Genetic Diversity, Compartmentalization, and Age of HIV Proviruses Persisting in CD4⁺ T Cell Subsets during Long-Term Combination Antiretroviral Therapy

The main barrier to HIV cure is the ability of a genetically diverse pool of proviruses, integrated into the genomes of infected CD4⁺ T cells, to persist despite long-term suppressive combination antiretroviral therapy (cART). CD4⁺ T cells, however, constitute a heterogeneous population due to their maturation across a developmental continuum, and the genetic “landscapes” of latent proviruses archived within them remains incompletely understood. We applied phylogenetic techniques, largely novel to HIV persistence research, to reconstruct within-host HIV evolutionary history and characterize proviral diversity in CD4⁺ T cell subsets in five individuals on long-term cART. Participants varied widely in terms of proviral burden, genetic diversity, and age distribution between CD4⁺ T cell subsets, revealing that proviral landscapes can differ between individuals and between infected cell types within an individual. Our findings expose each within-host latent reservoir as unique in its genetic complexity and support personalized strategies for HIV eradication.

ABSTRACT

The HIV reservoir, which comprises diverse proviruses integrated into the genomes of infected, primarily CD4⁺ T cells, is the main barrier to developing an effective HIV cure. Our understanding of the genetics and dynamics of proviruses persisting within distinct CD4⁺ T cell subsets, however, remains incomplete. Using single-genome amplification, we characterized subgenomic proviral sequences (nef region) from naive, central memory, transitional memory, and effector memory CD4⁺ T cells from five HIV-infected individuals on long-term combination antiretroviral therapy (cART) and compared these to HIV RNA sequences isolated longitudinally from archived plasma collected prior to cART initiation, yielding HIV data sets spanning a median of 19.5 years (range, 10 to 20 years) per participant. We inferred a distribution of within-host phylogenies for each participant, from which we characterized proviral ages, phylogenetic diversity, and genetic compartmentalization between CD4⁺ T cell subsets. While three of five participants exhibited some degree of proviral compartmentalization between CD4⁺ T cell subsets, combined analyses revealed no evidence that any particular CD4⁺ T cell subset harbored the longest persisting, most genetically diverse, and/or most genetically distinctive HIV reservoir. In one participant, diverse proviruses archived within naive T cells were significantly younger than those in memory subsets, while for three other participants we observed no significant differences in proviral ages between subsets. In one participant, “old” proviruses were recovered from all subsets, and included one sequence, estimated to be 21.5 years old, that dominated (>93%) their effector memory subset. HIV eradication strategies will need to overcome within- and between-host genetic complexity of proviral landscapes, possibly via personalized approaches.

IMPORTANCE The main barrier to HIV cure is the ability of a genetically diverse pool of proviruses, integrated into the genomes of infected CD4⁺ T cells, to persist despite long-term suppressive combination antiretroviral therapy (cART). CD4⁺ T cells, however, constitute a heterogeneous population due to their maturation across a developmental continuum, and the genetic “landscapes” of latent proviruses archived within them remains incompletely understood. We applied phylogenetic techniques, largely novel to HIV persistence research, to reconstruct within-host HIV evolutionary history and characterize proviral diversity in CD4⁺ T cell subsets in five individuals on long-term cART. Participants varied widely in terms of proviral burden, genetic diversity, and age distribution between CD4⁺ T cell subsets, revealing that proviral landscapes can differ between individuals and between infected cell types within an individual. Our findings expose each within-host latent reservoir as unique in its genetic complexity and support personalized strategies for HIV eradication.

INTRODUCTION

Although combination antiretroviral therapy (cART) suppresses human immunodeficiency virus 1 (HIV) replication (1, 2), it is not curative and must be maintained for life (3). This is due to a small pool of long-lived primarily CD4⁺ T cells that harbor integrated HIV in a transcriptionally reduced state and persist despite cART but that can be reactivated at any time to produce infectious virus (4–8). In general, the HIV reservoirs of individuals who initiated cART during chronic infection are genetically diverse (9–15), which is to be expected, given that seeding of the reservoir begins immediately following infection (16–19) and continues as long as HIV is actively replicating in vivo (20, 21). Recent longitudinal studies confirm this: proviral sequences dating as far back as transmission are present in many individuals’ reservoirs (20–22) though some are enriched for proviruses seeded around the time of cART (20, 22). Populations of cells harboring identical proviruses or identical integration sites also feature prominently in the reservoir, indicating that clonal expansion of latently infected cells also drives HIV persistence (15, 23–29).

HIV eradication will thus require an in-depth understanding of latent HIV genetic composition and persistence in CD4⁺ T cells, but this is complicated by the fact that CD4⁺ T cells mature along a program of development and thus constitute a heterogeneous population (30, 31). Upon encountering their cognate antigen, naive T (T_N) cells develop into effector and memory cell subsets which include, from least to most differentiated, stem-cell-like memory (T_SCM), central memory (T_CM), transitional memory (T_TM), effector memory (T_EM), and finally terminally differentiated (T_TD) cells (31). Though HIV DNA is reproducibly detected in all of these subsets during long-term cART (13, 32–38), it has been hypothesized that less differentiated memory T cell subsets may represent the most durable sites for long-term HIV persistence in peripheral blood (32, 34, 35, 37). This is an intuitive notion, given that the longevity of CD4⁺ T cell subsets (30) and the half-life of proviral DNA in these cells (34, 39) decrease with differentiation, but studies analyzing proviral sequences within CD4⁺ T cell subsets are limited and have yielded somewhat conflicting observations. Buzon et al. observed that proviruses isolated from less differentiated, longer-lived memory CD4⁺ T cells, in particular, T_SCM and T_CM cells, were phylogenetically most closely related to early pre-cART plasma sequences, suggesting that HIV strains circulating in early infection were more likely to persist in these cell subsets (34). Chomont et al. hypothesized that, due to their differential survival and proliferation dynamics, T_CM and T_TM cells may come to define distinct HIV reservoirs after long-term cART, with increased turnover of shorter-lived T_TM cells being associated with reduced within-subset proviral genetic distances compared to that of the longer-lived T_CM cells (32). However, counter to the notion that highly differentiated CD4⁺ T cell subsets represent less stable reservoirs, Imamichi et al. documented a proviral sequence that persisted for 17 years solely within T_EM cells, indicating that these subsets can harbor latent HIV for extended periods through proliferation (40). Only one study to our knowledge has applied formal tests for genetic compartmentalization to proviruses archived within CD4⁺ T cell subsets (29), while none have applied methods to infer latent HIV sequence age distributions within these subsets. Furthermore, T_N cells, despite generally harboring lower proviral burdens than memory T cell subsets (32, 33, 36), may nevertheless represent a key HIV reservoir (37, 38), but knowledge of proviral diversity within this subset remains limited (13).

