Dynamic antigen expression and cytotoxic T cell resistance in HIV reservoir clones

Abstract

Clonally expanded CD4⁺ T cells harbouring rebound-competent HIV persist lifelong during antiretroviral therapy^1–5. Latency is considered the principal barrier to viral eradication and has resisted pharmacological reversal^6,7, yet sustained immune pressure appears to erode reservoirs^8–15. Recent advances have yielded glimpses into exceptionally rare reservoir-harbouring cells, implicating prosurvival properties in persistence^16–18. Here we isolate and characterize authentic reservoir clones (ARCs) that robustly proliferate and accumulate while producing infectious virus, without overtly succumbing to cytopathicity. At any moment, only small fractions of ARCs expressed HIV proteins, a state associated with conserved host transcriptional programs but remarkably refractory to potent T cell stimulation. Nevertheless, sustained co-culture with a CD8⁺ cytotoxic T lymphocyte clone substantially culled proliferating ARCs, revealing time-integrated vulnerability to immune pressure. The corresponding ex vivo CD8⁺ T cell response was poorly cytotoxic, and in vivo erosion of ARCs occurred only slowly. A regulatory T cell ARC displayed pronounced cell-intrinsic resistance to cytotoxic T cells — a longstanding hypothesis now directly demonstrated — linked to low oxidative stress and reversed with deferoxamine¹⁹, a hypoxic stress inducer and FDA-approved therapeutic. Overall, we provide insights into the vulnerabilities of reservoir clones to potent, sustained cytotoxic T cell pressure and highlight intrinsic resistance pathways as actionable therapeutic targets, opening opportunities for advancing immune-based HIV cure strategies.

Insights into the vulnerabilities of HIV-1 reservoir clones to potent, sustained cytotoxic T cell pressure are revealed.

Main

HIV persists for life despite antiretroviral therapy (ART), in reservoirs of clonally expanded CD4⁺ T cells harbouring integrated proviruses, which fuel viral rebound if ART is interrupted^1–5. Efforts to reverse HIV latency for immune clearance have had limited success, probably reflecting the limited activity of first-generation latency-reversal agents^6,7. More potent approaches are entering clinical evaluation^20–22. Contrary to a model of complete latency, virological and immunological measures suggest ongoing immunosurveillance during ART: HIV-specific T cells show signs of recent antigen encounter^9,14,15, and reservoirs bear footprints of selection, including preferential reductions of intact versus defective proviruses^8,13. Residual reservoir cells become increasingly clonal and enriched for integrations in transcriptionally unfavourable genomic regions, including heterochromatin and zinc-finger (ZNF) genes^10,12,23,24. Some clones, however, continue to transcribe HIV — occasionally contributing to viraemia^25,26 — and can present features that suggest intrinsic resistance to elimination by cytotoxic T cells (CTLs)^{16–18,27,28}. However, the extreme scarcity of reservoir-harbouring cells (approximately 1–100 per 10⁶ CD4⁺ T cells) has precluded direct testing of persistence mechanisms and limited characterization of the interplay between latency, proliferation and immune surveillance.

Here we present a platform to isolate, expand and interrogate ARCs from ART-suppressed people with HIV (PWH), validated by stable multiomic characteristics over months in culture. Only rare fractions of ARCs express HIV antigen, and these converge on shared transcriptional programs. Stimulation of ARCs reveals the limited efficacy of ‘shocking’ provirus out of latency, yet proliferating ARCs nonetheless display vulnerability to sustained CTL pressure. In parallel, we uncover cell-intrinsic resistance to CTL-mediated killing in a subset of ARCs and identify mechanistic barriers to elimination that can be therapeutically reversed.

Isolation and expansion of ARCs

Although HIV is generally cytopathic, certain CD4⁺ T cell clones harbouring intact HIV proviruses survive proliferation following antigen stimulation^4,29,30. Building on these observations and our previous clone-enrichment approach^31–33, we developed strategies to isolate and expand pure ARCs. We isolated ARCs either by TCRVβ enrichment^32,33 or by unbiased limiting dilution (Fig. 1a). Cultures were screened for HIV Gag under antiretroviral suppression, ensuring that detected virus reflected pre-existing infected cells versus replication. Clonal purity was supported by near-1:2 provirus:RPP30 ratios and confirmed by single integration sites and TCR sequences (Fig. 1b). We isolated ten ARCs, including seven with intact proviruses and three with 5′ leader deletions, with some expanding to hundreds of millions of cells (Fig. 1c and Extended Data Fig. 1a,b). The two ARCs isolated by TCRVβ enrichment correspond to the provirus-enriched T cell lines previously described^31–33.

Fig. 1. Isolation and characterization of ARCs.

a, Schematic depicting isolation of ARCs via sorting for TCRVβ associated with prominent reservoir clones or by limiting dilution. Pure clones were derived by iterative subcloning of HIV Gag-p24⁺ wells. Schematic created in BioRender; Jones, B. https://biorender.com/vqgslip (2026). b, Intact proviral DNA assay supporting purity of isolated ARCs, indicated by a 1:2 provirus:RPP30 ratio. FAM, 6-carboxyfluorescein; VIC, 2′-chloro-7′-phenyl-1,4-dichloro-6-carboxyfluorescein. c, Summary of seven intact ARCs and three dARCs isolated from PWH. The sense (+) or antisense (−) orientation of integrated provirus relative to the host gene (only applicable to genic integration sites) is indicated. ARC-Chr16 and ARC-Chr22 have proviral integrations within non-genic human satellite II (hsat2) or hsat3 regions, respectively. d, Longitudinal analysis showing persistent detection of ARC-ZNF721 and ARC-Chr16 and dARC-COG5 over 12 years of ART, quantified by TCRVβ-specific and integration site-specific ddPCR. ddPCR quantitation of total HIV LTRs and intact HIV provirus (intact proviral DNA assay) is provided for comparison. The dashed lines indicate ARC isolation time points. e, ELISPOT demonstrating EBV-specific IFNγ and GzmB production by ARC-ZNF721. f, QVOA showing heterogeneous inducibility of viral outgrowth among individual ARCs and susceptibility of virus from ARC-Chr22 to autologous IgG. Data indicate mean values with error bars showing the 95% confidence interval calculated from 24 replicates per 6–7 dilutions (0.01 × 10⁶ to 1 × 10⁶ cells per replicate). Of note, no outgrowth was observed for ARC-Chr22 + autologous IgG and ARC-ZNF486 + control IgG. Schematic created in BioRender; Jones, B. https://biorender.com/d470c5j (2026). g, Flow cytometry analysis showing transmission of replication-competent HIV from carboxyfluorescein succinimidyl ester (CFSE)-labelled ARC-Chr22 to activated, CFSE-negative bulk CD4⁺ T cells after a 7-day co-culture without ARVs. For ARC-Chr22, the infection is shown in two ways: with a dual Gag stain and depicting CD4 downregulation in HIV-expressing cells. As expected, no transmission of infectious virus occurs from the dARC-COG5. Data are representative of two independent experiments (b,e–g) or combined from three separate experiments (d).

Source data

Extended Data Fig. 1. Characterization of authentic reservoir clones.

(a) Schematic of the HIV proviral genome HXB2 (top) aligned with proviral sequences for each clone (bottom). Clones are color-coded by donor. A 58-, 11- and 116-nucleotide deletion was present in the 5’ leader sequence of dARC-COG5, dARC-R3HDM2, and dARC-KDM2A respectively. Proviral integration sites were identified by linear amplification-mediated PCR (LAM-PCR). (b) Intact proviral DNA assay (IPDA) for ARC clones. Quantification of intact proviruses is shown, with FAM and VIC probes detecting ψ, env and the reference gene RPP30. A 1:2 ratio of HIV to RPP30 is consistent with one provirus per cell. Note that dARC-COG5 shows a lack of amplification at ψ, consistent with the proviral deletion depicted in (a). (c) Summary table of longitudinal dynamics over years on ART for donor OM5334 and donor 603, integrating the relative contribution of individual reservoir clones to total proviral DNA, intact proviral DNA, and corresponding TCR clonotype frequencies. (d) IFN-γ ELISpot assay showing cytokine responses of ARC clones to stimulation with CMV and EBV peptides or lysates, pertussis toxin, phytohemagglutinin (PHA), or DMSO control. The results indicate that ARC-Chr16 is EBV-specific while the specificities of ARC-ZNF486 and an OM5334 uninfected CD4⁺ T-cell clones were not determined. (e) Representative flow cytometry plots from transmission assays evaluating the capacity of indicated ARCs to transmit infection to uninfected target cells. Plots are gated on viable target cells (CFSE⁻), excluding CFSE⁺ ARCs, and show infection frequencies following co-culture with indicated ARCs versus uninfected control cells. Percentages indicate the proportion of newly infected cells expressing HIV-Gag protein. The top and bottom panels were performed on different days with different target cells. Two anti-HIV-Gag antibodies are used targeting different epitopes to maximize sensitivity and specificity. (f) Representative flow-cytometry plots from longitudinal transmission assays showing HIV Gag expression in target cells following co-culture with ARC-POLR1HASP or ARC-Chr22 at the indicated time points. Data are combined from four (b), two (c,e) or representative of two (d) independent experiments.

Source data

ARCs: selection, persistence and rebound

HIV integrates quasi-randomly, preferring actively transcribed genes³⁴. Under sustained immune pressure, intact proviruses preferentially reside in ZNF genes or heterochromatin, whereas defective proviruses persist in gene bodies^{10–12,23,24}. Consistent with this, six of seven intact provirus ARCs had integrations in ZNF genes or heterochromatic regions, whereas 5′ leader defective provirus ARCs (dARCs) were in genes (Fig. 1c). Thus, ARC isolation enriches for clones surviving prolonged selection.

In donor OM5334 (Supplementary Tables 1 and 2), we tracked two intact ARCs and one dARC across more than 12 years beginning near ART initiation using clone-specific TCR–CDR3 and integration-site assays alongside standard measures of total and intact HIV DNA. Each ARC constituted less than 0.2% of infected cells at ART initiation yet remained stable over 12 years (Fig. 1d and Extended Data Fig. 1c), contrasting with the marked early decline in overall intact proviruses and highlighting the exceptional resilience of these clones in vivo. ARC-ZNF721 (integration in ZNF721) showed features consistent with specificity to a persistent non-HIV pathogen. The presence of both infected and uninfected cells at the earliest time point suggests expansion before HIV infection, and longitudinal tracking showed that it comprised up to 34.6% of intact proviruses after 3.5 years of ART (Fig. 1d and Extended Data Fig. 1c). Functional testing demonstrated EBV specificity with robust granzyme B (GzmB) degranulation (Fig. 1e and Extended Data Fig. 1d), illustrating how ARCs can reveal immunological contexts associated with reservoir persistence.

To assess rebound potential, we subjected ARCs to quantitative viral outgrowth assay (QVOA)-like co-cultures with autologous or control IgG to measure neutralization sensitivity. Viral outgrowth was observed for each of the four ARCs tested, but with dramatically different levels (Fig. 1f). This may reflect both differences in frequencies of cumulative ARC cell reactivation over 7 days and, potentially, differences in viral replicative fitness. Of note, the ARC-Chr22 virus was previously detected in a QVOA from ex vivo CD4⁺ T cells¹¹. In the current experiment, autologous IgG effectively neutralized ARC-Chr22 virus, indicating that humoral immunity may be poised to limit rebound from some ARCs in vivo. In parallel, five ARCs transmitted infection to CD4⁺ target cells in co-culture, whereas dARC-COG5 did not (Fig. 1g and Extended Data Fig. 1e,f). Virus from ARC-Chr22 replicated robustly, whereas ARC-ZNF721-derived virus showed limited replication, potentially contributing to reduced QVOA detection.

Immunophenotypic programs in ARCs

We performed 5′ cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq)³⁵ on synchronously isolated ARC-Chr16, dARC-COG5 and an uninfected clone from OM5334 at 3 weeks of culture, with an additional sample of ARC-Chr16 collected after 5 months to assess stability. Clustering was mainly driven by clonal identity, with overlap between clones reflecting cell cycle genes including MKI67 (Fig. 2a–c and Extended Data Fig. 2a,b). ARC-Chr16 expressed high levels of FOXP3 transcripts and surface CD25/CD39, indicating a regulatory T (T_reg) phenotype. dARC-COG5 exhibited high IFNG transcript levels indicating a T helper 1 (T_H1) profile, whereas the uninfected clone expressed T_H2 IL4/IL5 transcripts (Fig. 2b and Extended Data Fig. 2a). Both dARC-COG5 and ARC-Chr16 expressed elevated PRF1 and IFNG transcripts compared with the uninfected clone (Fig. 2b). ARC-Chr16 contained MKI67-defined proliferative subclusters; however, the abundance and foundational T_reg profile were preserved, indicating stable clonal identity throughout prolonged culture (Fig. 2b,c and Extended Data Fig. 2b). Pathways enriched in dARC-COG5 included DNA replication and S-phase programs, consistent with a heightened proliferative signalling under resting conditions (Fig. 2c). Thus, ARCs exhibit diverse but stable foundational transcriptional programs during prolonged in vitro propagation.

Fig. 2. Multimodal profiling of immunophenotypic features and HIV expression of ARCs.

a, Uniform manifold approximation and projection (UMAP) clustering of weighted nearest neighbour-based integrated gene expression and surface protein data from OM5334-derived uninfected, dARC-COG5 and ARC-Chr16 clones after regression of TCR gene contributions. ARC-Chr16 is compared across early (3 weeks post-isolation) and later (5 months) time points. b, Heatmap of the top 15 DEGs from each clone in panel a including TCR genes. The red arrowheads indicate T_H lineage-related genes and the black arrowhead points to a perforin transcript. c, Top seven enriched Reactome pathways for ARC-Chr16 and dARC-COG5 compared with uninfected clones from DEGs with adjusted P < 0.05 from a one-sided hypergeometric test. d, Cytotoxicity module scoring (CD8 cytotoxic gene set from Szabo et al.⁵³) of ARCs versus synchronous uninfected clones. Numbers of uninfected clones were: one for COG5 and Chr16, four for ZNF721, two for Chr22 and three for ZNF486. P values are from a two-sided Wilcoxon test. The blue dashes represent mean values. e, Log-normalized levels of HIV expression and centred log-ratio-normalized surface CD4 protein in clones from panel a. Areas of cells with a HIV RNA^highCD4^low phenotype are circled. f, Summary table (top) showing ARC identity, donor identity, total cell numbers, HIV RNA⁺ cell numbers, HIV RNA⁻ cell numbers and frequency of HIV RNA⁺ cells within each clone that pass permissive HIV RNA count > 0 as HIV RNA⁺ cell filter. Summary table (bottom) showing the number of DEGs shared between HIV RNA⁺ versus HIV RNA⁻ cells across five ARCs (adjusted P < 0.05). g, Similar tables as panel f, but with stringent definition of HIV RNA⁺ cells as log-normalized HIV > 4 termed ‘productive HIV RNA⁺’ cells based on CD4 downregulation⁵⁴ and showing DEGs shared across the four ARCs. h, Heatmaps of total DEGs with selected genes (left) and selected Hallmark gene set DEGs (right) from pseudobulk pairwise analysis comparing productive HIV RNA⁺ cells versus productive HIV RNA⁻ cells. Data are from four combined experiments, and statistics in panels f–h are from a two-sided Wald test.

