Participants were recruited from a prospective observational study of HIV-1-infected individuals on ART who had blood and lymph node collected, and findings from this study were recently reported (31). In brief, a total of 16 participants treated with suppressive ART for at least 3 years were included in the current sub-study, 11 of which were enrolled at The Alfred Hospital, Melbourne, Australia, and 5 of which were enrolled at the University of California San Francisco (UCSF), California, USA. Leukapheresis samples were obtained from all 16 participants. Additionally, excisional lymph node biopsies were available for 4 participants. Fifteen of the participants were infected with HIV-1 subtype B, while one participant was infected with HIV-1 subtype CRF01_AE. These participants were chosen from the original cohort (n=21) (31) on the basis that they had sufficient cells available in storage to allow for sequencing of the provirus and the predicted size of the overall proviral reservoir. Clinical details of the participants in this study and the parent study are provided in Supplementary Table 1 , with the participant samples included in this study bolded.

Ethics approval for the study was obtained from the institutional review board at the Western Sydney Local Health District, which includes the Westmead Institute for Medical Research, the Human Research Ethics Committees at The Alfred and Avenue Hospitals in Melbourne, the University of Melbourne Ethics Committee, and the Institutional Review Board at UCSF. All participants provided written informed consent.

Leukapheresis and lymph node biopsy samples were processed, and cell sorting was performed as described in Rasmussen et al. (2022) (31). Briefly, for peripheral blood mononuclear cells (PBMCs) obtained by leukapheresis, memory CD4⁺ T-cells were first isolated by negative selection using the MojoSort Human CD4 memory T cell isolation kit (BioLegend, San Diego, CA, USA, Cat no 480064), followed by sorting. Lymph node mononuclear cells (LNMCs) were isolated from the lymph node biopsies, followed by sorting. The four peripheral blood and lymph node cell subsets obtained for the study following sorting were: double negative (CD3⁺CD4⁺CD45RA^-PD-1^-CTLA-4^-; DN), PD-1⁺CTLA-4^- (CD3⁺CD4⁺CD45RA^-PD-1⁺CTLA-4^-; PD-1⁺), PD-1^-CTLA-4⁺ (CD3⁺CD4⁺CD45RA^-PD-1^-CTLA-4⁺; CTLA-4⁺), and double positive (CD3⁺CD4⁺CD45RA^-PD-1⁺CTLA-4⁺; DP). Representative plots for the sorting strategy can be found in Supplementary Figure 1 , and in Rasmussen et al. (2022) (31). As shown in Rasmussen et al. (2022), to avoid collecting cells that overlap between the four cell populations, small gaps between the individual gates for each cell population were introduced during cell sorting (31). The purity of the sorted populations was also assessed on a subset of samples. The median purity was 90.1% (IQR 76.95-97.3%) (31).

As described in Rasmussen et al. (2022) (31), sorted cells were lysed using Buffer RLT (RNeasy Lysis Buffer; Qiagen) and stored at -80°C prior to extraction. Cell-associated DNA was extracted from the lysates using the Qiagen All-prep kit (Qiagen, Hilden, Germany).

Near-full-length proviral DNA amplicons were obtained from the extracted DNA for each cell subset using the full-length individual proviral sequencing (FLIPS) assay, as described in Hiener et al. (2017) (10, 32). Briefly, the extracted DNA was diluted to limiting dilution, followed by two rounds of a nested PCR targeting the 5’ and 3’ LTRs to amplify near-full-length proviral genomes. Extracted DNA from the 15 participants infected with HIV-1 subtype B was amplified using the subtype B-specific primers described in Hiener et al. (2017) (10). Extracted DNA from the single participant infected with HIV-1 subtype CRF01_AE was amplified as described in Wang et al. (2022) (33). Reagent concentrations and cycling conditions were unchanged for these primers. Amplicons are then prepared for next generation sequencing on the Illumina MiSeq platform (10).

Following sequencing, HIV-1 contigs were assembled using one of two assembly methods. The first method used a previously-designed custom workflow in CLC Genomics (Qiagen) (10). The second method is a custom pipeline for contig de novo assembly (34, 35), and can be found at https://github.com/jsede/virus_assembly. Briefly, the generated reads are QC trimmed using BBDuk v 37.98 (https://sourceforge.net/projects/bbmap/). MEGAHIT v1.1.3 (36) was then used to de novo assemble draft genomes, followed by mapping the reads onto the draft genome for confirmation using BBMap v37.98. This confirmation is visualised using Geneious Prime v.2020.0.3. The final majority consensus genome is then extracted, and this is used for further analysis.

