Progressive transformation of the HIV-1 reservoir cell profile over two decades of antiviral therapy

Summary HIV-1 establishes a life-long reservoir of virally infected cells which cannot be eliminated by antiretroviral therapy (ART). Here, we demonstrate a markedly altered viral reservoir profile of long-term ART-treated individuals, characterized by large clones of intact proviruses preferentially integrated in heterochromatin locations, most prominently in centromeric satellite/micro-satellite DNA. Longitudinal evaluations suggested that this specific reservoir configuration results from selection processes that promote the persistence of intact proviruses in repressive chromatin positions, while proviruses in permissive chromosomal locations are more likely to be eliminated. A bias toward chromosomal integration sites in heterochromatin locations was also observed for intact proviruses in study participants who maintained viral control after discontinuation of antiretroviral therapy. Together, these results raise the possibility that antiviral selection mechanisms during long-term ART may induce an HIV-1 reservoir structure with features of deep latency and, possibly, more limited abilities to drive rebound viremia upon treatment interruptions.


Correspondence mlichterfeld@partners.org
In brief Lian et al. show that following two decades of continuous antiretroviral therapy, the integration site profile of intact HIV-1 proviruses is heavily biased toward heterochromatin locations, likely as a result of immune selection mechanisms; such proviruses are less transcriptionally active and, possibly, less rebound competent.

INTRODUCTION
Antiretroviral combination therapy is highly effective in suppressing HIV-1 replication and is recommended as the standard of care for all people living with HIV (PLHIV). Through the inhibition of HIV-1 reverse transcriptase, integrase, or protease, antiretroviral drugs can effectively prevent infection of new viral target cells; however, they have no direct activity against already infected cells, some of which persist as a long-term viral reservoir and make HIV-1 infection an incurable disease. [1][2][3] These cells have traditionally been considered as ''latently infected'' lymphocytes that harbor ''transcriptionally silent'' proviruses, 4 but recent results have shown that a significant proportion of these cells can remain transcriptionally active, [5][6][7][8] consistent with the fact that antiretroviral agents do not directly inhibit viral gene transcription from chromosomal DNA. Ongoing viral gene transcription and protein translation may make HIV-1-infected cells visible to immune cells and increase their vulnerability to host-dependent immune activity or viral cytopathic effects; in contrast, true viral latency, defined by extremely limited or no residual viral transcriptional activity, can be viewed as an effective strategy to escape from host immune surveillance. Recent work suggested that in persons with natural control of HIV-1 (termed elite controllers [ECs]) the majority of genome-intact proviral sequences were integrated in heterochromatin locations that do not support effective viral transcription-a state that we and others have termed ''deep latency.'' [9][10][11] We proposed that such a distinct viral reservoir profile, colloquially referred to as a ''block and lock'' pattern, is the result of immune-mediated selection mechanisms that preferentially eliminate proviruses located in chromosomal positions more supportive of proviral transcriptional activity, while intact proviruses in repressive chromatin locations persist long term. A proviral reservoir landscape dominated by intact proviruses in heterochromatin positions may only be poorly rebound-competent in vivo and is likely to contribute to the remarkable ability of ECs to maintain a durable, drug-free remission of HIV-1 infection. In this work, we hypothesized that selection of viral reservoir cells with features of deep viral latency may also occur in antiretroviral therapy (ART)treated individuals, although presumably with weaker efficacy than in ECs. Reasoning that extended durations of suppressive ART may allow such selection forces to become more visible, we here conducted a detailed analysis of the HIV-1 reservoir profile for a group of persons who remained on suppressive ART for approximately two decades.

