Paired quantitative and qualitative assessment of the replication-competent HIV-1 reservoir and comparison with integrated proviral DNA

Results

To investigate the genetic and phenotypic complexity of the replication-competent reservoir, we modified the QVOA protocol to increase the number of unique outgrowth cultures and sequenced the emerging viruses. Unlike QVOA, where multiple dilutions are assayed, Q²VOA is performed using a single predetermined dilution that produces less than 30% positive wells to maximize the total number of individual viruses that can be sequenced. Based on Poisson distribution, this technique produces cultures that are likely to contain single replication-competent proviruses (Fig. 1).

CD4⁺ T lymphocytes were isolated from each of four chronically infected individuals who had been virologically suppressed by combination ART for 4–22 y at two time points 4–6 mo apart (Table S1). We tested 0.40–1.44 × 10⁸ CD4⁺ T lymphocytes from each ART-treated individual at each time point. On average, 13.5% of cultures were positive for p24. The number of cells yielding replication-competent viruses varied across individuals from 0.19 to 1.07 infectious units per million, which is similar to values obtained by others (3, 6) (Table 1).

To characterize the cultured viruses molecularly, we produced cDNA from culture supernatants and sequenced the env gene using primers that resulted in a clonal prediction score of 94 of 100 (silicianolab.johnshopkins.edu/cps) (21). Thus, there was a high probability that identical env sequences represented identical full-length genomes. We obtained a total of 234 env sequences from Q²VOA, of which 13.7% were excluded from further analysis due to the presence of short reads (3.8%) or the presence of reads producing an inconclusive consensus (9.8%). The phylogenetic analysis of the remaining 202 env sequences showed that the four individuals were infected with epidemiologically unrelated clade B viruses (Fig. S1). Phylogenetic analysis of individual sequences revealed the existence of a diverse viral population composed of multiple (bootstrap-supported) clusters for each of the four individuals (Fig. 2A).

To compare the diversity of viruses obtained from a “bulk” culture with the diversity of viruses derived by Q²VOA, we performed single genome analysis (SGA) on bulk culture supernatants established at the same time from the same four individuals. In contrast to Q²VOA, bulk culture supernatants were mainly monotypic and showed much reduced overall diversity (Fig. 2A). Thus, when multiple infected cells are reactivated in a single culture, the strain or strains with the fastest growth kinetics, or the greatest fitness in culture, dominate.

Individual replication-competent viruses obtained from different Q²VOA cultures frequently encoded identical env sequences. When Q²VOA-derived viruses obtained at the two time points were compared for each subject, typically less than half (40–52%) of their env sequences were unique (Table 2). The majority of sequences were identical to at least one other independently derived replication-competent virus obtained from the same subject. For example, of a total of 49 env sequences isolated from the two time points from B106, only 21 were unique. The majority (22) were identical to at least one other sequence that appeared at one of the two time points. We will refer to these repeated sequences irrespective of whether they appear at one or both time points as “clones” because they must originate from at least two different CD4⁺ T cells. This finding does not necessarily imply that the viruses are integrated in the same location in the genome because it is possible that identical viruses can infect different CD4⁺ T cells. The size of these clones ranged from two to 10 members, with a mean of 3.69 when all four individuals and both time points were considered (Fig. 2 B and C and Table 2). Fifty-four percent of all replication-competent viruses emerging in Q²VOA cultures were derived from expanded clones (Fig. 2 B and C and Table 2).

To determine whether the viral sequences obtained from the replication-competent reservoir remained stable over time, we compared the sequences from the two time points for each individual. Many branches in the phylogenetic trees contained sequences derived from both time points (Fig. 2 A–C and Table 2). The relationship between the sequences from both time points was formally assessed by determining their Genealogical Sorting Index (GSI), which quantitates the degree of phylogenetic association between sequences (23). GSI values, which range between 0 (complete interspersion) and 1 (complete monophyly), showed that for each of the four individuals, the sequences obtained by Q²VOA from both time points could not be segregated as distinct groups (Table 3). This finding demonstrates that the viral population emerging from the latent reservoir in four individuals was stable over the 4- to 6-mo time interval analyzed.