Generated proviral genomes were classified as genetically-intact or defective (10, 37). Briefly, genomes were first scanned for regions containing inversions, followed by large internal deletions of size >100bp. The remaining genomes of size >8800bp were scanned for frameshift mutations, premature stop codons or incorrect start codons within HIV-1 open reading frames (ORFs) using the Los Alamos database tool GeneCutter (https://www.hiv.lanl.gov/content/sequence/GENE_CUTTER/cutter.html) and the National Cancer Institute (NCI) Proviral Sequence Database (PSD) Proviral Sequence Annotation & Intactness Test (https://psd.cancer.gov/tools/tool_index.php) (38). Hypermutated sequences were identified using the Los Alamos tool Hypermut (https://www.hiv.lanl.gov/content/sequence/HYPERMUT/hypermut.html) (39). Any remaining genomes were scanned for defects in the packaging signal or major splice donor (MSD) site (position 744-745) (10, 40). Genomes with a mutated MSD were still classified as genetically-intact if they contained a cryptic splice donor site at HXB2 position 748-749 (10, 37, 41). The remaining genomes were classified as genetically-intact.

Genomes with genetically-intact p24, tat, rev, rev response element (RRE) or nef ORFs were classified using GeneCutter and the NCI PSD Proviral Sequence Annotation & Intactness Test (https://psd.cancer.gov/tools/tool_index.php) (38). Groups of two or more sequences within the same cell subset that were identified as 100% identical to one another were classified as an expansion of identical sequences (EIS). Genetically-identical sequences were identified using the Los Alamos database tool ElimDupes (https://www.hiv.lanl.gov/content/sequence/elimdupesv2/elimdupes.html). Tropism of proviral genomes were identified using the tool Web PSSM (https://indra.mullins.microbiol.washington.edu/webpssm/).

Comparisons of infection frequency or the proportion of proviruses with a particular characteristic between cellular subsets were made using logistic regression ( Supplementary Table 2 ). The correlated nature of this dataset (repeated observations/cells for each participant) was accounted for by using a mixed logistic regression with a random effect for the intercept and grouped by participant. When convergence or boundary issues arose with this analysis, the random effect for intercept, grouped by participant, was replaced with a fixed effect for participant. This is indicated in Table 1 and Supplementary Table 2 by p-values with a “*”. Effect modification was additionally investigated by testing the inclusion of an interaction term between participant and cellular subset in the fixed effect only (no random effect) model and is shown in the final column of Table 1 and Supplementary Table 2 . To further reduce issues associated with model convergence and complex effect modification between participants and cellular subsets, comparisons were made in a pairwise fashion rather than a single regression with all subsets included.

Supplementary Table 2 shows all comparison results across the four subsets (six pairwise comparisons) within two anatomic locations (peripheral blood and lymph node) for several different but not independent outcomes (overall infection frequency, intact infection frequency, proportion intact p24, etc). This results in 312 statistical comparisons. Importantly, these 312 comparisons are not independent to one another. For example, trends in infection frequency may match trends in intact infection frequency, or trends in peripheral blood may match trends in lymph nodes. Therefore, the ideal adjustment for multiple comparisons is unclear. As this is exploratory work, we address this with a false discovery rate calculation that estimates what proportion of comparisons below some p-value cut-off threshold (a) are likely to be spurious, calculated as

Where n_a is the number of observed p-values below the threshold a and N is the total number of p-values calculated (312 in Supplementary Table 2 ). Two columns of p-values (no effect modification, and participant effect modification) are treated as a single comparison, and if either of these p-values are below the threshold it will contribute to n_a .This false discovery rate will be artificially high due to the non-independence of comparisons, and so represents an upper bound (worst-case scenario) of spurious results. Due to this lack of independence of comparisons we also encourage readers to interpret p-values as a scale of the level of statistical evidence – with smaller p-values indicating stronger evidence and a lower possibility of spurious findings – as opposed to a categorical significant/non-significant based on a significance threshold (e.g. p=0.05). This is in alignment with modern practice of interpretation of statistical results (42).