Source data

Extended Data Fig. 2. ARCs exhibit distinct gene expression and surface protein expression patterns converging predominantly on cytotoxic genes and T_H1 profiles.

(a) Differentially expressed proteins (DEPs) from ARC-Chr16 early (3 weeks), ARC-Chr16 (5 months), dARC-COG5 and uninfected CD4⁺ T cell clone. Proteins shown pass filters of log2FC > 0.5 and P-value < 0.05 adjusted for multiple testing from a two-sided Wilcoxon Rank Sum test. (b) UMAPs showing MKI67 expression (left) and gating on ARC-Chr16 early and 5-month cells based on high (>1 log-normalized expression) or low (<1 log-normalized expression) of MKI67. (c) UMAP of WNN-based integrated gene expression (GEX) and surface protein (ADT) clustering of ARC-ZNF721 and uninfected clones. Genes shown pass filters of log2FC > 0.1 and P-value < 0.05 adjusted for multiple testing from a two-sided Wilcoxon Rank Sum test. (d) Heatmap of top 15 differentially expressed genes for each clone (DEGs) between ARC-ZNF721 and uninfected clones from (c). (e) Top 7 enriched REACTOME pathways in ARC-ZNF721 relative to uninfected clones. (f) UMAP showing KLRG1 protein expression of ARC-ZNF721 and uninfected clones. (g) DEPs from ARC-ZNF721 and uninfected clones passing filters as in (a). (h) DEPs from ARC-Chr22 and uninfected clones passing filters as in (a). (i) DEPs from ARC-ZNF486 and uninfected clones passing filters as in (a). (j) Intracellular cytokine staining for IL-4 and IFN-γ (top) and staining for CD25 and FoxP3 (bottom) after in vitro re-stimulation with PMA/Ionomycin and protein transport inhibitors for 4 h in clones described in (a). (k) Intracellular cytokine staining for IL-4 and IFN-γ after in vitro re-stimulation with PMA/Ionomycin and protein transport inhibitors for 4 h in ARC-ZNF486 and ARC-ZNF721 cells. All heatmaps show z-scored expression. Data are from four combined (a-i) or representative of two (j,k) independent experiments.

Source data

We extended this to ARC-ZNF721 and to ARC-Chr22 and ARC-ZNF486, each compared with synchronously isolated autologous uninfected clones. Across these, transcriptional, surface protein and functional assays revealed cytotoxic T_H1-like programs, including cytotoxic programs in T cell lines enriched for ARC-Chr22 and ARC-ZNF486 (ref. ³¹), expression of IFNG, cytotoxic effector genes (EOMES and GZMK) and IFN-stimulated genes (APOBEC3G and TRIM22) in ARC-ZNF721, and elevated levels of surface markers of cytotoxicity, activation and IFNγ signalling (KLRG1, CD56, HLA-DR and CD86; Extended Data Fig. 2c–i). Cytokine profiling confirmed IFNγ production in dARC-COG5, ARC-ZNF486 and ARC-ZNF721 (more than 97% IFNγ⁺ cells), whereas ARC-Chr16 exhibited a FOXP3⁺ regulatory phenotype (Extended Data Fig. 2j,k). Module scoring using a CTL gene set demonstrated significant enrichment of cytotoxicity genes across all ARCs relative to uninfected clones (Fig. 2d), agreeing with recent reports describing cytotoxic skewing of ex vivo HIV reservoirs^27,31.

Distinct profiles of HIV-expressing cells

Because HIV transcription depends on activation factors (NF-κB and AP-1)^36,37, latency is associated with resting cells. Ex vivo studies infer that cells harbouring intact proviruses can proliferate without viral reactivation and generate progeny that produce virus after subsequent stimulation^29,38, but direct interrogation has been limited by the scarcity of these cells. Quantification of the frequencies of cells actively transcribing HIV ex vivo has been achieved for three large intact provirus clones, yielding estimates of 1.2–8.8%³⁹. No study has directly linked measures of viral expression with the activation and proliferation status of a reservoir clone.

We leveraged ARCs to address these gaps, first by assessing the frequencies of HIV RNA⁺ cells and defining transcriptional correlates of productive viral expression. By CITE-seq, HIV transcripts were detectable in all ARCs when using a permissive definition (HIV RNA count > 0), with HIV RNA⁺ frequencies of: 54.8% for dARC-COG5, 9.7% for ARC-Chr16, 2.7% for ARC-ZNF721, 0.3% for ARC-ZNF486 and 79% for ARC-Chr22 (Fig. 2e,f). Pairwise pseudobulk intraclonal comparisons of HIV RNA⁺ versus HIV RNA⁻ cells revealed minimal overlap of differentially expressed genes (DEGs) across all ARCs: only HIV and STMN1 were consistently differentially expressed (Fig. 2f and Extended Data Fig. 3a–c), agreeing with recent observations using cell cultures enriched for reservoir clones³³.

Extended Data Fig. 3. Low concordance of differentially expressed genes across ARCs when HIV RNA+ cells are defined using a permissive filter.

(a) Venn diagrams comparing the shared-across-ARC DEG sets obtained from (i) productive HIV RNA+ versus productive HIV RNA- cells defined by a stringent filter (red circles; Log-normalized HIV expression > 4 based on co-surface CD4 downregulation) and (ii) total HIV RNA+ vs HIV RNA- cells defined by a permissive filter (blue circles; HIV RNA count > 0). Analyses are shown including ARC-ZNF486 (top) or excluding ARC-ZNF486 (bottom). Numbers indicate the size of each shared DEG set and the overlap between permissive and stringent definitions. (b–c) Heatmaps showing expression (row-scaled z-score) for the 248 shared DEGs identified under the stringent “productive HIV RNA+” definition, plotted for total HIV RNA+ and total HIV RNA- cells within each ARC, demonstrating minimal concordance across ARCs when using the permissive definition. Total 248 DEGs from using stringent “productive HIV RNA+” filter are shown (b) as well as genes from selected enriched Hallmark gene-sets (c). (d) Flow-cytometry gating strategy distinguishing ARC-Chr16 from an autologous uninfected clone using TCRvβ 18 staining (TCRvβ 18 usage for ARC-Chr16 and lack of usage for uninfected clone), followed by gating of ARC-Chr16 cells into latent (Gag⁻) and active (Gag⁺) populations based on HIV-Gag and CD4. Overlaid histograms show DAPI DNA content for latent vs active ARC-Chr16 populations across technical replicates; the uninfected clone is shown as a negative control for HIV-Gag. (e) Percentage of cells with 4 N DNA content of latent (Gag−), active (Gag+), and uninfected clone populations; bars show mean of triplicates. Error bars denote SD.

Source data

We applied an alternative approach informed by HIV-mediated CD4 downregulation (Fig. 1g and Extended Data Fig. 1f). Stratifying cells into CD4^low, we identified a threshold of HIV expression termed ‘productive HIV RNA⁺’ (log-normalized HIV counts > 4), which occurred at low frequencies across ARCs (1.1% for dARC-COG5, 2.8% for ARC-Chr16, 0.4% for ARC-ZNF721, 6.1% for ARC-Chr22 and 0.0% ARC-ZNF486; Fig. 2e,g). This revealed a conserved transcriptional program shared across four ARCs from two donors (248 DEGs). Pairwise pseudobulk differential expression analysis revealed consistent downregulation of CD3E together with upregulation of MHC class II (MHC-II) transcripts (HLA-DRA, HLA-DRB1 and HLA-DRB5), consistent with sustained TCR-linked activation in productive HIV RNA⁺ cells (Fig. 2h). Of note, the recently described host HIV restriction factor PTMA (encoding prothymosin-α) — which suppresses HIV transcription⁴⁰— was downregulated in productive HIV RNA⁺ cells across ARCs, compatible with relief of a transcriptional restriction.

DEGs were enriched in TNF–NF-κB signalling, G2M checkpoint, mitotic spindle and hypoxia. Of note, the AP-1 component genes FOS, JUN and FOSB, together with ATF3, were strongly upregulated within TNF-mediated NF-κB signalling, consistent with cooperative AP-1-mediated regulation of HIV expression. Downregulated genes were enriched for cell cycle (G2/M and spindle) and metabolic (hypoxia and glycolysis) programs.

Consistent with this transcriptional profile, HIV Gag⁺ ARC cells were enriched for 4 N DNA content, indicating accumulation at G2/M (Extended Data Fig. 3d,e), a known consequence of HIV Vpr activity⁴¹. Other notable DEG groupings are presented in Extended Data Fig. 4a–d. Among differentially expressed transcription factors, HMGB1 was consistently under-expressed, which was of interest given its reported role in suppressing HIV transcription⁴². Thus, productive HIV-expressing cells downregulate suppressors of HIV expression and resemble cells experiencing strong and sustained TCR-associated activation signals, yet exhibit suppressed proliferative programs.

Extended Data Fig. 4. Conserved transcriptional profile in productive HIV RNA+ cells across ARCs.

(a–d) DEGs passing FDR-adjusted P-value < 0.05 shared across dARC-COG5 and ARCs- Chr16, ZNF 721 and Chr22 between productive HIV RNA+ cells and productive HIV RNA- cells (passing log-normalized HIV expression > 4 for productive HIV RNA+) using a pair-wise two-sided Wald test belonging to groupings for transcription factors (a), cytokines and growth factors (b), cell differentiation markers (c) and protein kinases (d). Top annotations indicate ARC group and HIV RNA status; color indicates row-scaled expression (z-score).

ARCs remain largely latent despite activation

We then focused on HIV Gag protein expression as a prominent CTL target, adopting a dual anti-Gag staining to enhance specificity⁴³ (Fig. 3a). Across the panel of ARCs, basal HIV Gag expression ranged from 0.0% for ARC-ZNF486 to 3.4% for ARC-Chr22 (Fig. 3a,b), aligning with the relatively low and higher expression of these reservoir clones in enriched T cell lines^32,33. In parallel, we assessed HIV Gag expression in ARCs stimulated with anti-CD3/CD28, mimicking TCR engagement, or a combination of phorbol 12-myristate 13-acetate and ionomycin (PMA–I). Both stimuli activate the transcription factors NF-κB, NFAT and AP-1 and are considered to induce ‘maximal’ HIV reactivation^7,44,45. Although these stimuli modestly increased HIV Gag expression in some ARCs, in ARC-Chr22 they decreased frequencies of HIV Gag-expressing cells (Fig. 3b). Across all ARCs, most proviruses appeared remarkably latent, despite potent T cell activation (as evidenced by degranulation; Extended Data Fig. 5a). This corroborates recent results from highly clone-enriched lines, reporting limited responsiveness to a broader panel of latency-reversing agents³³.

Fig. 3. HIV expression and proliferation dynamics of ARCs in response to stimulation.

a, Flow cytometry quantification of HIV Gag in ARCs and an uninfected clone stimulated with anti-CD3/CD28 Dynabeads versus unstimulated. b, Summary data of Gag expression, extended to PMA–I stimulation (means of duplicates are shown). The boxed inset shows a magnified view of ARC-ZNF486. c, Correlation plot showing the relationship between 5′ LTR copies and readthrough copies in unstimulated, anti-CD3/CD28 Dynabead-stimulated, PMA–I-stimulated and bryostatin-stimulated cells from ARC-ZNF486, ARC-Chr16, ARC-POLR1HASP and ARC-ZNF721. Data points are mean values from duplicates for ARC-ZNF721 and ARC-ZNF486 and replicates of four for ARC-Chr16 and ARC-POLR1HASP. Spearman’s rank-order correlation and significance from a two-sided Spearman’s test are shown. d, ARC and uninfected clone proliferation following stimulation with anti-CD3/CD28 and feeder cells as assessed by CellTrace Violet (CTV) dilution on days 4 and 7 compared with unstimulated in ARC-ZNF721 and ARC-ZNF486 and an uninfected clone. e, Kinetics of HIV Gag expression, viral RNA release, cellular proliferation (inverse CTV mean fluorescence intensity (MFI)) and total cell counts over time following restimulation (ARC-Chr16, defective dARC-COG5 and uninfected control clone shown). Mean values from duplicates are shown. f, Proliferation of donor-derived T cells stimulated ex vivo. Integration site-specific PCR quantifies the infected clone by detecting the proviral integration site. TCR measures the rearranged TCR of the T cell clone (detects infected and uninfected subpopulations). Data show fold expansion normalized to baseline (day 0, left panels) and to contemporaneous RPP30 quantification (right panels), for ARC-ZNF486 and ARC-ZNF721 at the indicated sampling time points on ART. Data are from 12 replicates per condition. g, Summary plot comparing fold change of the ratio of ARC integration to TCR between ARC-ZNF721, ARC-Chr16, ARC-ZNF486 and dARC-COG5. Data are representative of two (a,b,e) or combined from three (c,d,f,g) independent experiments.

Source data

Extended Data Fig. 5. Cytotoxic profile of ARCs ZNF486 and ZNF721 and robust degranulation following anti-CD3/CD28 and PMA/ionomycin stimulation.

(a) Representative flow-cytometry plots showing intracellular granzyme A (GZMa) and granzyme B (GZMb) expression in ARC-ZNF721, ARC-ZNF486, and an uninfected CD4⁺ T-cell clone under unstimulated conditions or following stimulation with anti-CD3/CD28 or PMA/ionomycin. (b) Quantification of HIV transcripts detected by targeted RNA sequencing following stimulation, showing read-through, 5’ LTR, Ψ, env, and polyA signals in ARC-ZNF721, ARC-Chr16, ARC-ZNF486, and ARC-POLR1HASP. Values are shown as copies per cell. Right, percentage of HIV-Gag⁺ cells measured by flow cytometry under the same conditions. Bars depict mean values of duplicates in all clones except replicates of 4 for ARC-Chr16 HIV transcripts. (c) Fold change in 5’ LTR signal and percentage of Gag⁺ cells following stimulation with anti-CD3/CD28, PMA/ionomycin, or bryostatin in the indicated ARC clones. Bars depict mean values of duplicates in all clones except replicates of 4 for ARC-Chr16 and ARC-POLR1HASP 5’ LTR signal. (d) Percentage of mature HIV transcripts expressed as the ratio of polyA to 5’ LTR signal, following stimulation with anti-CD3/CD28, PMA/ionomycin, or bryostatin in the indicated ARC clones. Bars represent mean values of duplicates for ARC-ZNF721 and ARC-ZNF486 and replicates of 4 for ARC-Chr16 and ARC-POLR1HASP. (e) Representative flow-cytometry plots showing HIV-Nef and Gag staining on ARC-Chr16 and dARC-COG5, with an uninfected CD4⁺ T cell clone control. All data are from 3 combined independent experiments.