To examine the relationship between proviruses integrated into CD4⁺ T-cell DNA and the replication-competent viruses obtained by Q²VOA, we performed SGA on DNA isolated from primary CD4⁺ T lymphocytes from the same individuals at both time points. We obtained a total of 498 env sequences, of which 16.3% were excluded from further analysis due to the presence of hypermutated regions (2.4%), short reads (5.6%), or reads producing an inconclusive consensus (8.2%). The remaining 417 full-length env sequences, or 85–113 per individual, fell within the same four patient-specific clades as the Q²VOA-derived env sequences, indicating absence of sample mix-up or contamination (Fig. S1).

As previously reported for archived proviral DNA amplified from primary CD4⁺ T cells, we found unique sequences, as well as large groups of identical sequences, marking expanded cell clones (14, 15, 24, 25). For example, in B106, 67 of 113 env sequences obtained from the two time points were unique and the remaining 46 were members of clones. In addition, 36 of the 46 identical sequences overlapped between the two time points (Fig. 3 A and B and Table 4). Thus, like the Q²VOA-derived sequences, the archived proviral population was stable over the time interval analyzed.

Archived proviral DNA sequences were then compared with replication-competent Q²VOA-derived sequences. Because both the proviral DNA and the replication-competent viral sequences for any given individual were overlapping at the two time points, we combined each of the two sets of sequences (Fig. 4). As might be expected, given that the sample size is limited, we found relatively limited overlap between the archived proviral sequences and Q²VOA culture-derived sequences in phylogenetic trees, with multiple large clusters composed of sequences isolated from only a single source (Fig. 4A). The 417 archived proviral sequences contained 50 expanded clones, 12 of which were also found in the replication-competent outgrowth cultures. The relative frequency of expanded clones from these two sources differed: Some greatly expanded clones identified in primary CD4⁺ T-cell DNA represented only a small fraction of clones identified in the outgrowth cultures, and rare archived clones were disproportionally abundant among the reactivated latent replication-competent viruses (Fig. 4B). This finding is best illustrated in subject B155, where the largest archived proviral clone, which comprised 28 of 99 total sequences, was found only once among 52 replication-competent viruses (Fig. 4 B and C). In contrast, in B199, one archived proviral sequence, which appeared only once in a total of 120 sequences derived from primary CD4⁺ T-cell DNA, was found in 11 of 50 Q²VOA culture-derived sequences emerging from the latent reservoir. B115 provided the clearest example of the discrepancy between proviral DNA and cultured viruses, with no instances of matching sequences between 85 proviral sequences and 50 outgrowth viruses (Fig. 4 B and C). Although this discrepancy is most likely due to the prevalence of defective archived proviral sequences (6–9), differences in proviral accessibility to polymerase may also contribute.

To determine the extent to which sequences from replication-competent viruses and archived proviruses were compartmentalized, we calculated their GSI. This analysis demonstrated that archived proviral and Q²VOA culture-derived sequences were significantly segregated for every single individual analyzed (Table 5). Thus, proviral sequences derived from primary CD4⁺ T-cell DNA do not provide an accurate representation of viruses comprising the replication-competent reservoir.

To understand the relationship between the two groups of sequences better, we analyzed them by a mathematical model that describes every sequence by two variables (SI Methods). The first (p) is the frequency of a sequence in the proviral compartment, and the second (r) is its probability of reactivation in the active viral culture. These parameters were extracted from the experimental data using Bayesian inference methods. The data show that clone size is negatively correlated with the activation probability (r = −0.94, p = 3.4 × 10⁻³⁴) (Fig. 5). These results indicate that the larger the clone of archived proviral sequences, the lower is its probability of representing a replication-competent virus in the Q²VOA outgrowth culture.

