Tissue Origins and Immune Correlates of Viral Rebound in SIV-Infected Rhesus Macaques Following ART Discontinuation

Abstract

The majority of persons living with HIV-1 who discontinue antiretroviral therapy (ART) demonstrate viral rebound, but the tissue-level events that lead to rebound viremia remain poorly understood. Here we report the origin, dynamics, and correlates of viral rebound in 16 rhesus macaques (RMs) infected with molecularly barcoded SIVmac239M, treated with ART for 70 weeks, and necropsied on day 12 after ART discontinuation. Barcode analysis of plasma during daily post-ART sampling identified 1 to 38 rebounding viral lineages per animal, with a mean of 2.4 lineages contributing to initial rebound viremia. Analysis of barcode viral RNA expression in tissues revealed presumptive anatomic origin sites for 56 of 175 rebounding viral lineages, with enrichment in the gastrointestinal (GI) tract and GI-associated lymph nodes by mixed-effects logistic regression. Daily transcriptomic and proteomic profiling in peripheral blood following ART discontinuation showed up-regulation of pathways related to T cell signaling, cytokine responses, and cellular metabolism prior to detectable rebound viremia. These data show that viral rebound is initiated by oligofocal tissue expansion of a limited number of clonal lineages, followed by systemic dissemination and serial emergence of additional lineages from multiple tissues. Longer time to viral rebound was associated with metabolic, epigenetic, and immunologic signatures that suppress proviral reactivation. These findings advance our understanding of the tissue origins and host correlates of viral rebound and provide a framework for future studies examining GI-associated reservoir compartments, peripheral blood biomarker development, and host-directed strategies to prolong ART-free viral remission.

One Sentence Summary:

Gastrointestinal tissues harbor simian immunodeficiency virus lineages that rebound after antiretroviral therapy discontinuation in macaques.

INTRODUCTION

The replication-competent viral reservoir (RCVR) that persists despite extended antiretroviral therapy (ART) leads to viral recrudescence following ART discontinuation and remains the key obstacle to achieving an HIV-1 cure. The RCVR consists of latently infected CD4 T cells primarily in lymphoid tissues and is established early during acute infection and is not eliminated by ART, resulting in viral recrudescence following ART interruption in the vast majority of individuals (1, 3–5). However, the origin and pathogenesis of viral rebound following ART discontinuation is not well understood.

Current methods for quantifying virus that persists on ART do not accurately predict viral rebound due to the high prevalence of defective proviruses, incomplete latency, and additional confounding factors (1–5). Studies in humans with HIV-1 are typically limited by incomplete access to tissues, infrequent sampling, and assays primarily in peripheral blood, which do not provide a detailed assessment of tissue-level viral dynamics. It currently remains unclear if viral rebound primarily reflects stochastic reactivation of a limited number of provirus clones, simultaneous reactivation of multiple provirus clones, limited localized viral replication in permissive tissue or host microenvironments, or rapid systemic dissemination following initial reactivation events. Elucidation of the mechanisms and pathogenesis of viral rebound will address this gap in knowledge and will provide insights for next generation of HIV-1 cure strategies.

Nonhuman primate (NHP) models can address limitations of human studies by enabling controlled timing of infection and ART initiation, knowledge of the exact sequence of the challenge virus, frequent longitudinal sampling, and comprehensive tissue collections. Limitations of NHP models include the typically shorter duration of suppressive ART compared with humans as well as intrinsic differences between SIV and HIV-1 (3, 4). In addition, ART was initiated early in this study (day 6 to 12 following infection), which resulted in relatively low magnitude adaptive immune responses (1). Although this model parallels early treated HIV infection in humans, the tissue microenvironments driving viral reactivation in the context of minimal adaptive immunity may differ from those present in individuals who initiate ART later and develop more robust virus-specific responses. Using molecularly barcoded SIVmac239M (6, 7), we investigated viral rebound dynamics in rhesus macaques (RMs) following discontinuation of ART (6). Molecular barcoding involves genetic tags in the viral genome that do not impact viral fitness, enabling tracking of individual viral clonotype lineages across tissues and over time. Prior studies using barcoded SIV have demonstrated broad dissemination of viral lineages across lymphoid and mucosal tissues during acute infection with only a subset of these lineages contributing to rebound viremia following ART discontinuation (6, 7). This model provides an opportunity to interrogate the anatomical origins and tissue dynamics of rebounding viral lineages.

In this study, we defined the timing and number of barcode-defined viral lineages that contributed to viral rebound following ART discontinuation and identified presumptive anatomic origin sites for a subset of the rebounding clonal lineages. We also used by multi-omic analyses to identify early host inflammatory and metabolic pathways associated with viral rebound as well as signatures that define the time to viral rebound of peripheral blood following ART discontinuation. These analyses integrated viral lineage tracking with systemic host responses to provide a detailed picture of the origins and correlates of viral rebound.

RESULTS

ART suppressed viral replication and blunted SIV-specific immune responses in a manner dependent on timing of treatment initiation

Eighteen Indian-origin RMs were intravenously infected with molecularly barcoded SIVmac239M and randomly assigned to initiate ART on day 6, 9, or 12 post-infection (n = 6 per group), modeling progressively longer period of untreated acute infection before therapy (Fig. 1A). Animals remained on ART for 70 weeks with longitudinal virologic monitoring. Following ART initiation, plasma viremia was suppressed to <15 copies/ml at most timepoints after day 120 for animals in the day 6 and day 9 ART initiation groups (Fig. 1, A and B). Median time to viral suppression was 37 days in the day 6 group, 101 days in the day 9 group, and 265 days in the day 12 group, with day 12 animals requiring significantly longer to achieve suppression than day 6 animals (P = 0.0012), consistent with high plasma viral loads prior to ART initiation (Fig. 1, C to E). Occasional on-ART blips were also observed, particularly among animals in the day 12 group. Two RMs, one each from the day 6 (DHTG) and day 9 (DHPG) groups, were euthanized for procedure-related complications during the ART treatment period. Viral DNA (vDNA) and viral RNA (vRNA) in peripheral blood mononuclear cells (PBMCs) declined during ART, with vRNA showing a steeper decline over time than vDNA (Fig. 1, F and G). Intact proviral DNA in PBMCs, detected using an SIV intact proviral DNA assay (IPDA), was lower in RMs that initiated ART on day 6 compared with RMs that initiated ART on day 12 at both week 36 (P = 0.0013) and week 66 (P = 0.0019) post-infection (fig. S1) (4, 8). On-ART biopsies revealed early and widespread detection of vDNA and vRNA in lymphoid and GI tissues in all groups (Fig. 1, H and I; fig. S2).

Fig. 1. Early ART initiation limited peak viremia and accelerates viral suppression but did not prevent establishment of cell-associated viral reservoirs in lymphoid tissues of SIVmac239M-infected RMs.

(A to C) Plasma viral loads (PVL) are shown in individual animals infected with SIVmac239M and treated with ART beginning on day 6 (n = 6; red; A), day 9 (n = 6; blue; B), or day 12 (n = 6; green; C). Vertical dashed line is the time of ART initiation; horizontal dashed line indicates the limit of detection of 15 copies/ml. (D and E) Peak viral loads (D) and time to plasma viral loads <15 copies/ml (E) across groups. (F and G) Levels of cell-associated vDNA (F) and cell-associated vRNA (G) in individual animals were measured in PBMCs. (H and I) Levels of vDNA (H) and vRNA (I) in mesenteric and right axillary LN biopsies collected three days after ART initiation in each respective group. Black lines indicate mean ± SD. *P < 0.0332, **P < 0.0021; Kruskal-Wallis test.