Logistic regressions were conducted in R (43). Mixed logistic regressions were conducted using R library glmer (44). Spearman’s correlation tests were conducted using Prism software (GraphPad).

All sequences included in this manuscript have been deposited in GenBank (accession numbers OP700895-OP701628).

It is well-known that the majority of HIV-1 proviruses found within CD4⁺ T-cells of HIV-1-infected individuals are genetically-defective, and therefore unlikely to be replication-competent (40, 45). We therefore first investigated the distribution of genetically-intact proviruses within peripheral blood cells sorted based on PD-1 and CTLA-4 expression. Consistent with this known abundance of genetically-defective proviral genomes in the cells of ART-suppressed individuals, we observed a low number of genetically-intact genomes across the individual participants, ranging from 0-8 genetically-intact genomes per participant cell subset ( Table 2 ). We used the number of available cells per subset per participant, the total number of HIV-1 proviral genomes and the number of genetically-intact proviral genomes amplified from those cells ( Table 2 ) to calculate the infection frequency per 10⁶ cells of genetically-intact HIV-1 proviruses. Overall, we observed that DN and PD-1⁺ cells had a higher estimated infection frequency of genetically-intact genomes compared to CTLA-4⁺ and DP cells ( Figure 1 ). We found weak evidence for PD-1⁺ cells having a higher intact infection frequency compared to CTLA-4⁺ cells within this data (p=0.04; Figure 1 ). We also found evidence for DN cells having a higher intact infection frequency compared to CTLA-4⁺ cells (p=0.01) and DP cells (p=0.03) within this data ( Figure 1 ). This indicates that the expression of CTLA-4 on resting CD4⁺ T-cells may be a marker for a lower infection frequency of genetically-intact proviruses. We did find evidence for a difference in the infection frequency of genetically-intact proviruses between DN and PD-1⁺ cells, but these differences were highly variable across different participants (participant effect modification p=0.009; Table 1 ). This may indicate a participant-specific role for PD-1 expression in enrichment for genetically-intact proviruses. We did not observe evidence for any participant-dependent differences between DP cells and PD-1⁺ cells or DP cells and CTLA-4⁺ cells ( Supplementary Table 2 ).

As a surrogate for investigating replication-competency, Pardons et al. (2019) found that latently-infected cells from ART-suppressed HIV-1-infected individuals that are able to express HIV-1 p24 protein in response to stimulation were enriched in PD-1⁺ cells (27). However, recent studies have shown that genetically-defective proviruses are able to express the p24 protein, as well as other HIV-1 proteins (34, 46–50). We therefore investigated whether proviruses with particular genetic characteristics are enriched in CD4⁺ T-cells expressing PD-1 and/or CTLA-4 in order to determine whether PD-1⁺ cells are enriched for proviruses that have the capability to express specific HIV-1 proteins.

We first compared the proportion of cells containing certain types of genetic defects between the DN, PD-1⁺, CTLA-4⁺ and DP cell subsets. Note that the classification of proviruses as defective follows a process of elimination, with sequences classified as containing inversions, followed by large internal deletions, hypermutation, premature stop codons, frameshift mutations or defects in the cis-acting region (10, 37). This means that the classification of these defects is not mutually-exclusive. No significant differences were observed between the four cell subsets for inversions or hypermutation ( Table 1 ). When comparing the proportion of proviruses with large internal deletions between the four peripheral blood cell subsets, we observed evidence that any differences between subsets, particularly between PD-1⁺ cells and each of the other cell subsets, were largely dependent on participant-specific variation ( Table 1 ). Overall, however, for this data we observed that CTLA-4⁺ cells had the highest estimated proportion of proviruses with large internal deletions, followed by DP cells, DN cells and PD-1⁺ cells. We found strong evidence that the proportion of sequences carrying a large internal deletion was higher in CTLA-4⁺ cells than PD-1⁺ cells within this data (p=0.0001; Figure 2A ) and evidence that PD-1⁺ cells had a lower proportion of proviruses with large internal deletions compared to DN cells (p=0.03) and DP cells (p=0.04) within this data ( Figure 2A ). However, we did find evidence that each of these differences varied across participants (participant effect modification; CTLA-4⁺ vs PD-1⁺ cells: p<0.00001, DN vs PD-1⁺ cells: p=0.0002, DP vs PD-1⁺ cells: p=0.008; Table 1 ). This observation that CTLA-4⁺ cells had the highest proportion of proviruses with a large internal deletion is consistent with the low infection frequency of genetically-intact proviruses observed in CTLA-4⁺ cells. Altogether, this indicates that PD-1⁺ cells are more likely to harbour genomes that are full-length, and therefore more likely to express HIV-1 proteins such as p24, compared to PD-1^- and CTLA-4⁺ cells.

