Participant samples

Longitudinal cryopreserved PBMCs and plasma and/or serum samples stored at −80 °C were utilized in this study. ID107, and six other individuals, including two non-controller individuals ID104 and ID112, initially participated in the eCLEAR trial (EudraCT:2015-002234-53, ClinicalTrials.gov NCT03041012)⁴, and ID142 initially participated in the TITAN trial (EudraCT: 2018-001165-16, ClinicalTrials.gov NCT03837756)⁵. ID9254 initially participated in an open-label phase 1b study in people with HIV-1 on suppressive ART (EudraCT: 2016-002803-25, ClinicalTrials.gov NCT02825797)¹⁰. All individuals were found to harbor viruses that were sensitive to their respective bNAb treatment at enrollment and throughout follow-up, based on phenotypic and/or genotypic assessment methods^4,5,10 (Supplementary Table 6). Written informed consent was obtained from all individuals. All participants were offered financial compensation for lost income, time incurred for study procedures and/or for travel costs related to study visits. Continued ATI and leukapheresis for ID107 and ID142 was conducted under the trial protocol according to the National Committee on Health Research Ethics in Denmark (1-10-72-38-23). The decision to continue pausing ART was consensually made between the person or family, his physicians and the study team. Secondary use of samples from individuals who participated in eCLEAR was approved by the National Committee on Health Research Ethics in Denmark (1-10-72-110-16). For ID9254, the individual made the decision to remain off-ART after study completion except for a 3-day period of ART at week 34 after ART interruption, and is regularly monitored. Research blood samples were collected under an observational protocol approved by the Ethics Committee of the Medical Faculty of the University of Cologne (UKK 16-054).

Modified IPDA

Genetically intact proviruses were quantified using a modified intact proviral DNA assay (IPDA) as described previously^4,5,18. Primer–probe sets targeting the φ-region¹⁸ and the rev-response element within env¹⁷ were used to identify proviruses that are likely to be genetically intact in a duplex ddPCR assay using genomic DNA isolated from CD4⁺ T cells¹⁸. In parallel, two regions within the human RPP30 gene were also quantified. All primer and probe sets can be found listed in Supplementary Table 7. The output of both the HIV-1 and RPP30 assays were then used to estimate the number of intact HIV-1 copies per 10⁶ CD4⁺ T cells, based on the quantity of input DNA. A detailed methodology can be found in the Supplementary Notes.

Single-genome NFL proviral sequencing and matched integration site and proviral sequencing

Single-genome NFL proviral sequencing was performed using genomic DNA isolated from cryopreserved PBMCs as described previously²¹. In brief, isolated genomic DNA diluted to single-genome levels based on Poisson distribution statistics were subjected to single-genome amplification using Platinum Taq (Invitrogen, 10966018) and nested primers spanning NFL HIV-1 (HXB2 coordinates 638-9362). For a combined analysis of proviral sequences and corresponding chromosomal integration sites²¹, full-genome amplification was preceded by a multiple displacement amplification step with phi29 polymerase (REPLI-g Single Cell kit, QIAGEN,150345) per the manufacturer’s protocol. Proviral sequences were amplified from the unbiased WGA products using Platinum Taq (Invitrogen) and nested primers spanning NFL HIV-1 and were visualized by agarose gel electrophoresis (Quantify One and ChemiDOC MP Image Lab, Bio-Rad 12003154). Amplification products were subjected to Illumina MiSeq sequencing at the MGH DNA Core facility. Resulting short reads were de novo assembled and aligned to HXB2 to identify genomic defects using an automated in-house pipeline (https://github.com/BWH-Lichterfeld-Lab/Intactness-Pipeline). The presence or absence of hypermutations associated with APOBEC3G or APOBEC3F was determined using the Los Alamos HIV sequence Database Hypermut 3.0 program (https://www.hiv.lanl.gov/content/sequence/HYPERMUT/hypermutv3.html). Viral sequences that lacked all the defects described above were termed ‘intact.’ Proviral species that were completely sequence-identical were considered clonal.