We next assessed SIV-specific antibody and T cell responses during ART suppression. ART initiation at day 6 resulted in lower SIV Env-specific antibody titers than ART initiation on day 12 at week 66 post-infection (P = 0.0092; fig. S3, A to D). SIV-specific cellular immune responses were evaluated using interferon (IFN)-γ ELISPOT and intracellular cytokine staining assays. SIV-specific T cell responses were undetectable throughout follow-up in the day 6 ART initiation group and in most day 9 and day 12 ART initiation RMs at week 33 post-infection (fig. S3E), but Gag-specific responses were observed at week 66 in a subset of day 9 and day 12 ART initiation RM (fig. S3, F and G). At necropsy, Gag-specific IFN-γ⁺ and tumor necrosis factor (TNF)-α⁺ CD4 T cell responses were detected in the spleen in most animals (fig. S3, H and I). The low magnitude of SIV-specific T cell responses are consistent with observations in humans treated during acute HIV-1 infection, as early ART initiation limits antigen exposure and limits the development of virus-specific cellular immunity (1). SIV Env-specific antibody titers did not increase detectably following ART discontinuation (fig. S3B), which likely reflects the short duration of antigen exposure prior to necropsy and is similar to prior studies (1). These data suggest that later ART initiation may result in greater humoral and cellular immunity during suppressive ART (9).

Viral rebound involved oligoclonal lineages established during primary infection that actively replicated in tissues following ART discontinuation

At week 70, ART was discontinued in all 16 RMs, and plasma viral loads were monitored daily until day 12 following ART discontinuation to assess viral rebound dynamics. Two animals (TP5, DHHN) did not rebound by day 12 (Fig. 2A). Among the 14 animals that showed plasma viremia, time to viremia >50 copies/ml was observed between 6 and 12 days after ART discontinuation (Fig. 2A; fig. S4). The mean time to rebound viremia (>50 copies/ml) was 8.0 ± 2.2 days for animals initiating ART on day 6, 8.0 ± 2.2 days for the day 9 group, and 8.5 ± 2.4 days for the day 12 group (mean ± SD), with no differences between groups. Rebound viremia at necropsy was <10³ copies/ml in 3 of the 14 viremic RMs and was between 10³ and 10⁷ copies/ml in the remaining 11 RMs. Calculated plasma viral load rebound growth rates varied for individual RMs (mean 1.75, range 1.10 to 2.65 log RNA/day), with no significant differences between groups (Kruskal-Wallis test; 0.22; table S1).

Fig. 2. Viral rebound occurred in all day 12 ART-initiation animals and most early-treatment animals, with vRNA and vDNA persisting broadly across GI and lymphoid tissues at necropsy.

(A) PVL for 12 days following ART discontinuation at week 70 are shown for the day 6 group (n = 5; 4/5 rebounders), day 9 group (n = 5; 4/5 rebounders), and day 12 group (n = 6; 6/6 rebounders). The limit of detection is 15 SIV RNA copies/ml. (B and C) Tissue distribution of vDNA (B) and vRNA (C) at necropsy is shown as the number of viral copies per million cells (n = number of independent tissue samples per animal). Unfilled circles represent samples in which no replicates were positive for vDNA or vRNA; an imputed threshold value was plotted to indicate levels below the detection limit. Tissue comparisons by Wilcoxon rank-sum test. No differences between groups were observed.

Sequencing of rebound plasma virus showed 1 to 38 barcode lineages per animal at necropsy (fig. S4; table S1) with 1 to 4 lineages (mean 2.4) contributing to the initial timepoint of measurable rebound viremia. A total of 175 rebounding lineages were detected across all animals at necropsy. Individual RM growth curves and barcode proportions were used to calculate the average reactivation rate per animal (table S1) (6, 10). The overall rate for all animals was 1.4 (range 0.12 to 5.67) reactivation events per day that led to detectable rebound viremia, with no significant differences between groups (Kruskal-Wallis; P = 0.34).

Total vDNA and vRNA quantitation as viral copies per 10⁶ total cell equivalents was performed for tissue samples obtained at necropsy on day 12 following ART discontinuation. In tissue compartments, vRNA quantified total cell-associated SIV RNA detected by RT-qPCR (including polyadenylated unspliced and spliced transcripts). A standardized tissue collection protocol was used to obtain a median of 62 (range 51 to 85) independent tissue samples from each animal, including GI tissues, GI-associated and other lymphoid tissues, as well as non-GI/non-lymphoid tissues (table S2). Viral quantitation revealed broad distribution of vDNA (Fig. 2B) and vRNA (Fig. 2C) across multiple tissues, including GI tract and lymphoid tissues, with low levels of vDNA in non-GI/non-lymphoid tissues, such as reproductive organs, lungs, liver, and brain.

Barcode sequencing of vDNA and vRNA was employed to assess individual viral lineages replicating within necropsy tissues, with barcode sequences obtained from a mean of 39 (range 11 to 45) samples per animal, making it possible to demonstrate lineages identified in plasma viremia from primary infection prior to ART initiation as contributing to viral rebound in plasma and also present in tissues following ART discontinuation. Consistent with previous studies and a companion study (6, 10, 11), the representation of individual barcode lineages in plasma viremia during primary infection was predictive of the probability of those barcode lineages being represented in rebound viremia (fig. S5; table S3; P < 0.001, two-sided Wald test). To assess if rebounding barcode lineages showed evidence of viral replication in tissues compared to barcode lineages not identified in rebound viremia, we compared the cumulative vRNA and vDNA levels for all individual barcodes in all necropsy tissues to the contributions of individual barcodes in primary peak viremia, with rebounding lineages (heat map color) readily distinguishable from non-rebounding lineages (gray) (fig. S6). Rebounding barcode lineages had significantly higher total tissue vRNA and vDNA levels compared to non-rebounding lineages (log₁₀ vRNA median 4.9 vs. 1.9; vDNA median 3.6 vs. 1.7; P < 2×10⁻¹⁶, Wilcoxon rank-sum test). Furthermore, total vRNA and vDNA levels in tissues for individual barcodes predicted rebound viremia levels for those barcodes (table S4; P < 0.001, P < 0.001, two-sided Wald test), providing evidence that individual barcode lineages actively replicating in tissues following ART discontinuation contribute to rebound viremia.

Rebounding barcode lineages were enriched in GI tissues and GI-associated lymph nodes

The use of a barcoded virus enables the possibility of not only tracking the contributions of individual barcode clonotypes to rebound viremia, but also the identification of presumptive tissue sites of origin of the rebounding barcodes. In a companion study (11), RMs necropsied on ART were used to define the limits of barcode-level vRNA expression on ART and to establish a 99% predictive interval for vRNA expression in the absence of viral replication. Analysis of tissues from RMs necropsied 5 or 7 days after ART discontinuation revealed rare tissues with barcode vRNA exceeding this 99% prediction interval, identifying sites of initial viral replication after ART discontinuation. When those barcodes matched lineages detected during the earliest stages of rebound viremia (<30 copies/ml), the tissue origin site for these lineages could be identified.

In the present study, necropsies were performed 12 days after ART discontinuation with rebound plasma viremia <10³ copies/ml in 3 of the 14 viremic RMs and between 10³ and 10⁷ copies/ml in 11 of 14 viremic RMs, often with vRNA expression of barcode lineages present in rebound plasma viremia identified in multiple tissues. At necropsy, multiple rebounding barcode lineages were observed in the majority of animals, and inference of tissue origin was thus limited to the subset of barcodes with localized tissue enrichment of vRNA prior to widespread dissemination and broad re-infection of tissues. To infer presumptive tissue origin sites, we employed a machine-learning clustering approach to identify outlier tissue sites where vRNA expression compared to other tissues also expressing the same rebounding barcode vRNA exceeded the level defining disseminated replication (determined from non-outlier tissues of all rebounding barcodes), thereby implicating those individual tissue specimens as presumptive tissue origin sites (Figs. 3 to 5; figs. S7 to 17).