If the longest lived, least differentiated CD4⁺ T cell subsets constitute the most stable and privileged HIV reservoirs, one could hypothesize that, during long-term cART, the oldest within-host latent HIV sequences would tend to be found therein. Our recent development of a phylogenetic method to infer within-host latent HIV sequence ages allows this hypothesis to be tested (21). One could further hypothesize that, after years of cART, the most evolutionarily distinct within-host HIV lineages and/or the greatest overall proviral diversity will be found within the longest lived, least differentiated CD4⁺ T cell subsets as more rapid turnover in the shorter lived, more differentiated subsets may lead to more frequent elimination of unique proviral lineages therein. To this end, we investigated proviral diversity, genetic compartmentalization, and latent HIV sequence age distribution in CD4⁺ T cell subsets during long-term cART and interpreted this information in light of HIV’s pre-cART within-host evolutionary history in five HIV-infected individuals with long-term viremia suppression.

RESULTS

Proviral burden in CD4⁺ T cell subsets.

Five HIV-infected adults who initiated cART during chronic infection and for whom longitudinal archived pre-cART plasma samples were available were recruited for study (Table 1). At enrollment, participants had been receiving cART for a median of 9 years and had a median of 8 (interquartile range [IQR], 5 to 14) archived pre-cART plasma samples available for study. Peripheral blood mononuclear cells (PBMCs) were isolated at enrollment and sorted into CD4⁺ naive (T_N), central memory (T_CM), transitional memory (T_TM), and effector memory (T_EM) cell subsets based on their expression of CD45RA, CCR7, and CD27 (Fig. 1 and 2A) and assessed for purity as described previously in Chomont et al. (32) (see also Materials and Methods). CD4⁺ T cell subset frequencies varied markedly between participants: T_CM cells, for example, constituted from 14% to 34% (median, 32%; IQR, 22 to 33%) of total CD4⁺ T cells (Fig. 2B). Proviral DNA levels in CD4⁺ T cell subsets also varied. Nevertheless, in 4 of 5 participants, T_N cells harbored the lowest absolute proviral burdens (median, 890 HIV copies/million cells; IQR, 383 to 1,573 HIV copies/million cells) (Fig. 2C) and contributed the least to the total proviral burden in peripheral blood CD4⁺ T cells (median, 11%; IQR, 3 to 32%) (Fig. 2D), while memory subsets (T_CM, T_TM, and T_EM cells together) contributed the most to the total proviral burden in CD4⁺ T cells (median, 89%; IQR, 68 to 97%). This observation is consistent with studies identifying T_N cells as a minor yet potentially important HIV reservoir (37, 38).

TABLE 1.

Participant and HIV sequence data set details

Participantno.	Time infectedpre-cART (yr)	No. of pre-cARTplasma timepoints	Time oncART (yr)	Total no. of sequences (no. kept) by sourcea
				Pre-cARTplasma	TN cells	TCM cells	TTM cells	TEM cells
1	10	14	9	102 (93)	31 (29)	26 (24)	26 (24)	27 (16)
2	11	17	9	162 (130)	20 (18)	13 (13)	18 (15)	15 (2)
3	5	8	8	122 (113)	16 (13)	19 (15)	38 (35)	18 (15)
4	2	5	9	72 (60)	16 (16)	20 (16)	26 (16)	18 (8)
5	14	2	16	41 (38)	20 (20)	16 (16)	24 (11)	17 (10)

^aKept sequences refer to distinct sequences that were free from hypermutation, other defects, and within-host recombination.

FIG 1.

Gating strategy for sorting of CD4⁺ T cell subsets. Representative flow cytometry plot illustrating how participant CD4⁺ T cells were isolated after staining with the indicated antibodies. CD4 T cell subsets were defined as follows: T_N, CD45RA⁺ CCR7⁺ CD27⁺; T_CM, CD45RA⁻ CCR7⁺ CD27⁺; T_TM, CD45RA⁻ CCR7⁻ CD27⁺; T_EM, CD45RA⁻ CCR7⁻ CD27⁻.

FIG 2.

Frequency and latent HIV reservoir burden in CD4⁺ T cell subsets. (A) Schematic of CD4⁺ T cell differentiation, with corresponding cell surface markers below. All subsequent figures use this color scheme to denote specific CD4⁺ T cell subsets. (B) Frequency of CD4⁺ T cells within each subset. The “other” category denotes CD4⁺ T cells that express other combinations of the markers listed in panel A that are rare and have no reported function (80). (C) Proviral DNA burden, measured as the number of HIV Gag copies/million cells, in each CD4⁺ T cell subset. Error bars represent 95% Poisson confidence intervals calculated from merged replicate wells. (D) Proportion of the total proviral burden harbored by each CD4⁺ T cell subset.

Within-host HIV sequence characterization.

