Cells producing residual viremia during antiretroviral treatment appear to contribute to rebound viremia following interruption of treatment

During antiretroviral therapy (ART) that suppresses HIV replication to below the limit-of-quantification, virions produced during ART can be detected at low frequencies in the plasma, termed residual viremia (RV). We hypothesized that a reservoir of HIV-infected cells actively produce and release virions during ART that are potentially infectious, and that following ART-interruption, these virions can complete full-cycles of replication and contribute to rebound viremia. Therefore, we studied the dynamics of RV sequence variants in 3 participants who initiated ART after ~3 years of infection and were ART-suppressed for >6 years prior to self-initiated ART-interruptions. Longitudinal RV C2V5env sequences were compared to sequences from pre-ART plasma, supernatants of quantitative viral outgrowth assays (QVOA) of cells collected during ART, post-ART-interruption plasma, and ART-re-suppression plasma. Identical, “putatively clonal,” RV sequences comprised 8–84% of sequences from each timepoint. The majority of RV sequences were genetically similar to those from plasma collected just prior to ART-initiation, but as the duration of ART-suppression increased, an increasing proportion of RV variants were similar to sequences from earlier in infection. Identical sequences were detected in RV over a median of 3 years (range: 0.3–8.2) of ART-suppression. RV sequences were identical to pre-ART plasma viruses (5%), infectious viruses induced in QVOA (4%) and rebound viruses (5%) (total n = 21/154 (14%) across the 3 participants). RV sequences identical to ART-interruption “rebound” sequences were detected 0.1–7.4 years prior to ART-interruption. RV variant prevalence and persistence were not associated with detection of the variant among rebound sequences. Shortly after ART-re-suppression, variants that had been replicating during ART-interruptions were detected as RV (n = 5). These studies show a dynamic, virion-producing HIV reservoir that contributes to rekindling infection upon ART-interruption. The persistence of identical RV variants over years suggests that a subpopulation of HIV-infected clones frequently or continuously produce virions that may resist immune clearance; this suggests that cure strategies should target this active as well as latent reservoirs.