Proviruses with defects in the cis-acting region encompass those with deletions in any of the four stem loops that form part of the packaging signal within the 5’-LTR of the HIV-1 provirus, or those with point mutations in the major splice donor (MSD) site (10, 40). In this study, these genomes are genetically-intact in all regions of the HIV-1 genome except for the cis-acting region. Interestingly, the overall estimated proportion of proviral sequences with cis-acting defects was found to be highest in PD-1⁺ cells compared to DN, DP and CTLA-4⁺ cells within this data (p<0.00001 for all; Figure 2B ). However, this appeared to be driven by a single participant (PRA005). When this participant was removed, the DN to PD-1⁺ comparison was no longer significant, but the difference between PD-1⁺ and CTLA-4⁺, and PD-1⁺ and DP remained significant (p=0.0002 and p=0.026, respectively). PRA005 was also the only participant for which these types of sequences were identified in the CTLA-4⁺ cell subset ( Figure 2B ). We did not find evidence for differences in the overall proportion of proviruses with cis-acting defects between DN, CTLA-4⁺ and DP cells within this data (p>0.05 for all; Table 1 ). Altogether, these results suggest that proviruses with defects in the cis-acting region accumulate in PD-1⁺ cells compared to cells expressing CTLA-4. As all regions of the HIV-1 genome are intact except for the cis-acting region in these proviruses, this also indicates that PD-1⁺ cells containing these HIV-1 genomes also have a greater potential to express HIV-1 proteins compared to cell subsets that express CTLA-4.

Lastly, we investigated whether there is an enrichment for proviruses with an intact p24 ORF in cells expressing PD-1, assuming that defective proviruses with an intact p24 ORF are able to express p24 protein (34, 46, 47, 50). We therefore quantified the infection frequency of intact p24 per 10⁶ cells within the four CD4⁺ T-cell subsets. We found that the infection frequency of proviruses with an intact p24 ORF was highly variable between participants, with strong evidence that the observed differences in infection frequency between each of the cell subsets was highly dependent on participant-specific variation ( Table 1 ). Consistent with previous work (27), however, we found that PD-1⁺ cells had the highest overall estimated infection frequency per 10⁶ cells of intact p24 gene, followed by DN cells, DP cells and CTLA-4⁺ cells ( Figure 2C ). In general, we found strong evidence for PD-1⁺ cells having a higher intact p24 infection frequency compared to DN cells, CTLA-4⁺ cells and DP cells within this data (all p<0.00001; Figure 2C ). However, we also found strong evidence that these differences varied across participants (participant effect modification; p<0.00001 for PD-1⁺ vs DN and PD-1+ vs CTLA-4⁺, p=0.0005 for PD-1⁺ vs DP; Table 1 ). Moreover, we found evidence for a difference in the intact p24 infection frequency between DN and DP cells (participant effect modification p=0.0003), CTLA-4⁺ cells and DP cells (participant effect modification p=0.05) and DN and CTLA-4⁺ cells (participant effect modification p=0.06), though these differences were variable across different participants ( Table 1 ). Overall, this suggests that PD-1⁺ cells have high levels of proviruses with intact p24, but participant-specific differences play an important role in the accumulation of these proviruses in memory CD4⁺ T-cells.