Source data

Impaired HIV transcript completion in ARCs

To dissect the transcriptional basis of these observations, we used reverse transcription digital droplet PCR (ddPCR) to quantify HIV transcriptional initiation, read-through, elongation and polyadenylation in four ARCs⁴⁶ (Extended Data Fig. 5b). Transcriptional profiles were markedly heterogeneous across clones. Levels of 5′ long terminal repeat (LTR)-initiated transcripts varied widely and strongly correlated with transcriptional read-through, implicating basal non-HIV transcriptional activity at the integration locus as a major determinant of HIV expression (Fig. 3c). Of note, in ARC-ZNF721 fewer than approximately 33% of initiated transcripts matured to elongated and polyadenylated forms, consistent with a block to elongation and/or polyadenylation for this clone (Extended Data Fig. 5b–d). Stimulation with anti-CD3/CD28, PMA–I or bryostatin increased transcriptional initiation in three ARCs (Chr16, ZNF486 and POLR1HASP), but reduced the relative fraction of fully mature transcripts, indicating that increases in initiation do not necessarily lead to increases in translation-competent viral RNA. Together, these data indicate that ARC HIV transcription is shaped by basal transcription at the integration site but remains constrained downstream of initiation.

ARC proliferation uncoupled from latency

Given the limited induction of HIV, we asked whether ARCs proliferate under the same anti-CD3/CD28 conditions. In ARC-ZNF721, we observed robust proliferation that was uncoupled from latency reversal, with only 0.57% HIV Gag expression on day 7 (Fig. 3d). ARC-ZNF486 proliferated similarly, but with even less Gag expression. We extended this to a trio of CD4⁺ T cells clones isolated contemporaneously from OM5334: ARC-Chr16, dARC-COG5 and an uninfected clone. Following stimulation with anti-CD3/CD28, ARC-Chr16 displayed the highest increase in total cell counts, consistent with its proliferative transcriptional profile, yet only a fraction expressed HIV Gag (Fig. 3e). This frequency was stable and mirrored by the release of HIV RNA into supernatants. dARC-COG5 showed the highest levels of HIV Gag expression 24 h post-restimulation, which then declined steadily (Fig. 3e).

ARCs proliferate above average, but infected subsets lag

We compared the proliferation of ARC-ZNF486 and ARC-ZNF721 to autologous uninfected CD4⁺ T cells clones isolated in parallel. Across multiple subcloning wells, both ARCs proliferated at rates comparable with uninfected clones (Fig. 3d and Extended Data Fig. 6a–c). We then applied integration site-specific and TCR–CDR3-specific assays to ARC-ZNF486 (Extended Data Fig. 7b), as we had done with ARC-ZNF721, and tracked proliferation dynamics from total ex vivo CD4⁺ T cells from these donors stimulated with anti-CD3/CD28 to enable more robust comparisons of ARCs relative to average T cells. Relative to total CD4⁺ T cells, ARC-ZNF486 expanded at a similar rate (Fig. 3f and Extended Data Fig. 6d), whereas ARC-ZNF721 markedly outcompeted the bulk population, increasing by 8.7-fold in frequency by day 11 (Fig. 3f and Extended Data Fig. 6e). ARC-Chr16 and dARC-COG5 also expanded more than the average T cell, although less dramatically than ARC-ZNF721.

Extended Data Fig. 6. Proliferation and HIV-Gag expression of ARCs.

(a–c) Flow-cytometry of cell proliferation measured by CellTrace Violet dilution and HIV Gag expression in ARC-ZNF721 (b) and ARC-ZNF486 (c) together with autologous uninfected clones isolated and cultured in parallel. For participant OM5334, uninfected control clones (a; Uninfected 1, 8, and 5) are shown alongside ARC-ZNF721 subclones (different wells from subcloning plate) (b; wells 2–4). For participant 603, an uninfected control clone and ARC-ZNF486 subclones (c; wells 1–2) are shown. Cells were analyzed either without stimulation (Unstimulated) or following anti-CD3/CD28 Dynabead stimulation for 4 or 7 days (Day 4 and Day 7, respectively). Each column represents an independently derived clone/subclone (as labelled). Numbers indicate percentages of cells proliferating (x-axis) and/or expressing HIV Gag (y-axis). (d,e) LIST assay analyzed proliferation of donor-derived T-cells stimulated ex vivo. Results are analogous to those in Fig. 3f but use different blood draw time points for both donors 603 (2 weeks prior to an ATI) (d) and OM5334 (4.1 years and 6.0 years on ART) (e). Integration site (IS)-specific PCR quantifies the infected clone by detecting the proviral integration site. TCR measures the rearranged T-cell receptor of the T-cell clone (detects infected and uninfected subpopulations). RPP30 quantifies total T-cells. Data show fold-expansion normalized to baseline (day 0, top panel) and to contemporaneous RPP30 quantification (total T-cells) (bottom panel). Data are representative of 3 (a–c) or combined from 2 (d,e) independent experiments.

Source data

Extended Data Fig. 7. T cell responses and longitudinal dynamics of HIV reservoir clones.

(a) HIV-specific CTL responses measured by IFN-γ ELISPOT. For donor OM5334, shown are results of matrix pool mapping with overlapping 15mer peptides spanning HIV-Pol. The commonality of all peptide pools eliciting responses was the presence of the NETPGIRY epitope (highlighted in red). Responses were measured across multiple years on ART. For donor 603, responses to pools of overlapping 15mer peptides spanning the indicated HIV gene products are shown. These were measured at two time points spaced 25 weeks apart. Data shown are mean values from duplicates. (b) Longitudinal tracking of ARC-ZNF486, showing its relative contribution to the total TCR clonotype frequency and total infected cells over time on ART. (c) IFN-γ and granzyme B ELISPOT responses of PBMCs from donors with persistent low-level viremia (donors ‘F-07, C-03, T-13’) despite ART, measured in response to individual HIV peptides. Data shown are mean values of IFN-γ response from replicates of 2 for donors F-07 and C-03 and a replicate of 1 for T-13’ and mean values of granzyme B response from replicates of 4 in each donor. (d) Flow-cytometry assessment of HIV-Gag expression in ARC-Chr16 following CD4-based fractionation into CD4⁺ enriched and CD4⁻ enriched fractions (the latter are enriched for HIV-expressing cells). (e) HIV-Gag expression was monitored at days 0, 5, and 10 post-fractionation. Quantification of changes in the frequency of HIV Gag⁺ cells following CD4-based fractionation (CD4⁻ enriched vs CD4⁺ enriched) over days 0, 5, and 10. Data shown as mean values from duplicates. Data are from one (a,b), three combined (c) or representative of two (d,e) independent experiments.

Source data

Across all four ARCs or dARC tested, cells bearing the TCR clonotype (which include uninfected cells) expanded to greater numbers than the infected subpopulation, defined by the paired HIV integration site (Fig. 3g). This suggests appreciable death of the HIV-infected fractions of clones and/or of cell cycle arrest of HIV-expressing cells, as observed in Extended Data Fig. 3d,e. Because these depend on HIV expression, this suggests that a substantial fraction of ARCs reactivated over this period. Alternatively, these results might reflect virus-independent heterogeneity in proliferative potential within a T cell clone, with infection occurring in a relatively non-proliferative cell in every case.

Vulnerability of proliferating ARCs to CTLs

Shock and kill aims to pair latency reversal with immune effectors to eliminate reservoir cells⁴⁷. That only a small fraction of ARC cells expresses HIV Gag at any given time, even following strong stimulation, would seemingly undermine this strategy. However, in donor OM5334, the fraction of infected cells within each ARC-associated TCR clonotype declined steadily over the first 4.1 years of ART (Fig. 4a and Extended Data Fig. 1c), suggesting immune-mediated selection despite low instantaneous antigen expression. Proteome-wide mapping of ex vivo HIV-specific T cells in OM5334 revealed a single response, targeted to the HLA-B*18-restricted Pol-NETPGIRY epitope (Fig. 4b and Extended Data Fig. 7a). We isolated the corresponding CD8⁺ T cell clone for killing assays with autologous ARCs. Comparison of this CTL clone with the matched ex vivo response revealed a striking dissociation between cytokine production and cytotoxic capacity. Whereas the CTL clone produced IFNγ and GzmB in roughly equal measure, the ex vivo response produced IFNγ without detectable levels of GzmB (Fig. 4c), a key mediator of cytotoxic killing.

Fig. 4. Sustained CTL pressure reveals antigenic vulnerability and intrinsic resistance of HIV reservoir clones.

a, Longitudinal tracking of ARC-ZNF721, ARC-Chr16 and dARC-COG5 as a percentage of the total TCR clonotype. Data are from 12 replicates per condition. b, Longitudinal tracking of CTL responses using whole HIV gene products by ELISPOT. Data are from duplicates, with the line representing the mean. SFU, spot-forming units. c, ELISPOT plots showing IFNγ and GzmB responses to Pol-specific epitopes alongside anti-CD3/CD28 and CMV/EBV/Flu. d, GzmB response of PBMCs cultured with IL-15 to anti-CD3/CD28, CMV/EBV/Flu and the Pol-specific peptide. Data are from triplicates and shown as mean ± s.d. e, Representative slowcure assay flow cytometry plots showing HIV Gag expression and proliferation of ARC-ZNF721 with or without autologous HIV-specific CTL conditions (effector:target ratio 1:2). f, HIV Gag⁺ (top) and total (bottom) ARC-ZNF721 counts over 7-day co-culture CTLs (data are from duplicates, with the line representing the mean). g, Representative slowcure assay flow cytometry plots for ARC-POLR1HASP showing proliferation and HIV Gag expression with or without autologous HIV-specific CTL conditions (effector:target ratio 1:2). h, HIV Gag⁺ (left) and total (right) ARC-POLR1HASP counts over 7-day co-culture ± CTLs (data are from duplicates, with the line representing the mean). i, Representative flow cytometry plots showing results on day 7 of a slowcure of mixed populations of ARC–dARCs with uninfected clones. dARC-COG5 is identified by dual staining with TCRVβ 7.1 antibodies versus with TCRVβ 18 for ARC-Chr16. j, Summary results from panel i showing the frequencies of dARC-COG5 (left) and ARC-Chr16 (right; data are from duplicates, with the bar depicting the mean). k,l, Representative proliferation and HIV Gag expression (k) and quantitative data (l) for slowcure assays (data are from duplicates, with the bar representing the mean). dARC-COG5 is from the mixed culture in panel i, gated on TCRVβ 7.1 cells. The ARC-Chr16 and uninfected clones are from monocultures ± CTLs set up in parallel. Data are from one (a–d) or representative of two (e–l) independent experiments.

Source data

This cytotoxic defect was reversible: pre-treatment of peripheral blood mononuclear cells (PBMCs) with IL-15 increased GzmB response to peptide (Fig. 4c,d). We extended this to three PWH with persistent low-level viraemia originating from large expanded intact proviral clones²⁵ (repliclones; Supplementary Table 2). Although we were similarly able to readily detect IFNγ-producing responses against epitopes that were intact in the repliclones, GzmB production was largely absent (Extended Data Fig. 7c). These data reveal a dissociation between antigen recognition and cytotoxic competence in dominant HIV-specific CD8⁺ T cell responses, and demonstrate that impaired function can be pharmacologically restored.

Because the CTL clone exhibited greater cytotoxic capacity than the ex vivo CD8⁺ T cell response, and ARC activation and proliferation could be synchronously induced, we hypothesized that CTL-mediated reservoir erosion could be achieved in vitro, despite limited instantaneous antigen expression. We termed this experimental framework the ‘slowcure’ assay. We applied this to ARC-ZNF721, inducing proliferation with anti-CD3/CD28 and co-culturing cells with or without the Pol-specific CTL clone. As expected, overnight exposure to CTLs resulted in only a modest reduction in HIV Gag⁺ cells, with minimal effect on the total ARC population (Fig. 4e,f). By day 5, however, we observed substantial reductions in both HIV Gag⁺ cells and total ARC numbers, particularly among cells that had proliferated. By day 7, the presence of CTL reduced ARC numbers by more than 90%, despite fewer than 2% of cells expressing Gag at any individual time point.

We reasoned that dynamic antigen expression over time could expose a substantially larger cumulative fraction of cells to CTL-mediated killing than suggested by any snapshot. We developed a mathematical model describing transient latency reversal events in ARC-ZNF721, informed by: (1) temporal expression data (Fig. 4e,f), and (2) complementary perturbation experiments in which HIV-expressing cells were selectively depleted and HIV expression was monitored relative to non-depleted populations. Under the stimulation conditions of the slowcure assay, we estimated an average interval of 50 (42–63) days between latency-reversal events per cell, with a mean duration of expression of 0.79 (0.71–0.89) days, corresponding to an instantaneous frequency of approximately 1.5% (1.4–1.7%) HIV-expressing cells (Extended Data Fig. 7d,e and see Supplementary Tables 4–6 and Supplementary Figs. 2–8 in Supplementary Note 1). Integrating these dynamics over the 7-day co-culture period predicts that approximately 15% (12–18%) of ARC-ZNF721 cells transiently expressed Gag at least once, providing a quantitative framework through which limited instantaneous antigen expression can support substantial CTL-mediated reservoir depletion. At the same time, these estimates indicate that cumulative antigen exposure alone (at least as measured by Gag flow cytometry) cannot fully account for the more than 90% reduction in ARC abundance observed by day 7, implicating additional antigen-independent mechanisms in clone erosion.

Extending these results to a second ARC, ARC-POLR1HASP from donor 908 targeted with an autologous HLA-B*2705-restricted HIV Gag-KK10-specific CTL clone, we observed similar results to those from ARC-ZNF721, a more than 90% reduction in total ARC cells versus the no CTL condition (Fig. 4g,h). However, for a different ARC (Chr16) from donor OM5334, we observed a strikingly different outcome, with only modest reduction in either the frequencies of HIV Gag⁺ cells or total ARC cells at days 1, 4 and 7 of co-culture with the same HIV Pol-specific CTL clone that had been effective against ARC-ZNF721 (Extended Data Fig. 8a–c). The CTL epitope (Pol-NETPGIRY) was confirmed to be intact in the ARC-Chr16 virus. Results were suggestive of a modest CTL impact with a trend towards lower ARC numbers relative to a no CTL control at day 7, and MHC-I-blocking experiments suggested limited reductions in frequencies of HIV Gag⁺ cells (Extended Data Fig. 8c,d). However, in contrast to the result with ARC-ZNF721, the overall outcome was robust expansion of ARC-Chr16 despite the presence of CTLs (Extended Data Fig. 8c).

Extended Data Fig. 8. Limited susceptibility of proliferating ARC-Chr16 to CTL-mediated clearance in mixed-target co-cultures.