Introduction Antiretroviral therapy (ART) has modified the course of human immunodeficiency virus type-1 (HIV) infection from a largely fatal disease to a chronic, treatable infection with a nearnormal lifespan [1,2]. Treatments for HIV infection continue to improve, with less frequent dosing of medicines and fewer adverse reactions. However, despite individuals having HIV replication suppressed by ART for decades, ART does not cure the infection [3,4]. If ART is interrupted, infectious viruses resume replication. Given that cells capable of producing infectious viruses must be eliminated to cure HIV, many studies have focused on understanding the establishment of this reservoir [5][6][7][8] and the mechanisms that allow it to persist [9][10][11][12]. Studies have revealed essentially stable levels of HIV DNA in the blood [13]; although viral variants within the HIV DNA reservoirs appear dynamic [14], and clonally expanded infected cells increase in proportion over time [10,15]. Most infected cells harbor defective proviruses [16][17][18][19][20][21][22][23][24], and during effective ART, cells with infectious proviruses appear to progressively diminish in frequency [25,26], presumably due to lytic infection or elimination by cytotoxic-T-lymphocytes (CTL). An exception includes intact proviruses integrated in regions of the genome that are rarely transcribed [26,27], which are hypothesized to be "deeply" latent. Our findings suggest that another exception may be clones harboring infectious variants that despite the production of residual viruses, resist immune clearance and contribute to rebound viremia.
During ART-suppression, not all proviruses persist in a latent state. Among infected individuals with plasma HIV RNA levels below the limit-of-quantification of clinical assays (<20-50 copies per milliliter (c/mL)), virions are produced (mean 1.5 copies per milliliter), termed residual viremia (RV) [28,29]. RV are of interest as these arise from proviruses geneticallylinked to replication-competent variants from quantitative viral outgrowth assays (QVOA) [30,31], and as reported here, may contribute to rebound viremia. Analyses of RV, therefore, serve as a "window" to a segment of the reservoir capable of producing infectious virions, including variants that may escape clearance by lytic infection or immune elimination.
While current treatment regimens do not prevent the production of virions [28], ART blocks virions from infecting additional cells [15,29,32,33]. Low-levels of HIV replication during ART can be detected by genetic divergence of regions of HIV env and/or pol sequences [29], regions that are under immune and drug selection, respectively. RV sequences generally do not diverge from those found immediately pre-ART [28,34], consistent with suppression of full cycles of virus replication. While unique RV sequence variants have been observed to change in prevalence over time [14,30], RV include viral variants with identical sequences within and across time points [30,31,35]. The detection of an identical RV variant as a majority variant (>50% of total RV sequences) and at multiple timepoints over 2.8 years of ART [14,30] supports the hypothesis that RV are produced by proliferating clones. Biologic observations suggest that a single cell is unlikely to produce RV with identical sequences across multiple timepoints. An HIV infected cell undergoing lytic infection is estimated to release 50K virions [36], which could populate an individual's~3L of plasma or~11L of extracellular fluid with 17 or 4.5 virions/mL, respectively; and since a virion contains two copies of the HIV genome, twice this number would be expected in plasma diluted to single viral templates. However, lytic infection destroys the cell. Thus, as clones of HIV infected cells are known to exist [10,11,17,22], we reason that identical sequences, especially those detected across timepoints, likely come from clonal cell populations. We contend that RV arise from a subset of cells infected before the time of ART-initiation, which have been shown to proliferate and create clonal cell populations that persists during ART [32,33].
The recognition that HIV-infected cells that proliferate during ART [10,11,15,24,29] produce infectious RV [30,31] emphasizes that these clones are a barrier to curing HIV infection. To better understand this "active" HIV reservoir that persists and produces virions during ART-suppression, our study focused on the length of time that RV with identical sequences were produced, and the genetic linkage of these RV to sequence variants induced in QVOA of peripheral blood mononuclear cells (PBMC) collected during ART-suppression, or to plasma variants detected prior to ART-suppression or following ART-interruption. We hypothesized that HIV infected cells that persist during ART-suppression and contribute to RV are clonal and include infectious virus by virtue of detection of genetically identical variants across time, both within QVOA and within rebound viremia following ART-interruption.