Proviral integration site analysis

Integration sites associated with individual viral sequences were obtained by integration site loop amplification (ISLA) assays as previously described⁶⁷. DNA produced by WGA was used as a template. Resulting PCR products of the ISLA reaction were subjected to next-generation sequencing using Illumina MiSeq. MiSeq paired-end FASTQ files were demultiplexed; small reads (142 bp) were then aligned simultaneously to the human reference genome GRCh38 and the HIV-1 reference genome HXB2 using bwa-mem (v.0.7.19)⁶⁸. Biocomputational identification of integration sites was performed according to previously described procedures⁶⁷. The final list of integration sites and corresponding chromosomal annotations was obtained using Ensembl (v.113; www.ensembl.org), the UCSC Genome Browser (www.genome.ucsc.edu) and GENCODE (v.47; www.gencodegenes.org). Chromosomal coordinates of integration sites were indicated using the Hg38 reference genome nomenclature.

Quadruplex qPCR

Genetically intact proviral genomes were isolated and sequenced from PBMC samples sourced from ID9254 by Q4PCR as previously described²². In brief, CD4⁺ T cells were isolated from PBMCs by negative selection using the CD4⁺ T Cell Isolation kit (Miltenyi Biotec) and genomic DNA was isolated (Gentra Puregene Cell kit (QIAGEN)). Genomic DNA was then assayed by qPCR for the presence of HIV-1 gag to determine the limiting dilution of the DNA. Then, at the limiting dilution, NFL proviruses were amplified using primers targeting the 5′ and 3′ LTRs as described previously²². The products of these outer NFL PCRs were subjected to four-plex qPCR (targeting HIV-1 packaging signal, env, gag and pol) to identify amplicons that were most likely to contain a genetically intact provirus. NFL amplicons that were positive for at least two of four HIV-1 targets were used as template for the inner NFL PCR. These products were then subject to sequencing on the Illumina MiSeq platform. Amplicons were assembled using an in-house pipeline described previously (https://github.com/stratust/DIHIVA)²².

Quantitative viral outgrowth assay

qVOA was performed to quantify the frequency of inducible provirus for ID107 and ID142 as described previously^69,70,71,72, with minor modifications⁵⁹. A detailed description of laboratory methods can be found in Supplementary Notes. CD4⁺ T cells were isolated from PBMCs by negative selection and were plated at 50,000 cells per well in round-bottom 96-well culture plates. Cells were stimulated with PHA and then cultured for 12 days, before culture wells positive for inducible, infectious virus were identified using TZM-bl assays and quantified using limiting dilution analysis as previously described⁶⁶.

For ID9254, the similar quantitative and qualitative viral outgrowth assay (Q²VOA)²⁴ was used to quantify the inducible proviral reservoir, details of which can be found in the Supplementary Notes.

NFL sequencing of HIV-1 RNA from qVOA cultures

NFL (~8.7 kb; HXB2 coordinates 817-9501) HIV-1 RNA genomes were sequenced using a previously described assay based on single-genome sequencing: plasma-derived RNA using long-range sequencing (PRLS) assay²⁵. Approximately ten positive qVOA cultures were randomly chosen from ID107 timepoints at ATI weeks 29, 227 and 281. For ID142, two positive qVOA cultures were used for sequencing from ATI week 34. Detailed descriptions of methodology can be found in the Supplementary Notes. Following assembly of contigs, a consensus sequence was generated from at least three sequences per qVOA culture, and this was used to compare the virus present in individual qVOA wells to one another and to genetically intact proviruses generated by MIP-seq. For this comparison, all sequences were trimmed to the ~8.7-kb region sequenced by PRLS. Genetically identical sequences were identified using the Los Alamos webtool ElimDupes (https://www.hiv.lanl.gov/content/sequence/elimdupesv2/elimdupes.html). Phylogenetic trees were generated using PhyML⁷³, using the HKY85 nucleotide substitution model and a gamma rate of 4. Branch support was inferred using 1,000 bootstraps. Phylogenetic trees were visualized using ggTree⁷⁴ as described previously²⁵.