Finally, to characterize the viruses obtained by Q²VOA further, we tested virus-containing supernatants selected to be representative of the diversity in the phylogenetic trees for sensitivity to a panel of broadly neutralizing antibodies (bNAbs) that are currently in clinical development. As might be expected, there was significant variation in the sensitivity profile of each individual’s replication-competent reservoir. In each case, there was at least one example of coexisting sensitive and resistant viruses for any one tested bNAb (Fig. 6A). For example, in B155, all but one of the 10 viruses tested were resistant to 10-1074, a bNAb that targets the base of the V3 loop and surrounding glycans (26). As might be expected, the 10-1074–susceptible virus segregated from the others in the phylogenetic tree (Fig. 6A and Fig. S2). When results from all individuals were combined, a large range in sensitivities was observed against the panel of bNAbs (Fig. 6B). In contrast, supernatants obtained from the bulk cultures showed bNAb sensitivity that appeared to be representative of a single viral cluster in any individual (Fig. 6A). Thus, Q²VOA differs from bulk cultures in that it enables measurement of the genetic and phenotypic diversity of the replication-competent reservoir. Finally, the observation that bNAb neutralization-sensitive and -resistant clones coexist in the reservoir provides an explanation for the observation that combination therapy will likely be required to maintain suppression (27, 28).

SI Methods

We applied Bayesian inference methods to estimate jointly the frequency of integrated proviruses and their reactivation probability, an approach enabled by the combined proviral sequencing and viral outgrowth assay data. We model the probability of measuring a certain number k_DNA of integrated proviral sequences as binomial,

Here, W_DNA is the number of wells tested, C_DNA is the number of cells per well, and p is the frequency of CD4⁺ T cells with integrated provirus. Similarly, the probability of observing k_VOA-positive wells in the viral outgrowth assay is also binomial,

As above, W_VOA and C_VOA are the number of wells and number of cells tested per well in the viral outgrowth assay, respectively. Here, r is the probability that an integrated provirus reactivates and grows out in the VOA.

We assumed the following prior distributions:

where μ = max(k_DNA/(W_DNA × C_DNA), 10⁻⁷) are weakly informative (54), intended simply to prevent pathological inferences. This approach yields conservative parameter estimates. For example, posterior distributions of reactivation probabilities for large clones are nonzero even if no reactivation events are observed. In models that include data from multiple visits, we also included an informative prior to take into account the expected decay of the latent reservoir in individuals on uninterrupted ART,

Here, p_i and r_i are the frequency of integrated provirus and reactivation probability at visit i, respectively. Their product is therefore equivalent to the number of infectious units per million cells (IUPM), divided by 10⁶. We used λ = −0.007 mo⁻¹ as the IUPM decay rate and σ = 0.38 as the SD of its measurement, following previous estimates (3). The time between visits (in months) is t. We implemented all models in Stan (52) using the PyStan interface.

Discussion

A long-lived reservoir of HIV-1–infected cells that remain silent in infected individuals receiving ART is the primary impediment to HIV-1 cure (22, 29, 30). The reservoir is established early in infection (31) and has an estimated t_1/2 of 44 mo (2, 3). Several mechanisms have been proposed to account for the persistence of this reservoir, including low-level viral replication, the long t_1/2 of latently infected cells, and the proliferative expansion of these latently infected cells (5, 32, 33).

Arguments against persistent low-level viral replication include the observation that further intensification of ART has no measurable effect (34, 35). In addition, there is little evidence for viral evolution even after prolonged periods of ART (36). In contrast, the idea that the reservoir is maintained, at least in part, by the proliferation of latently infected cells is supported by the observation of increasing proportions of identical proviral sequences in circulating CD4⁺ T cells (15).