Fig. 3. A single rebounding barcode lineage in RM J639 was present across GI tissues and GI-draining lymphoid tissues at necropsy, whereas non-rebounding RM DHHN showed no detectable plasma rebound viremia over 12 days of post-ATI monitoring.

Data from n = 1 representative animal per category (non-rebounder: RM DHHN; single-barcode rebounder: RM J639); see fig. S7 to S17 for all viremic animals. (A) PVL dynamics following ATI for RM DHHN (top) and RM J639 (bottom); red lines indicate individual viral load trajectories. RM DHHN did not show viral rebound during the 12-day monitoring period. (B) Proportional distributions of viral barcode clonotypes in plasma viremia during primary infection prior to ART initiation for RM DHHN (top) and RM J639 (bottom). For J639, the barcode detected in rebound plasma (bottom left panel) is highlighted in red; the bottom right panel shows calculated rebound viral growth curves for the rebounding barcode lineage, with estimated time to a single copy in rebound viremia indicated. (C) Necropsy tissue distribution of vRNA and vDNA barcode clonotypes from primary infection for each animal. For RM J639 (bottom), the plasma rebound barcode (BC.2497) was detected in GI and lymphoid tissues; no presumptive tissue origin site was identified by DBSCAN-based outlier analysis. Grouped tissue categories: GI tract (blue), GI tract-draining LNs (red), non-GI lymph tissues (green), non-lymphoid tissue (purple). Tissue-associated SIV vRNA and vDNA values are expressed as copies per 10⁶ total cell equivalents. No formal statistical comparisons performed; data represent individual animals.

Fig. 5. Multiple rebounding barcode lineages in RM L681 arose from distinct GI and lymphoid tissue reservoirs, with the dominant lineage (BC.4653) contributing the largest proportion of rebound viremia.

Data from n = 1 animal (RM L681). (A) Plasma viral load dynamics following ATI; red line indicates the viral load trajectory of RM L681. (B) Left panel: Proportional distribution of viral barcode clonotypes in plasma viremia during primary infection prior to ART initiation, with barcodes found in rebound plasma highlighted. Middle panel: Rebound viral growth curves of each rebounding barcode lineage with estimated time to a single copy in rebound viremia indicated; red line indicates the dominant rebounding lineage (BC.4653); gray lineages correspond to clones detected in rebound plasma but without an identified presumptive tissue origin site. Right panel: Proportional distribution of rebound viral barcode clonotypes in necropsy plasma, with barcodes for which a tissue origin site could be identified highlighted. (C) vRNA and vDNA distribution of all rebounding barcodes (color-coded to panel B) in necropsy tissues. Open symbols indicate barcodes without an identified tissue origin site. Colored upward-facing triangles represent rebounding barcodes with a tissue origin site indicated by the large, filled symbol. Grouped tissue categories: GI tract (blue), GI tract draining LNs (red), non-GI lymph tissues (green), non-lymphoid tissue (purple), and blood (black). (D) Viral barcode SIV RNA copies in tissue sample grouped by barcode ID. Each point represents an individual barcode detected in the indicated tissue. Tissue-associated SIV vRNA and vDNA values are expressed as copies per 10⁶ total cell equivalents. Tissue origin sites identified by DBSCAN-based outlier analysis (see materials and methods).

Two animals (DHHN, TP5) did not exhibit plasma viral rebound and also did not have sufficient vRNA levels in tissues on day 12 following ART discontinuation for outlier analysis (Fig. 3A to C, top; fig. S7, top). Two animals (J639, L970) showed plasma viral rebound on day 11 following ART discontinuation with a single barcode vRNA in plasma viremia at necropsy (Fig. 3A, bottom; fig. S7, bottom). Animal J639 showed the same barcode (BC.2497; red) in primary viremia prior to ART initiation, in rebound plasma viremia on day 12 following ART discontinuation, and in GI and lymphoid tissues at necropsy, but no tissue origin site was identified (Fig. 3A to C, bottom). Animal L991 showed initial plasma viral rebound on day 12 following ART discontinuation with two barcodes (BC.897, BC.2604) that were identified in peak primary infection plasma and also detected in rebound plasma viremia (fig. S8). Tissue analysis detected vRNA and vDNA for both of these barcodes in multiple tissues.

Animal L604 showed measurable plasma viremia on day 6 following ART discontinuation (Fig. 4A). Three barcodes were identified in rebound viremia with a level of 180,000 copies/ml at necropsy (Fig. 4B). Barcode analysis of primary infection showed that these three barcodes were relatively abundant in primary viremia prior to ART initiation (Fig. 4B), although the order of detectable barcode emergence in plasma following ART discontinuation did not correspond to their abundances in plasma prior to ART initiation. Necropsy tissue analysis revealed that the first barcode detectable in rebound viremia (red, BC.15) was already widely disseminated, whereas the second (blue, BC.4095) and third (purple, BC.1192) barcodes identified in rebound viremia were more localized to the lower jejunum and perisplenic lymph node (LN), respectively (Fig. 4C), which were identified as tissue origin sites for the second (blue, BC.4095) and third (purple, BC.1192) detectable rebounding lineages, respectively (Fig. 4D).

Fig. 4. Two of three rebounding barcode lineages in RM L604 could be traced to tissue origin sites in the lower jejunum and perisplenic LN by DBSCAN-based outlier analysis.

Data from n = 1 animal (RM L604). (A) Plasma viral load dynamics following ATI; red line indicates the viral load trajectory of RM L604. (B) Left panel: Proportional distribution of viral barcode clonotypes in plasma viremia during primary infection prior to ART initiation, with the 3 barcodes found in rebound plasma highlighted. Middle panel: Calculated rebound viral growth curves for each rebounding barcode lineage with estimated time to a single copy in rebound viremia indicated; red line indicates the dominant rebounding lineage. Right panel: Proportional distribution of rebound viral barcode clonotypes in plasma at necropsy, with barcodes for which a tissue origin site could be identified highlighted. (C) vRNA and vDNA distribution of the three rebounding barcodes (color-coded to panel B) in necropsy tissues. No presumptive tissue origin site was identified for BC.15 (red open symbols). Colored upward-facing triangles represent rebounding barcodes with tissue origin sites (lower jejunum for BC.4095; perisplenic LN for BC.1192) indicated by the large, filled triangles. Grouped tissue categories: GI tract (blue), GI tract-draining LNs (red), non-GI lymph tissues (green), non-lymphoid tissue (purple), and blood (black). (D) Viral barcode SIV RNA copies in tissue sample grouped by barcode ID. Each point represents an individual barcode detected in the indicated tissue. Tissue-associated SIV vRNA and vDNA values are expressed as copies per 10⁶ total cell equivalents. Tissue origin sites identified by DBSCAN-based outlier analysis (see materials and methods).

Animal L681 also rebounded on day 6 following ART discontinuation, with 22 barcodes identified in rebound viremia of 170,000 copies/ml at necropsy on day 12 following ART discontinuation (Fig. 5A). All the barcodes in rebound plasma viremia were in the upper half of the primary viremia plasma barcode distribution prior to ART initiation (Fig. 5B). Barcode tissue analysis showed widespread distribution of rebounding lineages in the GI tract and GI-associated LNs, and other lymphoid tissues (Fig. 5C to D). The initial barcode detected in rebound plasma (red, BC.4653) was too widely distributed across tissues by day 12 to determine a presumptive tissue origin site. However, outlier analysis of 10 other secondary rebound barcode lineages allowed inference of tissue origin sites, including the duodenum for the third highest rebounding barcode at necropsy (blue, BC.2154) and two additional secondary rebounding lineages (purple, BC.25; pink, BC.2791), as well as a gastric LN for two more secondary rebounding barcodes (teal, BC.3084; green BC.2636). Multiple rebound barcode lineages were enriched in specific GI and GI-associated lymphoid tissues specimens, consistent with the aggregate 2.6-fold and 3.1-fold statistical enrichment of tissue origin sites in these compartments identified by mixed-effects logistic regression (11).