Single-genome sequencing of residual viremia

Characterization of residual viremia HIV-1 sequences in plasma was performed as described previously⁷⁵, with some modifications. In brief, plasma was first spun at 3,500g for 15 min at 4 °C to remove cell debris, lipids and fibrinogen. The supernatant was then transferred to new tubes and spun at 21,000g for 2 h at 4 °C. Viral pellets then underwent RNA extraction⁷⁶. HIV-1 RNA was used immediately for reverse transcription with SuperScript III in HIV-1 env using the primer env_RO 5′-GCARATGAGTTTTCYAGAGCA-3′ as previously described⁷⁷. cDNA was then diluted to the end point and used for single-genome sequencing of a partial region of the HIV-1 env gene (HXB2 position 6980-8036) as previously described⁷⁷. Primer sequences are listed in Supplementary Table 8. PCR products were sequenced by Sanger sequencing using the nested PCR primers (Azenta Life Sciences).

Single-genome sequencing of proviral env

Single-genome sequencing of proviral env was performed as described previously⁷⁸. In brief, following extraction of genomic DNA from resting CD4⁺ T cells, proviral env DNA was amplified using a two-step nested PCR protocol targeting HXB2 5983-8882 as described previously⁷⁸. Detailed methodology, including primer sequences and cycling conditions are listed in the Supplementary Notes.

Modified qVOA and single-genome env sequencing of positive cultures and plasma-derived genomes

Modified qVOAs (mQVOAs) were conducted as previously described²⁶. In brief, purified resting CD4⁺ T cells were seeded into qVOA culture wells at 200,000 cells per well with up to 107 replicate wells and 1 negative control well per qVOA arm. Resting CD4⁺ T cells were cultured with either no IgG, HIV-donor IgG (50 μg ml⁻¹) or autologous IgG (50 μg ml⁻¹). Cultures were carried out for 14 days as previously described, and assayed for HIV-1 p24.

Plasma viral RNA was extracted from p24⁺ qVOA supernatants using the Viral RNA Isolation kit (Zymo Research, R1041) as previously described⁷⁸. cDNA was synthesized using the env-specific primer OFM19 (5′-GCACTCAAGGCAAGCTTTATTGAGGCTTA-3′) with SuperScript III Reverse Transcriptase and RNaseOUT (Invitrogen) according to the manufacturer’s instructions, using the program 55 °C for 50 min followed by 85 °C for 10 min. Single-genome amplification (SGA) of HIV-1 env was then performed at limiting dilution as previously described⁴. A detailed methodology is provided in the Supplementary Notes.

For plasma HIV-1 RNA env sequencing, viral RNA extraction was performed as described previously for residual viremia env sequencing⁷⁶. cDNA was synthesized using Induro Reverse Transcriptase (New England Biolabs, M0681L). RNA and primer (OFM19) were first denatured at 65 °C for 5 min then snap-cooled at −20 °C. Reverse transcription was conducted in 20-µl reactions containing 1× Induro RT buffer, 1 mM dNTPs, 5 mM dithiothreitol (DTT) and 200 U Induro RT, and incubated at 55 °C for 50 min, followed by 95 °C for 10 min. SGA of env was performed using the same PCR conditions and primer sets described for mQVOA-positive wells, with a detailed methodology provided in Supplementary Notes.

Isolation of autologous IgG antibodies

For ID107 and ID142, autologous IgG antibodies were isolated from heat-inactivated plasma using a NAb Protein A Plus Spin Column (Thermo Scientific, 89956). IgG samples were dialyzed in 1× phosphate-buffered saline (PBS) solution, pH 7.2, at 4 °C to remove residual ART drugs. Final IgG concentrations were measured using Nanodrop 2000 Spectrophotometer (Thermo Scientific).

For ID9254, IgG were purified from heat-inactivated serum (56 °C for 40 min) using Protein G Sepharose 4 Fast Flow (GE Healthcare, 1706180x) followed by buffer exchange to PBS using Amicon Ultra Centrifual Filters (Merck Millipore) and sterile filtration using Ultrafree-MC columns (Merck Millipore).