Results
RV were genetically characterized by sequencing the C2V5 region of HIV env from plasma diluted to single viral templates (i.e., single genome sequencing, SGS) [37]. Three participants were studied from the time of acute HIV infection over a median of 13.2 years (Figs 1-3). Participants 83747 (hereafter referred to as "1"), 59530 ("2"), and 50877 ("3"), from the Seattle Primary Infection Cohort met inclusion criteria for this study: 1) ART initiation after~3 years of infection; 2) suppression of plasma HIV RNA to <50 copies/mL for �5 years; 3) a period of ART-interruption with documented rebound viremia with study specimens available; and 4) re-initiation of ART following virus rebound (S1 Fig). ART was first initiated after a median of 3.6 years of "chronic" infection ( Table 1). ART-suppression (VL<50c/mL) in Participants 1-3 spanned 8, 7.6, and 6.8 years prior to self-initiated ART-interruptions of 76 days, 20-31 days, and 3 years, respectively.
Plasma specimens with volumes ranging of 3-21 mL were used for SGS, including 3-5 "pre-ART" timepoints, 3-4 during "ART-suppression#1" timepoints (i.e., plasma HIV RNA <50c/mL), 1-3 "ART-interruption" timepoints (i.e., plasma HIV RNA >50c/mL following ART-interruption and shortly after re-initiation of ART), and 1-2 "ART-suppression#2/#3" timepoints (i.e., plasma HIV RNA <50c/mL); Fig 1A, Fig 2A, Fig 3A show when specimens were collected and analyzed. Negatively selected CD4 + T cells from specimens collected 6.6, 7.3, and 3.1 years following ART-initiation from Participants 1-3, respectively, were used in a 28-day QVOA (format shown in S2 Fig), with C2V5env SGS derived from supernatants of p24-antigen (+) wells. The total number of SGS are shown in Table 2. Plasma collected during ART-suppression yielded a mean of 3.3, 1.8, 1.5 SGS/mL from Participants 1-3, respectively (S3 Fig). RV levels were relatively consistent within each participant over time, except for two outlier timepoints when high frequencies of identical sequences were detected (Participant 2 at 7.3 years and Participant 3 at 6.3 years of infection when 90% (36/40) and 89% (8/9) of sequences were identical, respectively). Laboratory quality control included maximum likelihood phylogenetic analyses of all sequences generated in this project (S4 Fig) and within our laboratory; and these found a lack of cross-contamination between the 3 participants' specimens or with viral sequences from other projects.
In phylogenetic analyses, RV and QVOA sequences clustered with multiple pre-ART clades from each participant (Figs 1B, 2B and 3B). To better understand the timeframe when the RV-and QVOA-producing cells "seeded" the HIV reservoir that persists during ART (i.e., when clones were established), pairwise distances were compared between: 1) RV/QVOA sequences from ART-suppression#1 and pre-ART sequences; and 2) between RV sequences from ART-suppression#2/#3 and pre-ART and ART-interruption plasma sequences. The shortest pairwise distance to each unique RV/QVOA sequence was predicted to be the time the clone was established. Cell populations contributing to the RV throughout ART-suppres-sion#1 were estimated to have seeded the reservoir in large part just prior to ART-initiation for all 3 participants (Fig 4), although seeding occurred as early as 28 days following the estimated date of acute infection for Participant 2 (Fig 4C). After a year of ART, a greater proportion of RV were from clones established at earlier timepoints in Participants 2 and 3 (Fig 4C  and 4D). RV from ART-suppression#2/#3 included clones seeded prior to ART-initiation in Participants 1 and 3, and included viruses predicted to use CCR5 and CXCR4 as co-receptors (Fig 4A, 4B, 4D and 4E). However, during ART-suppressions#2/#3, RV were composed primarily of variants that evolved during ART-interruptions (Fig 1B, Fig 2B, Fig 3B), including 100% of RV variants from ART-suppression#3 in Participant 2 and 90% of the R5-tropic RV variants from ART-suppression#2 in Participant 3 (Fig 4C and 4D).
The genetic diversification (S5 Fig) and divergence (Fig 5) of plasma variants from the founder sequences of the 3 participants prior to ART-initiation revealed the expected timeordered increase in genetic distance and diversity. During ART, analyses of all RV sequences

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses suggested continued divergence in Participant 1 (Fig 5A and S5A Fig); however, this was due to the detection of X4 variants, which were not detected prior to ART-initiation. The R5 RV variants across the 3 participants did not diverge (Fig 5). Divergence of plasma sequences resumed upon ART-interruption in Participants 2 and 3 (Fig 5B and 5C).
The prevalence of each unique RV sequence variant fluctuated significantly over time in Participants 1 and 2 (p<0.0001, 2xC Fisher's exact test), demonstrating dynamic shifts in the cell populations producing RV. However, genetic heterogeneity of RV sequences was not found in Participant 3 (p = 0.131) likely due to our derivation of fewer RV sequences. The RV variants that persisted over years were detected as multiple populations of minority frequency variants in the three participants. Two RV variants were detected as majority frequency variants (>50%) at single timepoints across the 9 timepoints evaluated (at 7.3-7.8 and 13.1 years) in Participant 2 (Fig 6B).
To assess whether RV or rebound C2V5env sequences were glycosylated, which could promote escape from neutralizing antibodies and contribute to persistence [38, 39], the number