Frequencies of intact proviruses in persons undergoing long-term ART
We evaluated HIV-1 reservoir cells in study participants (n = 8) who stayed on continuous suppressive ART for a median of 20 (range: [18][19][20][21][22][23] consecutive years (with no more than 2 recorded plasma viremia blips [<100 copies/mL]) and were among the first generation of PLHIV benefiting from the roll-out of triple-drug antiretroviral combination therapy from 1996 to 2000 ( Figure S1). For an initial characterization of the HIV-1 reservoir profile in these long-term ART-treated individuals (LT-ART), we subjected their peripheral blood mononuclear cells (PBMC) to near full-length individual proviral sequencing (FLIP-seq), a protocol that relies on single-genome next-generation sequencing of HIV-1 proviral DNA, followed by biocomputational identification of genomeintact proviruses defined by the absence of lethal sequence variations. 12 These studies, involving a total of 496 individual proviral genomes from LT-ART individuals (n = 96 intact and n = 400 defective proviruses), demonstrated no specific difference in the frequency of genome-intact proviral sequences in LT-ART individuals relative to persons with moderate durations of suppressive ART exposure (median of 9 years) (m-ART). In comparison with ECs, the frequencies of intact proviruses in LT-ART study participants were significantly higher ( Figures 1A and 1B). However, the ratio of genome-intact proviruses to defective proviruses ( Figures 1C and 1D), and the average intra-individual genetic distance between intact proviruses ( Figure 1E), were more similar between LT-ART individuals and ECs, relative to persons with moderate ART treatment durations. Phylogenetic evaluations indicated that large proportions of intact proviruses from LT-ART individuals were part of sequence-identical clonal clusters that derive from clonal proliferation of virally infected cells; the proportion of intact proviruses that were part of such large clones was highest in LT-ART individuals ( Figures 1F and 1G), although such large clones of intact proviruses were also evident in ECs. Notably, clusters of sequence-identical genome-intact or replicationcompetent HIV-1 proviruses have also been observed in persons with more limited durations of ART; however, the general landscape of intact proviruses in such individuals was more diverse. [14][15][16][17][18][19] The number of HLA-class-I-associated sequence variations in intact proviruses did not differ profoundly between LT-ART individuals and ART-treated PLHIV with moderate treatment durations, and was lowest in ECs ( Figures 1H and S2).

Persistence of intact proviruses integrated in centromeres and in satellite DNA
To profile the chromosomal locations of intact proviruses, we used ''matched integration site and proviral sequencing'' (MIP-seq), an approach that focuses on phi29-catalyzed multiple displacement amplification of individual proviruses, followed by paired assessments of proviral sequences and corresponding proviral chromosomal integration sites. 16 In our cross-sectional analysis, we identified independent integration sites of n = 31 intact proviruses from LT-ART individuals. Most notably, these studies revealed that integration sites of intact proviruses from LT-ART individuals were profoundly biased toward centromeric satellite DNA, peri-centromeric micro-satellite DNA, or non-genic DNA, which generally represent disfavored targets for proviral integration during acute HIV-1 infection, 20,21 and can confer a state of deep latency 8,22,23 (Figures 2A and S3; Tables S1 and S2). In study participants LT01-03, the proviral reservoir landscape consisted almost exclusively of large clones of genomeintact proviruses integrated in the centromeric or peri-centromeric satellite/micro-satellite DNA of chromosome 21 (participant LT01), chromosome 10 and chromosome Y (participant LT02), or chromosome 18 (participant LT03). In study participants LT04-06, a more oligoclonal viral reservoir profile was observed; however, large clones of genome-intact proviruses integrated in satellite DNA were equally visible in these participants. Further, some intact proviral genomes in these three participants were integrated in highly repetitive satellite DNA regions that were present on multiple chromosomes and could not be definitively mapped to one specific chromosomal site. Notably, in study participant LT04, the integration of an intact provirus into noncentromeric micro-satellite DNA (on chromosome Y) was observed. Moreover, in participant LT05, an intact provirus was integrated in a large non-genic region on chromosome 16 in proximity to non-centromeric satellite DNA; this genomic region was previously shown to harbor intact proviruses in two ECs. 9 Collectively, the integration of intact proviruses in centromeric/pericentromeric satellite DNA was observed in 6/8 LT-ART study participants; moreover, the proportion of intact proviruses located in centromeric or peri-centromeric satellite/micro-satellite DNA among all independent intact proviruses detected in LT-ART individuals reached 32.26% ( Figure 2B). We also detected several additional intact proviruses that were located in non-genic DNA outside of centromeres or satellite/micro-satellite DNA. These proviruses were frequently positioned at a profound distance from host transcriptional start sites (TSSs) ( Figure 2C). Overall, the frequencies of intact proviruses in all non-genic locations were comparable between LT-ART individuals and ECs but significantly smaller in ART-treated persons with more limited ART duration (m-ART) ( Figures 2B and 2D).

