HIV-1 persistence in lymph node T follicular helper cells (TFH) is mitigated by functional virus-specic T cell responses during hyperacute-treated HIV-1 infection

HIV persistence in tissue sites despite ART is a major barrier to HIV cure. Detailed studies of HIV infected cells and immune responses in native lymph node (LN) tissue environment is critical for gaining insight into immune mechanisms impacting HIV persistence and clearance in tissue sanctuary sites. We compared HIV persistence and HIV-specic T cell responses in LN biopsies obtained from 14 individuals who initiated therapy in Fiebig stages I/II, 5 persons treated (Tx) in Fiebig stages III-V and 17 late Tx individuals who initiated ART in Fiebig VI and beyond. Using multicolor immunouorescence staining and in situ hybridization, HIV RNA and/or protein was detected in 12 of 14 Fiebig I/II Tx persons who were on suppressive therapy for 1 to 55 months, while all late Tx persons had persistent antigens. CXCR3+T follicular helper T cells harbored the greatest amounts of gag mRNA transcripts. Notably, HIV-specic CD8+ T cells responses associated with lower HIV antigen burden in LNs, suggesting that these responses may contribute to HIV suppression in LNs during therapy. These results reveal HIV persistence despite the initiation of ART in hyperacute infection and highlight the contribution of virus-specic responses to HIV suppression in tissue sanctuaries during suppressive ART. analyzed 64 excisional lymph node (LN) biopsies and paired peripheral blood (PB) samples obtained from a well pedigreed cohort of individuals, where some individuals initiated ART during hyperacute HIV infection, to investigate the impact of blunting peak viremia on the microanatomical location, cellular source and role of T cell responses on HIV persistence in LNs. Study participants were drawn from a unique HIV FRESH is a prospective study of uninfected 18-23-year-old women at high risk of HIV infection established the epicenter of the HIV epidemic where yearly incidence rates approach 10%. Despite vigorous prevention efforts, twice weekly monitoring for viral has identied and treated (Tx) persons at the onset of plasma viremia, allowing for immediate institution of ART in many cases resulting in peak plasma viral loads that are sometimes <1,000 RNA copies/ml and the preservation of CD4 + T cell numbers 18 . Our results show that despite ART-induced blunting of peak viremia 18 and augmentation of functional HIV-specic T cell responses 24 , HIV Gag p24 protein and viral RNA can persist in the LNs of Fiebig I/II Tx donors even after 4.5 years of fully suppressive ART, and these viral antigens are enriched in LN CXCR3 + Tfh cells. We also show that superior functioning T cell responses were with lower HIV antigen persistence in the LNs.


Introduction
Antiretroviral therapy (ART) does not eradicate HIV infection due in large part to early establishment and persistence of integrated proviruses in quiescent circulating and tissue reservoirs [1][2][3] , which are resistant to drug or immune-mediated clearance. Additional mechanisms of persistence that involve ongoing replication have been suggested 4 , including replication in germinal centers (GCs) within secondary lymphoid tissues, which have been described as sanctuary sites due to low CD8 + T cell in ltration 5 and suboptimal penetration of antiretroviral drugs 6 . Ongoing virus replication, and to a lesser extent viral gene expression in the face of ART are contested concepts, with studies of individuals initiated on ART during chronic infection yielding con icting results [6][7][8][9][10][11][12] . T follicular helper (Tfh) cells have been identi ed as a major source of persistent virus 13 , but the precise subset of these cells enriched for HIV transcription on therapy within sanctuary sites is unknown. HIV persistence on ART is underscored by the nearly inevitable rebound of plasma viremia when treatment is interrupted, even after years of suppression, and is the major barrier to HIV cure 14 . It has previously been shown that very early initiation of ART can lead to prolonged remission when treatment is interrupted 15,16 , but this is an infrequent occurrence. In most individuals, virus rebound occurs within weeks to months even in individuals initiated on ART during Fiebig stage I (hyperacute) HIV infection 14,17 , despite rapid suppression of viremia and dramatically lower numbers of latently infected cells in peripheral blood [18][19][20][21] . Intriguingly, virus rebound kinetics following treatment interruption are heterogenous, sometimes taking a year or more 14,22 . The underlying immunological and virologic mechanisms responsible for the diverse viral rebound kinetics remain unknown. For instance, prolonged therapy in SIV infected macaques initiating therapy within 6 days of infection, prior to detectable plasma viremia, led to apparent elimination of infection after 600 days of suppressive ART in some animals 23 .
However, such clearance of infection has not been observed in acute HIV infection, even in persons in whom therapy was initiated before detection of plasma viremia 20 . We previously showed that early treatment initiation enhances T cell functions in peripheral blood and limits viral diversity 24 , but it is not clear if functional responses occur in tissues and whether such responses play a signi cant role in HIV suppression during ART. Limited access to tissue samples from persons initiating therapy before peak viremia has impeded a better understanding of the impact of early therapy on the lymphoid reservoir.
Here we analyzed 64 excisional lymph node (LN) biopsies and paired peripheral blood (PB) samples obtained from a well pedigreed cohort of individuals, where some individuals initiated ART during hyperacute HIV infection, to investigate the impact of blunting peak viremia on the microanatomical location, cellular source and role of T cell responses on HIV persistence in LNs. Study participants were drawn from a unique hyperacute HIV infection cohort termed FRESH (Females Rising through Education, Support and Health). FRESH is a prospective study of uninfected 18-23-year-old women at high risk of HIV infection established at the epicenter of the HIV epidemic in South Africa, where yearly incidence rates approach 10%. Despite vigorous prevention efforts, twice weekly monitoring for viral RNA has identi ed and treated (Tx) persons at the onset of plasma viremia, allowing for immediate institution of ART in many cases resulting in peak plasma viral loads that are sometimes <1,000 RNA copies/ml and the preservation of CD4 + T cell numbers 18 . Our results show that despite ART-induced blunting of peak viremia 18 and augmentation of functional HIV-speci c T cell responses 24 , HIV Gag p24 protein and viral RNA can persist in the LNs of Fiebig I/II Tx donors even after 4.5 years of fully suppressive ART, and these viral antigens are enriched in LN CXCR3 + Tfh cells. We also show that superior functioning T cell responses were associated with lower HIV antigen persistence in the LNs.