Production of HIV-1 Env-expressing pseudoviruses

Env sequences of selected viral isolates were cloned into pcDNA 3.4 TOPO vector (Thermo Fisher Scientific, A14697) using either TOPO-TA Cloning (Thermo Fisher Scientific, 450071) or In Fusion Snap Assembly (Clontech, 638948) in accordance with the manufacturer’s instruction. When available, original PCR amplicons were directly cloned. Otherwise, for several isolates, we designed synthetic double-stranded DNA (gBlock) templates based on sequences derived from plasma, then amplified and cloned these using In Fusion Snap Assembly. The resulting Env-expression vector driven by a cytomegalovirus (CMV) promoter (12.5 µg) was transfected into HEK293T cells together with 15 µg pNL4-3-ΔEnv–eGFP packaging construct²⁷ in the presence of 2.5 µg pAdvantage construct (Promega, E1711) to boost protein expression. HEK293T cells were incubated at 37 °C, 5% CO₂ for 72 h, and culture supernatant containing the isolate-specific Env-expressing pseudoviruses was collected, centrifuged to remove cell debris, filtered through a 0.45-µm Steriflip (Millipore Sigma, SE1M003M00) and flash frozen.

In vitro pseudoviruses neutralization assay

Linear ranges for isolate-specific pseudoviruses were determined by titrating the pseudoviruses on TZM-bl cells (10,000 TZM-bl cells per culture well) with 3–6 technical replicates for each condition. In vitro neutralization assays were conducted with a pseudovirus concentration within the linear dynamic range. Dilutions of autologous IgG antibodies (10 ng ml⁻¹ to 100 µg ml⁻¹) were pre-incubated with pseudoviruses at 37 °C with 5% CO₂ for 90 min. TZM-bl cells in 50 µg ml⁻¹ DEAE-dextran were added and incubated at 37 °C with 5% CO₂ for 48 h. For each experiment, control wells were included containing pseudovirus and TZM-bl cells with no IgG (six wells per experiment) and TZM-bl cells with no pseudovirus or IgG (six wells per experiment). Viral infection was measured by luciferase production with Bright-Glo Reagent (Promega, E2610) 48 h later. For each pseudovirus, the percentage of infection (f_u) was calculated as a fraction of maximum infection (no antibodies present) after background signal was normalized. Both IC₅₀ and IIP values were calculated as previously described²⁷ to evaluate antibody potency and the in vivo efficacy of aNAbs. A detailed description of IIP calculations is provided in the Supplementary Notes.

In vitro neutralization assay using replication-competent viral isolates

For ID9254, neutralization sensitivity of replication-competent proviruses was assessed using a similar TZM-bl-based neutralization assay as described above for ID107 and ID142, but using replication-competent viral isolates generated using viral outgrowth assays. Replication-competent viral isolates were cultured from positive Q2VOA wells from the time point ATI week 102, as described previously¹⁰. Then, sensitivity of these viral isolates to purified IgG sourced from the time point ATI week 102, and the bNAbs 3BNC117 and 10-1074, was assessed by TZM-bl neutralization using serial threefold dilutions of purified IgG, as previously described^10,79.

Immunoassays

Lymphocyte proliferation, IFNγ ELISpot and AIM assays were performed on cryopreserved PBMCs that were thawed and washed with RPMI glutamine supplemented with penicillin–streptomycin and 10% FBS (cRPMI). PBMCs used for ELISpot and AIM were rested for 3 h at 37 °C, while PBMCs used for the lymphocyte proliferation assay were processed immediately. Detailed descriptions of the methodology of each immunoassay are provided in the Supplementary Notes.