The remaining RMs exhibited 5 to 38 barcodes in rebound plasma viremia, and outlier analysis inferred sites for tissue origins for a subset of these rebounding barcodes (fig. S7 to 17). In total, presumptive tissue origin sites could be assigned for 56 of the 175 total barcode lineages contributing to rebound plasma from all the 14 viremic RMs following ART discontinuation. The remaining 119 of 175 rebounding barcode lineages (68%) could not be assigned a presumptive tissue origin site at the day 12 necropsy timepoint.

Of the 56 barcodes for which presumptive tissue origin sites were identified, 20 were in the GI tract, 24 in GI-associated LNs, and 12 in non-GI lymphoid tissues (Fig. 6A). To test if there was an enrichment of rebound origin sites based on tissue type, we used mixed-effects logistic regression to assess if tissue group was a significant predictor of viral rebound among the 380 distinct tissue specimens from the 11 viremic RMs with at least one detectable tissue origin site. The odds of contributing barcodes to rebound viremia were 2.6-fold higher for GI tissues (P = 0.026) and 3.1-fold higher for GI-associated LNs (P = 0.004) relative to non-GI lymphoid tissues (table S5). GI or GI-associated lymphoid tissue origin sites were identified in 11 of the 11 viremic animals with at least one detectable tissue origin site. Additionally, GI tissues had the greatest number of tissue sites with more than one inferred tissue origin barcode, including two tissues containing three tissue origin barcodes and a third site with two tissue origin barcodes (Fig. 6B).

Fig. 6. GI tract and GI-draining LNs accounted for the majority of identifiable tissue origin sites of viral rebound across viremic animals.

Data from n = 14 viremic animals. (A) Number of barcode-defined rebound origin sites per RM, stratified by anatomic tissue group. Each point represents the total log₁₀ SIV DNA copies per tissue group (GI tract, GI-draining LNs, non-GI LNs, and non-lymphoid tissues). (B) Total number of barcodes originating from individual tissue samples. Fractions indicate the number of tissue origin sites identified out of the total number of tissues examined per group. Tissue origin sites identified by DBSCAN-based outlier analysis. Comparisons across tissue groups by Wald test.

These data illustrate the continuum of viral rebound following ART discontinuation, ranging from animals with no detectable rebound to animals with oligoclonal rebound involving multiple viral lineages. Because necropsies were performed 12 days after ART discontinuation, many rebounding lineages had already disseminated broadly across tissues, limiting the ability to infer definitive origin sites for all lineages.

Peripheral blood transcriptomic and proteomic signatures of immune and metabolic activation preceded detectable plasma viremia following ART discontinuation

We hypothesized that the early replication events within tissues leading to rebound viremia following ART discontinuation may be associated with systemic changes that would be detectable by transcriptomic signatures in peripheral blood prior to detectable plasma viremia. To characterize potential peripheral blood signatures of imminent viral rebound, we performed peripheral blood bulk RNA sequencing at the time of ART discontinuation on day 0, day 2, and then daily from day 4 to 12 following ART discontinuation. Transcriptomic analyses were aligned to the time when plasma SIV RNA first exceeded the commonly used clinical threshold of 50 copies/ml with sustained increases thereafter. Gene set enrichment analysis (GSEA) was conducted on differentially expressed genes (DEGs) among all rebounders, comparing transcriptomic profiles prior to ART discontinuation with the sampling timepoints following ART discontinuation but prior to plasma viremia >50 copies/ml (fig. S18).

This transcriptomic analysis revealed up-regulation of T cell, proinflammatory, cytokine, and innate immune pathways (Fig. 7A, fig. S19A), as well as pathways related to metabolism, cell cycle, and chromatin remodeling (Fig. 7B, fig. S19B; GSEA: false discovery rate [FDR] q < 0.25) prior to detectable rebound plasma viremia. We assessed the kinetics of transcriptomic changes over time, identifying up-regulation of key pathways, including γδ T cells, Interferon Regulatory Factor 7 (IRF7) activation, Toll-like receptor (TLR) signaling, T cell differentiation, metabolism, and cell cycle as early as 6 to 9 days before rebound plasma viremia of >50 copies/ml was reached (Fig. 7, C and D). Individual genes activated included canonical T cell signaling components (LCK, CD3D, CD247, ITM2A), major histocompatibility complex (MHC) class I antigen presentation mediators (CALR, PDIA3), cytotoxic granule proteins (GZMB, GZMK), and regulators of inflammatory signaling (JAK1, JAK2, STAT1, MAP2K6, FASLG). These findings suggest the onset of coordinated adaptive and innate immune responses prior to detectable viremia, consistent with peripheral detection of tissue-level viral activity (Fig. 7, E and F).

Fig. 7. Immune activation and metabolic pathway up-regulation in peripheral blood precede viral rebound following ART discontinuation in SIV-infected RMs.

(A and B) Transcriptomic analyses were performed on samples from n = 14 animals. Up-regulated transcriptomic pathways compared to baseline (day of ART discontinuation) in the last pre-rebound timepoint (plasma SIV RNA <50 copies/ml) include immune-related pathways (A) and metabolism, cell cycle, and chromatin remodeling pathways (B). For (A) and (B), transcriptomic analyses include samples collected at baseline (day 0), day 2, and days 4 to 12 following ART discontinuation, restricted to timepoints prior to plasma SIV RNA exceeding 50 copies/ml for each animal and aligned relative to individual rebound timing. (C and D) Time course analysis depicting up-regulated pathways relative to baseline for immune pathways (C) and metabolism, cell cycle, and chromatin remodeling pathways (D) across specific timepoints: baseline, days -9 to -6, days -5 to -4, days -3 to -2, and day -1 relative to the first timepoint with rebound viremia >50 copies/ml (defined as day 0 for this analysis). Only significant data (Benjamini-Hochberg-adjusted P < 0.05 by GSEA) are shown. Circle size reflects significance; color gradient represents the normalized enrichment score (NES). (E and F) Time course of genes corresponding to immune pathways (E) and metabolism, cell cycle, and chromatin remodeling pathways (F) as identified in (C) and (D), respectively.

To assess if the up-regulation of these pathways represented the early effects of viral replication in tissues following ART discontinuation or alternatively was related to the metabolic or other effects of discontinuation of antiretroviral drugs, we performed a control experiment using a previously published cohort of SIVmac251-infected, ART-suppressed RMs that initiated ART on day 0, day 1, day 2, or day 3 following SIV inoculation (12). In this prior study, ART was administered for 24 weeks and was then discontinued. Of the 20 animals, 9 animals exhibited viral rebound following ART discontinuation, and 11 animals did not rebound and were presumed to be uninfected as a result of post-exposure prophylaxis. We performed bulk RNA sequencing in this SIV-uninfected cohort following ART discontinuation and compared GSEA signatures with the SIV-infected cohort. These data suggest that the transcriptomic signatures related to metabolism, cell cycle, chromatin remodeling (fig. S20A), and immune responses (fig. S20B) following ART discontinuation in SIV-infected animals were associated with resumption of viral replication in tissues and impending viral rebound.

