Subjects included in this study are from a subset of patients recruited for our previous study (8). Briefly, LN samples from HIV⁺ donors were excised from palpable cervical LNs for clinical diagnostic workup in Mexico. HD samples were de-identified mesenteric or inguinal lymph nodes from the Cooperative Human Tissue Network (CHTN). Twenty-two HIV⁺ donor LNs and nine HD LNs were used for Mass CyTOF experiments. Considering sample availability and size, LN and PBMC samples from an independent, but semi-overlapping cohort (Tables S1, S5) were collected from LN and PBMC samples were collected from seven non-ART and one ART-treated HIV⁺ patients. Sample sizes were not pre-specified but dictated by the availability of the sample. All samples from HIV+ patients were de-identified and were obtained in accordance with the Declaration of Helsinki after obtaining written informed consent of participants and as part of protocol B03-16, which was reviewed and approved by the Research Committee and the Ethics in Research Committee of the National Institute of Respiratory Diseases “Ismael Cosío Villegas”, Mexico City.

Mass CyTOF data were collected as described in a previous publication (8), except that the cells were not stimulated prior to staining. Data were collected and normalized as described there. In analysis specific to this paper, cells were subset by drilling down on the gates defined in Figure S1 with FlowJo™ v10.8.0 Software (BD Life Sciences) and exported as individual FCS files. FCS files were then read into R using the “flowCore” package and analyzed using custom scripts. Dimensional reduction and clustering were done using the R package “Seurat”. Cluster annotations and rational are described in Table S2. Boxplots and UMAP plots were generating using the R package “ggplot2”. The heatmap in Figure 1 was made with the R package “pheatmap” (33). Statistical analyses were done using the R package “rstatix”. Final plots were compiled in Biorender.

Cell staining and sorting were performed as previously described (9). Briefly, cryopreserved LN cell suspensions were freshly thawed. A BD FACSARIA was used to sort live CD4⁺ T cells as CD3⁺CD4⁺CD8^-CD14^-CD19^-. Then, from the live CD4⁺ T cell population, naïve (CD45RO^-CXCR5^-CD57^-CCR7⁺), CXCR3⁺T_FH (CD45RO⁺CXCR5⁺CCR7^-PD1⁺CXCR3⁺), CXCR3^-T_FH (CD45RO⁺CXCR5⁺CCR7^-PD1⁺CXCR3^-), GC-T_FH (CD45RO⁺CXCR5⁺CD57⁺) and CXCR5^-T_FH (CD45RO⁺CXCR5^-PD1⁺ICOS⁺) populations were sorted directly into lysis buffer and stored at -80°C until further use.

Total RNA was purified using AllPrep DNA/RNA Mini and RNeasy MiniElute kits (Qiagen) according to the manufacturer’s instructions. Purified RNA quality was evaluated using Agilent RNA 6000 Pico Kit. Samples were then either supplemented with RNase inhibitor (RNase OUT, Thermofisher Scientific) and stored at -80°C or taken directly to reverse transcription.

TCRβ repertoire sequencing libraries were prepared and sequenced as described previously (8). Briefly, 30% of purified RNA was used for reverse transcription. Molecular identifiers (MIDs) were added to cDNA templates during reverse transcription using a TRBC-binding primer with 12 random nucleotides and a partial Illumina adapter (Table S4). PCR1 was carried out using multiplexed TRBV-binding primers. A second round of PCR was then used to append final Illumina sequencing adapters to the TCRβ junction-containing inserts. Final libraries were sequenced on an Illumina paired end 150x150 configuration at a minimum depth of 10 reads/cell. After sequencing, reads were clustered based on their MID and TCRβ sequence similarity. Consensus sequences were then built to correct PCR and sequencing errors as described (34). TCRβ sequencing information is summarized in Table S5. The CDR3 blast module MIGEC (35) was used for TRBV/J and CDR3 annotation. Bhattacharyya coefficient (36) was used to measure TCR repertoire similarity. Circos plots were generated using circlize R package (37).