To investigate whether differences in the infection frequency of proviruses with an intact p24 ORF were influenced by similar factors to those that affect enrichment of all HIV-1 proviruses, we first calculated the infection frequency per 10⁶ cells of all HIV-1 proviruses (intact and defective; Supplementary Figure 2 ). We then investigated whether there was a correlation between total infection frequency and infection frequency of an intact p24 gene ( Supplementary Figure 3 ). For DN, PD-1⁺ and DP cells, we found a strong positive association between the total infection frequency and infection frequency of proviruses with intact p24 (DN: r=0.78, 95% CI: 0.45, 0.92, p=0.0006; PD-1: r=0.96, 95% CI: 0.89, 0.99, p<0.0001; DP: r=0.78, 95% CI: 0.43, 0.92, p=0.001; Supplementary Figure 3A, B, D ). In contrast, we observed that for CTLA-4⁺ cells there was no correlation between the total HIV-1 infection frequency and the infection frequency of proviruses with an intact p24 gene (r=0.44, 95% CI: -0.11, 0.79, p=0.1; Supplementary Figure 3C ). This indicates that there is a strong association between levels of total HIV-1 provirus and the ability of a HIV-1-infected cell to express p24, particularly in PD-1⁺ and CTLA-4^- cells.

Finally, to determine whether proviruses with an intact p24 gene accumulate specifically in cells expressing PD-1, we calculated the proportion of sequences with an intact p24 gene in each subset for each participant. Overall, we found evidence for PD-1⁺ cells having a higher proportion of sequences with an intact p24 gene compared to CTLA-4⁺ cells (p=0.0001) and DP cells (p=0.003) within this data ( Figure 2D ), though, importantly, this difference varied across participants for PD-1⁺ vs CTLA-4⁺ (participant effect modification p=0.002; Table 1 ). This indicates that cells expressing CTLA-4 contain fewer proviruses with intact p24. We also found evidence for a difference in the proportion of proviruses with intact p24 between DN cells and PD-1⁺ cells, but this varied across participants (participant effect modification p=0.01; Table 1 ). This finding indicates that in addition to having a higher infection frequency of proviruses with an intact p24 gene, PD-1⁺ cells also have a higher proportion of proviruses with intact p24 compared to cells expressing CTLA-4 (both single-positive and DP), which may explain why PD-1⁺ cells are able to express high levels of p24 protein upon ex vivo stimulation.

The HIV-1 protein Nef has been shown to have important functions in immune escape by downregulating expression of surface MHC-I, thereby allowing an HIV-1-infected cell to escape detection by HIV-1-specific CD8⁺ T-cells (46, 48, 51–53). We therefore investigated whether cells expressing PD-1 and/or CTLA-4 are enriched for proviruses containing a genetically-intact nef ORF. When comparing the infection frequency of proviruses with an intact nef ORF between the four cell subsets, we did find strong evidence that observed differences between these cell subsets were highly dependent on participant-specific variation in the magnitude and direction of the difference ( Table 1 ). Overall, however, we observed that CTLA-4⁺ cells and DP cells had the lowest estimated infection frequency per 10⁶ cells of intact nef compared to the other cell subsets ( Figure 3A ). We found strong evidence that PD-1⁺ cells had a higher infection frequency per 10⁶ cells of intact nef compared to CTLA-4⁺ cells, DP cells and DN cells within this data (all p<0.00001; Figure 3A ), with evidence that these differences varied across participants (participant effect modification; PD-1⁺ vs CTLA-4⁺: p=0.0001, PD-1⁺ vs DP: p=0.0095, and PD-1⁺ vs DN: p<0.00001; Table 1 ). Similarly, we found evidence for DN cells having a higher infection frequency of intact nef compared to CTLA-4⁺ cells (p=0.002) within this data ( Figure 3A ), with evidence that this varied across participants (participant effect modification p=0.02; Table 1 ). Together, this indicates that cells expressing CTLA-4 have a lower infection frequency of HIV-1 genomes containing an intact nef compared to cell types that don’t express CTLA-4.

We next utilised the calculated infection frequency of all HIV-1 proviruses ( Supplementary Figure 2 ) to investigate whether intact nef infection frequency is correlated with total infection frequency for the four cell subsets ( Supplementary Figure 3 ). A significant positive correlation was found between the total infection frequency and intact nef infection frequency for DN cells (r=0.65, 95% CI: 0.21, 0.87; p=0.008) and PD-1⁺ cells (r=0.87, 95% CI: 0.64, 0.95; p=<0.0001) ( Supplementary Figure 3A, B ). In contrast, no significant correlation was observed between total infection frequency and intact nef infection frequency for CTLA-4⁺ cells (r=0.32, 95% CI: -0.25, 0.72; p=0.25) and DP cells (r=0.23, 95% CI: -0.33, 0.67; p=0.41) ( Supplementary Figure 3C, D ).