To confirm these findings, we performed plasma proteomic profiling using mass spectrometry at various timepoints relative to baseline. T-tests identified up-regulated and down-regulated proteins over time, and enrichment analysis using the Fisher Exact Overlapping Test was applied to determine the overlap of up-regulated proteins within the same pathway sets analyzed in the bulk RNA sequencing (Fig. 7; fig. S20). Consistent with the transcriptomic data, we observed up-regulation of T cell signaling, proinflammatory cytokines, and innate immune pathways (fig. S21A), as well as metabolism, cell cycle, and chromatin remodeling pathways (fig. S21B), prior to plasma viral rebound. Temporal analysis of proteomic pathways following ART discontinuation revealed early up-regulation (4 to 9 days prior to rebound viremia) of proinflammatory and cytokine pathways, including IL-12, IL-6, IL-1, and IL-8 signaling, interferons, and innate immune pathways such as macrophages and dendritic cell activation (fig. S21C) as well as metabolic signatures (fig. S21D).

Peripheral blood transcriptomic signatures prior to ART discontinuation distinguished animals with short versus long time to viral rebound

In a parallel analysis, we performed peripheral blood RNA sequencing prior to ART discontinuation to define transcriptomic pathways that predicted time to rebound. Shorter time to viral rebound was associated with up-regulation of CD4 and CD8 T cell activation pathways, IL-2 and IL-7 signaling, cortisol regulation, and SWI–SNF chromatin remodeling (Fig. 8, A and B). These data suggest that an increased inflammatory state at baseline together with a cytokine milieu and chromatin remodeling state associated with a permissive transcriptional environment facilitates rapid viral reactivation following ART discontinuation.

Fig. 8. Pre-ATI transcriptomic signatures stratified animals by time to viral rebound.

Data from n = 14 animals. (A) Gene expression at baseline (prior to ATI) was modeled using a Cox proportional hazards model to identify genes whose expression levels were associated with time to viral rebound. Genes were ranked by the direction and magnitude of their association with earlier rebound, and GSEA was performed using curated immune and biological pathway databases. Shown are pathways associated with shorter time to rebound (red gradient) or with delayed rebound (blue gradient), ranked by NES, with statistical significance indicated by −log₁₀(FDR q value). (B) Heatmaps display scaled expression (row z-scores) of selected gene sets and pathways stratified by time to viral rebound following ATI. Columns represent individual animals ordered from shortest to longest time to rebound; color intensity ranges from low expression (light green; early rebound, ~day 2) to high expression (dark green; delayed rebound, ~day 12). Rows correspond to genes belonging to curated functional modules, including CD4 T cell activation and cytokine signaling (CD4/TCR, IL-2, IL-7), chromatin remodeling (SWI–SNF), epigenetic and cell-cycle regulation (HDAC1 targets, ORC1 removal from chromatin), metabolic pathways (heme metabolism), immune regulation (regulatory T cells, IL-10, M2 macrophages), cytotoxic surveillance (NK cells, IL-15), and intracellular clearance (lysosome).

Cox proportional hazards modeling of pre-ATI gene expression identified two opposing transcriptomic program, one permissive for rapid viral reactivation and one associated with active suppression of proviral reactivation, that together define the immunology and metabolic state of the host at the time of ART discontinuation (Fig. 8, A and B).

DISCUSSION

We characterized viral dynamics of rebound viremia following ART discontinuation in SIV-infected RMs, using the barcoded virus SIVmac239M to track viral rebound at the level of individual barcode clonotypes and to identify tissue sites enriched for barcodes detected in rebound plasma viremia. Animals demonstrated a broad spectrum of rebound outcomes at necropsy from no detectable viral rebound to full rebound. We used extensive tissue sampling, vRNA and vDNA quantification, and barcode analyses along with a modified outlier analysis to identify tissue sites with elevated levels of individual barcode vRNA at necropsy as presumptive sites enriched for 56 of 175 barcode lineages contributing to rebound viremia. Moreover, we demonstrated inflammatory and metabolic changes in peripheral blood following ART discontinuation but prior to detectable rebound viremia that predicted imminent viral rebound and baseline profiles prior to ART discontinuation that predicted time to rebound.

Time to viral rebound did not differ among animals initiating ART on days 6, 9, or 12 post-infection. Prior macaque studies showed delayed or absent rebound with earlier ART initiation on days 4 to 5 post-infection (1), suggesting that the viral reservoir is largely seeded by day 6. Our data suggest a model in which individual viral clonotypes that contribute to viral rebound emerge from or amplify primarily in GI tissues and GI-draining LNs, representing early sites where these rebounding lineages expand prior to systemic dissemination. Because necropsies were performed 12 days after ART discontinuation, viral dissemination across tissues already begun in many animals. Viral rebound therefore appears to be a dynamic process that is initiated with one or a few clones in a limited number of tissues, followed by systemic expansion of the initial rebounding lineages while additional lineages also become reactivated in rapid succession, resulting in rebound viremia of increasing complexity and eventual generalized reinfection of tissues by dominant rebounding clones. Figure S4 demonstrates the concurrent presence of multiple rebound lineages in plasma at early post-ATI timepoints in the majority of viremic animals, suggesting overlapping reactivation events (13). These findings provide insight into the tissue-level dynamics contributing to viral rebound.

Our results focus on later stages of viral rebound and therefore complement those of a companion study using the same RM/SIVmac239M model (11) which focused on earlier necropsy timepoints (day 5 and day 7 following ART discontinuation) to define the earliest events in tissues leading to viremic rebound and to identify initial rebound barcodes and their anatomic origins, 96% of which were found in GI tract and GI associated LNs. These two studies thus suggest an important role for the GI tract and associated lymphoid tissue microenvironments in viral rebound, including both the initial clonotype lineages and the subsequent clonotype lineages that contribute to rebound, even in the context of systemic inflammation and immune activation (11, 14). The preferential enrichment of rebound origin sites in GI tissues and GI-associated lymph nodes is mechanistically consistent with the known properties of these compartments: GI-associated lymphoid tissue harbors the largest body reservoir of activated CCR5⁺ CD4 T cells, maintains chronic low-level immune activation even during suppressive ART, and provides a tissue microenvironment permissive for HIV/SIV replication and reactivation. These tissue-intrinsic features may therefore specifically favor both proviral persistence and early expansion of reactivating lineages in GI-associated compartments following ART discontinuation.

Barcodes with higher primary infection levels prior to ART initiation generally correlated with those found more frequently in rebound plasma viremia after ART discontinuation. These findings suggest that early viral replication dynamics for individual viral barcode clonotypes can impact viral seeding of tissues during acute infection and eventually contribute to viral rebound. However, although more highly represented barcodes had a higher probability of contributing to viral rebound, the most highly represented barcode in both primary infection and in tissues did not, suggesting that additional factors may influence which proviruses reactivate, such as the cellular and local tissue microenvironment (2, 7, 15).

Multi-omic profiling of peripheral blood showed early up-regulation of immune activation and metabolic pathways following ART discontinuation but prior to detection of rebound plasma viremia, presumably reflecting viral expression and replication occurring within tissues. These findings align with previous studies showing early up-regulation of interferon-stimulated genes, antiviral restriction factors, and inflammatory responses following ART discontinuation (16). For example, a human ATI study comparing viremic controllers and non-controllers (17) identified up-regulation of FCGR1A/3A/3B, IFIT1/2/3, MX1/2, and STAT1 among the earliest pre-rebound signatures; our NHP transcriptomic analysis similarly identified early up-regulation of STAT1 and IRF7-regulated transcriptional programs, suggesting that convergent interferon-driven antiviral responses precede detectable viremia across both NHP and human ATI settings and may represent a conserved host response to early tissue-level viral reactivation. Metabolic shifts, including glycolysis and mitochondrial function, also suggest a systemic response to viral replication in tissues before virus in blood can readily be measured. The proteomic signatures identified in our RM model closely parallel data from recent human studies of ART discontinuation, in which individuals with impending viral rebound exhibited increased immune activation, interferon signaling, and apoptosis, whereas virologic controllers displayed metabolic shifts and reduced inflammation (18, 19).