RNA-Seq libraries were prepared using a protocol modified from SMART-seq2 (SSII) (38). Briefly, 2ul of purified total RNA was reverse transcribed using a poly-T primer fused to the ISPCR handle (Table S6; RT_dT30VN). Second-strand synthesis was then done using the SSII template-switching oligo (Table S6; SSII_TSO). The following program was used: 42°C RT for 90min, 10 cycles of 50°C for 2min, 42°C for 2min and then 70°C for 15min. cDNA amplification was then done using KAPA HIFI and the ISPCR handle (Table S6; RT_TSPCR) with the following program: 98°C initial denature for 3min, 16 cycles of 98°C denature for 20sec, 64°C annealing for 30sec, 72°C extension for 6min and 72°C final extension for 5min. The PCR product was then purified using AmpureXP beads (Agencourt), according to the manufacturer’s protocol. Purified PCR product was then diluted to 0.1 – 0.3 ng/ul and tagmented using the Nextera XT kit (Illumina) with a reduced reaction volume. Briefly, 1.25 ul of diluted sample was used in a 5ul total reaction volume and fragmented for 10min at 55°C, and then held at 10°C. The reaction was then neutralized by adding 1.25 ul of NT buffer. Final libraries were then generated from the tagmentation product using Nextera adapters (Table S6) and the following program: 72°C for 3 min, 95°C denature for 3sec, 12 cycles of 95°C denature for 10sec, 55°C annealing for 30sec, 72°C extension for 1min and 72°C final extension for 5min. Indexed libraries were then purified using Ampure XP beads according to manufacturer’s instructions and quantified with an Agilent High Sensitivity DNA kit. The final libraries were pooled and sequenced on Illumina HiSeq (150x150) with a minimum depth of about 200 reads per cell. Each library was split into two technique replicates and sequenced independently to ensure the reliability of RNA sequencing results.

Sequencing reads were aligned to human reference genome GRCh38.p5 using RSEM (39). Differentially expressed genes were analyzed using DESeq2 (40). GSVA, a non-parametric unsupervised method that quantifies the relative enrichment of selected pathways, was performed using the R package, GSVA (41). Gene Set Enrichment Analysis (GSEA) on selected pathways or gene sets were performed with the R package fgsea (42). Tonsil Tfh signatures (GSE50391, CXCR5^highCD45RO⁺ versus CXCR5^- tonsil samples) were created using the GEO2R online tool (https://www.ncbi.nlm.nih.gov/geo/geo2r/). Metascape (43) and NetworkAnalyst (44) was used to determine GO pathway enrichment.

HIV transcripts were quantified from bulk RNA-Seq following the stepwise procedures (Figure S19) similar to the method outlined in Chang et al. (45). Briefly, reads that could not be aligned to the human genome were mapped to the HIV genome (NC_001802.1) with the gapped aligner software HISAT2 (46), accounting for splice junction events. Quantification of HIV transcript expression was calculated by normalizing the HIV genome mappable reads to human genome mappable reads and the input cell number. For a more reliable quantification of HIV copies in each library, the correlation between two replicates was first evaluated (Figure S20), then the HIV copy numbers of both technical replicates were averaged.

A paired sample two-tailed Student’s t-test was used for pairwise comparisons, with a P value less than 0.05 as a cutoff to determine statistical significance. These comparisons include CXCR3⁺/CXCR3^-T_FH percentages in Figure S10C, TCR repertoire similarities in Figures 4B, C and HIV copies in Figure 5A. For gene set enrichment analysis, CAMERA (47) (Correlation Adjusted MEan RAnk gene set test) was used to calculate enrichment significance. Pearson correlation was used to determine the degree of association.