We next characterized HIV RNA sequences from longitudinal archived plasma samples collected prior to cART initiation, as well as proviral sequences from T_N, T_CM, T_TM, and T_EM cells isolated during long-term cART, using subgenomic single-genome amplification and sequencing. The nef gene was selected based on its relatively high within-host diversity, its richness in phylogenetic signal, and its representativeness of within-host HIV evolution and diversity relative to the rest of the viral genome (21); nef also represents the gene most likely to be intact within proviruses sampled during long-term cART (15, 41). All further references to “sequences” in this text refer to nef sequences. The numbers of HIV RNA and proviral DNA sequences collected from each participant (total and intact/distinct) are reported in Table 1. Of note, we observed identical HIV sequences more frequently among the more differentiated CD4⁺ T cell subsets though this did not achieve statistical significance (Fig. 3). Participant 2 was remarkable in that this participant’s T_EM cells harbored the highest proviral burden of all CD4⁺ T cell subsets, but 93% (14/15) of the intact proviral sequences recovered from T_EM cells were identical, such that only two distinct sequences were isolated from this subset (Table 1). Together, these observations are consistent with reports that the more differentiated latently infected CD4⁺ T cell subsets tend to be more clonally expanded (13, 29).

FIG 3.

Sequence distinctness in CD4⁺ T cell subsets. Data points denote the frequency of distinct (unique) proviral sequences per subset per participant. Frequencies across subsets were compared using the Friedman rank sum test. CM, central memory; TM, transitional memory; EM, effector memory.

Proviral ages and genetic compartmentalization in CD4⁺ T cell subsets.

Our main goal was to characterize within-host proviral landscapes within CD4⁺ T cell subsets during long-term cART and to interpret this information in the context of HIV’s within-host pre-cART evolutionary history (Fig. 4; see also Fig. 8). For each participant, we used Bayesian methods via MrBayes (42) to infer a random sample of between 15,000 and 30,000 within-host phylogenetic trees from alignments of pre-cART plasma HIV RNA and post-cART proviral sequences and used these trees to investigate the diversity, genetic compartmentalization, and age distribution of proviruses persisting in CD4⁺ T cell subsets. Conditioning analyses across a distribution of trees mitigates the inherent uncertainty in reconstructing within-host HIV evolutionary relationships and is particularly useful for phylogenetic tests (e.g., for genetic compartmentalization [43] and evolutionary distinctiveness [44]) that are sensitive to tree topology.

FIG 4.

Participant 1. (A) Plasma viral load and sampling history. Yellow shading denotes periods on cART. (B) The phylogeny with the highest likelihood derived from Bayesian inference, rooted using root-to-tip regression. (C) Highlighter plot showing amino acid differences with respect to the consensus of the sequences collected at the earliest plasma HIV RNA sampling time point. (D) Linear model (dashed gray diagonal) relating plasma HIV RNA collection dates to their respective distances from the root of the phylogeny shown in panel B. Thin gray lines show the phylogenetic relationships between the sequences. HIV RNA nef evolutionary rate (ER), expressed in estimated nucleotide substitutions per site per day, is shown in the bottom right corner. (E) The 95% highest posterior density (HPD) intervals of the integration date estimates of the proviral sequences, derived from all phylogenies generated by the Bayesian analysis. The points represent mean values. The distribution of mean values across the CD4⁺ T cell subsets is compared using a Kruskal-Wallis test (P value at bottom right). (F) P values of proviral genetic compartmentalization across the CD4⁺ T cell subsets, derived from AMOVA and Slatkin-Maddison (where the latter values are derived from all phylogenies generated by the Bayesian analysis). Here, the circle represents the overall mean P value, the x represents the mean weighted by the likelihood of the phylogeny tested, and the bars represent the 95% HPD interval.

FIG 8.

Participant 5. (A) Plasma viral load and sampling history. Yellow shading indicates cART, and pink shading denotes nonsuppressive dual therapy (also in panel D). (B) The phylogeny with the highest likelihood derived from Bayesian inference, rooted using the HXB2 reference strain as an outgroup. (C) Highlighter plot showing amino acid differences with respect to the consensus of the sequences collected at the earliest plasma HIV RNA sampling time point. (D) Linear model (dashed gray diagonal) relating plasma HIV RNA collection dates to their respective distances from the root of the phylogeny shown in panel B. Gray lines show the phylogenetic relationships of the sequences. HIV RNA nef evolutionary rate (ER) could not be determined for this participant due to insufficient molecular clock signal. (E) The 95% highest posterior density (HPD) intervals of root-to-tip patristic distances, expressed as the number of estimated nucleotide substitutions/site, derived from all phylogenies generated by the Bayesian analysis. The points represent mean values. The distribution of mean values across the CD4⁺ T cell subsets is compared using the Kruskal-Wallis test (P value at bottom right). (F) P values of proviral genetic compartmentalization across the CD4⁺ T cell subsets, as described in the legend of Fig. 4F.

Participant 1.

Participant 1, for whom a limited number of proviral sequences were previously characterized from PBMCs (21), was diagnosed with HIV in August 1996 and did not initiate suppressive cART until August 2006 (Fig. 4A). Using single-genome amplification, we isolated 102 plasma HIV RNA sequences over an ∼10-year period pre-cART and 110 proviral sequences from CD4⁺ T cell subsets collected after ∼10 years of cART (Table 1 and Fig. 4A). Optimal rooting of phylogenies inferred from these sequences, where the root represents the most recent common ancestor of the within-host data set, revealed strong molecular clock signal, as demonstrated by the robust linear relationship between the divergence from the root and the collection date of longitudinally sampled pre-cART plasma sequences (representative phylogeny, sequence diversity plot, and linear model are shown in Fig. 4B to D). Averaged over all within-host phylogenies, the mean pre-cART plasma HIV RNA nef evolutionary rate was estimated as 4.24 × 10⁻⁵ (95% highest posterior density [HPD], 2.64 × 10⁻⁵ to 5.94 × 10⁻⁵) substitutions per nucleotide site per day for this participant. Proviral sequences sampled from CD4⁺ T cell subsets after ∼10 years on cART were interspersed throughout the tree (Fig. 4B), consistent with the notion that HIV sequences are continually seeded into the reservoir during uncontrolled infection and can persist for a decade or longer (20, 21, 45). Identical proviral sequences were also observed, including seven in T_EM cells (Fig. 4B), consistent with clonal expansion of latently HIV-infected cells (15, 23–29).