Given that these molecular signatures emerged prior to detectable plasma viremia, they likely reflected systemic inflammation and metabolic reprogramming triggered by viral expression and local viral replication in tissues. These signatures provide insight into host immune and metabolic responses associated with impending viral rebound. Although the clinical utility of predicting viral rebound shortly before detectable viremia remains uncertain, these findings highlight biological pathways associated with viral reactivation that may inform future studies aimed at identifying biomarkers or therapeutic targets related to viral rebound. One of the earliest up-regulated pathways identified in our transcriptomic analysis was the γδ T cell pathway, detectable up to 9 days prior to measurable rebound viremia. γδ T cells are an innate-like lymphocyte population enriched in mucosal compartments, including the GI tract, and can respond to viral antigens in an MHC-independent manner. Their early peripheral blood activation may therefore serve as a sentinel of tissue-level viral activity in the GI compartment preceding systemic viremia, consistent with the GI-enriched tissue origin sites identified by barcode analysis. Future studies examining the mechanistic role of γδ T cells in mucosal antiviral surveillance during early viral rebound may provide additional insight into the tissue-level dynamics of viral reactivation. A fully integrated multi-omics approach could further illuminate the coordinated interplay among these modalities in future studies.

In a separate analysis, the time to viral rebound was predicted by competing factors prior to ART discontinuation: inflammatory processes that promote viral reactivation and host metabolic, epigenetic, and immunologic mechanisms that actively suppress viral reactivation. These data highlight host mechanisms that may influence the pathogenesis of viral rebound and contribute to developing strategies to achieve ART-free virus remission. For example, rather than focusing exclusively on latency reversal strategies designed to activate and eliminate the viral reservoir, these findings highlight host pathways associated with the maintenance of proviral latency. The metabolic, epigenetic, and immunoregulatory programs associated with delayed rebound include heme oxygenase-1 up-regulation through heme metabolism pathways, which reduces reactive oxygen species with antiviral properties; HDAC1-mediated epigenetic repression and ORC1-regulated cellular licensing, which promote proviral latency; IL-10, regulatory T cell, and M2 macrophage programs, which are anti-inflammatory and may counteract cytokine-mediated latency reversal; and proteasomal and lysosomal clearance pathways, which may eliminate cells with early viral protein expression. NK cell and IL-15-mediated immunosurveillance may further contribute to elimination of cells with early replication events. Conversely, animals with shorter time to rebound showed baseline enrichment of CD4 T cell activation, IL-2/IL-7 signaling, and SWI-SNF chromatin remodeling, consistent with a transcriptional environment permissive for rapid proviral reactivation. Together, these data suggest that longer time to viral rebound reflects an actively maintained restrained state rather than passive virologic quiescence while on ART. These data suggest that stabilization of proviral latency through modulation of host cellular states or tissue microenvironments may represent a complementary conceptual approach for achieving durable ART-free remission. Future studies will be required to determine whether such pathways can be selectively modulated within relevant tissue compartments without compromising systemic immune function.

Our study has several limitations. First, because necropsies were performed on day 12 following ART discontinuation, viral dissemination had already occurred in many animals, limiting the ability to identify tissue sites enriched for all rebounding lineages. Furthermore, we cannot rule out the possibility that some inferred origin sites could represent early amplification sites rather than the primary reactivation site. Second, although we evaluated multiple tissues at necropsy, these tissues still only reflected a fraction of the total GI and lymphoid tissues in the animals. Third, our transcriptomic and proteomic profiling was limited to peripheral blood and thus may not reflect the microenvironment in relevant tissues. Fourth, although the barcoded SIV model allows high-resolution clonal lineage tracking, we cannot exclude the potential for barcode-associated fitness effects that could influence lineage dominance, the limited ability of this approach to detect recombination, and its extrapolation to HIV-1 in humans that requires cautious interpretation. Fifth, the duration of suppressive ART in this macaque study although extensive (70 weeks) is still shorter than the typical duration of ART in people living with HIV-1. Sixth, cell-associated vRNA measurements detect total transcriptional activity, including from both replication-competent and transcriptionally-active-but-replication-defective proviruses, and therefore may overestimate the abundance of cells capable of contributing to productive rebound viremia; however, our comparative outlier analysis, which identifies tissues with elevated vRNA for a given barcode relative to the distribution across other tissues, is less sensitive to this concern than relying on absolute vRNA counts. Seventh, GI tissue specimens were collected as bulk mucosal biopsies without separation of discrete intramural compartments. Future studies incorporating compartment-level GI sampling could provide additional anatomical resolution regarding which specific GI microenvironments harbor rebounding viral lineages.

In summary, our findings indicate that viremic rebound following ART discontinuation appears to involve oligoclonal and oligofocal reactivation of viral clonotypes within tissues, within GI tissues and GI-associated LNs frequently representing early sites of viral expansion, followed by systemic dissemination and broader tissue involvement of these and additional rebounding lineages. Transcriptomic and proteomic analyses of peripheral blood revealed up-regulation of inflammatory and metabolic pathways associated with both time to rebound prior to ART discontinuation and imminent viral rebound following treatment interruption. Together, these findings provide insight into the tissue-level dynamics and host responses associated with viral rebound following ART discontinuation. These findings advance mechanistic understanding of the tissue origins and host correlates of viral rebound and provide a framework for future studies examining GI-associated reservoir compartments, peripheral blood biomarker development, and host-directed strategies to prolong ART-free viral remission.

MATERIALS AND METHODS

Study Design

Eighteen outbred, Indian-origin adult male and female RM (Macaca mulatta), housed at AlphaGenesis, were intravenously infected with 5,000 infectious units (IUs) of a previously titrated stock of molecularly barcoded SIVmac239M at week 0 (6, 7, 20). Intravenous inoculation was used to ensure infection with numerous barcodes (average 342 lineages), which was essential for the subsequent high-resolution tracking of clonal lineages. The cohort consisted of seventeen males and one female animal (TP5). The predominance of male animals reflects standard availability of age- and MHC haplotype-matched cohort animals at the time of enrollment; sex was not a primary experimental variable in this study. The single female animal is included in all analyses. Future studies should incorporate balanced sex representation to assess potential sex-dependent differences in reservoir dynamics and rebound kinetics. None of the animals were hormonally treated. Prior to enrollment, animals were screened and confirmed to be free of simian retrovirus D, STLV-1, Herpes B, and Mycobacterium tuberculosis. MHC class I genotyping was performed to exclude the presence of protective alleles (Mamu-A01, Mamu-B08, and Mamu-B17). Demographics are provided in table S2. All RMs were housed and cared for under protocols approved by the Institutional Animal Care and Use Committee (IACUC). This study was conducted in accordance with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) 2.0 guidelines.

Animals were randomly assigned to one of three groups (n = 6 per group) and started on ART at either day 6, 9, or 12 following challenge, reflecting early primary viremia, mid-exponential primary viremia, and near-peak primary viremia, respectively, to allow different levels of RCRV seeding. The ART regimen consisted of daily subcutaneous injections of tenofovir disoproxil fumarate (TDF; 5.1 mg/kg/day), emtricitabine (FTC; 40 mg/kg/day), and dolutegravir (DTG; 2.5 mg/kg/day) pre-formulated in a 15% (v/v) kleptose solution at pH 4.2 (21). Animals were treated with daily ART for 70 weeks with longitudinal plasma viral loads and PBMC vDNA and vRNA measurements. Biopsy sampling prioritized lymph nodes and GI tissues. Two RMs were euthanized for procedure-related complications and were not followed through the ART discontinuation phase of the study.