Discussion
Residual viremias provide a window into the segment of the HIV reservoir that actively produces virions during ART-suppression. Our longitudinal analyses of RV sequences reveal that this "active" HIV reservoir is dynamic, and that a subset of RV are genetically linked to ARTsuppression viruses induced in QVOA and/or ART-interruption viruses in "rebound" plasma; hence to the infectious HIV reservoir. These findings indicate that, in addition to intact proviruses in the latent reservoir, the cells that produce RV appear to be a critical target for cure strategies.
Genetic distances and sequence identity between RV, QVOA, and rebound viruses have been examined in multiple studies to gain insights into the infectious reservoir. We found relatively few identical variants between RV and QVOA and few between RV and rebound viruses, which aligns with others' findings [30, 35, 40] and [22], respectively. While our detection of RV that shared identical sequences with replicating viruses detected in QVOA and rebound viruses were not common, these observations support the hypothesis that the infectious reservoir that contributes to rebound viremia also produces RV. We detected identical sequence variants, albeit relatively few, between QVOA-derived and rebound viruses; similarly, we observed that many QVOA and rebound viral sequences intermingled in phylogenetic analyses, suggesting shared genetic lineages. Studies by others [41] have sought but not found sequence linkages between these virus populations, with one study reporting that viral outgrowth assays (VOA) and rebound viruses clustered separately in phylogenetic analyses of twelve individuals' sequences [41]. The linkages between QVOA-derived and rebound viruses in our study may reflect greater sampling of an extensive viral reservoir compared to the latter study. Overall, our study appears to provide novel evidence that clonal cell populations within

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses the persisting HIV reservoir produce RV and infectious variants that contribute to rebound viremia upon ART-interruption.
The HIV-infected cell reservoir producing RV and viruses detected in QVOA in our participants appeared to have been seeded throughout the pre-ART period. The genetic distances between RV and pre-ART viruses appeared to transition across the time of ART-suppression, progressing from RV genetically similar to variants just prior to ART-initiation to variants from across the pre-ART period. This transition is consistent with the idea that the majority of RV detected immediately after ART-initiation come from recently infected cells; however, following the expected decay of these cells [33, [42][43][44][45], a greater proportion of RV is then detected from long-lived cell populations. Others' examinations of the time for reservoir seeding include a recent study of viral sequences that predicted seeding of the infectious reservoir occurs across the pre-ART period [46]; similar to findings in our participants. Another group observed seeding of the majority of on-ART QVOA-derived viruses 1 year prior to ART-initiation [42]. This latter pattern suggests that the population of cells replicating immediately prior to ART did not decay rapidly, and suggests that in the individuals studied that this population of infected cells was perhaps renewed due to imperfect adherence to ART. Our observation

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses The time of pre-ART (red gradient colors) and ARTinterruption (blue gradient colors) that HIV replication is predicted to have seeded the reservoir that persisted during ART-suppression and produced each unique RV or QVOA-derived virus is shown with different gradient colors. A total of 4, 3, 4 pre-ART and 1, 1, 3 ART-interruption timepoints were used to infer reservoir seeding that following ART-interruption and re-suppression (i.e., during ART-suppression-#2/3), the RV included new variants not detected during the ART-suppression-#1, provides data in support of the hypothesis that recent HIV replication partially resets and/or obscures the previously seeded long-lived reservoir. The prevalence of each RV variant fluctuated across ART-suppression. Identical RV sequence variants observed to persist over time, were detected over a median of 3.9 years, including after ART-interruption and during ART-re-suppression. The recurrent detection of identical RV sequences at multiple timepoints is consistent with the production of RV by clonal cell populations; as one cell would be unlikely to persist over this timespan with repeated production of virions at a burst size [36] required to produce RV. Our finding that RV sequence variants fluctuate over time confirm those of others' studies [14,30,40,44]. This variation in the representation of RV variants may be due to diverse antigens periodically stimulating different clonal cell populations that then produce virions [14], differential CTL immune recognition and elimination of RV-producing clones [24, 47,48], and/or a gain or loss of gene function due to the location of the HIV integration site [49]. The specific clone producing persistent viremia during ART was previously identified by viral sequence and HIV integration site in an individual with cancer, with the provirus integrated in a repetitive element [22], which impedes the study of virally modulated gene expression. Given that we detected multiple RV variants that persisted over years, we hypothesize that clones producing RV are selected either for escape from CTL recognition or dysfunctions in immune elimination; both paths to "immune evasion" imply that RV-producing clones represent a particularly challenging but critically important reservoir to target for cure strategies.
The contribution of RV variants to rebound viremia was not associated with whether the clone was detected frequently or rarely among RV sequences. This suggest that multiple or disparate factors likely contribute to RV production and viral rebound, or that our sampling was too sparse to detect weak associations.
While previous studies of low-level intermittent viremias (plasma HIV RNA 50-400 c/mL) [29, 50] occasionally detected evidence of HIV replication, genetic divergence of our participants' sequences was detected only after frequent or prolonged ART-interruptions. No significant increase in viral divergence was observed among plasma sequences imputed to use the CCR5 co-receptor during the first period of ART suppression in all 3 participants, suggesting the lack of ongoing viral replication during effective ART. However, the absence of detectable divergence despite viremia during ART-interruptions of Participants 1 and 2 demonstrates that divergence is insensitive to short periods of low-level virus replication, and emphasizes that a lack of divergence does not provide certainty that HIV replication did not contribute to RV. The detection of multiple RV sequences predicted to use the X4 co-receptor that cluster closely in the phylogenetic analyses of Participant 1, and in multiple individuals in previous studies of low-level viremias [29, 50] leads us to speculate that X4 tropism is selected for persistence and/or replication during ART-suppression.
The increase in PNLGS in the RV of two participants provide a weak suggestion that escape from antibodies may select HIV variants that persisted and were expressed during ART. However, the significance of this observation is uncertain given our small study population and the absence of a pattern across the three participants. time for the RV/QVOA-producing proviruses in Participants 1-3, respectively. ART-interruption timepoints for Participants 1 and 2 included 1 timepoint after ART reinitiation when viral levels were above the limit-of-quantification of our assay (50c/mL). Red dotted line indicates the timepoint ART was initiated. Black solid line indicates the timepoint ART was interrupted. Total (n) number of unique R5 and X4-tropic RV and QVOA-derived sequences plotted are shown above the bar. https://doi.org/10.1371/journal.ppat.1008791.g004