We first set out to determine if, in concordance with SIV infection, CXCR3⁺T_FH were inflated in human LN during HIV infection. We used a 29 marker Mass CyTOF panel to survey T cells from 22 HIV⁺ and 9 healthy donor LNs (Characteristics of the donors are listed in Table S1). We compared the frequencies of four T_FH populations using the gating strategies described in Figure S1. Two of these populations, denoted as MEM T_FH and CXCR3^-T_FH, are defined as CD45RO⁺CXCR5⁺ and CD45RO⁺CXCR5⁺PD1⁺CXCR3^- CD4⁺ T cells, respectively. These two gating schemes are frequently used when characterizing T_FH (8, 32, 48). The remaining two populations, CXCR3⁺T_FH and CXCR5^-T_FH, are defined as CD45RO⁺CXCR5⁺PD1⁺CXCR3⁺ and CD45RO⁺CXCR5^-PD1⁺ICOS⁺ CD4⁺ T cells, respectively. CXCR3⁺T_FH have been characterized in rhesus macaques (RMs) and human blood (27, 31, 32), while CXCR5^-T_FH were recently described as functional T_FH in human LN and blood during HIV infection (9). Amongst the four populations analyzed, CXCR3⁺T_FH and CXCR5^-T_FH were significantly enriched in HIV⁺ patient LNs versus healthy donor LNs (Figure 1A). It should be noted, however, that all LNs from HIV⁺ donors were cervical, whereas HD LNs were derived from a mix of distinct anatomical locations (Table S1). Regardless, given their high frequency over steady state, we suspected these two T_FH subsets may play a role in HIV infection. To directly address this question, we compared the frequencies of T_FH populations with each patient’s plasma viral load (pVL) (Figure S2). Interestingly, the only metric correlated to pVL is the ratio of CXCR3⁺T_FH to CXCR3^-T_FH (Figure 1B), suggesting that donors with greater CXCR3⁺T_FH populations may be better at controlling virus.

To gain insight into the role and interplay of these T_FH populations in HIV infection, we sought to delve deeper into the Mass CyTOF data. To do so, we subset the data by drilling down on four T cell populations (CXCR3⁺T_FH, CXCR3^-T_FH, CXCR5^-T_FH and Naïve) described in Figure S1, effectively sorting these populations in silico. We then projected the in silico sorted populations onto two dimensions using Uniform Manifold and Projection (UMAP). At a first glance, cells derived from HIV⁺ donors occupied several distinct locations on the UMAP (Figure S3). Additionally, although modest heterogeneity was observed between the three in silico sorted T_FH populations, each tended to occupy different regions at different densities (Figure S4). We thus decided to use unsupervised clustering to gain unbiased insight into each of the populations and their relationships with one another. Twelve clusters emerged, several representing canonical CD4⁺ T cell and T_FH phenotypes (Figure 1C). Select signature markers (Figure S5), as well as the heatmap in Figure 1C were used to annotate clusters 0-8. Clusters 9-11 were not considered in downstream analyses due to their insignificant sizes. To evaluate the impact of ART on phenotype distribution, we next calculated the frequencies of ART⁺ and ART^- patients in each cluster (Figure S6). Although some clusters show some trending, we found no significant phenotypic skewing and thus proceeded to group both ART⁺ and ART^- patients as HIV⁺ in subsequent analyses of this dataset.

To determine unique features of HIV infection, we compared the distribution of T cells in HIV⁺ and healthy LN-derived cells in each cluster (Figure 1D). In agreement the elevated expansion of the T_FH compartment seen in the literature (6, 7), we found that cluster 5, denoted as proliferating T_FH based on high expression of Ki67 and BLyS, was significantly enriched in HIV⁺ donor LN cells (p < 0.001). Surprisingly, even though cluster 2 had the highest and most homogeneous expression of CXCR3 (Figure S5), we found no difference between HIV⁺ and healthy cells. This might reflect that many of the CXCR3⁺T_FH in HIV infection take on diverse phenotypic programs. Specifically, given the proximity of cluster 2 with the NAÏVE cluster, it is possible that the CXCR3⁺T_FH within cluster 2 represent early T_FH entering the LN as previously described (20), whereas CXCR3⁺T_FH taking on alternative phenotypic programs may represent other stages of the T_FH lifecycle. In line with this argument, we reasoned that cluster 5, in addition to being enriched in HIV⁺ donors, might also have signatures unique to HIV infection. To test this hypothesis, we subsampled cluster 5 to have equal numbers of HIV⁺ and healthy donor cells and ran a likelihood-ratio test on each marker (49). We found that CXCR3 expression in cluster 5 was in fact significantly higher in HIV⁺ than healthy donors (Figure 1E). Given the HIV-intrinsic upregulation of CXCR3 in cluster 5, we reasoned that the in silico sorted populations might have differential distributions within cluster 5. As expected, CXCR3⁺T_FH were significantly enriched in cluster 5 over all other populations in HIV⁺ patients, but not healthy donors (Figure S7). Concordantly, CXCR3⁺T_FH from HIV⁺ donors occupied cluster 5 at a significantly higher frequency than from healthy donors, whose predominant population was contrastingly CXCR3^-T_FH (Figure S8). In addition to CXCR3, Granzyme A, a marker indicative of cytotoxic function in T-helper cells (50), was also enriched in cells from HIV⁺ donors within cluster 5 (Figure 1E). It is possible that, although CXCR3⁺T_FH generally reside in a more quiescent, immature state in the steady state, HIV infection triggers them to activate, proliferate in the LN, and take on phenotypic signatures unique to HIV infection.