Integration of intact proviruses in KRAB-ZNF genes
In addition to intact proviruses in centromeric satellite/micro-satellite DNA, we identified several large clones of intact proviruses in Kr€ uppel-associated box zinc-finger protein (KRAB-ZNF) genes, most of which were present in a defined region on chromosome 19. These genes are densely packed with the heterochromatin protein 1 (HP1), 25 which promotes a chromatin compartment that effectively excludes RNA polymerase from nuclear DNA. 26 Moreover, ZNF gene bodies are loaded with repressive H3K9me3 histone modifications that bind to HP1 and are deposited by the human silencing hub (HUSH) complex, which has a known role for transcriptional repression of endogenous and exogenous retroviruses. [27][28][29] We observed  Coordinates of chromosomal integration sites obtained by integration site loop amplification (ISLA) and corresponding gene name (if applicable) are indicated. Symbols indicate sequences generated by FLIP-seq, MIP-seq, PRIP-seq or from quantitative viral outgrowth assays (qVOAs). Proviruses integrated in highly repetitive satellite DNA could not always be definitively mapped to specific chromosomal locations; a detailed list of integration sites is shown in Table S1. *Sequences differ by 1 or 2 base pairs from adjacent clonal sequences. LADs, lamina-associated domains. (B) Proportions of intact proviruses with indicated integration site features in LT-ART individuals and comparison cohorts. (C) Chromosomal distance between integration sites of intact proviruses to most proximal host transcriptional start sites (TSSs), as determined by RNA-seq in CD4 T cells from reference datasets in total, effector-memory (EM), or central-memory (CM) primary CD4 T cells or from genome browser (GB). 16 Box and Whisker plots show median, interquartile ranges and minimum/maximum. (D) Proportions of genome-intact proviral sequences in structural compartments/subcompartments A and B, as determined by Hi-C seq data. 24 Integration sites not covered in the reference dataset were excluded. large clones of intact proviruses integrated in KRAB-ZNF genes in participants LT06, LT07, and LT08; in LT07 (Figures 2A and  S3), almost the entire detectable intact proviral reservoir consisted of one clone integrated in a ZNF gene on chromosome 19. In total, 23.08% of all independent intact proviruses detected in LT-ART individuals were integrated in ZNF genes; most of them were located on chromosome 19 ( Figure 2B). Notably, 2 of the intact proviruses integrated in ZNF genes (in participants LT06 and LT07) were integrated in regions predicted to be part of lamina-associated domains (LADs), based on DNA adenine methyltransferase identification (DamID) assays conducted previously. 30 LADs are located at the inner sides of nuclear membranes, represent infrequent targets for HIV-1 integration, and contain mostly transcriptionally silent heterochromatin. 31 In addition, integration sites of intact proviruses in ZNF genes were positioned in immediate proximity to inhibitory H3K9me3 histone modifications, as determined by alignment to chromatin immunoprecipitation sequencing (ChIP-seq) data from primary CD4 T cells ( Figure S4); integration in proximity to H3K9me3-enriched chromatin was associated with resistance to HIV-1 reactivation in in vitro assays in our previous work. 8 Transcriptional repression of HIV-1 proviruses integrated in ZNF genes has also been suggested by other investigators. 32,33 However, transcriptional behavior of intact proviruses in ZNF genes may be influenced by the precise proviral positioning relative to epigenetic chromatin features, and not all intact proviruses in ZNF genes may be in a state of "deep latency".