Spectral flow cytometry intracellular cytokine staining

Cell stimulation and staining

A 27-color spectral flow cytometry-based intracellular cytokine staining assay was used to characterize HIV-1-specific memory CD4⁺ and CD8⁺ T cell responses. PBMCs were thawed and rested overnight at 37 °C. The next day, cells were stimulated with HIV-1 peptide mix (Pol, Gag, Nef or Env) or negative control (DMSO) together with co-stimulatory antibodies, secretion inhibitors and anti-CD107a antibody, and were incubated for 6 h at 37 °C. Next, cells were stained with viability dye, blocked and stained for markers of lineage/differentiation (CD3, CD4, CD8, CD16, CD19, CD45RA, CCR7, CD27 and CD95), chemokine receptors (CCR4 and CXCR5), cytokines (IFNγ, TNF and IL-2), proliferation (Ki67), transcription (FOXP3, T-bet and TCF1), effector proteins (perforin, GzmB, GzmK and granulysin) and immune checkpoints (TIGIT and PD-1). Details of all antibodies used are listed in Supplementary Table 9. For intracellular markers, cells were fixed and permeabilized using the eBioscience Foxp3/Transcription Factor Fixation/Permeabilization Concentrate and Diluent (Invitrogen, 00-5521-00) according to the manufacturer’s instructions and stained overnight at 4 °C. The next day, cells were washed and acquired within 3 h on a five-laser Sony ID7000 Spectral Analyser (SONY Biotechnologies).

The following controls were used for unmixing: unstained PBMCs for autofluorescence spectrum characterization, single-stained PBMCs with LIVE/DEAD Fixable Blue Dead Cell Stain (Invitrogen, L23105) and single-stained beads (UltraComp eBeads Plus Compensation Beads; Invitrogen, 01-3333-42) for all fluorochrome conjugated antibodies. When switching to a new antibody lot, we checked whether the spectral curved matched our existing single-stain control and if deviations were found, we acquired a new single-stain control for the unmixing.

A detailed description of all post-acquisition analysis steps can be found in Supplementary Notes.

Single-cell transcriptome, surface protein and TCR sequencing of AIM-sorted T cells

Sample size

Two PICs and two non-controllers with sufficient longitudinal material were included. Following quality filtering, 40,942 cells from 16 samples were retained (Supplementary Fig. 5), with a median of 2,094 cells per sample (interquartile range 1,573–3,572).

Sorting

The AIM assay was performed as described above, except that cells were stimulated with a combined HIV-1 Gag, Pol and Nef peptide pool at a final total peptide concentration of 0.7 μg ml⁻¹. After stimulation and staining, T cells were enriched using the Pan T cell Isolation kit (Miltenyi Biotec, 130-096-535) and sorted on a MACSQuant Tyto. Cells coexpressing 4-1BB and PD-L1, or 4-1BB and CD69 were collected.

Library preparation, sequencing and read alignment

After sorting, samples were stained with Total-Seq C Human Universal Cocktail v.1.0 (BioLegend, 399905; Supplementary File 1). Single-cell libraries were prepared using Chromium Next GEM Single Cell 5′ v.2 reagents (10x Genomics, 1000265, 1000252 and 1000541) according to the manufacturer’s instructions (CG000330 Rev F). Library quality was assessed by TapeStation D5000 Screen Tape (Agilent, 5067-5592, G2991AA) and Qubit dsDNA High Sensitivity Assay (Thermo Fisher Scientific, Q32851 and Q33238). Libraries failing quality control were excluded (Supplementary File 1). Sequencing was performed on an Illumina NovaSeq 6000 at the Department of Molecular Medicine, Aarhus University Hospital. Reads were aligned with Cell Ranger v.8.0.1 (10x Genomics) to a custom GRCh38-2020-A reference supplemented with autologous or subtype-matched HIV-1 sequences (Los Alamos National Laboratory (LANL) HIV Database, https://www.hiv.lanl.gov/, accession numbers K03455 and MH705134). TCR and ADT reads were processed using the human V(D)J reference (GRCh38) and the Total-Seq-C Human Universal Cocktail v.1.0 barcode list (Supplementary File 1).

Transcriptomic analysis

Cells with ≤500 genes, ≤1,000 unique molecular identifiers (UMIs) or ≥7.5% mitochondrial content were excluded. Doublets were identified using scDblFinder (v.1.18.0)⁸⁰. Cells with >3 productive TCR chains were additionally flagged as doublets and removed. Ambient RNA contamination was assessed using SoupX (v.1.6.2)⁸¹, no correction was applied (levels 1). Filtered singlet datasets were integrated and clustered to isolate high-quality T cells (Supplementary File 1).