Our ability to estimate a within-host rate of HIV evolution from longitudinal pre-cART plasma HIV RNA sequences allows us to estimate the integration date of a given proviral sequence from its root-to-tip genetic distance (21). Moreover, our reconstruction of thousands of trees per participant allows us to compute a 95% highest posterior density (HPD) around this date estimate (21; see also Materials and Methods) (Fig. 4E). Distinct proviral sequences within a given CD4⁺ T cell subset differed widely in terms of their estimated integration dates. The oldest provirus isolated from participant 1’s T_EM cells, for example, was estimated to have integrated in 1998 (i.e., it was almost 18 years old at time of sampling) while the youngest was estimated to have integrated in 2007. Furthermore, the estimated integration date distributions of distinct proviruses differed significantly between CD4⁺ T cell subsets in this participant (Kruskal-Wallis test, P = 0.002) (Fig. 4E). However, in contrast to our hypothesis that the longer lived, less differentiated memory T cell subsets would harbor the oldest proviruses, the three memory subsets did not differ in terms of their proviral age distributions (P > 0.2 for all pairwise comparisons), and T_N cells harbored the youngest proviruses on average (P < 0.05 in all pairwise comparisons). While some proviral integration date point estimates were after cART initiation, in all cases the 95% HPD estimates extended into the pre-cART period (Fig. 4E), and it is also notable that this participant experienced a major viremic episode at 1 year post-cART, which could have led to reservoir reseeding (Fig. 4A). These observations therefore do not constitute conclusive evidence of ongoing HIV replication during cART.

We next investigated whether distinct proviral sequences within CD4⁺ T cell subsets displayed population structure or genetic compartmentalization. In actively replicating viral populations, compartmentalization often arises due to restricted gene flow between compartments, but in the context of the HIV reservoir, it may arise due to differential seeding of within-host HIV strains into, or differential subsequent persistence within, distinct CD4⁺ T cell subsets. Differential clonal expansion dynamics between latently infected CD4⁺ T cell subsets could also lead to the detection of population structure if identical sequences are retained in the analysis (46, 47); to avoid this, we analyzed distinct sequences per subset only. As recommended (48), we applied two tests capable of detecting population structure across the four subsets (T_N, T_CM, T_TM, and T_EM cells): analysis of molecular variance (AMOVA) (49), a genetic-distance-based test, and the Slatkin-Maddison (SM) test, a tree-based test (43). Both of these tests have been applied to within-host HIV data sets (47, 48) including the latent reservoir (46, 48, 50). AMOVA yielded a P value of 0.004, while application of the SM test to all 30,000 within-host phylogenies yielded an overall mean P value of 0.057, a weighted mean P value of 0.011 (where the latter takes the tree likelihoods into account), and a 95% HPD of 0.000 to 0.182 (Fig. 4F). These results indicate that proviral sequences archived within participant 1’s CD4⁺ T cell subsets exhibit modest yet statistically significant compartmentalization, which is consistent with our detection of significant differences in proviral age distributions between subsets.

Participant 2.

Participant 2, for whom a small number of proviruses isolated from PBMCs were also previously characterized (21), was diagnosed with HIV in April 1995, initiated incompletely suppressive dual ART in July 2000, and initiated suppressive cART in August 2006 (Fig. 5A). We isolated 162 plasma HIV RNA sequences over an ∼9.5-year period pre-cART and 66 proviral sequences from different CD4⁺ T cell subsets collected ∼10 years post-cART (Table 1). Optimal rooting of participant 2’s phylogenies (Fig. 5B, representative tree, and Fig. 5C, diversity plot) yielded a mean pre-cART plasma HIV RNA nef evolutionary rate estimate of 1.55 × 10⁻⁵ (95% HPD, 8.83 × 10⁻⁶ to 2.29 × 10⁻⁵) substitutions per nucleotide site per day (Fig. 5D shows representative data). Proviruses isolated from CD4⁺ T cell subsets were interspersed throughout the phylogeny and included sequences from T_N, T_CM, and T_EM cells that branched close to the root, consistent with the notion that, even after years of cART, multiple CD4⁺ T cell subsets still harbor proviruses archived near the time of HIV transmission (16). Notably, only two distinct proviral sequences were isolated from this participant’s T_EM cells, both of which fell within a clade close to the tree root, yielding proviral integration date estimates of 1995 and 1996, respectively (Fig. 5E). The isolation of the former sequence 14 times supports clonal expansion as a major driver of HIV persistence (23–26, 51, 52) where clones may arise very early in infection (45). In contrast, HIV sequences isolated from the other T cell subsets exhibited broader age distributions (e.g., proviruses isolated from T_N cells dated to between 1994 and 2006). Overall, T_EM cells harbored the oldest proviruses in this participant, but this was not statistically significant (Kruskal-Wallis test, P = 0.163) (Fig. 5E), in part because we recovered only two distinct sequences from T_EM cells. Despite this, both genetic distance and tree-based tests indicated that distinct proviruses isolated from participant 2’s CD4⁺ T cell subsets exhibited modest genetic compartmentalization: AMOVA yielded a P value of <0.001, while the weighted average P value from SM tests was 0.037 (95% HPD, 0.00 to 0.202) (Fig. 5F).

FIG 5.

Participant 2. The panels are as described in the legend of Fig. 4. Pink shading in panels A, D, and E denotes a period of nonsuppressive dual therapy.

Participant 3.