To better understand the relationship between the phenotypic signatures that the in silico sorted T_FH populations (Figure S1) take on specifically in HIV infection, we did pairwise comparisons of the distributions of CXCR3⁺, CXCR3^-, and CXCR5^- T_FH populations within HIV+ donors across each cluster (Figure 1F). Reflecting the cellular plasticity of T_FH, gating down on only a handful of markers unsurprisingly revealed modest heterogeneity within each population. As expected, however, cluster 2, defined by the highest expression of CXCR3, bore a significantly large proportion of CXCR3⁺T_FH compared to all other populations. In line with previous studies (8, 9), CXCR5^-T_FH were significantly skewed toward activated states in clusters 5 and 6 (Proliferating T_FH and EOMES⁺T_FH, respectively), while CXCR3⁺ and CXCR3^-T_FH both tended to exist in cluster 7 (Activated GC-T_FH). Although statistical significance was only reached between CXCR3⁺T_FH and CXCR3^-T_FH in cluster 2, we noticed that CXCR3⁺T_FH often existed at a frequency between CXCR3^-T_FH and CXCR5^-T_FH, suggesting this population might exist as a transitional state between the canonical T_FH and the recently characterized CXCR5^-T_FH. Given this possibility, we next evaluated the median position of each HIV+ donor’s populations on the same UMAP coordinates as in Figure 1C (Figure 1G). At first glance, the Naïve population is clearly distinct from the T_FH populations, which group together on the right side of the UMAP. We also noted that, as suggested in Figure 1F, CXCR3⁺T_FH appear at an intermediate position between CXCR3^-T_FH and CXCR5^-T_FH. Statistical analysis of each population also revealed that all populations, except CXCR3⁺T_FH and CXCR5^-T_FH, were distinct over UMAP_1 (Table S3). Additionally, hierarchical clustering of each population from each donor revealed a grouping of CXCR3⁺T_FH with CXCR5^-T_FH more than with CXCR3^-T_FH (Figure S9). In concordance with our hypothesis, these data also posit CXCR3⁺T_FH as an interim population between CXCR3^-T_FH and CXCR5^-T_FH.