Only a relatively small number of intact proviruses (19.35% of all independent proviruses) in LT-ART individuals were located in active transcription units other than ZNF genes. Such integration sites were almost exclusively observed for intact proviruses from one single study participant (LT06). For a systematic analysis of integration sites from LT-ART individuals, we aligned integration site coordinates of intact proviruses to genome-wide maps of host gene transcription (determined by RNA-seq in primary CD4 T cell subsets) and of 3D chromatin interactions (determined by Hi-C) from reference datasets. 16,24 These studies demonstrated an enrichment of intact proviruses from LT-ART individuals in marked distance to host TSS ( Figure 2C) and in the 3D heterochromatin compartments B2 and B4 ( Figure 2D). Overall, our results indicate that the proviral integration site profile in LT-ART individuals is characterized by disproportionate frequencies of intact proviruses integrated in heterochromatin locations and closely resembles the integration site landscape in ECs. 9 Transcriptional activity of intact proviruses To further evaluate whether proviruses in centromeric satellite/ micro-satellite locations are indeed transcriptionally repressed and in a state of deep latency, we performed parallel assessments of HIV-1 RNA, integration sites and proviral sequences (PRIP-seq) from single HIV-1-infected cells, using a previously described protocol 8 ; such a direct ex vivo analysis can evaluate proviral transcriptional activity in response to physiological reactivation signals that occur in vivo. These experiments demonstrated that the large genome-intact proviral clone integrated in centromeric satellite DNA of chromosome 18 in LT03 was completely transcriptionally silent across all 5 member sequences; an additional defective proviral sequence integrated in centromeric satellite DNA on chromosome 16 was also tran-scriptionally silent ( Figure 3A). In contrast, multiple other defective clonal and non-clonal proviral sequences with integration sites in genic chromosomal locations were detected; most of these defective proviruses transcribed HIV-1 long-LTR and/or more elongated HIV-1 RNA transcripts ( Figure 3A). In particular, we noted one large defective proviral clone that was integrated in the SMURF2 gene in immediate proximity to activating epigenetic chromatin features and displayed high-level proviral transcriptional activity (Figures 3A and 3B). PRIP-seq assays with additional proviruses integrated in satellite DNA (n = 34), described in our previous work, 8 also suggested effective transcriptional repression conferred by chromosomal integration into such distinct chromatin locations. As an additional analysis step, we performed traditional quantitative viral outgrowth assays (qVOAs) that evaluate infectious viral particle production after in vitro stimulation with non-physiological reactivation agents and high-dose exogenous cytokine support, using an assay protocol that, in our previous work, readily permitted detection of replication-competent viruses in persons with shorter durations (<2.3-13 years) of ART 12,35 and in ECs. 9 A combined calculated estimate of n = 307 genome-intact proviruses from a combined total of 15.7 million isolated CD4 T cells from 5 study participants (LT02, LT03, LT06, LT07, and LT08; Table S3) was tested; viral outgrowth was observed for none of these proviruses after 14 days of in vitro culture and for n = 2 proviruses (from study participants LT02 and LT03) after 21 days of culture ( Figure 2A). These results, combined with data from in vitro models of viral latency, 22,23 support our hypothesis that HIV-1 proviruses integrated in satellite DNA or in heterochromatin regions are in a state of deep latency and difficult to reactivate. However, we explicitly do not claim that such proviruses are completely resistant against any type of viral reactivation signals and have reached a state of irreversible latency.

Longitudinal selection of intact proviruses in deep latency
To better evaluate the mechanisms leading to the specific reservoir profile in LT-ART individuals, we conducted a longitudinal analysis of integration sites for intact proviruses in a subgroup of n = 5 participants from whom PBMC samples collected at prior time points were available ( Figures 4A and S5; Tables S1 and S2). We noted that intact proviruses after relatively shortterm ART (T1) were frequently located in introns of highly expressed genes, which represent the preferred target sites for HIV-1 integrase. After moderate time periods of ART (T2), intact proviruses were again mostly integrated in genic locations; proviruses in centromeric satellite/micro-satellite DNA were detectable at slightly increased frequencies at this time relative to the earlier analysis time point ( Figure 4B). Although the majority of intact proviruses during earlier stages of treatment were detected as singlets, larger clones of intact proviruses were noted after moderate durations of ART; these proviral clones were frequently integrated in genes associated with or involved in regulation of cell proliferation, such as Reticulocalbin 2 (RCN2) (participant LT01), Thrombospondin Type Laminin G Domain And EAR Repeats (TSPEAR) (participant LT01), or Integrin Subunit Alpha L (ITGAL) (participant LT03). At the final time point (T3), the integration site landscape of intact proviruses was strongly biased toward heterochromatin regions, as described above. Together, these longitudinal evaluations suggested a progressively increasing chromosomal distance of intact proviruses to host TSS, coupled with an enrichment for intact proviruses integrated in 3D chromatin compartments with heterochromatin features (Figures 4C and 4D). Importantly, intact proviruses detected in centromeric satellite DNA and, to a lesser extent, in ZNF genes detected at early treatment time points, frequently persisted long term and were detectable at the final time point of follow-up. In contrast, proviruses in genic positions tended to disappear over time and were rarely detected at multiple time points. Together, our work supports the notion that intact proviruses in heterochromatin positions have a selection advantage that permits long-term persistence over two decades of ART. In contrast, intact proviruses in more permissive chromatin seem to be actively selected against.