Participant 3 was diagnosed with HIV in 2002 and initiated suppressive cART in May 2007 (Fig. 6A). We isolated 122 plasma HIV RNA sequences over an ∼5-year period pre-cART and 91 proviral sequences from CD4⁺ T cell subsets collected ∼9 years post-cART (Table 1). Optimal rooting of participant 3’s phylogenies (Fig. 6B, representative tree, and Fig. 6C, diversity plot) yielded a mean pre-cART plasma HIV RNA nef evolutionary rate estimate of 5.33 × 10⁻⁵ (95% HPD, 3.18 × 10⁻⁵ to 7.85 × 10⁻⁵) substitutions per nucleotide site per day (representative data are shown in Fig. 6D). Proviral sequences were again interspersed throughout the tree, where the four proviruses closest to the root were isolated from T_TM and T_EM cells and dated to various times in late 2002 and 2003. This participant harbored an identical HIV sequence that was isolated from T_EM cells and repeatedly from T_N and T_CM cells, suggesting that latently infected cells can both differentiate and clonally expand to sustain the HIV reservoir. Overall, the proviruses isolated from participant 3’s CD4⁺ T cell subsets did not significantly differ in terms of their estimated integration dates (Kruskal-Wallis test, P = 0.910) (Fig. 6E), and support for genetic compartmentalization was not consistent between tests: whereas the weighted average P value for SM was 0.015 (HPD, 0.000 to 0.217), AMOVA yielded a P value of 0.151 (Fig. 6F). Taken together, our findings indicated that while participant 3 harbored diverse proviruses, there was no strong evidence that these differed in genetic composition or age between CD4⁺ T cell subsets.

FIG 6.

Participant 3. The panels are as described in the legend of Fig. 4.

Participant 4.

Participant 4 was diagnosed with HIV in 2005 and initiated suppressive cART in November 2006 (Fig. 7A). We isolated 72 plasma HIV RNA sequences at 5 time points between September 2005 and November 2006, prior to cART initiation, and 80 proviral sequences from CD4⁺ T cell subsets collected ∼9.5 years post-cART (Table 1). Consistent with cART initiation in relatively early infection, within-host HIV diversity was relatively limited (Fig. 7B, representative tree, and Fig. 7C, diversity plot). Despite this, there was sufficient molecular clock signal in the data to infer a mean pre-cART plasma HIV RNA nef evolutionary rate of 4.30 × 10⁻⁵ (95% HPD, 1.44 × 10⁻⁵ to 7.67 × 10⁻⁵) substitutions per nucleotide site per day (representative data are shown in Fig. 7D). Proviral sequences were again interspersed throughout the tree. Notably, one HIV sequence, which branched close to the root, was repeatedly isolated from all memory CD4⁺ T cell subsets (Fig. 7B). Overall, proviruses isolated from the different CD4⁺ subsets did not differ from one another in terms of their estimated integration dates (Kruskal-Wallis test, P = 0.263) (Fig. 7E), and no evidence of genetic compartmentalization was detected by either test (Fig. 7F).

FIG 7.

Participant 4. The panels are as described in the legend of Fig. 4.

Participant 5.

Participant 5 was diagnosed with HIV in 1985, was subsequently exposed to nonsuppressive ART (though full treatment history is unavailable), and initiated cART in January 1997. Only two pre-cART plasma samples were available, from July and October of 1996, drawn while the participant was receiving nonsuppressive dual ART (Fig. 8A). A total of 41 plasma HIV RNA sequences were isolated from these two samples, and 77 proviral sequences were isolated from CD4⁺ T cell subsets collected nearly 20 years after cART initiation (Table 1). However, due to the limited time span of plasma sampling, participant 5’s data set lacked sufficient molecular clock signal to root the tree using standard approaches (see Materials and Methods). We therefore outgroup-rooted the tree using the HIV subtype B consensus sequence HXB2 (Fig. 8B, representative tree, and Fig. 8C, diversity plot); however, the molecular clock signal in the data still did not meet our predefined significance criterion (Fig. 8D). We thus cannot reliably estimate the within-host HIV evolutionary rate or proviral integration dates for participant 5.

Nevertheless, the data still yield useful insights. First, the identification of two highly divergent proviral sequences that branched close to the root, one from T_CM cells and the other from T_EM cells, suggests that these sequences were seeded into the reservoir well before the first plasma sampling in 1996; however, it should be noted that within-host HIV sequences can sometimes evolve toward the subtype consensus (53, 54), so this observation alone does not prove that these proviruses are old. The remainder of proviruses were interspersed with the plasma HIV RNA sequences, suggesting that they were seeded into the reservoir around the time of cART initiation. Moreover, the root-to-tip distances of HIV sequences isolated from the CD4⁺ T cell subsets were significantly different; consistent with our original hypothesis, proviral sequences from T_N and T_CM cells were, on average, closer to the root (suggesting they were older) whereas those from the more differentiated T cell subsets were more distant from the root (suggesting that they were younger) (Kruskal-Wallis test, P < 0.001) (Fig. 8E). Both ANOVA and SM detected highly significant genetic compartmentalization between proviruses archived in different CD4⁺ T cell subsets (AMOVA, P < 0.001; SM weighted mean, 0.005; HPD, 0.000 to 0.0057) (Fig. 8F). Identical proviral sequences were also observed, including one sequence isolated 13 times from T_TM cells and one sequence isolated from both T_TM and T_EM cells (Fig. 8B).

No consistent trend in the proviral ages of unique versus duplicated sequences.

Clonal expansion of CD4⁺ T cells is a major driver of HIV persistence (23–26, 51, 52), but the relationship between clonal expansion and proviral age remains unknown. To investigate this, we calculated the average proviral ages of unique versus duplicated sequences identified in participants 1 to 4 (proviral dating could not be performed for participant 5). We found no consistent evidence, however, to support duplicated (presumably clonally expanded) proviral sequences as being on average younger or older than unique ones (paired t test, P = 0.384) (Fig. 9).

FIG 9.

Mean estimated proviral integration year of unique versus duplicated sequences. Data points correspond to the mean proviral integration year of distinct sequences, stratified by whether they are unique (i.e., observed only once) or duplicated (i.e., observed more than once). The P value was calculated using a paired t test between the mean proviral integration year of unique versus duplicated sequences.

No consistent patterns in proviral diversity or distinctiveness among CD4⁺ T cell subsets.