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses Given the extremely low levels of viremia under study, only a limited, albeit highly variable region of the viral genome (the~600bp C2V5 region of env) was assessed, which may have overestimated "identical" populations due to sequence differences outside of this region. To increase the power of this analysis, we studied individuals likely to have substantial viral diversity when ART was initiated, i.e., approximately three or more years of untreated infection. The Clonal Prediction Score Calculator [51] (https://silicianolab.johnshopkins.edu/cps/#/) estimated that identical sequences over the examined region of the viral genome provided an 83% certainty that the sequences were identical throughout the genome of persons who initiated ART during chronic infection. Our QVOA studies only sampled infectious viruses induced from PBMC. Examination of other tissues where replication is hypothesized to occur, such as lymphatic tissues [52], may have provided further insight as to the viral variants contributing to rebound following ART-interruption.
Our findings suggest that despite years of effective ART, certain cell clones expanded by proliferation are selected to persist and some of these produce RV, even after ART-interruption. The persistence of these virus-producing clones for at least 8.2 years during effective ART implies that immune elimination of infected cells may be hampered due to CTL-escape mutations or dysfunctional immune responses. Multi-faceted strategies that use immunotherapies may be needed to cure HIV infection. These would need to include strategies to target proviruses that escape CTL via different mechanisms, including latent proviruses integrated in "silenced" regions of the human genome, and latency-reversed or RV-producing cells not effectively targeted by host CTL responses.

Study participants
This study was a retrospective observational study based on prospectively-followed participants from the Seattle Primary Infection Clinic (PIC) cohort who were enrolled from 1992-2015 [53][54][55][56]. Participants were selected based on the following inclusion criteria: 1) initiated ART during chronic infection; 2) ART-suppressed for �5 years with all plasma HIV RNA loads <50c/mL; 3) multiple plasma specimens available prior to and during ART-suppression; and 4) a treatment interruption with plasma specimens available following re-initiation of treatment. At the time of cohort entry, all participants had either acute or early HIV infection [56]. ART was initiated based on the availability of potent antiretroviral therapy, the strength of consensus treatment recommendations, and personal preferences. Self-initiated treatment interruptions for the 3 participants included in our study were for reasons including insurance lapse and a decision made in conjunction with a clinical care provider.