For comparative purposes, we also conducted a longitudinal analysis of the proviral integration site profile for defective proviruses in the same subgroup of our study participants for whom intact proviruses had been longitudinally analyzed (n = 5). Generally, we observed that the proviral integration site landscape of defective proviruses differed profoundly from intact proviruses; specifically, defective proviruses were predominantly integrated in genic regions and there was only one single defective provirus that was located in satellite DNA (in LT03) (Figures 5A and 5B). These data are consistent with previously described preferences for HIV-1 integration site locations. [36][37][38][39] In addition, the chromosomal distance between host TSS and defective proviruses remained relatively stable over time, in apparent contrast to intact proviruses ( Figure 5C). Together, these results demonstrate striking differences in the longitudinal evolution of intact versus defective proviruses. Such differences suggest that immune selection mechanisms may differentially affect intact and defective proviruses, possibly because host selection mechanisms may be able to distinguish intact proviruses that are able to produce fully functional viral particles from defective proviruses that, depending on the type of defect, may only produce HIV-1 RNA and proteins. A distinct trajectory of intact and defective proviruses during ART is also suggested by prior studies that focused on a quantitative longitudinal analysis of proviral frequencies during ART. [40][41][42][43] Integration site profile of persons with drug-free control of HIV after treatment interruption Remarkably, the integration site profile in LT-ART individuals closely resembles the block and lock viral reservoir configuration in persons with natural viral control, 9 suggesting that PLHIV may, over prolonged periods of antiretroviral treatment, mount a level of selection pressure against viral reservoir cells that reduces their ability to effectively fuel rebound viremia in case of treatment interruptions. Consistent with this notion, longer antiretroviral treatment durations were associated with a higher probability for post-treatment, drug-free control in one prior study involving individuals identified during primary infection. 44 We believe that this question deserves future investigation in dedicated clinical studies that will selectively evaluate the effects of ART discontinuation in PLHIV displaying signs of a block and lock profile for intact proviruses. To address this here, we used MIP-seq to longitudinally profile the integration site landscape of intact proviruses in two persons who had served as placebo recipients in a prior therapeutic vaccination study 45 and developed drug-free control of HIV-1 after discontinuation of ART 46 (Figures 6A and S6). These persons resembled post-treatment controllers (PTCs) described in a previous study. 47 Both study (C) Chromosomal distance between integration sites of intact proviruses to most proximal host TSS, as determined by RNA-seq in CD4 T cells from reference datasets in total, effector-memory (EM), or central-memory (CM) primary CD4 T cells or from genome browser (GB). 16 Box and Whisker plots show median, interquartile ranges and minimum/maximum. (D) Proportions of genome-intact proviral sequences in structural compartments/subcompartments A and B, as determined by Hi-C seq data. 24 Integration sites not covered in the reference dataset were excluded. For T3, data from 2017 and 2020 from LT01 were pooled. persons had received ART for approximately 6.6 years prior to stopping suppressive antiviral therapy. In study person 04, we observed three different clones of intact proviruses at baseline (immediately prior to stopping ART), all of which were integrated in or surrounded by centromeric satellite DNA; additionally, two intact proviruses in genic locations were also detected at this time ( Figure 6B; Table S1). Several spikes of plasma viremia were noted after treatment interruption; however, these spikes were self-limited and did not lead to prolonged viremia. Spikes of plasma viremia became less obvious during the subsequent  Table S1. p values were calculated by a chisquare test in (B) and by an FDR-adjusted two-sided Kruskal-Wallis nonparametric test in (C), adjusted for multiple comparison testing where applicable. n reflects the number of integration sites.

