig. 1A), and subject 576 with slow CD4 decline (Fig. 1B). Other subjects had differences in peaks between the two regions (see rapid CD4 decline subjects 641 and 582 and slow CD4 decline subjects 815 and 150), indicating that there is likely independent evolution of the two domains most of the time.

The total change and HTA index values did not show the same trend in some cases, i.e., the total change value increased while the HTA index stayed low. Examples include subject 1048 (V1-V2 and V4-V5 at 7.5 years), subject 778 (V4-V5 at 4.5 years), subject 179 (V1-V2 at 1.3 years), and 822 (V1-V2 at 3.2 years) (Fig. 1). These examples illustrate the difference between these two measures of sequence evolution, in that there can be a great deal of change in the HTA pattern due to changes in band intensity or loss of bands, but the HTA index remains low because there are few if any new bands emerging. We are most interested in new bands emerging because they are due to sequence evolution and virus escape. These differences between the two measures were the exceptions, however, because the HTA index correlated strongly and significantly with total change for both regions (V1-V2, r = 0.48; V4-V5, r = 0.41; P < 0.001 in both cases). Similar correlations were observed between the HTA index and entropy (V1-V2, r = 0.54; V4-V5, r = 0.60; P < 0.001 in both cases). This suggests that the trends observed in our novel index follow the trends seen in more-standard, yet less-sensitive, measures of change.

V1-V2 and V4-V5 evolution occurs at lower rates in subjects with rapid loss of CD4 cells.

To determine whether there were differences between the subjects with slow and rapid loss of CD4 cells, we compared mean total change and mean HTA index values between these two groups (Fig. 2). The mean total change for V4-V5 was significantly higher for subjects with slow CD4 loss, but the mean total change for V1-V2 was not different between the two groups (Fig. 2A). This was surprising because we observed the same stabilization of bands in subjects 778 and 1048 in both V1-V2 and V4-V5. However, the mean HTA index was significantly different between the two groups for both env regions, with higher mean HTA indices in the subjects with slow CD4 loss (Fig. 2B). Therefore, the subjects with rapid CD4 loss had lower new-band emergence (measured by HTA index) than the subjects with slow CD4 loss but equivalent changes in the relative abundance of existing bands (measured by both methods). Since long intervals (>0.8 year) could have skewed this analysis with excess low HTA indices, the long interval HTA index values were removed and the data were reanalyzed. The difference between the two groups of subjects was still significant for both V1-V2 and V4-V5 (not shown).

FIG. 2.

Box plots comparing total change (A) and HTA index (B) for the V1-V2 and V4-V5 regions in subjects with rapid or slow loss of CD4 cells. The boxes represent the 25th to 75th percentile values, and the lines within the boxes represent the median values. The whiskers represent the high and low values, with outlier values represented by circles. The means are represented by squares. All P values are two-tailed with Satterthwaite's approximation to correct degrees of freedom for unequal variances; only P values that were significant are shown. Because percent change values (expressed as a decimal) are bounded between zero and one, they were logit transformed prior to analysis. Because HTA index could be normalized via log transformation, t tests were conducted on the log-transformed means.

Levels of V1-V2 and V4-V5 variant turnover are correlated over time.

Overall, most of the intervals had similar levels of total change for V1-V2 and V4-V5. The total changes in V1-V2 and V4-V5 within the same subject were significantly correlated with one another over time (Fig. 3A, r = 0.55, P < 0.0001). This suggests that the viral population constantly turns over for both of these regions and that the two regions stabilize concurrently. Moreover, the HTA indices from V1-V2 and V4-V5 within the same subject were significantly correlated over time (Fig. 3B, r = 0.39, P = 0.001), reflecting the high levels of evolution in both regions until stabilization. To better define the high levels of change in the V1-V2 and V4-V5 variant populations, we compared them to a separate analysis of the V3 region of env in these same subjects; V3 tends to be more conserved over time, especially while the virus is only using CCR5 for entry (30) and sits between V1-V2 and V4-V5 in the env sequence. The V3 HTA analysis showed much less change in variant populations over time (mean total changes of 31.6% for V3, 79.9% for V1-V2, and 83.7% for V4-V5; mean HTA indices of 0.71 for V3, 13.2 for V1-V2, and 12.0 for V4-V5), with the appearance of new V3 bands associated with the emergence of variants that use CXCR4 for entry (J. A. E. Nelson, unpublished data). Comparisons of total change and HTA index values showed no significant correlation between either V1-V2 and V3 or V4-V5 and V3. These data suggest that at least three separate selection pressures were at work simultaneously against these three (V1-V2, V3, and V4-V5) regions during chronic infection.

FIG. 3.

Total change and HTA index comparisons of V1-V2 and V4-V5 regions for each interval. Total change (A) and HTA index (B) for the two regions were plotted for each 6-month interval (A) or every interval (B). Black lines of correlation are shown with shaded regions of 95% confidence intervals.

DISCUSSION

In the present study, we have used HTA analyses specific for the V1-V2 and V4-V5 regions of the env gene to determine the extent of viral variant turnover among HIV-infected men enrolled in the San Francisco Men's Health Study. Our analyses of semiannual samples from eight men with slow loss of CD4 cells and six men with rapid loss of CD4 cells showed that the majority of viral variants present at a given time point were ultimately lost within the subsequent 6 months of infection and replaced by previously absent viral variants. This high rate of turnover occurred over a wide range of CD4 counts and occurred in all subjects studied.

Analysis of total change in HTA pattern revealed significantly more change in V4-V5 HTA patterns among subjects with slow CD4 loss, while analysis with a new HTA index algorithm, with its emphasis on the emergence of new bands, showed significantly higher indices for both V1-V2 and V4-V5 regions in the subjects with slow CD4 loss. Statistical analyses showed that CD4 count was correlated with total change and HTA index values only when low CD4 counts were included in the data set, indicating that low CD4 counts under 150/μl were associated with reduced changes in the HTA pattern and reduced emergence of new variants. Sustained low CD4 counts in two subjects with rapid CD4 loss were associated with stabilization of V1-V2 and V4-V5 variant populations as reflected by both low total change and low HTA index values.

The present study is the first to use HTA to follow the evolution in two separate regions of env during the advanced stages of chronic infection. Most previous HTA and sequencing studies have focused on only one region of env (8, 18, 24, 36, 37). A recent study by Frost et al. (13) analyzed more than one region of env by bulk sequencing full-length env in samples from acute infection and 1 year later. The sequence changes were compared to neutralization activity against autologous virus measured over the first 3 years of infection and found that multiple amino acid substitutions were associated with neutralization escape. Our study complements the findings of Frost et al. by showing that sequence turnover continues until late in infection. Thus, immune selection against HIV appears to continue until late in infection.

Our observation that both V1-V2 and V4-V5 populations stabilized at sustained low CD4 counts is evidence that immune selection against Env wanes at low CD4. The rapid turnover of env variant populations when CD4 counts are above 150/μl suggests that the immune system continues to be able to develop new responses against escape variants. Previous studies have shown evidence of both selection and genetic drift in driving HIV evolution (1, 13, 18, 24, 39), but the stabilization of env variants in two subjects in the present study argues against genetic drift as a significant force in env evolution and in favor of continually emerging immune responses that decline at persistently low CD4 counts. Our data also suggest, however, that these continually emerging immune responses against Env are largely ineffective in reducing viral burden, since the virus is constantly escaping as the viral loads remain high. It remains to be determined whether the selective pressure on the env gene plays a role in the equilibrium between HIV-1 and its human host and whether this selective pressure can be directed to be more effective in controlling the virus.