At week 70, ART was discontinued and animals were monitored daily for 12 days. Blood was collected on days 0, 2, and 4 to 12 after ART discontinuation. Animals then underwent comprehensive necropsy on day 12. A median of 62 tissue samples per animal were collected from all major organ systems (table S2). Comprehensive GI sampling covered the esophagus, stomach, duodenum (proximal and distal), jejunum (proximal and distal), ileum, cecum, ascending colon, and rectum, as well as mesenteric and regional lymph nodes draining the GI tract. GI tissue specimens were collected as bulk mucosal biopsies; Peyer’s patches were not dissected out as anatomically separate specimens, and lamina propria and lymphoid follicle compartments within individual GI biopsy specimens were not distinguished at tissue collection. Necropsy tissues were assessed for total vDNA and vRNA levels and clonotypic barcode representation in vDNA and vRNA. Immunologic and virologic assays were performed blinded.

Antiretroviral Therapy (ART) regimen

The pre-formulated antiretroviral therapy (ART) cocktail was provided by Gilead Sciences and contained 5.1 mg/ml of tenofovir disoproxil fumarate (TDF), 40 mg/ml of emtricitabine (FTC), and 2.5 mg/ml of dolutegravir (DTG) dissolved in 15% (v/v) kleptose (1), which was adjusted to a pH of 4.2. The ART cocktail was administered daily to the study animals by subcutaneous injection at 1 ml/kg body weight for a period of 70 weeks.

Plasma Viral Load Assays

Plasma SIV vRNA was measured essentially as described (22). Viral load was measured at multiple time points throughout duration of the study, including prior to the initiation of ART and after the discontinuation of ART. vRNA was measured using a gag targeted real time assay with a threshold of 15 SIV RNA copies/ml (23).

Quantitative Evaluation of Cell-Associated vDNA and vRNA

For PBMCs and tissue specimens, levels of vRNA and vDNA were measured essentially as described using assays targeted to gag (24, 25). For tissue specimens, RT-qPCR measurements of vRNA detect total cell-associated SIV RNA, including polyadenylated unspliced and spliced viral transcripts, and do not distinguish transcript classes. Cell-associated vRNA and vDNA levels are reported as copies per 10⁶ total cell equivalents, as determined by host genomic DNA quantification.

Intact Proviral DNA Assay (IPDA)

To isolate CD4 cells from frozen PBMCs, we used the EasySep NHP CD4+ T Cell Isolation Kit (Stem Cell), and total cellular DNA was extracted from the isolated CD4 cells using the QIAmp DNA Blood Mini Kit (Qiagen). DNA concentration was measured by NanoDrop 2000 (Thermo Fisher Scientific). To quantify intact proviral SIV DNA, we performed the SIV-specific digital droplet PCR (ddPCR) intact proviral DNA assay as described previously (26). Briefly, up to 300 ng sample DNA was added to a master mix containing 2x ddPCR Supermix for Probes (no dUTP, Bio-Rad), 600 nM of each primer, and 200 nM of each probe for each 22 μl reaction. Cell equivalents were determined by quantifying copies of macaque RPP30 in parallel, allowing for the absolute quantification of intact proviral SIV DNA. ddPCR outputs were analyzed using QuantaSoft Analysis-Pro (Bio-Rad) (27).

Barcode Sequencing and Enumeration

To characterize barcode clonotype populations in plasma during primary SIV infection prior to ART initiation, during off-ART rebound viremia, and from vRNA positive PBMCs and biopsy and necropsy tissue specimens, barcode sequence analysis was performed on viral nucleic acid samples as previously described (10) including both vDNA and vRNA (cDNA) sequencing for PBMCs and tissue samples. Barcode sequences were compared to the >10,000 lineages detected in the viral stock (22). The barcode region is amplified using primers in conserved flanking sequences to generate a short amplicon (361 bp), which is then sequenced directly by Illumina technology (10). While no PCR-based assay can be assumed to be entirely bias-free, particularly for lineages near the detection threshold, we have not observed reproducible enrichment of particular barcodes attributable to the assay itself across the extensive use of this platform. Furthermore, as part of this analysis, a 1/input lower limit of detection is used as a threshold for valid barcodes. Finally, although recombination cannot be ruled out, because the viral genome is otherwise identical, recombination outside the short barcode region would not alter lineage assignment. Only a recombination breakpoint occurring within the barcode itself would be expected to obscure detection of a lineage and given the very short barcode region (10 bp) and the limited duration of replication both pre-ART and during early rebound, we expect this to have minimal impact on the present analysis.

Outlier Analysis to Identify Tissue Origin Sites of Rebounding Barcode Lineages

To distinguish the rebound origin site for an individual rebounding lineage expressing vRNA in multiple different tissues, we used Density-Based Clustering of Applications with Noise (DBSCAN) to cluster tissues based on the barcode vRNA and vDNA levels to identify a single outlier site with elevated vRNA expression that did not cluster with any other tissues. The analysis was performed using the “DBSCAN package” in R (version 4.4.2) for each rebounding barcode with vRNA and vDNA in at least 7 samples. The algorithm iteratively clusters points into closely packed, dense regions, separated by regions of lower density, and identifies outliers without assigning them to any cluster. The minimum number of points required to form a dense cluster was chosen as k = 3 for one-dimensional analysis (based on vRNA only) and k = 4 for two-dimensional analysis (based on vRNA vs. vDNA). For each barcode, the k-nearest neighbors (kNN) distance method was used to determine epsilon, the maximum distance between two points for them to be assigned to the same cluster. The kNN-distance (total distance of a point to its k-nearest neighbors) was calculated for each point, which were then rank-ordered, resulting in a generally piecewise linear relationship with a single-breakpoint (defining epsilon), which we estimated automatically by fitting a piecewise linear regression using the “Segmented” package in R. Using this process, we identified all samples that were not assigned to a cluster (vRNA or vRNA vs. vDNA) for each rebounding barcode lineage (with vRNA+ and vDNA+ in at least 7 samples), designated as cluster-outliers. We then compared vRNA expression in these cluster-outliers to non-outlier tissues to identify sites with elevated local replication compared to background vRNA levels or replication in secondary sites. For each lineage, we rank-ordered all samples based on their vRNA levels and defined the sample with the highest vRNA level as the putative rebound origin if (1) it was a cluster-outlier and (2) its vRNA (log₁₀) distance to the next-highest sample exceeded a threshold level η associated with disseminated replication or background vRNA expression, defined as the 99^th quantile of the distribution of vRNA differences between non-outlier samples across all lineages (η = 0.52 log₁₀ for the primary analysis with k = 4; re-derived under each k setting in sensitivity analysis). The origin sites were categorized into anatomical groups to assess patterns of tissue-specific initiation of viral rebound. To assess robustness of presumptive origin-site identification to clustering parameter choice, the full analysis was repeated across k = 3 to 5, with epsilon re-estimated using the same automated kNN procedure and η re-derived from non-outlier tissue distributions under each setting.