(a) Representative flow cytometry plots showing HIV-Gag protein expression and proliferation (CellTrace Violet dilution) of ARC-Chr16 and an uninfected CD4⁺ T-cell clone (bottom panels) in the presence or absence of an HIV-Pol-specific CTL clone at days 1, 4, and 7 of co-culture. (b) Flow cytometry analysis showing ARC (CD4) and HIV-specific CTL clone (CD8) proportions at initiation of co-culture (top: −CTL condition, bottom: +HIV-spec CTL clone). (c) Quantification of total cell numbers over time for ARC-Chr16 and uninfected control clones cultured with or without HIV-specific CTL (normalized to counting beads). Data points represent mean values from duplicates. These data confirm that the presence of CTL does not broadly impair CD4⁺ T-cell proliferation in these co-cultures and that expanding ARC-Chr16 cells are not efficiently eliminated. (d) Percentage of HIV-Gag⁺ cells in ARC-Chr16 co-cultures at days 1, 4, and 7 in the presence or absence of HIV-specific CTL clone, and with or without MHC-I blockade or isotype control antibodies. Bars represent mean values from duplicates, indicating modest MHC-I-dependent CTL-mediated reductions in HIV-Gag⁺ ARC-Chr16 cells. (e) Mixed-target CTL co-culture assay. Within viable CD3⁺CD8⁻ lymphocytes, CellTrace Yellow intensity was used to distinguish three target populations present in the same culture: dARC-COG5 (CTY negative), ARC-Chr16 (CTY dim), newly in vitro infected targets (CTY bright). Upon addition of CTL (bottom panel), HIV-Gag⁺ cells are largely eliminated from the dARC population, remain largely unchanged in the ARC-Chr16 population, and are reduced in the newly in vitro infected targets (replicates and stats in Fig. 5b). (f) Daughter plots corresponding to the gates defined in e (No CTL) showing dual HIV-Gag⁺ gates used to determine frequencies of infected cells. An uninfected staining control is shown on the left. (g) Intracellular BCL-2 staining of the indicated target populations (inclusive of HIV-Gag⁺ and HIV-Gag⁻) under no CTL condition. (h) Identification of ARC-Chr 16 via TCRvβ staining. (i–l) Response of ARC-Chr16 and uninfected control cells to BCL-2 inhibition at increasing concentrations, shown as changes in ARC frequency among viable cells, total viable cell numbers, and BCL-2 mean fluorescence intensity in surviving cells. Data are shown as mean ± SD values from triplicates. All data are representative of two independent experiments.

Source data

To investigate this finding, we compared ARC-Chr16 and dARC-COG5 in slowcure assays, including internal controls of autologous uninfected CD4⁺ T cell clones. For dARC-COG5, we used a partially subcloned ARC precursor containing both the defective clone and an uninfected clone, ensuring identical cell culture histories. For ARC-Chr16, this was simulated by spiking in a parallel history autologous uninfected clone, distinguishing clones by TCRVβ segments. Over the 7-day CTL co-culture, dARC-COG5 was potently and specifically eliminated, whereas ARC-Chr16 showed only modest selective elimination (Fig. 4i,j). Monocultures of ARC-Chr16 and uninfected clones, analysed alongside TCRVβ 7.1-gated dARC-COG5, further corroborated the relative resistance of ARC-Chr16 to CTL-mediated killing (Fig. 4k,l). Thus, despite transient HIV Gag expression, ARC-ZNF721 and dARC-COG5 were highly vulnerable to CTL elimination, whereas ARC-Chr16 was only modestly impacted. The persistence of an HIV Gag⁺ population in ARC-Chr16 under sustained CTL pressure suggests that a barrier other than latency limits CTL-mediated elimination of this clone. Applying the same antigen-dynamics modelling framework to ARC-Chr16 inferred a slower rate of latency reversal but longer-lived expression episodes than ARC-ZNF721 (see Supplementary Tables 4–6 and Supplementary Figs. 2–4 in Supplementary Note 1), indicating that reduced antigen availability alone is unlikely to explain the limited effect of CTLs on ARC-Chr16.

Cell-intrinsic CTL resistance

To determine whether the limited susceptibility of HIV Gag-expressing ARC-Chr16 to CTL-mediated killing was due to viral or cellular factors, we tested whether the Pol-specific CTL clone could kill bulk autologous CD4⁺ T cells newly infected with ARC-Chr16-derived virus. By differentially labelling these newly in vitro infected cells, and co-culturing with dARC-COG5 and ARC-Chr16, we assessed susceptibility to CTL of HIV Gag⁺ cells in these three target populations in competitive killing assays (Fig. 5a, co-culture 1). CTLs efficiently eliminated HIV Gag⁺ dARC-COG5 cells but had no detectable effect on HIV Gag⁺ ARC-Chr16 cells, aligning with their respective results in the slowcure assays (Fig. 5b). The majority (more than 75%) of bulk CD4⁺ T cells newly infected with virus from ARC-Chr16 were efficiently eliminated by CTLs, indicating that the lack of elimination of HIV Gag⁺ ARC-Chr16 was attributable to cellular properties rather than virus. Approximately 25% of newly infected HIV Gag⁺ bulk CD4⁺ T cells survived at both 1:10 and 1:2 effector:target ratios (Fig. 5c). This is typical of a phenomenon that we have recently described, resulting from a subset of CTL-resistant CD4⁺ T cells^17–19,48. Thus, ARC-Chr16 exhibited CTL resistance associated with cellular rather than viral properties.

Fig. 5. Cell-intrinsic resistance of a reservoir clone to elimination by CTL.

a, Schematic of competitive killing assays with ARC–dARC and autologous newly infected CD4⁺ T cells with ARC-derived virus using different intensities of cell trace dye distinguishing targets. Schematic created in BioRender; Jones, B. https://biorender.com/v55w3ck (2026). b,c, Quantitative data from competitive killing measuring CTL susceptibility of HIV Gag⁺ cells from dARCs or ARCs (b) or CD4⁺ T cells newly infected with an ARC-derived virus (c). The Kruskal–Wallis test and Dunn’s multiple-comparison test (two-sided) were used for statistical analyses, and data are from six replicates (dARC-COG5, ARC-ZNF721, ARC-Chr16 and newly infected cells) or five replicates (ARC-POLR1HASP and newly infected cells) shown as mean ± s.d. d, Quantification of CTL killing of peptide-pulsed or unpulsed ARC-Chr16 and dARC-COG5 ± MHC-I-blocking antibody (Ab; data are from duplicates, with the bar depicting the mean). e, Summary plots showing effect of treatment of cells from ARC-Chr16 with DMSO or the BCL-2 inhibitor ABT-199 or the BCL-xL inhibitor A1331852 (A-852) for 24 h before co-culturing with Pol-specific CTLs on Gag⁺ cell frequency in comparison with cells treated with DMSO and no CTL. Data are from triplicates (mean ± s.d.). f, Violin plots showing expression of HIF1A, CD69 protein and module scores of Hallmark hypoxia and Reactome ‘KEAP1–NFE2L2 pathway’ gene sets on ARC-Chr16 and dARC-COG5. The blue dash denotes mean expression. P values shown are from a two-sided Wilcoxon test. g, Representative flow cytometry plots of a peptide pulse killing assay from ARC-Chr16 cells either pre-treated with 10 μM DFO or DMSO for 72 h before differential CTV staining based on whether pulsed with a Pol-SV15 peptide or mock pulsed and cultured overnight with Pol-specific CTL at a 3:1 effector:target ratio. FSC, forward scatter. h, Summary plots of ARC-Chr16 data from panel g showing triplicates as mean ± s.d. i, Summary plots of killing assays testing the effects of DFO (72 h pre-treatment versus DMSO as vehicle) on memory CD4⁺ T cells from two donors in vitro infected with HIV-JRSCF and co-cultured with or without Gag-TW10-specific CTL clones. Symbols denote two and three replicates across donors (mean ± s.d.). j,k, Slowcure assay evaluating DFO on viral RNA release, proliferating cell counts and HIV Gag⁺ ARC cells (ARC-Chr16 (top) and dARC-COG5 (bottom) with or without CTL): representative flow cytometry at day 7 (j), and quantitative data from days 1, 4 and 7 (k; data are from duplicates, with the line representing the mean). l,m, Slowcure assay evaluating DFO and ABT-199 on proliferating cell counts and HIV Gag⁺ ARC cells (dARC-R3HDM2 ± CTL): representative flow cytometry at day 7 (l), and quantitative data of proliferating and total ARC counts from days 1, 4 and 7 (m; data are from duplicates, with the line representing the mean).

Source data

We performed competitive killing assays with two other ARCs (ZNF721 and POLR1HASP) to determine whether CTL resistance is a universal property of ARCs. In co-cultures with ARC-ZNF721 and ARC-Chr16 (Fig. 5a, co-culture 2), CTL again failed to impact Gag-expressing ARC-Chr16 while modestly reducing Gag-expressing ARC-ZNF721, suggesting intermediate CTL susceptibility of the latter (Fig. 5b). In comparing relative susceptibility to CTLs between ARC-POLR1HASP and cells newly in vitro infected with virus from this ARC (Fig. 5a, co-culture 3), HIV Gag-expressing cells were killed to a similar degree to autologous CD4⁺ T cells infected with ARC-derived virus (Fig. 5b,c). Thus, CTL susceptibility varies markedly across ARCs and can be dictated by cell-intrinsic properties.

To investigate the ability of ARC-Chr16 to survive CTL attack, we asked whether this was tied to a deficiency in HIV antigen processing and/or presentation. For both ARC-Chr16 and dARC-COG5, cells were pulsed with cognate HIV Pol peptide and cultured with or without CTL. The CTL clone degranulated in response to peptide, but whereas dARC-COG5 was efficiently eliminated in a peptide and MHC-I-dependent manner, ARC-Chr16 was only modestly impacted, indicating resistance downstream of CTL recognition (Fig. 5d and Extended Data Fig. 9a–c). We found that ARC-Chr22 also exhibited CTL resistance and confirmed moderate resistance in ARC-ZNF721, whereas ARC-ZNF486 and ARC-ZNF561 were highly susceptible (Extended Data Fig. 9d,e). One of the uninfected clones used as a comparator to ARC-ZNF561 resisted CTL killing, further supporting that relative CTL susceptibility is a variable and clone-specific features of CD4⁺ T cells.

Extended Data Fig. 9. Intrinsic susceptibility of HIV reservoir clones to CTL-mediated killing and lack of protection by MHC-I downregulation.

(a) Schematic of a two-clone competitive killing assay in which an ARC and an autologous uninfected CD4⁺ T-cell clone, isolated and maintained in parallel, are differentially labeled, peptide-pulsed, and co-cultured with CTL. (b) Validation of CTL clone specificity and functionality. CTL clones are consistently tested within 2 days of use by assessing degranulation (CD107a) in response to peptide stimulation. Representative results are shown for the HIV Pol-specific CTL clones used throughout this study and a CMV-specific CTL clones as a comparator (with CMV peptide). (c) Quantification of CTL killing of peptide-pulsed/unpulsed ARC-Chr16 and dARC-COG5 labeled with differential CFSE intensities ±MHC-I blocking antibody in duplicates (mean ± SD). Shown are representative flow-cytometry plots corresponding to the summary data shown in Fig. 5d. Schematic created in BioRender; Jones, B. https://biorender.com/xmat3rp (2026). (d) Results of two-clone peptide-pulse competitive killing assays comparing ARC-Chr22 (n = 6, distinct subcloning wells) with six distinct autologous uninfected CD4⁺ T-cell clones (n = 6). Each line represents an independent ARC-uninfected clone pairing. ARC-Chr22 consistently contributes a higher fraction of surviving infected cells, indicating relative resistance to CTL-mediated killing. (e) Competitive killing assays comparing individual ARC with three autologous uninfected CD4⁺ T-cell clones per ARC, distinguished by differential cell-labeling intensities. In a peptide-pulsed setting, ARC-ZNF486 (highly latent) is more susceptible to CTL-mediated killing than all three comparator uninfected clones. ARC-ZNF561 shows intermediate susceptibility, whereas ARC-ZNF721 exhibits moderate resistance relative to each comparator alone. Data are shown as mean ± SD values from triplicates. (f) Flow-cytometry analysis of MHC-I expression gated on target populations from competitive killing assays featured in Fig. 5a,b. ARC-Chr16 shows downregulation of MHC-I on HIV-expressing cells, whereas dARC-COG5 does not. Cells newly infected in vitro with ARC-Chr16-derived virus display heterogeneous MHC-I downregulation. Analysis of surviving infected cells shows no enrichment for MHC-I-low cells among survivors (highlighted by red circles), suggesting that the degree of MHC-I downregulation induced by ARC-Chr16 does not confer protection from CTL-mediated killing under these conditions. (g) Overnight killing assays testing the effects of deferoxamine (DFO, 72 h treatment) on memory CD4⁺ T-cells in vitro-infected with HIV-JRSCF and co-cultured with/without Gag-TW10 specific CTL clones. Representative flow cytometry plots after the killing assay from one of two donors. Summary results from both donors are shown in Fig. 5i. All data are representative of two independent experiments.

Source data

Overcoming CTL resistance

To overcome CTL resistance in ARC-Chr16, we first evaluated the potential role of the pro-survival factor BCL-2 (ref. ¹⁷). Although both ARC-Chr16 and the CTL-vulnerable dARC-COG5 exhibited higher levels of BCL-2 than bulk CD4⁺ T cells, expression was highest in dARC-COG5, arguing against BCL-2 as the dominant driver of resistance in ARC-Chr16 (Extended Data Fig. 8e–g). Consistent with this, pharmacological inhibition of BCL-2 or BCL-xL did not sensitize ARC-Chr16 to CTL-mediated killing (Fig. 5e). Of note, using an impure ARC-Chr16, we did observe that the BCL-2 and BCL-xL antagonists each induced a degree of death in the ARC, but not in the co-cultured uninfected bystanders (Extended Data Fig. 8h–l). Thus, although the previously reported sensitivity of reservoir cells to BH3 inhibitors was observed, this did not underlie the CTL resistance for this particular ARC.

We considered the role of MHC-I and observed that for HIV Gag⁺ cells, MHC-I expression was downregulated on both ARC-Chr16 and newly infected targets, but not dARC-COG5, with the defect impacting HIV Nef expression (Extended Data Fig. 5e). Although this may contribute to the particular sensitivity of dARC to CTL, it cannot explain the discordance between ARC-Chr16 and newly infected targets, as survivors from the latter did not show selection for MHC-I^low cells (Extended Data Fig. 9f), nor does MHC-I downregulation in HIV Gag⁺ cells explain the insensitivity of peptide-pulsed total ARC-Chr16 to CTLs.

To identify other mechanisms by which ARC-Chr16 may resist CTLs, we cross-referenced CITE-seq analysis of ARC-Chr16 and dARC-COG5 with recent multiomic profiling of cells that survive CTL killing in an in vitro infection system¹⁹. In this model, relatively quiescent (CD69^low) cells with reduced expression of the hypoxia-promoting HIF1A and lower oxidative stress response preferentially survived CTL co-culture, probably due to the role of reactive oxygen species (ROS) in death downstream of a granzyme–perforin hit⁴⁹. Concordantly, ARC-Chr16 showed significantly lower levels of HIF1A, surface CD69 expression as well as oxidative stress response modules than the CTL-vulnerable dARC-COG5 (Fig. 5f).

We tested whether the FDA-approved iron chelator deferoxamine (DFO), which induces hypoxic stress and ROS accumulation¹⁹, could overcome resistance in ARC-Chr16. In peptide-pulsed killing assays, DFO markedly enhanced CTL-mediated killing of ARC-Chr16 (Fig. 5g) and newly HIV-infected bulk CD4⁺ T cells (Fig. 5h,i and Extended Data Fig. 9g). In slowcure assays, CTLs alone modestly limited ARC-Chr16 expansion (3.6-fold increase), whereas DFO plus CTL nearly abrogated clonal expansion (1.1-fold; Fig. 5j,k). This was mirrored by viral RNA measurements, which rose 40-fold in monoculture, 7.6-fold with CTL alone and 1.8-fold with the combination (Fig. 5k).