Ethics statement
The University of Washington Institutional Review Board approved this study, and all participants provided written consent. Participants initiated treatment under the care of their clinicians as recommended by guidelines and personal preferences.

Study schema
Participants' study visits and specimen collections were as frequent as weekly for the month following enrollment, monthly for three months, at eight-week intervals through the differences ( � p<0.05) in divergence of the RV/QVOA-derived viruses relative to the first RV timepoint as determined by 2-sample Wilcoxon rank-sum test are indicated. Also indicated are significant differences between the ART-interruption and RV after re-suppression relative to the latest RV timepoint during ARTsuppression #1. https://doi.org/10.1371/journal.ppat.1008791.g005

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses remaining first year of observation, and then at a minimum of six-month intervals, although the frequency varied somewhat within the cohort. Multiple longitudinal plasma specimens were used to derive viral RNA and 1 timepoint during ART suppression (6.5, 7.25, 3.08 year after ART initiation), where leukapheresis specimens were available, was used for the QVOA for Participants 1-3, respectively.

Plasma RNA extraction
RNA was extracted from 1-21mL aliquots of participants' plasma or 50-140uL of QVOA supernatants. When plasma contained HIV RNA at <50c/mL, virus particles were pelleted by centrifugation at 14,000rpm for 90 min at 4˚C, with removal of all but~140uL of plasma over the pellet. The Qiagen QIAmp Viral RNA Mini kit (Qiagen, Catalog # 52906) was then used as directed by the manufacturer to extract the RNA. When viruses were pelleted, lysis buffer was added directly to the pellet and remaining plasma and once resuspended, was transferred to a new 1.5mL Eppendorf tube for the extraction, followed by~500uL of PBS used to wash any remaining virions from the tube. A positive control plasma sample containing >100,000 copies/mL of HIV RNA was similarly processed with each extraction. RNA was eluted into 60uL of AVE buffer (Qiagen) and immediately DNase-treated to degrade contaminating HIV DNA using Invitrogen DNase I, Amplification Grade (Catalog # 18068-015) following the manufacturer's instructions (1uL of 10X buffer was added to the eluted RNA followed by 1uL DNase I). The mixture was incubated at room temperature for 15 min, the enzyme was then deactivated with the addition of 1uL 25mM EDTA and a 10 min incubation at 65˚, and then put on ice. RNA from the QVOA supernatants were similarly DNase I-treated with the exception of 6/28 (21.4%) QVOA p24Ag(+) wells from all 3 participants when DNase I was not available. RNA was immediately converted to cDNA (see below).

cDNA synthesis
All 60uL of DNase-treated RNA was converted into cDNA using the Invitrogen Superscript IV (SSIV) enzyme kit (Catalog # 18090200). Briefly, to the 60uL RNA, 5.5uL of 10mM dNTP, 5.5ul of each of the 20uM of HIV pol-specific forward RTA primer (HXB2 positions 3328-3303) and env-specific reverse BH2 primer (HXB2 positions 7697-7725) were added and the mixture incubated in a thermocycler at 65˚for 5 min and then immediately put on ice for at least 1min. Subsequently, 2.75uL of molecular grade water, 22uL 5X RT buffer (Invitrogen, catalog # 18090200, 5.5uL 0.1M DTT, 2.75uL Promega RNasin Plus RNAse inhibitor, 10,000U (Catalog # N2615), and 1uL of the SSIV enzyme was added and incubated at 55˚for 90min, 80˚for 10min, and then held at 12˚prior to storing at -20˚.