OPEN ACCESS
Article follow-up period, and the person decided to self-initiate ART at day 1,262 after the initial treatment discontinuation. Over time, a progressive longitudinal accumulation of intact proviruses in centromeric satellite DNA was noticed, while intact proviruses in genic locations were no longer detectable, likely as a result of continuous host selection mechanisms during treatment interruption ( Figure 6B). A large genome-defective proviral clone inte-grated in the Vav Guanine Nucleotide Exchange Factor 1 (VAV1) proto-oncogene was noted at multiple time points in this individual; proviral integration in this gene can lead to insertional mutagenesis supporting cell-autonomous clonal T cell proliferation 48 ( Figure 6C). In study person 30, we observed two clones of intact proviruses integrated in non-genic chromosomal regions in immediate proximity to micro-satellite DNA at baseline; additionally, smaller numbers of intact proviruses in typical genic chromatin locations were also present before treatment interruption ( Figure 6B; Table S1). Spikes of plasma viremia were smaller after ART discontinuation in this person (in the setting of less frequent viral load testing compared with study person 04); however, an accumulation of intact proviruses in heterochromatin regions, paired with a de-selection of intact proviruses in genic locations, was also noted in this person. A large number of defective proviruses were identified in person 30, with the vast majority of these proviruses located in typical genic chromosomal locations ( Figure 6C).
Using the same analytic approach, we analyzed the integration site profile of intact proviruses in three participants from the same study 45 who rebounded to high-level viremia following treatment interruption. In these individuals, the majority of intact proviruses detected prior to treatment interruption were integrated in genic locations ( Figures 7A and 7B), and no integration of intact proviruses within or close to centromeres, micro-, or macro-satellite DNA was observed, resulting in a markedly different integration site profile of intact proviruses in PTC and rebounders ( Figure 7C). Together, these observations suggest that (1) a viral reservoir landscape enriched for intact proviruses integrated in heterochromatin locations has a weaker ability to fuel sustained viral rebound after ART interruption and that (2) PTCs show evidence for accelerated selection of intact proviruses in heterochromatin, likely as a result of stronger antiviral selection forces and, possibly, higher levels of proviral transcriptional reactivation of intact proviruses in euchromatin during off-treatment periods. Such transcriptional reactivation may expose the reservoir cells to antiviral immune effects.

DISCUSSION
Previous data suggested that it may require 70 or more years of continuous ART to induce a clinically significant decay of HIV-1 reservoir cells; 49 however, these calculations are based on theoretical extrapolations of experimental results from in vitro outgrowth assays, which may not capture the true biological behavior of HIV-1 reservoir cells in vivo. The growing number of PLHIV who started treatment approximately 20 years ago when triple-drug combination antiretroviral treatment became first available makes it possible to empirically test the evolution and persistence of HIV-1 during two decades of continuous suppressive therapy. Our data suggest that over such extended periods of time, intact proviruses integrated in permissive chromatin are more likely to be recognized and eliminated by innate or adaptive immune effector cells, or by cell-intrinsic immune recognition pathways that result in cell death. 50 In contrast, the small subset of intact proviruses integrated in repressive chromatin locations seems to have a longitudinal selection advantage and persists long-term; integration in such repressive chromosomal segments might protect against immune-mediated elimination and can be regarded as a biomarker of selection pressure that viral reservoir cells have been exposed to in vivo. These selection effects seem particularly visible during the second decade of suppressive ART, during which the reservoir profile appears to undergo a marked transformation. Notably, evidence of selection of proviruses with specific chromosomal integration site features was also noted in other retroviral diseases, specifically in the context of Human T-lymphotropic virus type 1 (HTLV-1) infection; 51 however, proviral selection may also be influenced by cell-intrinsic host features of virally infected cells that are independent of the chromosomal integration site, as recently suggested by studies involving single-cell transcriptional profiling of HIV-1 reservoir cells. 