Up until this point, each participant’s data set was examined separately for evidence that their CD4⁺ T cell subsets harbored genetically distinct proviral populations. In closing, we attempted to glean more general inferences about proviral genetic composition within CD4⁺ T cell subsets during cART by analyzing data across individuals. In particular, we sought to test the hypothesis that less differentiated memory T cell subsets, by virtue of their longer life-spans, tend to harbor the most genetically diverse and/or distinctive proviral populations. First, we quantified the phylogenetic diversity of distinct proviral sequences isolated from each participant’s CD4⁺ T cell subsets (55). This provides a measure of the overall genetic variation within each subset. Second, we quantified the average evolutionary distinctiveness of these same sequences (44). This provides a measure of the overall level of topological isolation of sequences within each subset (e.g., subsets rich in sequences that branch deeply within the tree would yield a high distinctiveness score). However, comparisons across participants yielded no significant differences in within-host proviral phylogenetic diversity (Friedman test, P = 0.392) or mean evolutionary distinctiveness metrics across subsets (Friedman test, P = 0.178) (Fig. 10A and B). Overall, no consistent trends were apparent that would allow us to generalize with respect to CD4⁺ T cell differentiation stage and the genetic properties of proviral sequences archived therein.

FIG 10.

Cross-participant analysis of proviral diversity and distinctiveness within CD4⁺ T cell subsets. (A) Normalized phylogenetic diversity in the number of nucleotide substitutions per site per tip of distinct sequences in each participant, stratified by specific CD4⁺ T cell subset. (B) Mean evolutionary distinctiveness of distinct sequences in each participant, stratified by specific CD4⁺ T cell subset. Evolutionary distinctiveness is a unitless measurement. P values were calculated using a Friedman rank sum test.

DISCUSSION

We observed marked interindividual differences in the extent to which proviruses persisting within distinct CD4⁺ T cell subsets during long-term cART differed in their genetic composition and estimated ages. Two participants (3 and 4) displayed no evidence of proviral genetic compartmentalization across CD4⁺ T cell subsets, while three participants (1, 2, and 5) did. For participants 1 and 2, however, the extent of compartmentalization was modest; only participant 5 demonstrated strong compartmentalization (this is evident from the phylogenies shown in Fig. 8B, where proviruses from the less differentiated T_N and T_CM cell subsets are largely concentrated in subclades closer to the root, while those from the more differentiated T_TM and T_EM cell subsets, the latter in particular, are largely concentrated in subclades more divergent from the root). However, it is relevant that participant 5 experienced frequent viremic episodes during cART, which may have influenced reservoir dynamics compared to those with a scenario of sustained long-term viremia suppression.

While some level of proviral genetic compartmentalization was detected in 3/5 (60%) of participants, a frequency that is broadly consistent with the recent report by De Scheerder et al. (29), its manifestation was inconsistent between individuals. Notably, we found no consistent evidence to support our original hypothesis that, of the memory CD4⁺ T cell subsets, T_CM cells would tend to harbor the oldest, most genetically diverse and/or distinctive proviral populations by virtue of their increased stability as a reservoir. In all but participant 5, T_CM cell-derived proviral sequences were distributed throughout the phylogenies and intermixed with sequences from other subsets. Moreover, in the four participants for whom proviral ages could be estimated (participants 1 to 4), T_CM cell proviral ages were not significantly different from those of other memory subsets (all pairwise comparisons, P > 0.8) (data not shown). Other reasonable hypotheses exist: for example, as HIV preferentially infects HIV-specific T cells (56) one could posit that, over time, repeated antigen-driven stimulation of latently infected T cells would lead to their differentiation, such that the oldest proviruses would eventually accumulate in the most differentiated CD4⁺ T cell subsets. Alternatively, given that HIV is capable of infecting resting CD4⁺ T cells including T_N cells (36, 57), one might hypothesize that, if most of these fail to subsequently encounter their cognate antigen, T_N cells may come to harbor the oldest proviruses over time. However, we observed no overall patterns that could support any such generalizations. In fact, other than our observation that identical proviruses tended to be enriched in more differentiated T cell subsets (Fig. 3), which is consistent with the findings of others (13, 29), we did not observe any consistent overall relationship between T cell differentiation stage and the age, diversity, and/or distinctiveness of proviral sequences archived therein.

These results extend our understanding of proviral composition in peripheral blood CD4⁺ T cells during long-term cART. The observation that proviral sequences were interspersed throughout phylogenetic reconstructions of pre-cART HIV evolutionary history, where the oldest sequences were estimated to have integrated more than 20 years prior to sampling, confirms that the reservoir is continually seeded during uncontrolled infection and that these lineages can persist for decades (32, 45). Our frequent isolation of identical proviral sequences, sometimes representing very old viral lineages (e.g., a 21.5-year-old sequence was repeatedly recovered from participant 2’s T_EM cells) and in other cases isolated from multiple CD4⁺ T cell subsets (e.g., a 10-year-old sequence was repeatedly recovered from participant 4’s T_CM, T_TM, and T_EM cells), further supports CD4⁺ T cell differentiation and clonal expansion as mechanisms to sustain the HIV reservoir (13, 15, 23–29).

Some caveats and limitations of our study should be noted. First, as we performed subgenomic sequencing, we do not know whether our sequences derive from intact, replication-competent proviruses; in fact, it is likely that most do not, given that the vast majority of proviruses persisting during long-term cART are defective (40, 58–60). It is further conceivable that, during uncontrolled HIV infection, cells harboring replication-competent proviruses may not persist as long as those harboring defective ones, as proposed in a recent study that revealed that a major portion of the replication-competent reservoir dates to around the time of cART initiation (22). Indeed, participant 5’s proviral landscape differed from the others in that the majority of recovered proviruses were interspersed with plasma HIV RNA sequences isolated prior to cART initiation (Fig. 8B), suggesting that this individual's reservoir is skewed toward sequences seeded around the time of cART initiation (22, 34) (though proviral seeding order should be interpreted with caution as the trees of this participant were outgroup rooted). This participant also experienced frequent episodes of uncontrolled viremia during cART, which could have led to gradual clearance of cells harboring older proviruses through repeated restimulation. A second caveat is that because we performed subgenomic proviral sequencing without integration site determination (61), we cannot conclusively state that identical proviral sequences derive from cellular differentiation and/or clonal expansion events; in fact, there is no optimal subgenomic HIV region that can be used to identify unique full-genome HIV sequences 100% of the time (62). A third caveat is that stem cell memory CD4⁺ T (T_SCM) cells, which are a rare yet important HIV reservoir (34), would have been sorted into the T_N cell subset as they also express CD45, CCR7, and CD27. Due to the rarity of T_SCM cells, however, we do not anticipate that these will have substantially skewed our genetic characterization of T_N cells, which is supported by Zerbato et al. who reported that depletion of T_SCM from T_N cells via CD95 had no effect on total proviral levels measured in T_N cells (36). A final caveat is that only peripheral blood CD4⁺ T cells were studied, while the majority of memory CD4⁺ T cells reside in tissues (31).