In parallel, dARC-COG5 — already enriched for oxidative stress programs — was efficiently eliminated by CTLs alone, and DFO induced cell death even in the absence of CTL. This supports a model in which baseline oxidative stress establishes a death threshold: dARC-COG5 resides near this threshold and is readily eliminated, whereas ARC-Chr16 requires metabolic sensitization for effective CTL-mediated killing. Extending slowcure assays to an additional dARC (Uniprot ID: R3HDM2), we tested DFO, the BCL-2 inhibitor ABT-199 and CTLs individually and in combination (Fig. 5l,m). CTLs curtailed expansion of this dARC, whereas DFO alone induced marked cell death. Although ABT-199 had minimal effect as a single agent, it significantly enhanced CTL-mediated clearance in combination. These results reveal substantial heterogeneity in CTL susceptibility across ARCs and demonstrate that cell-intrinsic resistance mechanisms are both mechanistically identifiable and selectively reversible.

Discussion

By isolating ARCs ex vivo, we have established a platform for immunological, virological and functional interrogation of persistent HIV reservoirs at clonal resolution. ARCs are diverse yet consistently skew towards cytotoxic phenotypes, aligning with ex vivo observations^27,31. We functionally confirmed this skew by demonstrating high basal granzyme expression and granzyme release upon stimulation. Although prolonged culture may shift aspects of cellular state, ARCs retain distinctive and stable multiomic profiles over months and recapitulate key signatures of ex vivo reservoir cells, supporting their use for mechanistic studies.

We have shown that productive HIV-expressing (CD4-downregulated) cells converge on a conserved transcriptional program across independent ARCs, revealing a coordinated interplay between host activation pathways and virus-driven perturbations. This program integrates sustained TCR-linked signalling with suppression of proliferative and metabolic pathways and highlights candidate regulatory nodes — such as the HIV restriction factor PTMA — that may be leveraged to enhance HIV expression and sensitize infected cells to elimination. Of note, this convergence occurs despite substantial interclone heterogeneity in basal transcriptional and proliferative programs, indicating that productive HIV expression drives cells into a shared and constrained host state. Future studies pairing chromatin accessibility with transcriptional profiling of Gag⁺ versus Gag⁻ cells will be required to define the epigenetic features gating productive HIV expression within individual clones.

Despite limited instantaneous HIV expression, three of four ARCs exhibited pronounced time-integrated vulnerability to CTL-mediated attack during proliferation, supporting a model in which interception of transient HIV expression enables gradual reservoir erosion when CTLs are sufficiently potent. The marked erosion of rebound-competent reservoirs observed in ‘elite controllers’ may reflect this mechanism, given their hallmark robust CTL responses¹². By contrast, ex vivo HIV-specific CD8⁺ T cell responses from donor OM5334 and donors with repliclones showed impaired cytotoxicity, suggesting that endogenous responses may be poorly positioned to exploit dynamic antigen exposure. Strategies that enhance CTL cytotoxic capacity, including IL-15-based interventions, may therefore be required to translate transient antigen expression into meaningful reservoir reduction in vivo.

The magnitude of CTL-mediated reductions in ARC accumulation exceeded that predicted by cumulative antigen expression alone, probably reflecting additional antigen-independent mechanisms such as CTL-derived antiproliferative cytokines, including IFNγ. This is consistent with modest reductions observed in uninfected CD4⁺ T cells during co-culture. However, our definition of antigen expression (Gag⁺CD4⁻) may underestimate the fraction of ARCs presenting antigen, as HIV transcripts were detected in a larger population of cells. Antigen presentation below the threshold of Gag detection, as well as bystander mechanisms such as Fas–FasL engagement, may therefore contribute to clone erosion under sustained CTL pressure.

Even under conditions of extreme CTL pressure, a subset of ARC consistently persisted through multiple rounds of proliferation, as well as non-proliferating survivors. Further study is needed to understand the basis of both of these modalities of survival, and whether complete elimination can be achieved through either improved latency reversal or CTL sensitization strategies. In addition, evaluating CTL efficacy against ARCs independent of TCR stimulation, for example, using latency reversal with Tat-containing lipid nanoparticles, will be critical^20,22.

It is important to assess to what extent ARCs represent the broader reservoir. Our approach may preferentially isolate reservoir clones adept at proliferating while expressing HIV Gag protein, which may represent among the most refractory reservoir subsets that retain the ability to re-seed replication. Indeed, we observed integration site characteristics in ARCs (in ZNF genes and heterochromatic regions) converging with those of reservoir clones persisting long-term under immune pressure in vivo and notably, the remarkable 12-year in vivo stability of the OM5334 ARCs relative to the total intact HIV reservoir in this donor¹². Alternative cloning strategies — such as screening by HIV PCR rather than p24-based detection — could unveil additional ARC subtypes. Although low-level spontaneous production of infectious virus supports the rebound potential of ARCs, in vivo assessment will require linkage to analytical treatment interruption studies.

We directly demonstrate potent cell-intrinsic CTL resistance in a reservoir-harbouring clone: here a T_reg (ARC-Chr16), alongside moderate resistance in ARC-ZNF721 relative to dARC-COG5 and newly infected CD4⁺ T cells. Such resistance was initially hypothesized based on the minimal effect on the reservoir with combinations of potent latency-reversing agents and CTLs in ex vivo studies¹⁸. Proposed mechanisms have included GzmB inhibitor (SERPINB9), inhibitory ligands^16,27 and overexpression of the anti-apoptotic protein BCL−2 (refs. ^17,50,51), inhibitors of which facilitate reservoir reductions ex vivo and in non-human primates^17,52. Our direct demonstration of intrinsic CTL resistance validates these hypotheses and provides a framework for systematic identification of resistance pathways across reservoir clones.

Building on the precision approach used against ARC-Chr16, further study is needed to determine the degree to which DFO or other agents can broadly sensitize HIV reservoirs to elimination, opening avenues for cure strategies beyond latency reversal. Given that resistance to CTLs also limits cancer immunotherapies, our findings open avenues for cross-disciplinary insight. Pathways such as BCL-2 and EZH2 are already established therapeutic targets in oncology, whereas DFO remains unexplored in immunotherapy contexts. Ultimately, the ARC platform enables not only mechanistic dissection of HIV persistence but also a framework for probing how human T cells balance proliferation, immune resistance and vulnerability under sustained cytotoxic pressure.

Methods

Data reporting

No statistical methods were used to pre-determine samples sizes used in this study. The experiments were not randomized. Investigators were not blinded to sample allocation and outcome assessment in individual experiments.

Study design and participants

Peripheral blood or leukapheresis samples were obtained from individuals living with and without HIV. Participants living with HIV who had been suppressed on ART to undetectable levels for at least 2 years were recruited from the Maple Leaf Medical Clinic, the Rockefeller University and the University of Pittsburgh, or through a clinical protocol approved by the Institutional Review Board of the US National Institutes of Health (NIH). Leukapheresis products were collected in accordance with approved protocols, and all participants provided written informed consent for subsequent testing. Samples from deidentified individuals negative for HIV were obtained from the New York Blood Center, the Gulf Coast Regional Blood Center or STEMCELL Technologies. The study was approved by the Institutional Review Boards of the University of Toronto, Rockefeller University and Weill Cornell Medicine.

Media and reagents

The following culture media were used: R10, which is RPMI-1640 (Thermo Fisher) supplemented with 10% FBS (Gibco) and 1% penicillin–streptomycin (Gibco), 1% HEPES buffer (Gibco), 1% l-glutamine (Gibco); ‘clone medium’, which is R10 supplemented with 100 U ml⁻¹ of IL-2, 1 µM nevirapine, 1 µM emtricitabine, 1 µM tenofovir, 10 µM enfuvirtide and 10 µg ml⁻¹ 10-1074 bNAb, each from the NIH AIDS reagent program; and R10-50 + IL-15, which is R10 supplemented with 50 U ml⁻¹ of IL-2 and 10 ng ml⁻¹ of IL-15. Recombinant human IL-2 and IL-15 were obtained from National Cancer Institute.

p24 ELISA

High-binding 96-well plates (655061, Grenier Bio-One) were coated with 100 µl per well of mouse anti-HIV p24 monoclonal antibody (AIDS and Cancer Virus Program, Frederick National Laboratory for Cancer Research) diluted in DPBS according to the manufacturer’s instructions and incubated overnight at 4 °C. The next day, coating solution was removed, and wells were blocked with 200 µl of blocking buffer (PBS + 1% BSA) for 30 min at room temperature. Blocking buffer was then aspirated, and 100 µl of culture supernatant from each well was added along with 10 µl of lysis buffer (distilled water + 10% Triton X-100). Plates were incubated at 37 °C for 1 h.

Following incubation, wells were washed five times with 300 µl of wash buffer (PBS + 0.1% Tween-20). Rabbit anti-HIV p24 polyclonal antibody (AIDS and Cancer Virus Program, Frederick National Laboratory for Cancer Research), diluted in primary antibody diluent (RPMI-1640 supplemented with 10% FBS and 2% normal mouse serum), was added at 100 µl per well. Plates were incubated for 1 h at 37 °C, then washed again as described.

Next, 100 µl of goat anti-rabbit IgG (H + L) secondary antibody (Novus Biologicals), diluted 1:20,000 in secondary antibody diluent (RPMI-1640 supplemented with 2% normal mouse serum, 5% normal goat serum and 0.01% Tween-20), was added per well and incubated for 1 h at 37 °C. After a final wash step, 100 µl of 1-Step Ultra TMB-ELISA Substrate Solution (Thermo Fisher Scientific) was added to each well and incubated at room temperature for 30 min in the dark. The reaction was stopped by adding 100 µl of Stop Solution for TMB substrate (BioLegend) per well.

Plates were read at 450 nm using a SpectraMax microplate reader (Molecular Devices) with SoftMax Pro software. Clone medium was used as a negative control. Wells with absorbance value more than 1.5 times the negative controls were considered p24 positive.

Intact proviral DNA assay

Intact proviral DNA assay (IPDA) was performed as previously described⁵⁵, on DNA from CD4⁺ T cells populations as a routine purity check every three stimulation cycles. Genomic DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen) with the elution step using nuclease-free water. DNA concentrations were quantified using the Nanodrop Spectrophotometer (Thermo Scientific). Each ddPCR reaction was composed of 1× ddPCR Supermix for Probes (no dUTPs; Bio-Rad), 900 nM primers (Integrated DNA Technologies), 250 nM probes (Thermo Fisher Scientific), 22.5 ng of genomic DNA and nuclease-free water. Primer and probe sequences are given in Supplementary Table 3. For each sample, the HIV reaction (HIV Ψ/HIV gag Back-up with HIV env/HIV env Back-up) and RPP30 reaction (RPP30 and RPP30 Shear) were run in duplicate. For newly isolated ARCs, two different sets of HIV reactions using different primer–probe combinations (HIV Ψ with HIV env or HIV gag Back-up with HIV env Back-up) were run in duplicate to determine the primer–probe compatibility. Droplets were prepared using an Automated Droplet Generator (Bio-Rad), cycled at 95 °C for 10 min; 45 cycles of (94 °C for 30 s, 59 °C for 1 min) and 98 °C for 10 min, then analysed with the QX600 Droplet Reader (Bio-Rad) using QuantaSoft software (v2.1; Bio-Rad).

ARC isolation

Limiting dilution method

Memory CD4⁺ T cells (mCD4) were isolated from the PBMCs of PWH by negative selection using the EasySep Human Memory CD4⁺ T Cell Enrichment Kit (STEMCELL Technologies), plated (10 cells per well, 96-well plates).

TCRVβ sorting method

Cells were enriched for T cell populations bearing HIV reservoir clones of interest by flow cytometry sorting, using previously described methods³². In brief, CD4⁺CD45RA⁻ T cells were isolated from PBMCs by negative selection using magnetic separation (130-096-533 and 130-045-901, Miltenyi). Fc receptor was blocked by incubating cells with Fc-blocking reagent (130-059-901, Miltenyi). CD4⁺CD45RA⁻ T cells were stained for live/dead detection at 1:1,000 with Fixable Viability Dye eFluor 780 (65-0865-14, Invitrogen) and Brilliant Violet 605 anti-human TCR (747979, BD), PerCP/Cy5.5 anti-human CD4 (317428, BioLegend) and Pacific Blue anti-human CD3 (300431, BioLegend). Cells were also stained, at 1:100, with antibodies binding to TCRβ variable chains specific to the clone: FITC anti-human TCR Vβ17 (ARC-ZNF486; IM1234, Beckman Coulter). As there is no available commercial antibody for the TCRβ variable chain for ARC-Chr22 (TRBV5-4) we enriched by labelling with all antibodies from Beta Mark TCRVβ Repertoire Kit (IM3497, Beckman Coulter) and sorted for the negative population. Five cells per well were sorted by FACSymphony S6 using FACSDiva software (v9.5.1; BD Biosciences) in 96-well U-bottom culture plates. For expansion of plated cells, cells were stimulated with irradiated allogeneic HIV-negative PBMCs (1 × 10⁶ ml⁻¹) in clone medium for 21 days. Feeder PBMCs were depleted of CD8⁺ T cells and NK cells using the EasySep Human CD8 Positive Selection Kit II and the EasySep Human CD56 Positive Selection Kit II (STEMCELL Technologies) before being co-cultured with mCD4 T cells from PWH. At the end of stimulation cycle, p24 ELISA was performed on all plates. Cells from p24 ELISA-positive wells were plated in serial threefold dilution (96-well plates), stimulated at the same condition as aforementioned, and subjected to p24 ELISA after 21 days. After the second round of plating and ELISA, cells from p24 ELISA-positive wells were expanded under the same stimulation condition. Clonal purity of isolated infected populations was determined using IPDA and further serial dilution subclonings were performed as needed.

Near-full-length HIV sequencing (FLIP-seq)

Genomic DNA was extracted from the clones using a QIAGEN DNeasy kit. DNA was diluted to single-genome levels based on Poisson distribution statistics and ddPCR results, and was subjected to single-genome amplification using Invitrogen Platinum Taq and nested primers spanning near full-length HIV (HXB2 coordinates 638–9632). All primers have been previously described⁵⁶ and can be found in Supplementary Table 3. PCR products were visualized by agarose gel electrophoresis. All near-full-length sequences (more than 8,000 bp) were subjected to Illumina MiSeq sequencing. Resulting short reads (142 bp) were de novo assembled and aligned to HXB2 to identify large deleterious deletions, out-of-frame indels, premature/lethal stop codons, internal inversions or packaging signal deletions, using an automated in-house pipeline written in R scripting language as previously described⁵⁷. The presence or absence of APOBEC-3G/3F-associated hypermutations was determined using the Los Alamos HIV Sequence Database Hypermut 2.0 program. Viral sequences that lacked all mutations listed above were classified as intact. Sequences were aligned to the HXB2 reference genome using Geneious Prime (v2025.1) and manual editing where appropriate. The 5′ and 3′ ends of the sequences not covered by the near-full-length PCR were extracted from the corresponding integration site reads (HXB2 coordinates 1–681 and 9686–9719, respectively). A short 53-bp section of the 3′ LTR (HXB2 coordinates 9633–9685) was inferred based on the corresponding 5′ LTR sequence. To examine phylogenetic distances between sequences, a neighbour-joining tree containing all clone proviral sequences and the HXB2 reference sequence was prepared using MEGA (v7.0).