52,53 The integration site profiles of intact proviruses shown here in LT-ART-treated persons closely resemble those of ECs, 9 raising the possibility that once reservoir cells show footprints of intense immune selection pressure, an attenuated viral reservoir profile with a weaker ability to drive sustained rebound viremia may have been reached. In support of that, our work demonstrated that the transcriptional activity of proviruses integrated in satellite DNA and heterochromatin regions, when assessed directly ex vivo by the PRIP-seq assay, is profoundly reduced relative to proviruses in alternative locations. Furthermore, in vitro outgrowth assays with cells from LT-ART persons demonstrated poor susceptibility of proviruses integrated in heterochromatin locations, although repetitive rounds of in vitro stimulation may have more effectively reactivated viral replication. 39 However, it remains uncertain to what degree in vitro viral outgrowth assays can serve as an appropriate model for in vivo viral outgrowth, as host immune factors-including virus-specific T and B cells-are typically not included in such assays and results from in vitro viral outgrowth assays do not correlate with kinetics of viral rebound during antiretroviral treatment interruptions. Moreover, the responsiveness of proviruses to reactivation signals is likely to differ profoundly between artificial in vitro culture conditions and physiological in vivo systems. In fact, prior studies have shown that proviral outgrowth in in vitro assays, typically involving stimulation with Phytohaemagglutinin (PHA) and highdose IL-2, can follow a stochastic pattern, 39,54,55 while in vivo viral outgrowth may be more prominently influenced by epigenetic features at the proviral integration site. Therefore, restriction of proviral outgrowth in vivo that is mediated by the proviral chromosomal location and the surrounding epigenetic microenvironment may not be appropriately captured by in vitro viral outgrowth assays.
Whether integrations site profiles of intact proviruses do influence the propensity for viral outgrowth in vivo will, in our opinion, require future prospective clinical studies in which treatment is selectively interrupted in persons in whom a blocked and locked integration site pattern is noted. In this manuscript, we provided a retrospective analysis of integration site locations in persons with post-treatment control during a prior study. 46 Owing to the small number of such persons that have been described in the literature, this analysis was limited to two study persons; however, within these individuals, integration of intact proviruses in repressive heterochromatin was already noted prior to treatment interruption and progressively increased after treatment cessation. In contrast, a more ordinary chromosomal integration site landscape with predominance of intact proviruses integrated in genes was noted in three persons with rapid viral rebound during treatment interruption. Although larger studies will be necessary to further investigate this in the future, our work suggests that it might not only be the presence or absence of antiviral immune responses that determines drug-free control of HIV-1. Rather, we argue that the degree of selection that has occurred in viral reservoir cells may be a distinguishing feature of persons who maintain drug-free undetectable viremia, with or without prior ART. Future technologies designed to evaluate integration sites of much larger numbers of intact proviruses will be informative to determine whether enrichment in heterochromatin locations is also detectable in subdominant intact proviral species that cannot be efficiently captured with currently available experimental approaches but may contribute to viral rebound. Additionally, it will be critical to investigate whether the progressive selection of proviruses in heterochromatin during LT-ART is associated with specific immune mechanisms. Indeed, recent work suggests a role for CD8 T cells in influencing viral transcriptional behavior 56 and an accelerated decline of intact proviruses after treatment with broadly neutralizing antibodies. 57 Collectively, our work reveals strong evidence that HIV-1 reservoir cells are subject to host selection mechanisms that can, over extended periods of ART, profoundly transform the viral reservoir landscape. Whether such selection mechanisms truly act in favor of the host and translate into clinically significant benefits remains to be determined in future clinical studies.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

DECLARATION OF INTERESTS
The authors declare no competing interests.

INCLUSION AND DIVERSITY
One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in their field of research or within their geographical location. One or more of the authors of this paper self-identifies as a member of the LGBTQ+ community.