RNA counts were normalized and scaled in Seurat v.5 (LogNormalize, ScaleData), with cell-cycle effects regressed out. Principal-component analysis (PCA) was performed on highly variable genes (2,000 during preprocessing, 1,500 in the final dataset), excluding nonprotein-coding and mitochondrial genes. PCA embeddings were batch-corrected using Harmony (v.1.2.3)⁸².

Clustering was performed using FindNeighbors (20 neighbors and 15 principal components) and FindClusters (Leiden)⁸³ with clustree⁸⁴-optimized resolutions of 0.6 (doublet discrimination), 0.8 (contaminants removal) and 0.7 (final). UMAP embeddings were generated using RunUMAP. Clusters were annotated manually using differentially expressed genes (DEGs). Differences in cluster proportions were assessed using Scanpro⁸⁵ with 100 bootstrapped runs, reporting the median adjusted P value (Supplementary File 1).

Virus-specific activation scores were calculated using AddModuleScore with a published gene set^32,33: ACTB, ACTG1, CD82, CRTAM, CTNNA1, EGR2, GNG4, GZMB, HSP90AB1, HSPA8, IFNG, IL2RA, MIR155HG, NR4A2, PKM, TAGAP, TNFRSF18, TNFRSF9, XCL1 and XCL2.

DEGs for cluster annotation were identified using FindAllMarkers (log₂fold change (FC) ≥ 0.25 and expression ≥ 25%) with the Wilcoxon rank-sum test during preprocessing. For the final analysis, MAST (v.1.30.0)⁸⁶ was used with donor origin as a covariate. For analyses of clusters 1, 3 and 10, genes with log₂FC ≥ 0.5, expression ≥ 10% and adjusted P 87 on GO Biological Process terms (C5: GO:BP, human; minimum gene set size of 15). Terms with adjusted P q 88 and simplify (clusterProfiler). All DEGs and enriched GO terms are reported in Supplementary File 1.

Surface protein analysis

DSB-normalized (v.2.0.0)⁸⁹ counts were used for marker selection, protein-positivity thresholding and visualization. Centered log-ratio (CLR)-normalized counts were used for differential testing with FindAllMarkers using the Wilcoxon rank-sum and receiver operating characteristic tests. Following previous studies^90,91, only markers with Z > 2 versus isotype controls were tested. Markers with adjusted P  0.7 were retained for visualization (Supplementary File 1). Hierarchical clustering was performed using hclust on Euclidean distances calculated with dist (stats v.4.4.0).

TCR analysis

Clonotypes were defined by shared VJ usage and identical CDR3 nucleotide sequences in productive αβ pairs, without merging three- and two-chain clonotypes. At least one chain was recovered in 33,059 cells (80%), and a productive αβ pair in 26,345 cells (64%). Clone frequency ranks were calculated per participant. Clonality was quantified using the Gini coefficient³⁵ (DescTools v.0.99.56), based on bootstrapped, downsampled repertoires (n = 100). Clusters with fewer than 50 TCRs were excluded. Clonal overlap was quantified using the Morisita–Horn index³⁵ in R (v.4.4.0).

HIV-participant-derived xenograft mouse model

All animal procedures were conducted according to a protocol approved by the Weill Cornell Medical College Institutional Animal Care and Use Committee (protocol 2018-0027). Detailed standardized operating procedures for this model have recently been published⁹². NOD.Cg-Prkdc_scid Il2rg_tm1Wjl/SzJ mice (stock 005557), commonly referred to as NSG mice, were purchased from The Jackson Laboratory. Female (6–8-week-old) NSG mice were used in all studies. Mice were co-housed in ventilated cages with wood chip bedding and maintained in a temperature-controlled environment with a 12-h light–dark cycle at the facilities of Weill Cornell Medical College. The temperature of the holding rooms was maintained between 70–74 °F. The relative humidity was maintained between 30–70%.