Integration site sequencing (PRISM-seq)

Genomic DNA was extracted from the ARCs using a QIAGEN DNeasy kit. Subsequently, 2.5 μl of genomic DNA sample was subjected to multiple displacement amplification with phi29 polymerase (REPLI-g Single Cell Kit, 150345, Qiagen) at 30 °C for 4 h. DNA produced by whole-genome amplification was digested with DraI, SspI and HpaI. The digested DNA was purified, followed by ligation of genome-walker (GW01) adapter. Viral–host junction regions were amplified using a nested PCR strategy (see Supplementary Table 3 for primers)²³. Resulting PCR products were subjected to Illumina MiSeq sequencing. MiSeq paired-end FASTQ files were demultiplexed; small reads (142 bp) were then aligned simultaneously to human reference genome T2T and HIV reference genome HXB2. Biocomputational identification of integration sites was performed using in house R-based pipeline⁵⁸. In brief, chimeric reads containing both human and HIV sequences were evaluated for mapping quality based on (1) HIV coordinates mapping to the terminal nucleotides of the viral genome, (2) absolute counts of chimeric reads, and (3) depth of sequencing coverage in the host genome adjacent to the viral integration site. The final list of integration sites and corresponding chromosomal annotations was obtained using Ensembl (v86; www.ensembl.org), the UCSC Genome Browser (www.genome.ucsc.edu) and GENCODE (v39; www.gencodegenes.org).

Quantitative viral outgrowth assay (QVOA)

QVOAs were performed to measure the frequency of inducible ARCs. ARCs were removed from ARV-containing media 2 h before assay setup. For each ARC, cells were mixed with background cells at defined ratios: an autologous HIV-negative CD4⁺ T cell clone for ARC-Chr16 and autologous CD4-depleted PBMCs for all other ARCs. CD4⁺ T cells were removed using the EasySep Human CD4 Positive Selection Kit II (STEMCELL Technologies) and replaced by irradiated MOLT-4/CCR5 cells (NIH HIV Reagents Program, Division of AIDS, NIAID, NIH). ARCs were spiked in at 100–2,000 cells per 1 × 10⁶ background cells. ARC–background mixtures were then plated in 12–24 replicate wells per 6–7 dilutions, spanning 0.01 × 10⁶ to 1 × 10⁶ cells per well in 24-well plates (total cells per ARC: 27–44 × 10⁶). Cells were stimulated with either 1 μg ml⁻¹ phytohaemagglutinin (PHA; ARC-ZNF721 and ARC-Chr16) or anti-CD3/CD28 (ARC-ZNF486 and ARC-Chr22), each together with 1 × 10⁶ irradiated feeder cells from people without HIV per well. QVOA cultures were established in the presence of 50 μg ml⁻¹ autologous IgG purified using the Melon Gel IgG Spin Purification Kit (Thermo Fisher Scientific) or control IgG from people without HIV (Sigma-Aldrich). Donors 603 and 5104 were previously enrolled in clinical trials in which they received bNAb infusions. The autologous IgG used for this study was purified from pre-bNAb infusion plasma samples to avoid confounding effects. After 24 h, culture media were replaced to remove PHA or anti-CD3/CD28, and 0.2 × 10⁶ MOLT-4/CCR5 cells were added to each well. Media were changed every 3 days, and IgG was re-added at each medium change. Supernatants were collected on day 7 and assessed for p24 antigen by ELISA. Wells with absorbance values exceeding that of the lowest p24 standard (3.25 pg μl⁻¹) were designated as p24 positive. The frequency of inducible cells was calculated using the infectious units per million algorithm developed by the Siliciano laboratory⁵⁹. MOLT4/CCR5 cells were not tested for mycoplasma contamination.

Integration site and TCRVβ ddPCR assays

We designed ddPCR assays that quantify total HIV LTR, the integration site of a specific provirus and the corresponding TCRβ CDR3 of the clone containing that provirus. The primers and probe used to amplify the R-U5 region of the HIV LTR have been previously described^4,5. Integration site-specific and TCRβ CDR3-specific primers and probes were designed for individual reservoir clones as previously described^5,6. In brief, integration site-specific probes were designed to span the host–5′ LTR junction with corresponding forward and reverse primers on the host genome and the U3 region of the HIV LTR, respectively. Similarly, TCRβ CDR3-specific probes were designed to span the junctions between the V-D-J regions with corresponding forward and reverse primers on the V and J regions, respectively. The specificity of each primer and probe set was validated using DNA from the corresponding reservoir clone and compared with DNA from the CD4⁺ T cells of HIV-negative donors and DNA from other infected and uninfected clones. A complete list of primers and probes can be found in Supplementary Table 3. Primers and probes were used at final concentrations of 900 nM and 250 nM, respectively. ddPCR was performed with the Bio-Rad QX600 platform using the following cycling conditions: 95 °C for 10 min, 50 cycles of 95 °C for 30 s and 56 °C for 1 min, 98 °C for 10 min, hold at 4 °C. RPP30 ddPCR was performed in parallel for cell equivalent normalization as previously described⁷. Because the HIV LTR-specific primers and probe do not distinguish between the 5′ and 3′ location of the amplified R-U5 region, a correction factor of 1/2 is applied to LTR copy number values before cell equivalent normalization.

Proliferation of ARCs in ex vivo CD4⁺ T cells

Total of 10⁷ CD4⁺ T cells were plated at 10⁶ cells per millilitre clone medium and stimulated with anti-CD3/CD28 Dynabeads at a 5:1 bead:cell ratio. Twenty-four hours post-stimulation, the cells were counted and media were added to reach a final cell density of 10⁵ cells per millilitre. At days 3, 5, 7, 9 and 11 post-stimulation, precise total cell counts were recorded, 10⁷ cells were set aside for genomic DNA extraction and media were added to reach a final cell density of 10⁵ cells per millilitre. Cell viability remained at more than 90% for the duration of the 11-day culture. LIST ddPCR was used to quantify total HIV LTRs and the integration site and TCRβ of specific ARCs at each culture time point. Expansion dynamics of total CD4⁺ T cells was determined by relative changes in cell counts over the duration of the culture. Changes in the frequency of the ARC integration site and TCRβ to total RPP30 cell equivalents over the duration of the culture were used to calculate expansion dynamics of cells harbouring the ARC integration site or TCRβ.

ELISPOTs

To perform the IFNγ and GzmB ELISPOT assays, capture antibodies (clone 1-D1K for IFNγ and clone MT28 for GzmB; Mabtech) were first diluted in PBS according to the manufacturer’s instructions and plated at 100 µl per well into multiscreen IP 96-well PVDF plates (MilliporeSigma); coated plates were subsequently incubated overnight at 4 °C to allow optimal antibody binding to the membrane surface, ensuring reliable cytokine capture upon stimulation. Following this incubation period, the plates were washed five times with PBS to remove any unbound antibody and to prepare the wells for cell seeding; subsequently, 200,000 cells for ARCs or PBMCs or 20,000 cells for CTL clones were added per well in R10 medium containing the relevant antigens or controls. For Fig. 4c,d only, PBMCs were pretreated for 18 h with R10 medium supplemented with IL-15 at 0.1, 1 or 10 ng ml⁻¹ as indicated before adding to the capture plates and addition of antigens or controls. The antigens being used were CMV-pp65 peptide pool at 1 μg ml⁻¹ per peptide (STEMCELL Technologies), CMV/EBV/Flu peptide pool at 1 μg ml⁻¹ per peptide (JPT Peptide Technologies), HIV-1 subtype B consensus peptides at 1 μg ml⁻¹ per peptide (either individual peptides, matrix pools or whole-gene product pools as indicated; NIH AIDS Reagents Program), CMV-infected cell extract, EBV-infected cell extract and negative control cell extract (for EBV-infected cells) at 10 µl per 200,000 cells (all three from Virusys). PHA at 0.25 µg ml⁻¹ or ImmunoCult Human CD3/CD28 T Cell Activator at 25 µl ml⁻¹ (STEMCELL Technologies) served as a positive control to confirm assay responsiveness, whereas 0.5% DMSO was used as a negative control to determine background signal levels.

The plates were then incubated at 37 °C for 17–20 h in a humidified incubator, allowing T cell activation and cytokine secretion to occur; after this stimulation phase, wells were washed five times with PBS to eliminate residual cells and media, thus minimizing nonspecific staining during detection. A biotinylated detection antibody (clone 7-B6-1 for IFNγ and clone MT8610 for GzmB; Mabtech) in 100 µl PBS at the manufacturer’s recommended concentration was then added to each well and incubated at room temperature for 2 h; after this period, plates were again washed five times to remove unbound antibody before the addition of streptavidin-ALP at 1 µg ml^–1 in 100 µl PBS per well (Mabtech), which was incubated for 1 h at room temperature to allow sufficient binding and amplification of the biotin signal. The plates were washed again five times and developed using AP Conjugate Substrate Kit (Bio-Rad) for 15 min; during this time, spot formation occurred where IFNγ had been secreted by activated T cells and captured by the plate-bound antibody complex. Following development, plates were thoroughly washed to halt the enzymatic reaction, air dried overnight to stabilize the spots, and subsequently analysed using a CTL ELISPOT reader (ImmunoSpot), which quantitatively measured the number and intensity of cytokine-producing cells.

HIV transmission assay

To determine whether virus produced by ARCs was replication competent, we co-cultured ARCs with activated CD4⁺ ‘target cells’ in the absence of ARVs and assessed infection. CD4 cells were isolated from PBMCs of HIV-negative donors via negative selection (EasySep Human CD4⁺ T Cell Isolation Kit, STEMCELL Technologies), cultured in R10-50 + IL-15 (1 × 10⁶ ml⁻¹), and activated for 72 h with ImmunoCult Human CD3/CD28 T Cell Activator at 25 µl ml⁻¹ (STEMCELL Technologies). ARCs were washed to remove ARVs labelled with CellTrace CFSE at 5 µM according to the manufacturer’s protocol (CFSE, Thermo Fisher Scientific), cultured in R10-50 + IL-15 for 24 h for additional ARV washout, and then co-cultured with target cells at ratios ranging from 1:10 to 1:100 ARC:target cells in R10-50 + IL-15. In control conditions, we cultured with a CFSE-labelled uninfected CD4⁺ T cell clones that had been isolated and maintained in parallel to the ARC. Negative controls were activated uninfected CD4⁺ T cells not co-cultured with infected populations. At days 3 and 7, samples were stained for flow cytometry using two antibodies to HIV Gag to enhance sensitivity and specificity (see below for details). The presence of HIV Gag⁺ populations within the CFSE-negative target cell population comprised evidence for viral transmission.

Flow cytometry, including HIV Gag staining

Cells were stained for viability using the live/dead dye in MACS buffer (PBS + 2% FBS + 1 mM EDTA) following the manufacturer’s instructions and with other surface antibodies depending on the experiment, including: CD3 OKT3 (1:100; BioLegend), CD4 OKT4 (1:100; BioLegend), CD8 SK1 (1:100; BioLegend), CD107a H4A3 (1:100; BioLegend), CD56 HCD56 (1:100; BioLegend), CD69 FN50 (1:50; BioLegend), CD71 CY1G4 (1:100; BioLegend), HLA-A,B,C W6/32 (1:800; BioLegend), IL-17 TC11-18H10.1 (1:100; BioLegend), IL-4 MP4-25D2 (1:50; BioLegend), IFNγ B27 (1:100; BioLegend), IL-22 2G12A41 (1:50; BioLegend), GzmA CB9 (1:50; BioLegend), GzmB QA18A28 (1:100; BioLegend), TCR Vb18 BA62.6 (1:12; Beckman Coulter), TCR Vb7.1 ZOE (Beckman Coulter) and Beta Mark TCR Vbeta Repertoire Kit (1:12; Beckman Coulter). After staining, cells were washed twice with FACS buffer. Cells were then fixed and permeabilized using the Cytofix/Cytoperm kit (BD Biosciences) according to the manufacturer’s instructions. Intracellular HIV p24 Gag protein was stained using anti-HIV p24 antibody (clone KC57-RD1 or KC57-FITC, Beckman Coulter; and clone 28B7, MediMabs) diluted in Perm/Wash buffer. The combination of these two antibodies targeting different HIV Gag epitopes has been shown to improve the specificity and sensitivity of detection⁴³. Cells were incubated for 30 min at room temperature in the dark, washed twice with Perm/Wash buffer and resuspended in FACS buffer for acquisition. Samples were acquired on a BD Symphony and analysed using FlowJo software (BD Biosciences). Representative gating strategies for all flow cytometry results are provided under Supplementary Fig. 1.

T cell stimulation

To assess HIV reactivation, 100,000 cells per ARC and dARC were stimulated with T cell activators anti-CD3/CD28 Dynabeads (Thermo Fisher) at 25 µl of beads per million cells or PMA (50 ng ml⁻¹) plus ionomycin (1 μg ml⁻¹) (PMA–I), or bryostatin-1 (10 nM), in clone medium. Stimulations were performed for 24 h at 37 °C with 5% CO₂. After stimulation, cells were stained for viability, fixed and intracellularly stained with anti-HIV Gag antibodies conjugated to APC and RD-1, and analysed by flow cytometry. HIV reactivation was quantified as the percentage of Gag⁺ cells, with dual staining used to identify latency reversal. Data were analysed using FlowJo.

Co-evaluation of proliferation and HIV-gag expression

Latently infected CD4⁺ T cells were cultured in clone medium. To track proliferation, cells were labelled with CellTrace Violet (Thermo Fisher Scientific) according to the manufacturer’s protocol. In brief, cells were incubated with 5 µM CellTrace Violet in PBS for 20 min at 37 °C in the dark, followed by quenching in R10 medium and a 5-min incubation at room temperature.

Cells were plated at 1 × 10⁶ per well in U-bottom 96-well plates in clone medium and either left unstimulated or stimulated with 1 µg ml⁻¹ of soluble Ultra-LEAF purified anti-human CD3 (clone OKT3, 317325, BioLegend) and anti-human CD28 (clone 28.2, 302933, BioLegend) antibodies. Irradiated allogeneic PBMCs (from fresh buffy coat, irradiated at 30 Gy) were added as feeder cells at a 1:1 ratio. Cells were cultured for 4 and 7 days at 37 °C with 5% CO₂, with media replenished as needed. At the end of culture, cells were harvested, fixed and permeabilized using the Cytofix/Cytoperm kit (BD Biosciences) according to the manufacturer’s instructions. Intracellular HIV Gag was detected using APC-conjugated anti-HIV p24 antibody (clone 28B7, MediMabs) diluted in Perm/Wash buffer. Cells were incubated for 30 min at room temperature in the dark, washed twice with Perm/Wash and resuspended in FACS buffer.