Nevertheless, a strength of our study is the availability of archived plasma dating back to the mid-1990s, which allows us to interpret proviral diversity in context of HIV’s within-host evolutionary history and make inferences regarding reservoir longevity and dynamics. Another strength is our use of Bayesian methods (42) to generate a distribution of phylogenies instead of inferring a single tree via maximum likelihood (20–22) or neighbor-joining (25, 32) methods, allowing us to account for the uncertainty in reconstructing the evolutionary relationships between sampled viral sequences.

In conclusion, our study underscores the genetic complexity of proviruses persisting in CD4⁺ T cell subsets during long-term cART and the unique individuality of each participant studied. Other than our observation that identical proviral sequences tended to be observed more often in more differentiated memory CD4⁺ T cell subsets, we observed no consistent relationship between T cell differentiation stage and the age, diversity, and/or distinctiveness of proviral sequences archived therein. Our findings underscore the importance of developing cure strategies that will take intra- and interindividual reservoir genetic diversity into account, including personalized strategies capable of eradicating HIV from a variety of cell types.

MATERIALS AND METHODS

Study participants and ethics statement.

Five participants who initiated cART a median of 10 years following HIV diagnosis (range, 2 to 14 years) and who subsequently received cART for a median 9 years (range, 8 to 16 years) were recruited through the British Columbia Centre for Excellence in HIV/AIDS (Table 1). Participants provided a single large blood donation from which peripheral blood mononuclear cells (PBMCs) were isolated by standard density gradient separation (Histopaque-1077; Sigma), counted (TC20 automated cell counter; Bio-Rad), and cryopreserved at –150°C in 90% fetal bovine serum and 10% dimethyl sulfoxide (DMSO). A median 8 (range, 2 to 17) longitudinal pre-cART plasma samples per participant, dating back to 1996, were also available for analysis. The study was approved by the Providence Health Care/University of British Columbia and Simon Fraser University research ethics boards. All participants provided written informed consent.

Isolation of CD4⁺ T cell subsets.

CD4⁺ T cells were purified from a minimum of 350 million cryopreserved PBMCs by negative magnetic selection (StemCell Technologies), after which naive (T_N), central memory (T_CM), transitional memory (T_TM), and effector memory (T_EM) CD4⁺ T cells were isolated as described in Chomont et al. (32). Briefly, purified CD4⁺ T cells were labeled with a live/dead dye (Invitrogen) and Alexa Fluor 700 (AF700)-labeled mouse anti-human CD3 antibody, pacific blue (PB)-labeled mouse anti-human CD8, allophycocyanin (APC)-labeled mouse anti-human CD4 antibody, APC-Cy7-labeled mouse anti-human CD45RA antibody, phycoerythrin (PE)-Cy7-labeled rat anti-human CCR7 antibody, and PE-labeled mouse anti-human CD27 antibody (all from BD Biosciences) and separated using a FACSAria cell sorter (BD Biosciences) (Fig. 1). Purity of all isolated subsets was verified by flow cytometry after staining for CD3, CD4, CD8, CD45RA, CCR7, CD27, and viability, and percent purity was confirmed as follows: T_N median, 95.9% (IQR, 95.6% to 98.2%); T_CM median, 98.3% (IQR, 97.3% to 99.5%); T_TM median, 98.9% (IQR, 98.9% to 99.6%); and T_EM median, 98.5% (IQR, 98.5% to 99.4%).

HIV reservoir quantification.

For each participant, genomic DNA was extracted from a median of 3.1 ×10⁶ (IQR, 2.1 × 10⁶ to 4 × 10⁶) purified CD4⁺ T cells per subset using a Purelink Genomic DNA Mini kit (Invitrogen). The number of HIV Gag copies/million CD4⁺ T cells was determined using droplet digital PCR (ddPCR) as previously described (63), where HIV Gag and human RPP30 reactions were conducted in parallel, and copy numbers were normalized to the quantity of input DNA to determine the number of HIV copies/million CD4⁺ T cells. Briefly, in each ddPCR reaction mixture, between 9 and 13 ng (RPP30) or between 131 and 1,152 ng (HIV Gag) of genomic DNA was combined with 10 units of XhoI restriction enzyme (New England Biolabs), ddPCR Supermix for Probes (no dUTPs) (Bio-Rad), primers (final concentration, 900 nM; Integrated DNA Technologies), probe (final concentration, 250 nM; Integrated DNA Technologies), and nuclease-free water. Primer and probe sequences were as follows: RPP30 forward primer, GATTTGGACCTGCGAGCG; RPP30 probe, VIC-CTGACCTGA-ZEN-AGGCTCT-3IABKFQ; RPP30 reverse primer, GCGGCTGTCTCCACAAGT; HIV Gag forward primer, TCTCGACGCAGGACTCG; HIV Gag probe, FAM-CTCTCTCCT-ZEN-TCTAGCCTC-3IABKFQ; HIV Gag reverse primer, TACTGACGCTCTCGCACC. Droplets were prepared using an Automated Droplet Generator (Bio-Rad) and cycled as follows: 95°C for 10 min; 40 cycles of 94°C for 30 s and 53°C for 1 min; and 98°C for 10 min. Droplets were analyzed on a QX200 Droplet Reader (Bio-Rad) using QuantaSoft software (version 1.7.4; Bio-Rad), and data from a median of 4 replicate wells (IQR, 4 to 6 wells) were merged prior to analysis.

Single-genome HIV amplification and sequencing.