To investigate whether proviruses with an intact nef ORF accumulate specifically in cells that don’t express CTLA-4, we compared the proportion of proviruses with intact nef between the four cell subsets. Complementing what was observed for the infection frequency of intact nef, we found strong evidence for DN cells having an overall higher proportion of proviruses with intact nef compared to CTLA-4⁺ cells within this data (p=0.0006; Figure 3B ), with evidence that this varied across participants (participant effect modification p=0.006; Table 1 ). We also found evidence that PD-1⁺ cells have an overall higher proportion of proviruses with intact nef compared to CTLA-4⁺ cells (p<0.00001), DP cells (p=0.0001) and DN cells (p=0.002) within this data ( Figure 3B ), with evidence that this varied across participants for PD-1⁺ vs CTLA-4⁺ (participant effect modification p=0.004) and for PD-1⁺ vs DN (participant effect modification p=0.007) ( Table 1 ). Interestingly, we also observed weak evidence for DP cells having a higher proportion of proviruses with intact nef compared to CTLA-4⁺ cells within this data (p=0.03; Figure 3B ). We found no evidence for a difference in the proportion of proviruses with intact nef in DN cells compared to DP cells (p>0.05; Table 1 ).

The HIV-1 viral proteins Tat and Rev are essential in the production of partially-spliced and unspliced RNA transcripts during HIV-1 proviral expression, and the Rev response element (RRE) is essential in ensuring that partially-spliced and unspliced RNA transcripts are able to leave the nucleus of an infected cell for translation (54). Previous studies of anti-PD-1 and anti-PD-1/anti-CTLA-4 combination immune blockade in HIV-1-infected individuals and SIV-infected non-human primates have indicated an increase in cell-associated unspliced RNA expression in response to this treatment (16, 19, 22–24). We therefore investigated whether these results may be due to increased levels of proviruses with intact tat/rev/RRE ORFs (referred to as tat/rev/RRE⁺) in cells expressing PD-1 and/or CTLA-4. Though we did find significant evidence for a strong contribution of participant-specific variation to any enrichment of proviruses with intact tat/rev/RRE⁺ ORFs within specific cell subsets ( Table 1 ), overall we observed that the estimated infection frequency of proviruses with intact tat/rev/RRE was highest in PD-1⁺ cells, followed by DN cells, DP cells and CTLA-4⁺ cells ( Figure 4A ). We found strong evidence that PD-1⁺ cells had the highest infection frequency of tat/rev/RRE⁺ proviruses compared to DN cells, CTLA-4⁺ cells and DP cells within this data (p<0.00001 for all; Figure 4A ), however with strong evidence that these differences did vary across participants for PD-1⁺ vs DN (participant effect modification p=0.0001) and PD-1⁺ vs CTLA-4⁺ (participant effect modification p=0.0005) ( Table 1 ). We found no evidence for a difference between DN and DP cells, or CTLA-4⁺ cells and DP cells within this data (p>0.05 for both; Table 1 ). Altogether, this indicates that PD-1⁺ cells are enriched for intact tat/rev/RRE compared to cells that don’t express PD-1 or those that express CTLA-4, which may lead to increased transcription initiation, reflected by higher levels of cell-associated unspliced RNA.

To further investigate the role of PD-1 as a marker for tat/rev/RRE⁺ proviruses, we utilised the calculated overall proviral infection frequency within each of the cell subsets ( Supplementary Figure 2 ) to look for correlations between the total infection frequency and the infection frequency of tat/rev/RRE⁺ proviruses in the four cell subsets ( Supplementary Figure 3 ). We found a significant positive correlation between the total infection frequency and the infection frequency of tat/rev/RRE⁺ proviruses in PD-1⁺ cells (r=0.78, 95% CI: 0.45, 0.92, p=0.0006; Supplementary Figure 3B ) and DP cells (r=0.56, 95% CI: 0.06, 0.84, p=0.03; Supplementary Figure 3D ). The correlation between total infection frequency and the infection frequency of tat/rev/RRE⁺ proviruses in DN cells approached significance (r=0.5, 95% CI: -0.01, 0.8, p=0.05; Supplementary Figure 3A ), while no significant correlation was observed between total infection frequency and the infection frequency of tat/rev/RRE⁺ proviruses in CTLA-4⁺ cells (r=0.25, 95% CI: -0.32, 0.68, p=0.37; Supplementary Figure 3C ). This indicates that PD-1 may be a marker for HIV-1-infected cells with tat/rev/RRE⁺ proviral genomes, while CTLA-4 may be a marker for reduced frequency of tat/rev/RRE⁺ proviruses during suppressive ART.