Our findings in the Mass CyTOF dataset prompted us to gain deeper insight into the similarities and differences of these T_FH populations in HIV infection. Although Mass CyTOF can be useful for delineating cell states due to its ability to analyze large numbers of single cells rapidly and robustly for a selected panel of surface markers, it failed to provide an unbiased analysis of the transcriptome differences between samples. Additionally, due to the destructive nature of Mass CyTOF, TCR sequences, and thus clonal relationships, cannot be obtained. To circumvent these limitations, we sorted five populations of T cells for bulk RNA sequencing (RNA-Seq) from 7 HIV⁺, non-ART-treated donors using the gating strategy illustrated in Figure S10. We excluded the one ART-treated donor in this section as we were unsure how treatment would affect a highly sensitive assay such as RNA-seq. We focused specifically on five populations. The first two, CXCR3⁺T_FH (CD45RO⁺CXCR5⁺CCR7^-PD1⁺CXCR3⁺), and CXCR3^-T_FH (CD45RO⁺CXCR5⁺CCR7^-PD1⁺CXCR3^-), were sorted to delineate the specific effects of CXCR3 expression on the state of LN T_FH. GC-T_FH (CD45RO⁺CXCR5⁺PD1⁺CD57⁺) and CXCR5^-T_FH (CD45RO⁺CXCR5^-ICOS⁺PD1⁺) were sorted as two distinct reference points between two subsets that have been demonstrated to provide B cell help (8, 9). CD57-expressing T_FH were sorted as a proxy for GC-T_FH as they have been shown to be a major subset of active GC cells (51–53), which serves as a positive control for T_FH function. An initial analysis on CyTOF and Flow Cytometry data shows comparable expression of CD57 between CXCR3⁺T_FH and CXCR3^-T_FH populations (Figures S11, S12). Naïve CD4+ T cells CD45RO^-CXCR5^-CCR7⁺) were sorted as a non-T_FH control.

Given a clear enrichment of CXCR3⁺T_FH within HIV⁺ patient LNs in the Mass CyTOF dataset, we wanted to understand their transcriptional characteristics in HIV infection. Of particular concern, existing studies with animal LN (27) and human blood (31) show contradicting results on their potential for B cell help. We thus set out to explore the transcriptional landscape of CXCR3⁺T_FH in relation to the other, better characterized, T_FH populations described above using RNA sequencing (RNA-Seq). We projected the RNA-Seq data from each population onto 2-dimensional space using Principal Component Analysis (PCA). As expected, the naïve population was distinct from all the other populations based on PC1. The T_FH populations, although similar on PC1, were separated based on PC2 (Figure 2A). In concordance with our analysis on the Mass CyTOF data, CXCR5^-T_FH were distinguishable from GC-T_FH and CXCR3^-T_FH on PC2, while CXCR3⁺T_FH situated in the middle. These data again suggest CXCR3 expression on T_FH may be indicative of a transitional state between the canonical GC-T_FH and the recently described motile CXCR5^-T_FH (9). Further advocating for CXCR3⁺T_FH as a transitional state, we also noticed CXCR3 expression in both flow cytometry and Mass CyTOF was higher in CXCR5^-T_FH than CXCR5⁺T_FH (Figure S13). To better understand the transcriptome differences among various T_FH populations, we next used Metascape (43) to evaluate the pathways enriched in genes contributing to PC2 (Figure S14). Interestingly, genes that mostly negatively correlated with PC2 (upregulated in CXCR5^-T_FH while downregulated in GC-T_FH) were enriched in cell division and migration. However, genes that most positively correlated with PC2 (upregulated in GC-T_FH while downregulated in CXCR5^-T_FH) were enriched on pathways inhibiting cell proliferation and migration (Table S7). For example, CTLA4 transmits inhibitory signal to T cells proliferation (54), SMAD3 mediates inhibition of CD4 T-cell proliferation (55), MCC blocks cell cycle progression from G0/G1 to S phase (56). In support of our CyTOF dataset, CXCR3⁺T_FH and CXCR5^-T_FH appear to be in a more activated, motile state than their canonical GC-T_FH counterpart.

To evaluate the differences between CXCR3⁺ and CXCR3^-T_FH more directly, we next analyzed differentially expressed genes (DEGs, Figure 2B) between the two populations. In total, 86 out of 12,515 genes were identified as significant DEGs (Table S8). Hierarchical clustering based on those DEGs further demonstrated the similarity of CXCR3^-T_FH with GC-T_FH and CXCR3⁺T_FH with CXCR5^-T_FH (Figure 2C). A significant upregulation of canonical T_H1 genes (CXCR3, IFNG, CXCR6 and GBP1) in CXCR3⁺T_FH also indicated their unique T_H1-like program. Additionally, upregulation of NELL2 and MKI67 in CXCR3⁺T_FH suggested an increased proliferative capacity compared to their CXCR3^- counterpart. Interestingly, CXCR3⁺T_FH also upregulated ITGB7, encoding a subunit of the integrin α4β7, that plays a role in leukocyte adhesion and can serve as a homing receptor bound by HIV (27). Thus, it is possible that CXCR3⁺T_FH could be similarly, or even more, susceptible to HIV infection than other T_FH subsets. Surprisingly, IL10, encoding an unconventional cytokine (IL10) in T_H1 cells, was also highly expressed by CXCR3⁺T_FH. Since previous studies support IL10 as a key player in the establishment and perpetuation of HIV persistence (57), the upregulation of IL10 in CXCR3⁺T_FH may result in enhanced persistence of HIV after infection.