DNA sequencing and phylogenetic tree analyses
The~610 bp C2V5env gene amplicons from the 2 nd round nested PCR were directly sequenced by the Sanger method using DR7 and DR8new or env7656 primers. All chromatograms were evaluated to ensure the lack of any double peaks and thus confirm single genome sequences (sequences with >1 double peak were eliminated from analyses). The forward and reverse sequences were aligned to a reference HXB2 sequence (Genbank, K03455.1) using Geneious R11.1.5 (https://www.geneious.com) and base calls reviewed for bi-directional consistency. Codon-based DNA sequence alignments were generated in Geneious using the MUS-CLE algorithm (version: 3.8.425) [61] and manually refined. Sequences with significant levels of APOBEC-induced G-A hypermutations were determined by Hypermut 2.0, LANL [62]. UniSeq (https://indra.mullins.microbiol.washington.edu/cgi-bin/UniSeq/uniseq.cgi) was used to remove identical sequences from the alignment prior to creating phylogenetic trees in DIVEIN [63] (https://indra.mullins.microbiol.washington.edu/DIVEIN/diver.html) using the GTR substitution model, and best of NNI and SPR tree improvement, and bootstrapping. MEGA version 7 [64] was used to format the output phylogenetic tree from DIVEIN. DIVEIN [63] was used to estimate pairwise distances and divergence of unique sequences relative to the imputed founder. C2V5env sequences that were significantly hypermutated by Hypermut 2.0, LANL or contained stop codons were removed from analysis. Mean and standard deviations were determined using GraphPad Prism v6.07. Two-sample Wilcoxon rank sum test was performed to evaluate significant changes in RV divergence comparing the first ART-suppression timepoint to each subsequent RV timepoint. Two-sample Wilcoxon rank sum test was also performed comparing sequences at the latest RV timepoint during ART-suppression #1 relative to all subsequent timepoints.

Reservoir "seeding"
For each unique ART-suppressed sequence, the distance between that sequence and each pre-ART and ART-interruption sequence from the same participant was examined, using pairwise

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses distances obtained from the divergence analysis above. The estimated time of seeding of the provirus contributing to the RV was estimated as the pre-ART or ART-interruption timepoint with the smallest pairwise distance relative to the RV. In case of a tie, the seeding time was estimated to be the earliest time at which we observed pre-ART sequences tied for the smallest distance.

Cellular tropism
To predict viral co-receptor use, or "tropism", the plasma and QVOA viral sequences were codon-aligned relative to the V1-V3 region (7,110-7,218bp) of HXB2 (Genbank, K03455.1). Any insertions found in the alignment of plasma virus sequences and sequences of viruses induced from QVOA that were not found in HXB2, were removed from tropism analysis performed using PSSM [65]. Sequences with a percentile >95 or that contained stop codons were excluded from tropism analysis. Tropism predictions are reported as previously described [65]; scores � -7 as CCR5-tropic, -7 to -3 as dual-tropic, and � -3 as CXCR4-tropic.

N-linked glycosylation
Unique C2V5env nucleotide sequence alignments were input into the N-glycosite program [66] and evaluated using the default settings. Output files were used to create the bar graphs in S7 Fig. Sequences that were significantly hypermutated or contained stop codons were removed from alignments.

Statistical analyses
Fisher's exact test was used to determine whether the distribution of observed variants during ART suppression was homogeneous over time. For each participant, we determined the most common variant (MCV) by pooling variant counts during the ART suppression period. In the case of a tie, the MCV was the variant with the greatest number of counts during the entire study period. Each participant thus contributed a 2xC contingency table, where C is the number of timepoints (non-ordered) during ART suppression for that participant, and observations at each timepoint were dichotomized as being MCV or non-MCV.
To determine the degree to which viruses from pre-ART contributed to residual viremia during ART suppression and re-suppression, we determined the proportion of sequences seen during ART suppression or re-suppression that were also seen pre-ART for each participant. We calculated 95% exact binomial confidence intervals for each proportion reported.

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses HXB2 N-linked glycosylation sites (n = 12 sites) [67]. Red and green boxes represent the V3 and V4 regions in the HXB2 C2V5env sequence, respectively. Only unique C2V5env gene sequences that did not contain stop codons and that were not significantly G-A hypermutated are shown. Panels D-F show the mean +/-standard deviation of N-linked glycosylation sites among all unique plasma and QVOA sequences. The number of unique sequences included in the analysis is shown in the key. � p<0.05, 2-sample Wilcoxon rank sum test, adjusted for multiple comparisons using the Holm method, was performed to compare levels of glycosylation among RV during ART-suppression #1 relative to sequences from all other timeframes.

PLOS PATHOGENS
HIV residual viremias linked to replicating viruses