To determine whether proviruses that are tat/rev/RRE⁺ specifically accumulate in cells expressing PD-1, we compared the proportion of proviruses that are tat/rev/RRE⁺ between the four cell subsets. We again observed evidence that differences in the proportion of tat/rev/RRE⁺ proviruses between the four cell subsets were participant-specific ( Table 1 ). Overall, however, we did find evidence that PD-1⁺ cells had the highest proportion of proviruses that are tat/rev/RRE⁺ compared to DN cells (p=0.0001), CTLA-4⁺ cells (p<0.00001) and DP cells (p<0.00001) within this data ( Figure 4B ), with evidence that these differences varied across participants for PD-1⁺ vs DN (participant effect modification p=0.02) and PD-1⁺ vs CTLA-4⁺ (participant effect modification p=0.04) ( Table 1 ). We also found evidence that DN cells had a higher proportion of proviruses that are tat/rev/RRE⁺ compared to CTLA-4⁺ cells within this data (p=0.006; Figure 4B ), with evidence that this varied across participants (participant effect modification p=0.03; Table 1 ). Interestingly, we also observed evidence that DP cells have a higher proportion of proviruses with intact tat/rev/RRE compared to CTLA-4⁺ cells within this data (p=0.03; Figure 4B ). We found no evidence of a difference in the proportion of proviruses with intact tat/rev/RRE between DN and DP within this data (p=0.46; Figure 4B ). This further indicates that PD-1 expression may be a marker for proviruses with intact tat/rev/RRE ORFs.

The role of cellular proliferation in the maintenance of the HIV-1 reservoir during suppressive ART is well-known (10, 25, 37, 48, 55–64). Groups of two or more sequences within a cell subset that are 100% identical to one another are defined as an expansion of identical sequences (EIS) (37). When we investigated the proportion of sequences within each cell subset that were part of an EIS in the peripheral blood, we found evidence that all differences between cell subsets were affected by participant-specific variation ( Table 1 ). Overall within these participants, however, we observed that PD-1⁺ cells have the highest estimated proportion of sequences that are part of an EIS, followed by DP cells, DN cells and CTLA-4⁺ cells ( Figure 5 ). We found evidence that PD-1⁺ cells had a higher proportion of sequences part of an EIS compared to CTLA-4⁺ cells (p=0.02) and DN cells (p=0.001) within this data ( Figure 5 ), with strong evidence that these differences varied across participants (participant effect modification p<0.00001 for both; Table 1 ). We also found weak evidence that PD-1⁺ cells had a higher proportion of sequences part of an EIS compared to DP cells (p=0.05; Figure 5 ) within this data, with evidence that this difference varied across participants (participant effect modification p=0.01; Table 1 ). We found evidence for a difference in the proportion of sequences that were part of an EIS between DN and CTLA-4⁺ cells (participant effect modification p=0.0001), DN and DP cells (participant effect modification p=0.03), and CTLA-4⁺ cells and DP cells (participant effect modification p=0.008), though these differences were highly dependent on the participant ( Table 1 ). Altogether, this suggests that PD-1⁺ cells harbour a high level of identical proviral sequences, which is likely to contribute to the persistence of higher levels of HIV-1 proviruses with certain intact ORFs.