HIV RNA was extracted from 0.5-ml pre-cART plasma aliquots using a NucliSENS EasyMag system (bioMérieux), while genomic DNA was extracted from purified CD4⁺ T cells using an Invitrogen genomic DNA isolation kit. Single-genome amplification of a subgenomic HIV fragment (nef) was performed by limiting dilution using high-fidelity enzymes and sequence-specific primers optimized for amplification of multiple HIV group M subtypes. For HIV RNA extracts, cDNA was generated using Expand reverse transcriptase (Roche). Next, cDNA as well as genomic DNA extracts was endpoint diluted such that ∼25 to 30% of the resulting nested PCRs, performed using an Expand High Fidelity PCR system (Roche), would yield an amplicon. Negative PCR controls were included in every run. Primers used for cDNA generation/1st round PCR were Nef8683F_pan (forward; 5′-TAGCAGTAGCTGRGKGRACAGATAG-3′) and Nef9536R_pan (reverse; 5′-TACAGGCAAAAAGCAGCTGCTTATATGYAG-3′). Primers used for the 2nd round of PCRs were Nef8746F_pan (forward; 5′-TCCACATACCTASAAGAATMAGACARG-3′) and Nef9474R (reverse; 5′-CAGGCCACRCCTCCCTGGAAASKCCC-3′). Amplicons were sequenced on an ABI 3130xl or 3730xl automated DNA sequencer using BigDye, version 3.1 chemistry (Applied Biosystems). Chromatograms were base called using Sequencher, version 5.0 (Gene Codes).

Alignment and phylogenetic inference.

Sequences were aligned using MAFFT, version 7.313 (64), and manually inspected and edited using AliView, version 1.23 (65). Sequences exhibiting gross defects (e.g., large deletions), mixed bases, hypermutated sequences (identified using Hypermut, version 2.0 [66], with the consensus of a participant’s earliest sequences used as the reference), or within-host recombinants (identified using RDP4 Beta 95 [67]) were discarded. Duplicate sequences (identified using seqinr, version 3.4-5, of the R package [68]) were discarded for the purpose of inferring phylogenies. We used jModeltest, version 2.1.10 (69), to select the substitution model with the lowest Bayesian information criterion (BIC) for each participant except participant 4, for whom we used the best substitution model lacking gamma rate heterogeneity (as models featuring the latter yielded trees that overestimated within-host branch lengths). For each participant, Markov chain Monte Carlo (MCMC) methods were used to build a random sample of phylogenies. Two parallel runs for each participant with MCMC chains of 10 or 20 million generations sampling every 1,000 generations were performed in MrBayes, version 3.2.7 (42), using the substitution model selected as described above and default priors. Convergence of chains was assessed by ensuring the deviation of split frequencies was <0.01 (using MrBayes [42]) and that the effective sampling size of all parameters was >200 (using Tracer, version 1.7.1 [70]). We discarded 25% burn-in from each chain and parallel chains combined in R, version 3.6.0 (71), resulting in 30,000 phylogenies per participant for participants 1, 3, and 4 and 15,000 phylogenies per participant for participants 2 and 5. Phylogenies were plotted using the R package ggtree, version 1.17.4 (72, 73).

Age reconstruction, compartmentalization, and phylogenetic statistics.

We used a published within-host phylogenetic approach to infer proviral sequence ages (21), with the following modifications. Instead of inferring a single maximum likelihood phylogeny per participant, we analyzed all of the phylogenies inferred by the Bayesian method described above. Briefly, each phylogeny was rooted using root-to-tip regression implemented in the R package ape, version 5.3 (74). Then, a linear regression was fitted to each phylogeny that related the collection date of each plasma HIV RNA sequence to its divergence from the root using R (71). We assessed molecular clock and model fit by using a delta Akaike information criterion (ΔAIC) (75), computed as the difference between the AIC of the null model (a zero slope) and the AIC of the linear regression. We determined a within-host data set to have a sufficient molecular clock signal if at least 50% of the trees produced a linear regression with a ΔAIC of at least 5 (a threshold that, in our experiments, corresponds to a P value of 0.0081 when a log-likelihood ratio test is used) and with a root date estimate whose lower limit for the 95% confidence interval was before the first sampled time point. Each linear regression was then used to estimate the integration date of each proviral sequence from its divergence from the root.

Sequence compartmentalization was assessed using two methods: analysis of molecular variance (AMOVA), a genetic distance-based test (49), and Slatkin-Maddison (SM), a tree-based test (43). For the SM test and all other subsequent tests that specifically assessed proviral composition within CD4⁺ T cell subsets, the within-host phylogenies were pruned to retain only proviral sequences, using the R package ape (74). AMOVA was implemented in the R package pegas, version 0.11 (76), while SM was implemented in the R package slatkin.maddison, version 0.1.0 (77). Phylogenetic diversity (55) and evolutionary distinctiveness (44) were computed using custom scripts implemented in R. Phylogenetic diversity was calculated by summing the edge lengths of a pruned phylogeny retaining only sequences from a particular subset and then dividing by the number of sequences in the subset. Prior to assessing phylogenetic distinctiveness, all within-host phylogenies were first outgroup rooted (OGR) with the evolutionary placement algorithm (EPA) implemented in RAxML, version 8.2.12 (78) with the HIV-1 subtype B reference strain HXB2 (GenBank accession number K03455) as the outgroup. Phylogenetic distinctiveness was then computed as described in May (44) with the R package ape (74). Paired t tests, Kruskal-Wallis rank sum tests, and Friedman rank sum tests were computed using R (71). Computer parallelization was performed using GNU Parallel (79).

Data availability.

Previously deposited sequences in this study were obtained from GenBank and have the following accession numbers: MG822918, MG822919, MG822923 to MG822933, MG822935 to MG822997, MG822999 to MG823016, MG823018 to MG823039, MG823041 to MG823050, MG823052 to MG823101, MG823106 to MG823112, MG823114 to MG823123, MG823125 to MG823143. GenBank accession numbers for new sequences used in this study are MN600002 to MN600712.