To infer the functional potential of human LN-derived CXCR3⁺T_FH, we next compared them to CXCR3^-T_FH specifically on T_FH-related genes. We created a T_FH signature gene set using previously published RNA-Seq data from human tonsil samples (31). Both CXCR3⁺T_FH, CXCR3^-T_FH and GC-T_FH appeared similar when hierarchically clustered on tonsil T_FH signature genes (Figure S15A). GSEA and GSVA based on the same gene set also revealed no significant differences between these two cell populations (Figures S15B, C). Detailed analysis on several manually curated T_FH-related genes also suggested CXCR3⁺ and CXCR3^-T_FH bear similar T_FH marker gene expression patterns (Figure S15D). For example: FOXO1, a negative regulator of BCL6 ( 58), was downregulated in both populations. Additionally, BCL6, MAF and CD84, key transcriptional regulators of T_FH differentiation, were similar in both CXCR3⁺ and CXCR3^-T_FH. B3GAT1 (an enzyme necessary for the production of CD57), which is specifically expressed by active GC-T_FH (8, 51–53), was also comparably expressed in both populations. As expected, CXCR5 and its upstream regulator ASCL were both similarly expressed in CXCR3⁺ and CXCR3^-T_FH. Taken together, these observations reveal that CXCR3⁺T_FH have a similar T_FH transcriptional program to CXCR3^-T_FH, suggesting a functional overlap between the two populations.

In summary, global transcriptome analysis depicts CXCR3⁺T_FH unique from GC-T_FH, biasing toward a T_H1-like program. It also revealed CXCR3⁺T_FH are similar to the recently described CXCR5^-T_FH. Focusing specifically on T_FH marker genes, significant differences between CXCR3⁺ and CXCR3^- T_FH were not observed, supporting the paradigm that human LN-derived CXCR3-expressing T_FH may still bear T_FH function.

To investigate deeper into the functional program of CXCR3⁺T_FH, we used NetworkAnalyst (44) to identify GO biological pathways enriched within this population. Only two GO pathways were significantly enriched within CXCR3⁺T_FH when compared with CXCR3^-T_FH. Interestingly, both pathways were related to T cell migration (Figure 3A). We visualized the differentially expression genes of these two pathways in detail, as expected, most of the genes were upregulated in CXCR3⁺T_FH (Figure 3B; Figure S16), which suggested a possibility of this cell subset to be more motile.

Our previous study (9) suggested that T_FH can downregulate CXCR5 expression and accumulate as CXCR5^-T_FH in LNs during HIV infection. The same study also found that these CXCR5^-T_FH provide B cell help and have a propensity to migrate into the periphery. The combination of CXCR3⁺T_FH also bearing a cell migration signature, as well as existing in an intermediate cell state between CXCR5^-T_FH and GC-T_FH based on both Mass CyTOF and RNA-Seq data, led us to hypothesize that CXCR3⁺T_FH may be the immediate relative of CXCR5^-T_FH. To test this hypothesis more directly, we compared the transcriptomes of CXCR3⁺ and CXCR3^-T_FH using GC-T_FH and CXCR5^-T_FH as reference gene sets. Specifically, we first identified 938 DEGs from 12,559 total shared genes between GC-T_FH and CXCR5^-T_FH. Among these DEGs, 514 were upregulated in CXCR5^-T_FH (Gene set 1) and 424 were upregulated in GC-T_FH (Gene set 2). We then performed GSEA analysis to compare CXCR3⁺ and CXCR3^-T_FH on Gene Set 1 and 2 (Figures 3C, D). As hypothesized, CXCR3⁺T_FH followed the CXCR5^-T_FH program, while CXCR3^-T_FH more closely followed the GC-T_FH program.