Samples were acquired using a BD Symphony flow cytometer and analysed with FlowJo software (BD Biosciences). Proliferation was assessed by dilution of CellTrace Violet, and latency reversal was quantified as the frequency of p24⁺ cells.

HIV RNA quantitative PCR

HIV RNA copy numbers were measured in 30 μl of cell culture supernatant using the previously described integrase single-copy assay protocol with minor modifications⁶⁰. In brief, QIAcube HT automated system (Qiagen) and QIAamp 96 Viral RNA Kit-10 (Qiagen) were used for RNA extraction and elution in 80 μl of water. A validated RNA standard was used for quantification of RNA samples. For each reaction, 12.5 μl of 2× buffer and 1 μl of N2 polymerase from AgPath-ID One-Step RT-PCR Reagents kit (Thermo Fisher) were mixed with 8.5 μl of extracted RNA or serially diluted RNA standards, 1 μl of 10 μM forward and reverse primers, and 1 μl of 6.25 μM probe, in a total volume of 25 μl per well. Samples and standards were run on each 96-well plate using a QuantStudio Pro 7 Real-Time PCR System (Thermo Fisher Scientific). The cycling parameters were 50 °C for 10 min, 95 °C for 10 min, followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. Cycle threshold values of samples and standards were analysed by Design & Analysis Software (v2.8.0; Thermo Fisher Scientific) to determine HIV RNA concentration. Primers and probes used in this assay are given in Supplementary Table 3.

Reverse transcription ddPCR

DNA and RNA were extracted from ARCs and treated with different latency reversal agents using the AllPrep DNA/RNA Mini Kit (Qiagen) according to the manufacturer’s protocol. A previously described ddPCR assay was adopted and modified to quantify levels of different HIV transcripts⁶¹. In brief, 40 µl cDNA reactions were generated using SuperScript III First-Strand Synthesis System (Invitrogen) as following: 4 μl 10× SuperScript III buffer, 4 μl 50 mM MgCl₂, 2 μL random hexamers (50 ng μl⁻¹), 2 μl dT15 (50 μM), 2 μl dNTP (10 mM), 2 μl RNAseOUT (40 U μl⁻¹), 2 μl SuperScript III reverse transcriptase (200 U μl⁻¹), 5 μl RNA template and 17 μl dH2O. cDNA generation reactions were run at 25 °C for 10 min, 50 °C for 1 h and 85 °C for 5 min. cDNA was stored at 4 °C and had to be used within 24 h after generation. Several ddPCRs were run concurrently from generated cDNA to measure HIV transcripts containing the 5′ LTR, Psi, Env and polyadenylation (polyA). We additionally measured read-through transcripts that initiated upstream of the HIV promoter. All primers and probes used are listed in Supplementary Table 3. The Psi and Env primer and probe sequences are the same as those used for the IPDA. cDNA reactions were diluted with additional 110 μl dH2O to make 150 μl cDNA per sample. 20 μl RT-ddPCR comprised of 1× ddPCR Supermix for Probes (no dUTP; Bio-Rad), 900 nM primers (Integrated DNA Technologies), 250 nM probes (Thermo Fisher Scientific), 5 μl diluted cDNA and nuclease-free water. For each RT-ddPCR, an equivalent ddPCR on genomic DNA was performed as control and for normalization. ddPCRs on genomic DNA using the RPP30 primer and probe systems were also performed similarly to IPDA for analysis and normalization. Transcripts per infected cells were analysed according to Yukl et al.⁴⁶ with the assumption that eluted RNA and DNA were equivalent of total sample cells and normalized with RPP30 and HIV-specific genomic DNA ddPCR results.

Competitive ARC killing assay

This assay was designed to parse viral from host factors in the susceptibility of resistance of an ARC to killing by HIV-specific CTL. For quality control of ARCs and dARCs, the day before an assay, these were assessed by flow cytometry staining for the TCR Vb segment of a given ARC or dARC to confirm more than 95% purity. This was corroborated by IPDA. ARCs were stained and analysed by flow cytometry for intracellular Gag ensuring sufficient spontaneous production to yield a readout. For generation of ARC-Chr16 viral stock, ARC-Chr16 was washed three times and resuspended in R10-100 at 1–2 × 10⁶ cells per millilitre in 4 ml for 24 h. Supernatants were harvested, and virus was concentrated by PEG-it, following the manufacturer’s instructions. Virus was frozen at −80 °C in 5 × 50 ml aliquots and two aliquots were used for the infection. For preparation of target cells, CD4⁺ T cells were enriched from donor OM5334 by negative selection using the EasySep Human Memory CD4⁺ T Cell Enrichment Kit (STEMCELL Technologies) and plated at 5 × 10⁶ cells per millilitre in R10-100 (no ARVs) along with ImmunoCult Human CD3/CD28 T Cell Activator at 25 µl ml⁻¹ (STEMCELL Technologies). After 48 h, these cells were infected by adding the ARC-Chr16 viral stock and incubating at 37 °C for 2 h. Cells were then washed with R10-100, plated at 5 × 10⁶ cells per millilitre in fresh R10-100 and incubated for 3 days, at which point they were harvested for the competitive killing assay. These are termed newly in vitro infected cells. For labelling of target cells, on the day of co-culture, dARCs, ARCs and newly in vitro infected cells were each washed and then labelled separately with different concentrations of CellTrace Yellow (CTY; Thermo Fisher Scientific) according to the manufacturer’s instructions. Newly in vitro infected cells were labelled with 5 µM CTY, ARC-Chr16 with ten times less (0.5 µM) CTY, and the dARC treated identically but without CTY. Small aliquots of each were analysed by flow — separately and mixed — to ensure that staining intensities could be distinguished. For quality control of HIV-specific CTL clone, the CTL clone was taken 3 weeks after its most recent restimulation cycle. Specificity and functionality was confirmed by an overnight ‘specificity check (see above) using the cognate HIV Pol peptide. Labelled targets and CTLs were co-cultured with each other at a range of effector:target ratios ranging from 1:2 to 1:32 with a no CTL control in triplicate — approximately 2 × 10⁵ total cells per well in a 96-well plate in clone medium, along with 1:100 anti-CD107a APC-Cy7 (BioLegend) to allow assessment of degranulation. Co-cultures proceeded overnight at 37 °C. Cells were then surface stained with the following antibodies: MHC-I Pe/Dazzle (1:100), CD4 Alexa-Fluor-700 (1:100), CD3 BV785 (1:100), CD8 BV605 (1:100; all from BioLegend) and live/dead aqua (1:200; Thermo Fisher). Cells were fixed and permeabilized with the Cytofix/Cytoperm kit (BD Biosciences) and then labelled intracellularly for: HIV Gag (KC57, Beckman Coulter), HIV Gag (28B7, MediMabs), BCL-2 BV421 (BioLegend) and CD8 BV605 (re-labelled for internalized CD8). Cells were then fixed with paraformaldehyde and analysed on an Attune NxT flow cytometer.

Evaluation of dual Gag and Nef expression on ARCs

Cells were stained with MHC-I Pe/Dazzle (1:100), CD4 Alexa-700 (1:100), CD3 BV785 (1:100), CD8 BV605 (1:100; all from BioLegend) and live/dead aqua (1:200; Thermo Fisher) for 20 min at 37 °C. They were fixed using the FOXP3 fixation and permeabilization buffer (Thermo Fisher eBioscience) following the manufacturer’s instructions. Cells were intracellularly stained with HIV Gag (KC57; 1:500, Beckman Coulter), HIV Gag (28B7; 1:150, MediMabs) and an anti-human Nef rabbit polyclonal antiserum (ARP-2949, NIH HIV Reagent Program) for 30 min in 1× permeabilization buffer (Thermo Fisher). Cells were then washed two times and stained with a Alexa Fluor 594-conjugated donkey anti-rabbit IgG (minimal cross-reactivity; BioLegend) at 1:800 concentration for 30 min. Cells were then washed and analysed on an Attune NxT flow cytometer.

CD4-based enrichment or depletion of ARCs

Multiple day sampling

ARC cells were stained with 20 µl of anti-human CD4 microbeads (Milteny Biotec) per 1 × 10⁷ cells in 80 µl of MACS buffer at 4 °C for 15 min. Cells were then washed in 1 ml of MACS buffer and centrifuged at 400g for 5 min. Stained cells were resuspended in 500 µl of MACS buffer for up to 1 × 10⁷ cells. Of cells, 500 µl was loaded onto an MS MACS column on a miniMACS magnetic separator on a MACS multi stand magnet (Milteny Biotec). Cells were washed on the magnet three times with 500 µl of MACS buffer and flow through was collected as Gag⁺-enriched cells (CD4⁺ depleted). The cells bound to the column were then removed from the magnet and flushed with 1 ml of MACS using a plunger into a collection tube as Gag⁺-depleted cells (CD4⁺ enriched). Prepared cell fractions were then analysed at day 0 by flow cytometry for Gag expression and also analysed at day 5 and day 10 after being plated at 1 × 10⁶ per well in U-bottom 96-well plates in clone medium and stimulated with 1 µg ml⁻¹ of soluble Ultra-LEAF purified anti-human CD3 (clone OKT3, BioLegend) and anti-human CD28 (clone 28.2, BioLegend) antibodies.

High-resolution day 5–7 timecourse

Gag⁺-enriched cells were isolated as described above and compared with total untouched ARCs. Cells were restimulated with 1 µg ml⁻¹ of soluble Ultra-LEAF purified anti-human CD3 (clone OKT3, BioLegend) and anti-human CD28 (clone 28.2, BioLegend) antibodies in the presence of irradiated feeders at a 1:1 ratio with ARCs. Cells were sampled to measure changes in Gag expression at 1, 3, 6, 12, 24 and 48 h timepoints post-stimulation, simulating a time period between days 5 and 7 from slowcure assays.

5′ CITE-seq on the 10X Genomics platform

Cells were stained with DNA oligo-barcoded antibodies and loaded onto the 10X Chromium single-cell workflow (10X Genomics). Approximately 0.05–0.3 × 10⁶ cells per sample were resuspended in R10 medium and incubated with TotalSeq-C Human Universal Cocktail V1.0 (BioLegend) according to the manufacturer’s instructions for 30 min at 4 °C. In some cases, the TotalSeq-C hashing antibodies TotalSeq-C0251 anti-human Hashtag 1 clones LNH-94, 2M2 (BioLegend), TotalSeq-C0251 anti-human Hashtag 3 clones LNH-94, 2M2 (BioLegend) and TotalSeq-C0251 anti-human Hashtag 4 clones LNH-94, 2M2 (BioLegend; all at 0.5 μg ml⁻¹) were spiked into individual samples, which allowed for multiplexing of ARCs and uninfected clone combinations into the same 10X lane. After staining, cells were washed three times in R10 medium, followed by centrifugation (400g for 5 min at 4 °C) and supernatant aspiration. After the final wash, cells were resuspended in PBS and filtered through 70-µm cell strainers. Stained cells from each sample were pooled and loaded into the 10X Chromium Single Cell Immune Profiling workflow (CG000330, revision G). Libraries were generated according to the 10X Chromium Single Cell Immune Profiling workflow instructions, pooled and sequenced on an Illumina NovaSeq 6000 or NovaSeqX.

Single-cell data alignment

FASTQ files from the 10X gene expression, surface protein and TCRα/β libraries were processed using Cell Ranger v7.0.1 or v9.0.1 (10X Genomics), using either the count pipeline or the multi-pipeline (when TCR libraries were available) with default parameters.

RNA transcripts from ECCITE-seq were mapped to a modified version of Cell Ranger’s GRCh38-2020-A reference package (GRCh38/GENCODE v32 annotations), with the addition of the HIV HXB2 reference sequence. TCRα/β libraries were mapped to Cell Ranger’s cellranger-vdj-GRCh38-alts-ensembl-7.1.0 reference package.

Initial cell-level or gene-level filtering

For cell filtering, cell barcodes with less than 200, more than 8,000 genes expressed or more than 10% mitochondrial genes expressed were removed. For gene-level filtering, genes were filtered such that only those expressed at a level of less than 200 unique molecular identifiers or in less than 3 cells were removed.

These cell-level-filtered and gene-level-filtered gene or barcode matrices were used for downstream analysis.

Seurat normalization

The count matrices of all modalities were loaded into a Seurat v5 object (Seurat v5.1.0). For gene expression (GEX) data, we normalized the count data using LogNormalize and scaled the resulting normalized values using ScaleData. For all other modalities (ADT, HTO and TCRα/β), we normalized the count data using centred-log ratio (margin = 2) and scaled the resulting normalized values using ScaleData.

Hashing antibody demultiplexing

To separate individual samples from pooled cell libraries, samples were demultiplexed by their HTO using Seurat HTODemux with the positive quantile parameter set to 0.999.

Multimodal WNN analysis

We applied weighted-nearest neighbour (WNN) analysis on our 5′ CITE-seq data, enabling the integrative analysis of GEX and ADT modalities in the same cells as previously described²⁰. In brief, k-nearest neighbour graphs (k = 20) were first constructed, and the neighbours of each cell in each modality were identified using FindMultiModalNeighbors in Seurat V5. For graph construction, the first 15 principal components of the log-normalized values from the GEX modality and the first 12 principal components of the centred-log ratio values from the ADT modality were used. The dimensionally reduced molecular profiles of the k-nearest neighbours of each cell were then averaged and compared with the actual values to obtain the residuals in an RNA–RNA, protein–protein and RNA–protein manner. Residuals were converted into affinity-based similarities using UMAP. Ratios of the within-modality and cross-modality affinities were softmax transformed to obtain the cell-specific modality weights. The weighted similarity then formed the basis of the construction of the WNN graph visualized on a WNN integrated UMAP. Clusters were called using the FindClusters function at a resolution of 0.5.

Differential expression and pathway analysis

Differential gene expression analysis across WNN clusters was performed using FindAllMarkers on cells defined as ARCs or uninfected clones based on hashing information and TCR sequences. The logfc.threshold (minimum log₂ fold change) was set to 0.1 and min.pct (minimum proportion of cells expressing the given gene) was set to 0.01. Only genes with an adjusted P < 0.05 were further considered. Row normalized Z-scores produced using scale.data were used for visualization on the heatmap focusing on the top 15 DEGs. Similarly, differentially expressed proteins were identified using FindAllMarkers with filters of log₂ fold change > 0.5 and P < 0.05 adjusted for multiple testing from a Wilcoxon rank-sum test. Row normalized Z-scores produced using scale.data were used for visualization on the heatmap. Genes passing the above differential gene expression filters were used for pathway enrichment analysis. ClusterProfiler was used to profile clone populations on Reactome pathways. Reactome terms were considered significant if the false discovery rate-adjusted P < 0.05.

Module scoring on ARCs and uninfected clones

The AddModuleScore function of Seurat V5 was used for scoring gene expression modules. For Reactome or Gene Ontology gene sets, the msigdbr package was used to pull selected gene sets from the Molecular Signatures Database (MSigDB).

As above, the AddModuleScore function of Seurat V5 was used for scoring gene expression module of the ‘CD8 cytotoxic’ functional state gene set from Szabo et al.⁵³.