Archived proviral integration site analysis provided further support for clonal expansion of infected T cells (8, 24, 25). In all cases examined, a large proportion of the archived integrated proviruses were found to belong to expanded clones of CD4⁺ T cells that share a unique integration site. However, as expected from the observation that the vast majority of integrated proviruses are defective (7, 9), many of the proviruses found in clonally expanded T cells are also defective (8). The only exception was found in a singular individual with metastatic squamous cell carcinoma, which showed a large expanded clone of replication-competent virus (37). However, the integration site of the provirus in this individual was ambiguous and could not be mapped with certainty. Thus, the precise contribution of expanded cell clones bearing replication-competent proviruses to maintaining the latent reservoir is not defined well.

To examine the diversity of the latent but replication-competent reservoir, we developed a modification of current quantitative viral outgrowth protocols to include a qualitative measure of the reservoir of replication-competent proviruses. Q²VOA maximizes the yield of limiting dilution cultures to obtain viruses originating from a single cell. Consistent with the absence of viral evolution in individuals on suppressive ART, Q²VOA revealed that the latent replication-competent viral population is similar between time points 4–6 mo apart (12, 13, 38). This interval is relatively short in the life of an infected individual, and it will be interesting to evaluate the reservoir by Q²VOA over longer time intervals.

Fifty-four percent of the viruses emerging in Q²VOA belonged to expanded cell clones bearing the same env sequences, and these viruses have a high probability of also being identical in the rest of their genomes [prediction score of 94 of 100; available at silicianolab.johnshopkins.edu/cps (21)]. The isolation of identical sequences from multiple wells on different plates across two time points suggests that these replicating viruses do not accumulate significant mutations during 14 d in culture. The lack of immunological pressure under culture conditions may account for this finding.

There are several nonmutually exclusive explanations for the prevalence of expanded CD4⁺ T-cell clones carrying replication-competent viruses in the latent reservoir. Such cells could be clonally expanded as part of normal homeostatic processes, in response to antigen, or as a result of HIV-1 integration into and disruption or activation of genes that regulate cell division (24, 25). Alternatively, identical viral sequences could represent a burst of independent viral integration events into unrelated CD4⁺ T cells. Mapping the precise integration sites of the replication-competent proviruses is required to distinguish definitively between viral clones produced by CD4⁺ T-cell clonal expansion and by bursts of proviral integration.

Clonally expanded CD4⁺ T cells with shared proviral integration sites were detected in samples of 1–2 × 10⁶ cells (8, 24, 25), which are far smaller than the samples assayed in Q²VOA. Only the largest of these previously described expanded cell clones were amenable to further characterization, and all 75 proviruses that were examined molecularly showed gross defects that rendered them replication-incompetent (8). The finding that large proviral clones are typically defective is entirely consistent with the observation that proviral clone size is inversely related to the probability of reactivation. The larger the archival proviral clone, the less likely it was to be represented in the replication-competent latent reservoir. This observation is also in agreement with the finding that most clonally expanded proviruses integrated into CD4⁺ T-cell DNA are defective (6–9).

The relative stability of clones in the replication-competent reservoir has significant implications for assessing efforts aimed at HIV-1 eradication. In particular, Q²VOA may be a sensitive assay with which to measure changes in the reservoir that result from immune-targeted interventions (39–42). Viruses isolated from the replication-competent latent reservoir demonstrated a broad range of neutralization sensitivity to bNAbs, and may also show differing levels of sensitivity to immune-based interventions. The coexistence of neutralization-sensitive and -resistant clones in the same patient may be important to consider in development of clinical strategies using bNAbs in the future. However, whether the viruses detected in Q²VOA represent species that are preferentially reactivated upon termination of ART in infected individuals remains to be determined.

In conclusion, we find clones of replication-competent viruses in the latent reservoir. The viral sequences of these clones show limited overlap with archived proviral sequences in primary CD4⁺ T cells, suggesting that the two compartments may be under different types of selective pressure in vivo. Finally, if the reservoir is maintained by CD4⁺ T-cell division, it may facilitate immunotherapy-directed efforts for HIV-1 cure because T-cell activation is thought to be associated with expression of HIV-1 antigens (39).

Methods