Together with our observations in Mass CyTOF, global transcriptome, differential gene expression, gene set enrichment, and pathway analyses, we conclude that CXCR3⁺T_FH likely exist in a phenotypically intermediate state between canonical GC-T_FH and the more motile and peripheral CXCR5^-T_FH.

After observing that CXCR3⁺T_FH upregulated cell migratory pathways and skew phenotypically toward CXCR5^-T_FH, we suspected that changes in CXCR3 expression on T_FH might facilitate transitions to and from GC-T_FH and CXCR5^-T_FH cell states. To test this hypothesis, we used bulk T cell receptor (TCR) repertoire sequencing to measure their in vivo clonal relationship with various T cell populations. Because of the immense diversity generated by V(D)J recombination, TCR sequences can be thought as unique ‘ID-Cards’ (59), where cells sharing a given TCR sequence are likely to be the progenies of the same ancestor. By comparing the overlap of TCR sequences among samples, a clonal lineage across tissues and cell states can be inferred.

We thus compared the TCR repertoire similarity across the five LN T cell populations using Bhattacharyya Coefficient (36) as the similarity index. Bhattacharyya coefficient (BHA) measures the overlap of clonotypes between two T cell populations, while taking size of clones into consideration. In contrast to previous findings in healthy human tonsils (23), CXCR3⁺T_FH, CXCR3^-T_FH and GC-T_FH were strikingly similar in their repertoires across all donors (Figures 4A, C; Figure S17). This suggests that these three populations are likely capable of transitioning in and out of each of these states with ease, which may be a phenomenon intrinsic to HIV infection. In concordance with transcriptomic data, however, CXCR3⁺T_FH were significantly more related to CXCR5^-T_FH than CXCR3^-T_FH across all donors (Figures 4A, B). For two out of the seven donors, we were able to compare the CXCR3⁺ and CXCR3^-T_FH repertoires with the cT_FH repertoires in the blood (60) (Figures 4D–F; Figure S18). Circos plots representing the overlap of TCRs between cT_FH and LN populations indicate that CXCR3⁺T_FH are also more clonally related to cT_FH than their CXCR3^-T_FH counterpart (Figure 4E), similar to a recent finding in healthy tonsils (23). A similar trend was observed considering clone size weighted BHA index (Figure 4F). In combination, high clonal overlap amongst all CXCR5-expressing T_FH in the LN, and the elevated relationship of CXCR3⁺T_FH with CXCR5^-T_FH and cT_FH, suggests that this population sits as a bridge between secondary lymphoid tissue and the periphery.

T_FH are major reservoirs for HIV (61), complicating their role in controlling virus through the coordination of GC reactions. Furthermore, CXCR3⁺T_FH have been speculated to maintain a dynamic HIV reservoir due to high expression of HIV co-receptors (i.e. CCR5 and α4β7) (62). High viral load in this highly motile cell type could either exacerbate systemic spread of virus or expose viral antigens to the immune system. Given our result that CXCR3⁺T_FH frequency negatively correlates with pVL (Figure 1C), we speculated a high level of HIV copies in CXCR3⁺T_FH might be a mechanism for improving clearance of virus.

To investigate this idea, we mapped the bulk RNA-Seq reads from each T cell population to the HIV genome to quantify their relative HIV infection intensities (see methods for details). We detected comparable levels of HIV transcripts in both CXCR3⁺ and CXCR3^-T_FH, although both were significantly higher than Naïve cells (Figure 5A; Figure S21). Similar to experiments in RMs (27), HIV abundance also negatively correlated with repertoire diversity (Figure 5B), reflecting the propensity of proliferating and antigen-experienced cells to bear a higher HIV burden. Given that the two T_FH subsets discussed here share a similarly high HIV burden, the consequence of the migratory potential of CXCR3⁺T_FH is likely a complex balance between limiting systemic spread of virus and exposing it to novel immune compartments.