Pseudobulk analysis of HIV RNA⁺ versus HIV RNA⁻ cells

Single-cell RNA sequencing data from HIV-infected T cell clones were analysed to identify transcriptional differences between cells actively transcribing HIV (HIV RNA⁺) and cells lacking sufficient HIV transcription (HIV RNA⁻). Two parallel analyses were performed with different thresholds for defining HIV RNA⁺ cells: a permissive threshold (HIV counts > 0 for total HIV RNA⁺ cells) and a stringent threshold (log-normalized HIV counts > 4 for productive HIV RNA⁺ cells). The stringent threshold was derived by stratifying cells by CD4 surface protein expression (CITE-seq) and examining the distribution of HIV RNA in CD4-low cells, which are enriched for active infection.

Cells from infected clones were aggregated into pseudobulk samples by HIV expression status using Seurat::AggregateExpression(). Differential expression was performed using DESeq2 with ARC clone identity as a blocking factor. Genes with adjusted P < 0.05 were considered differentially expressed. For visualization, batch effects were removed from VST-transformed data using limma::removeBatchEffect(). Gene set enrichment analysis was performed using fgsea::fgseaMultilevel() with genes ranked by DESeq2 Wald statistic. MSigDB Hallmark gene sets were tested, and pathways with adjusted P < 0.05 were considered significantly enriched. The most highly enriched pathways were used to visualize gene sets with shared differential expression across ARCs.

Generation of HIV-specific CTL clones

CTL clones were generated as previously described. PBMCs from PWH were plated at 10⁷ cells per millilitre in 1 ml of X-VIVO serum-free medium (Lonza) supplemented with HIV peptide pools (NIH AIDS reagent program) at a final concentration of 10 µg ml⁻¹ per peptide and incubated overnight. T cells producing IFNγ in response to peptide were enriched using the Miltenyi Biotec IFNγ Secretion Assay-Cell Enrichment and Detection Kit (PE; 130-054-201), following the manufacturer’s instructions. Cell separation was carried out using the OctoMACS magnet and MACS MS Columns (Miltenyi Biotec). Enriched cells were subjected to four 5× serial dilutions in 10 ml R10-50-15 each plated on a full 96-well plate. Feeder media were added containing 1 × 10⁶ irradiated (5,000 rad) PBMCs per millilitre from fresh buffy coats (HIV-negative donors), along 100 ng ml⁻¹ each of Ultra-LEAF purified anti-human CD3 antibody (clone OKT3, 317325, BioLegend) and Ultra-LEAF purified anti-human CD28 antibody (clone 28.2, 302933, BioLegend). After 3 weeks, wells from the lowest dilution plate showing growth were screened for responsiveness to the target peptide using the ‘CTL specificity check’ (see below).

CTL specificity check

CTLs were assessed for peptide specificity 4 weeks after isolation. Cells were resuspended in R10-50 medium supplemented with 10 ng ml⁻¹ IL-15 and plated into 96-well plates. To each well, 2 µl of anti-CD107a (LAMP-1) PE-conjugated antibody (BioLegend) was added. The cognate peptide was added to half of the wells at a final concentration of 10 µg ml⁻¹. All wells were brought to a final volume of 200 µl with media. Cells were incubated at 37 °C for 4 h to allow for degranulation marker expression. Following stimulation, cells were stained with a surface antibody master mix composed of 0.5 µl anti-CD3 (BioLegend), 0.5 µl anti-CD4 (BD Biosciences), 0.5 µl anti-CD8 (BioLegend), 0.3 µl of viability dye (Thermo Fisher Scientific) and 48.2 µl of MACS buffer per well. Plates were covered and incubated at room temperature for 20 min, then washed twice with MACS buffer. Cells were fixed in 4% paraformaldehyde (VWR) and analysed by flow cytometry to determine antigen-specific CD8⁺ T cell degranulation.

BCL-2 and BCL-xL antagonist treatments

The BCL-2 inhibitor ABT-199 and the BCL-xL inhibitor A-1331852 were both obtained from MedChemExpress (HY-15531 and HY-19741, respectively). Both were dissolved to 10 mM stocks in DMSO and added to cell cultures at the indicated concentrations. For the experiments shown in Extended Data Fig. 7, cells were treated with the inhibitors for 24 h. For the experiment in Fig. 5k,l, the inhibitors were left in over the course of the slowcure assay.

Slowcure assay

Authentic HIV reservoir clones were co-cultured with autologous HIV-specific CTL clones at indicated effector:target ratios in R10 medium (RPMI 1640 supplemented with 10% FBS, 1% penicillin–streptomycin, 1% HEPES and 1% l-glutamine). Before co-culture, ARCs were labelled with CTV (Thermo Fisher Scientific) according to the manufacturer’s protocol. In brief, ARCs were incubated with 5 µM CTV in PBS for 20 min at 37 °C in the dark, followed by quenching with five volumes of R10 medium and a 5-min incubation at room temperature. Irradiated allogeneic feeder cells were prepared from freshly isolated PBMCs obtained from healthy donors and irradiated with 30 Gy using a gamma irradiator to prevent proliferation. These feeder cells were added at a 1:1 ratio with the number of reservoir clone cells alone, or with the combined number CTL and reservoir clone cells. Technical duplicates for each condition were set up at the start of culture. Co-cultures were maintained in U-bottom 96-well plates at 37 °C with 5% CO₂ for 1, 5 and 7 days for readout. Supernatant from culture were also harvested for viral load measurements. Media were replenished as needed to maintain cell viability. In some cases, reservoir clones alone or in combination with CTLs were treated with 10 µM of DFO in CTL sensitization experiments.

At the end of the incubation period, cells were harvested and spiked with CountBright Absolute Counting Beads (Thermo Fisher) and stained for surface markers including TCRVβ specificity when available to more accurately track clonal dynamics in culture and intracellular HIV p24 Gag protein using APC-conjugated anti-p24 antibody (clone 28B7; ARP-1475, NIH HIV Reagent Program) in combination with a FITC-conjugated anti-p24 antibody (clone KC57; 6604665, Beckman Coulter). Staining was performed following fixation and permeabilization with the eBioscience Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher) according to the manufacturer’s instructions. Proliferation of ARCs (or uninfected CD4 cells) was assessed by dilution of CTV fluorescence intensity using flow cytometry (BD Symphony A5), and the total counts of ARCs and p24⁺ cells as well as p24⁺ frequency were quantified to evaluate clone propagation and survival and viral expression. Data analysis was performed using FlowJo software (BD Biosciences).

In vitro infection of memory CD4⁺ T cells

Memory CD4⁺ T cells were magnetically isolated from thawed PBMCs using the EasySep Human memory CD4⁺ T Cell Isolation Kit (STEMCELL) according to the manufacturer’s instructions. Purified memory CD4⁺ T cells were then activated using anti-human CD3/CD28-coated magnetic beads (one bead for two cells; 11131D, Gibco) in R10 medium. After 3 days, cells were maintained at a concentration of 10⁶ cells per millilitre in medium containing 50 IU of human recombinant IL-2. At day 7, the cultured memory cells were infected with HIV JRCSF. The HIV JRCSF plasmid was obtained from BEI-resources and virus was produced as previously described¹⁹. One-quarter of the cells were infected by the addition of 200 units of virus per 1 × 10⁷ cells in 1 ml and centrifugation at 1,500g for 2 h at 37 °C. Following this ‘spinoculation’, the infected cell cultures were mixed in flasks containing the remaining three-quarters of the cells at 2 × 10⁶ cells per millilitre in medium containing 50 IU of human recombinant IL-2. At day 10, the cells were crowded in 96-well round-bottom plates at a density of 2 × 10⁵ cells per 200 µl per well to enhance cell-to-cell transmission of HIV. On day 13, cells were cryopreserved in FBS with 10% DMSO and stored in liquid nitrogen for future experiments.

In vitro-infected memory CD4⁺ T cell killing assays

Infected memory CD4⁺ T cells were thawed and rested for 2 days at 2 × 10⁶ cells per millilitre in medium containing 50 IU of human IL-2. Before co-culture with CTLs, cells expressing high levels of CD4 were depleted by magnetic selection using the EasySep Human CD4 Positive Selection Kit II (STEMCELL); this step enriches for CD4⁻ infected cells (as HIV-infected cells downregulate CD4). Cells infected with HIV JRCSF were labelled with CellTrace far red (dilution 1:20,000). TW10-Gag specific CTLs were left unlabelled and were added to each well at a 1:2 effector:target ratio along with anti-CD107a APC-Cy7 (dilution 1:200). Killing assays were performed in R10 medium supplemented with 50 IU of human recombinant IL-2 and 0.51 ng ml⁻¹ of human recombinant IL-15 for 16 h at 37 °C. In DFO treatment experiments, cells were pre-treated with 10 µM DFO or DMSO for 72 h and then washed three times in R10 medium before co-culture.

Intracellular cytokine staining

To confirm functional T helper subtypes among ARCs and uninfected clones, cells were reactivated for 4 h with the eBioscience Cell Stimulation Cocktail (plus protein transport inhibitors; Thermo Fisher) following the manufacturer’s instructions. Cells were stained with surface antibodies targeting CD4 (OKT4, BioLegend), CD8 (SK1, BioLegend), CD71 (CY1G4, BioLegend) and CD69 (FN50, BioLegend), along with a fixable live/dead stain (Thermo Fisher) in PBS for 30 min at 37 °C. Cells were washed and fixed with 1× eBioscience FoxP3/transcription factor fixation/permeabilization solution (Thermo Fisher) for 45 min at 4 °C. This fixation solution was washed away and cells were stained intracellularly with anti-HIV Gag-p24 (KC57, Beckman Coulter), anti-IL-17 (TC11-18H10.1, BioLegend), anti-IL-4 (MP4-25D2, BioLegend), anti-IFNγ (B27, BioLegend) and anti-IL-22 (2G12A41, BioLegend) antibodies in 1× FoxP3/transcription factor permeabilization solution (Thermo Fisher) for 30 min at 37 °C. Cells were washed one more time before analysis on an Attune NxT flow cytometry instrument.

Peptide pulse killing assays

In DFO treatment experiments, ARCs were pre-treated with 10 µM DFO or DMSO for 72 h before three washes in R10 medium. ARCs were mock pulsed or peptide pulsed or with a Pol peptide pool or HIV optimal peptide Pol-NETPGIRY at 10 μg ml⁻¹ or 2 μg ml⁻¹ for 16 h at 37 °C, respectively. ARC-Chr16 cells pulsed with Pol-NETPGIRY were stained with 5 µM CTV, whereas mock pulsed cells were stained with 1/10 concentration of dye in PBS for 20 min at 37 °C in the dark, followed by quenching in R10 medium and a 5-min incubation at room temperature. After washing in R10, peptide pulse and mock pulse conditions were mixed at approximately a 1:1 ratio and either incubated with or left without Pol-NETPGIRY specific CTLs at a 1:1 or 3:1 effector:target ratio for 16 h at 37 °C. Where indicated, the killing assay co-cultures were supplemented at the initiation of co-culture with either an MHC-I-blocking antibody (anti-HLA-A/B/C, clone W6/32, H463, Thermo Fisher Scientific; 10 µg ml⁻¹) or a matched mouse IgG1 isotype control (554121, BD Biosciences; 10 µg ml⁻¹). We performed a peptide-pulsed killing assay in which one ARC was co-cultured with three autologous uninfected CD4⁺ T cell clones in a single well. Each clone was uniquely labelled using distinct CellTrace dye intensities to enable simultaneous tracking. Target cells were pulsed with the cognate peptide alongside non-pulsed bystanders and cultured in the presence or absence of HIV-specific CTLs. Relative survival of each clone was quantified following co-culture, allowing direct comparison of CTL-mediated elimination under identical conditions.

DNA content analysis by flow cytometry

DNA content of ARCs and uninfected clones was measured to assess distributions of cells in cell cycle. Cells were surface stained with an antibody targeting CD4 (OKT4, BioLegend) along with a fixable live/dead stain (Thermo Fisher) in PBS for 30 min at 37 °C. Cells were washed and fixed with 1× eBioscience FoxP3/transcription factor fixation/permeabilization solution (Thermo Fisher) for 45 min at 4 °C. This fixation solution was washed away and cells were stained intracellularly with anti-HIV Gag-p24 (KC57, Beckman Coulter) in 1× FoxP3/transcription factor permeabilization solution (Thermo Fisher) for 30 min at 37 °C. Cells were washed for an additional time in fixation/permeabilization solution, resuspended in this buffer with 100 μg ml⁻¹ of RNase A, and incubated at 37 °C for 30 min. DAPI was then added to a final concentration of 1 μg ml⁻¹ for a further 10-min incubation. Cells were then directly analysed by flow cytometry.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-026-10298-w.

Author contributions

R.B.J., I.A.T.M.F., A.H. and N.L.B. conceived the study. R.B.J., I.A.T.M.F., A.H., N.L.B., T.T.H., N.L.L., Y.R., C.B., G.Q.L., F.L., R.S., I.G.M., M.S.G., A.L.H. and S.G. designed the experiments. I.A.T.M.F., A.H., N.L.B., R.B.J., T.T.H., N.L.L., Y.R., C.B., V.K.P., S.V., E.N., P.S., A.D., M. Canis, C.O., F.L., R.S., I.G.M., S.S., E.S., E.W., M.S.G., A.L.H., A.R.T. and S.G. performed the experiments. I.A.T.M.F., A.H., N.L.B., R.B.J., T.T.H., G.Q.L., A.D., P.D.B., R.S., I.G.M., S.S., E.S., E.W., M.S.G., A.L.H. and S.G. analysed and interpreted the data. C.B., M.C.N. and M. Caskey contributed essential reagents or analytical tools. M. Caskey, S.M., T.-W.C., C.K. and E.K.H. provided clinical samples or study participant data. A.H., P.Z., D.B., G.Q.L. and V.K.P. performed the bioinformatic analyses. R.B.J., P.D.B. and M.C.N. supervised the research. R.B.J., A.H. and I.A.T.M.F. wrote the manuscript with input from all authors. All authors reviewed, edited and approved the final manuscript.

Peer review

Peer review information

Nature thanks the anonymous reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Data availability

All data necessary to evaluate and understand the conclusions of this article have been provided in this article or the supplementary information associated with it. Individual data points used for summary data display items are all included in the Source data provided with this paper. The proviral sequences reported in this article have been deposited in GenBank under accession numbers PX992673–PX992682. The 5′ CITE-seq data reported in this article have been deposited in the Gene Expression Omnibus database and made publicly available under accession number GSE303197. Other data are available from the corresponding author on reasonable request. Source data are provided with this paper.

Code availability

This paper does not report any original code. Code used for analysis of CITE-seq data will be made available on request.

Competing interests

R.B.J. has served as an advisor to ViiV Healthcare and received payment for this role. G.Q.L. receives Merck & Co. funding through her institution via investigator-initiated grant programs but does not have personal financial interests with Merck. M.C.N. serves on the Celldex Scientific Advisory Board. The other authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41586-026-10298-w.

Supplementary information

The online version contains supplementary material available at 10.1038/s41586-026-10298-w.