In brief, mCD4⁺ T cells were isolated from PBMCs by negative selection and NSG female mice were engrafted with 5 × 10⁶ mCD4⁺ T cells. Three weeks post-mCD4⁺ T cell engraftment, the mice were bled weekly to quantify human cell counts and viral load. Spontaneous viral rebound occurred in two mice and therefore plasma from these two viremic mice was used to infect the remaining mice. A week following two rounds of infection, mice were divided across two groups for equivalent CD4⁺ T cell counts and viral loads (Supplementary Table 10). One group of mice received an engraftment of 5 × 10⁶ autologous mCD8⁺ T cells. Viral load was quantified weekly. A detailed methodology is provided in the Supplementary Notes. The statistical comparison between the plasma HIV-1 RNA load of the individual mice at week 11 post-mCD4⁺ T cell engraftment in the –CD8⁺ T cells group (n = 11) and the +CD8⁺ T cells group (n = 7) was performed using an unpaired t-test (two-tailed) on log-transformed data in GraphPad Prism. Normal distribution of log-transformed data was confirmed using a Shapiro–Wilk test.

We also included data from a PDX-HIV model using PBMCs from a person living with HIV-1 who did not exhibit signs of virological control in the absence of ART (clinical characteristics shown in Supplementary Table 11). This participant was recruited from Whitman Walker Health, Washington DC, through a protocol approved by the George Washington University Institution Review Board. Secondary use of PBMCs was approved by the Weill Cornell Medicine Institutional Review Board. The individual gave written informed consent.

Memory CD4⁺ T cells were isolated by negative selection from PBMCs and 5.2 million live cells were engrafted to NSG mice via tail vein injection. All mice (n = 11) were infected with 10,000 TCID₅₀ HIV-JRCSF and one group of mice (n = 5) received 5.0 million live memory CD8⁺ T cells on day 41 post engraftment. Blood was collected by tail nick on a weekly basis.

Identification of unique viral variant following viral rebound in ID142

Plasma-derived HIV-1 RNA genomes were sequenced from ID142 ATI week 142 plasma using the PRLS assay as described above for positive qVOA cultures²⁵. Genetically intact plasma-derived genomes were identified using the GeneCutter tool on the Los Alamos HIV database (https://www.hiv.lanl.gov/content/sequence/GENE_CUTTER/cutter.html). Phylogenetic trees including all genetically intact genomes sourced from post-rebound plasma and sequences derived from pre-rebound time points using MIP-seq and PRLS were generated as described above for qVOA-derived NFL sequences.

To check for the possibility of superinfection with a genetically unrelated viral strain causing the viral rebound in ID142, two analyses were performed, details of which are in the Supplementary Notes.

Phylogenetic trees of env sequences isolated from pre-rebound and post-rebound time points by MIP-seq, PRLS or SGA were generated using this nucleotide alignment using PhyML⁷³, using the GTR substitution model and a gamma rate of 4. Branch support was inferred with 1,000 bootstraps. An amino acid alignment was also generated, and the Highlighter tool⁹³ was used to visualize genetic changes in specific env regions, with the lowest diversity sequence in the alignment used as the reference.

HLA-matched wild-type and experimentally verified escape mutations within HIV-1 Gag, Pol and Nef were initially identified as previously described, utilizing the Los Alamos ‘Best-defined CTL/CD8⁺ Epitope Summary’ (https://www.hiv.lanl.gov/content/immunology/tables/optimal_ctl_summary.html), Los Alamos ‘CTL/CD8⁺ Epitope Variants and Escape Mutations’ table (http://www.hiv.lanl.gov/content/immunology/variants/ctl_variant.html) and the Biopython SeqUtils package (https://github.com/biopython/biopython/tree/master/Bio/SeqUtils)⁹⁴. Additional amino acid changes between pre- and post-rebound Gag, Pol and Nef sequences were cross-referenced with the Los Alamos HIV-1 database tool HIV Molecular Immunology Database Search (https://www.hiv.lanl.gov/mojo/immunology/search/ctl/form.html). Mutations were visualized using Highlighter plots⁹³, using a consensus sequence of all pre-rebound variants as the reference.

bNAb sensitivity screening

All sequences reported in this manuscript for the PICs were subjected to a genotypic assessment for bNAb sensitivity using the bioinformatic tool bNAb-ReP (v.1.1-4)⁹⁵, with a threshold for bNAb sensitivity of >0.8 for the prediction score of each single sequence and threshold of >90% of sequences from each individual having a prediction score of >0.8.

Reporting summary

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