In addition to the peripheral blood samples, we also had access to lymph node biopsies for four participants. Memory CD4⁺ T-cells from the lymph node were sorted into the same four cell subsets (DN, PD-1⁺, CTLA-4⁺ and DP cells) as investigated in the peripheral blood, and FLIPS was utilised to genetically-characterise the HIV-1 provirus within these cell subsets. We note that the low number of cells available for sequencing from these cell subsets in the lymph node compared to what was available from the peripheral blood ( Table 2 ) has restricted the depth of our analyses. When we compared the total HIV-1 infection frequency per 10⁶ cells within the lymph node, similar to the peripheral blood, we observed that many differences in infection frequencies between the four cell subsets were affected by substantial participant-specific variation ( Table 1 ). Overall, however, we found that the estimated total infection frequency was higher in DN, PD-1⁺ and DP cells compared to CTLA-4⁺ cells ( Figure 6A ). We found evidence that, overall, CTLA-4⁺ cells had a lower total infection frequency of HIV-1 in the lymph node compared to DN cells (p=0.006), PD-1⁺ cells (p=0.001) and DP cells (p=0.0002) within this data ( Figure 6A ), with evidence that these differences varied across participants for PD-1⁺ vs CTLA-4⁺ (participant effect modification p=0.03) and DN vs CTLA-4⁺ (participant effect modification p=0.05) ( Table 1 ). We also found evidence for a difference in the total infection frequency between DN and DP cells, and PD-1⁺ and DP cells, but these differences were highly variable across participants (participant effect modification p<0.00001 for both; Table 1 ). We found no evidence for a difference in the total infection frequency between DN cells and PD-1⁺ cells in the lymph node within this data (p=0.61; Table 1 ). Altogether, this suggests that CTLA-4 expression in the lymph node is a marker for lower levels of HIV-1 provirus, with a potential role for co-expression with PD-1 leading to increased levels of HIV-1 provirus in some participants.

We were only able to identify genetically-intact proviruses within the lymph node for two out of four participants, with these intact proviruses found in the PD-1⁺ and DP cell subsets ( Figure 6B ). We therefore found no statistically significant difference in the infection frequency of genetically-intact provirus, or proportion of proviruses that were genetically-intact, between cell subsets expressing PD-1 and/or CTLA-4 in the lymph node ( Table 1 ). Similarly, we were unable to identify any overall statistically significant differences in the infection frequency of proviruses with an intact p24 ORF between the four cell subsets within this data, though we observed that DN, PD-1⁺ and DP cells had a higher estimated infection frequency of provirus with an intact p24 ORF compared to CTLA-4⁺ cells ( Figure 6C ). We did find evidence for a difference in the intact p24 infection frequency between DN and DP cells (participant effect modification p=0.001) and PD-1⁺ and DP cells (participant effect modification p=0.02), but these differences were highly dependent on the participant ( Table 1 ).

When considering the infection frequency of proviruses with an intact nef gene, we found evidence that, overall, PD-1⁺ cells have a higher intact nef infection frequency compared to DP cells within this data (p=0.006; Figure 6D ). We did not observe any statistically significant difference in the intact nef infection frequency between any other cell subset. Similarly, for the infection frequency of proviruses with intact tat/rev/RRE, we only observed a significantly higher infection frequency of tat/rev/RRE⁺ proviruses for PD-1⁺ cells compared to DP cells within this data (p=0.01; Figure 6E ).

In summary, our findings suggest that in the lymph node, CTLA-4⁺ cells have a lower HIV-1 infection frequency compared to CTLA-4^- cells, and particularly PD-1⁺ cells. These PD-1⁺ cells also have higher levels of proviruses with intact p24 and intact nef ORFs, which may be indicative of the presence of genetically-intact genomes in PD-1⁺ cells.

The 312 comparisons presented in Table 1 and Supplementary Table 2 will have higher family wise type 1 error rates than that suggested by any single p-value. To investigate this, we calculated worst-case false discovery rates at different p-value value thresholds to help understand the possibility of spurious results. These are “worst-case” as all comparisons in Supplementary Table 2 are treated as independent comparisons, when many of these are in fact correlated (e.g. intact infection frequency and intact p24, nef, tat, or rev ORFs). We find that for a p-value thresholds of 0.05, 0.02, 0.01 and 0.001, the worst-case false discovery rate (expected proportion of potentially spurious results) is 0.13, 0.066, 0.039, and 0.005, respectively. These worst-case false discovery rates are not considered unacceptably high. However, we stress that due to the exploratory nature of our study and the possibility of spurious results, our findings presented here have identified areas that require further investigation in future studies in order to confirm the distribution of HIV-1 proviruses in CD4⁺ T cells expressing PD-1 and/or CTLA-4 in HIV-1-infected individuals on long-term ART.