Regression Analyses

All regression analyses were performed in R using the lme4 package for mixed effects (tables S3 to S5). For Model 1 (table S3), where we determine the effect of pre-ART replication on probability of rebound at the barcode-level, we performed mixed effects logistic regression analyses to investigate for all 16 evaluable animals whether a barcode’s pre-ART peak plasma viral load across was predictive of whether it rebounded (i.e. detected in rebound plasma). Repeated observations within individual animals were accounted for by including a random effect on the intercept (assumed normally distributed with a zero mean). For Model 2 (table S4), to determine the effect of rebound plasma viral loads on barcode vDNA and vRNA levels in tissues, we performed mixed-effects linear regression on the 175 rebounding lineages from the 14 viremic animals to investigate if the barcode viral load (log₁₀) in rebound plasma was predicted by either its total vDNA or vRNA levels (log₁₀) across all tissues. Repeated observations within individual animals were accounted for by including a random effect on the intercept (assumed normally distributed with a zero mean). For Model 3 (table S5), where we determine the effect of tissue group on probability of rebound at the tissue-level, we performed mixed effects logistic regression analysis to investigate for the 11 animals with at least one identified tissue origin site whether tissue type was predictive of the probability of rebound at the tissue-level. Clustering of observations within individual animals was accounted for by including a random effect on the intercept (assumed normally distributed with a zero mean). The null model (without tissue type as a covariate) was compared to the full model using the likelihood ratio test via ANOVA.

Antibody ELISA

SIV-specific antibodies were assessed using a serial Enzyme-Linked Immunosorbent Assay (ELISA) to measure the reactivity of serum antibodies to the SIVmac239 gp160 recombinant protein. To perform the ELISA, serum samples were diluted and added to 96-well plates that were coated with the SIVmac239 gp160 recombinant protein before being washed with phosphate buffered saline (PBS) + 0.05% Tween 20 and blocked with Blocker Casein (Pierce). The plates were incubated for 1h, allowing SIV-specific antibodies in the serum to bind to the protein. After another wash to remove unbound antibodies and incubation with rabbit anti-mouse IgG horseradish peroxidase (Thermo Fisher Scientific), plates were washed and developed with SureBlue (KPL Laboratories) and stopped with TMB Stop Solution (KPL Laboratories. Plates were read on a VersaMax microplate reader (Molecular Devices) and absorbance at a wavelength of 450 nm was recorded. Endpoint titers were determined as the reciprocal of the highest serum dilution that delivered an optical absorbance value above the value of the negative control sera.

Analysis of Cellular Immune Responses

SIV-specific cellular immune responses were assessed using IFN-γ ELISPOT assays. These assays were performed essentially as described, with some modifications to allow for the assessment of cellular immune breadth. To estimate the breadth of virus specific cellular immune responses, PBMC IFN-γ ELISPOT assays were performed using sub-pools of peptides spanning the SIVmac239 Env, Gag, and Pol proteins, providing a simultaneous measurement of T cell responses to multiple epitopes. The limit of detection of this assay was 5 spots/million cells (8). Flow cytometric staining was performed to further characterize the phenotype and functional properties of the SIV-specific T cells. This was accomplished using predetermined titers of monoclonal antibodies at concentrations, suggested by the manufacturer (Becton Dickinson), against a panel of surface and intracellular markers, including CD3 (SP34; Alexa Fluor 700), CD4 (OKT4; BV510, BioLegend), CD8 (SK1; APC-Cy7), CD14 (M5E2; BUV737), CD16 (3G8; BV650), CD25 (PE-Cy7; M-A251), CD28 (L293; PerCP-Cy5.5), CD38 (APC; HB-7), CD56 (NCAM16; BV786), CD69 (TP1.55.3; PE-TexasRed; Beckman Coulter), CD95 (DX2; BV711), CCR5 (3A9; PE), CCR7 (3D12; BV421), HLA-DR (BUV-395; G46–6), Ki67 (B56; FITC), and PD-1 (EH21.1; BV605).

Transcriptomic Profiling

Transcriptomic Bulk RNA sequencing analysis was conducted on whole blood samples collected in PAXgene tubes, with sequencing by the NHP Genomics Core at Emory Yerkes National Primate Research Center. RNA sequencing was carried out using the Illumina NextSeq 500/550 High Output v2 kits (150 cycles), following the manufacturer’s protocol. The sequencing reads were aligned to the reference genome using the STAR aligner, and downstream differential expression analysis was conducted with the DESeq2 package in R. This analysis identified normalized expression counts and differentially expressed genes (DEGs) between timepoints post-ATI compared to baseline (the day of treatment interruption), with statistical significance defined by adjusted P values (Benjamini-Hochberg correction) <0.05. Visualization of normalized expression counts was achieved Principal Component Analysis (PCA) and heatmaps. GSEA was applied to evaluate biological pathways associated with DEGs, leveraging curated pathway sets such as Biological Themes (BTM), C2 pathways, and in-house compiled pathways. Single-cell Linear Expression Analysis (SLEA) was employed to perform a Spearman correlation between gene expression and time to viral rebound, offering insights into how specific genes or pathways may influence the timing of rebound post-ATI. Further, a univariate Cox model was applied to identify pathways significantly associated with time to rebound.

Proteomic Profiling

Proteomic analysis was conducted on plasma samples, with sequencing performed at the Pacific Northwest National Laboratory (PNNL) using advanced mass spectrometry techniques to measure a panel of less than 2,000 proteins. Data visualization included the use of Principal Component Analysis (PCA) and heatmaps to depict expression patterns across time points. For statistical analysis, T-tests were employed to compare the levels of up-regulated and down-regulated proteins at various time points against baseline measurements, identifying significant shifts in protein expression. Additionally, enrichment analysis was conducted using the Fisher Exact Overlapping Test, using the same pathway sets applied in the Bulk RNA sequencing analysis.

Statistical Analysis

Statistical analyses were performed using GraphPad Prism Version 10.0.3 (GraphPad Software) and R (version 4.4.2). Group comparisons were made with Kruskal-Wallis tests, and associations were assessed using two-sided Spearman correlations. Mixed-effects regression models were used to identify predictors of barcode rebound, Cox proportional hazards models were used for time-to-rebound analyses, Wald tests were applied for barcode distribution comparisons, and Wilcoxon rank-sum tests were used for tissue vRNA and vDNA comparisons.

For virologic and immunologic data, n refers to the number of animals; for barcode-level analyses, n refers to the number of unique barcodes detected per animal; and for tissue analyses, n refers to the number of independent tissue samples. All assays were performed in at least technical duplicate, and PBMC and tissue analyses included biological replicates across animals. Data are reported as mean ± SD unless otherwise indicated.

Statistically significant thresholds were set at an adjusted P < 0.05 (Benjamini-Hochberg correction) for transcriptomic and proteomic analyses, q < 0.25 for GSEA, and P < 0.05 for all other tests. Animals were randomized into ART initiation groups, and assays were performed blinded to treatment group. Two animals were excluded from analyses due to unrelated euthanasia during ART; no other exclusions were applied.

For mixed-effects logistic regression models, model fit was assessed using the corrected Akaike information criterion (AICc) and residual diagnostics. The proportional hazards assumption for Cox models was verified using Schoenfeld residual tests. Non-parametric tests (Kruskal-Wallis, Wilcoxon rank-sum, and Spearman correlation) were used where normality could not be assumed given small sample sizes; normality was assessed with the Shapiro-Wilk test. For proteomic analyses, t-tests were applied following log-transformation of protein abundance values; homogeneity of variance was confirmed by Levene’s test.

Supplementary Material

List of Supplementary Materials

Figure S1 to S21

Tables S1 to S5

MDAR Reproducibility Checklist

Data, code, and materials availability:

All data associated with this study are in the paper or supplementary materials. This study did not generate new unique reagents. Bulk RNA sequencing data generated in this study are available at the NCBI Gene Expression Omnibus (GEO) under accession number GSE294867. All materials used or generated in this study are commercially available or will be supplied upon reasonable request. The molecularly barcoded SIVmac239M construct and the intact proviral DNA assay (IDPA) primers and probes are available upon request, subject to institutional approvals and standard material transfer agreements. Further information and requests for resources and reagents should be directed to and will be fulfilled by the corresponding author (dbarouch@bidmc.harvard.edu).