Hyperacute HIV infection as a model to interrogate antigen persistence in lymph nodes
To determine the impact of immediate initiation of ART in hyperacute HIV infection (before peak viremia) on HIV clearance from sanctuary sites, we studied 14 women aged 18-26 who initiated ART during hyperacute HIV infection (Fiebig I/II Tx) and achieved full suppression of plasma viremia within a median of 15 days (range, 6 to 33). LNs were obtained by excisional biopsy after treatment for a median of 370 days [range, 19 to 1647]. All remained fully suppressed except for one donor who had a transient viral load blip prior to LN excision. Five additional individuals identi ed in Fiebig stages III-V of infection and started on ART one day after diagnosis were also included. Three additional control groups were included: 13 HIV negative (HIVneg) donors; 17 individuals who initiated treatment in Fiebig VI and beyond (late Tx); and 15 untreated individuals whose duration on infection is unknown (unTx). Detailed characteristics of the cohorts are in Tables 1 and S1. 95% of the study participants were females.
Long-term persistence of HIV Gag p24 antigen in germinal centers (GCs) of individuals initiating antiretroviral therapy during hyperacute HIV-infection To investigate HIV persistence in LNs of individuals initiating ART in Fiebig stages I/II, we measured HIV Gag p24 antigen in excisional LN biopsies by multicolor immuno uorescence staining of formalin-xed para n-embedded LNs and imaging of tissue sections (Fig. S1). The transcription factor BCL-6 was used to identify active GCs 25 (Fig. S1) and images were quanti ed for Gag p24 content using the algorithm for area measurements in TissueQuest (TissueGnostics) 26 . Fig. 1A shows a representative image of HIV Gag p24 LN staining for a participant who was diagnosed in Fiebig stage I, initiated ART within 48 hours and achieved persistent plasma viremia suppression within 33 days. The LN sample shown was obtained after 479 days of uninterrupted ART treatment with undetectable viremia, and depicts HIV Gag p24 antigen within a GC, which was present in 5 of the 7 GCs examined in this LN (Fig. S1E).
Gag p24 staining and imaging were conducted on a total of 14 LNs from fully suppressed Fiebig I/II Tx donors obtained at a median of 370 days (range, 19 to 1647 days) post-ART initiation. 12 of 14 (86%) of these Fiebig I/II Tx donors had detectable HIV Gag p24 in at least one GC, and overall 42 of 55 GCs evaluated (76%) were postive for Gag p24. In two Fiebig I/II Tx donors there was no detectable HIV Gag p24 despite examining more than more than 4 GCs (Fig. 1B). In those with detectable HIV Gag p24, quantitative image analysis revealed no correlation between the amount of HIV Gag p24 present in the LN tissue section and treatment duration prior to LN excision ( Fig. 1C) or peak plasma viral load (data not shown). Notably, regardless of treatment duration, Fiebig I/II Tx donors had signi cantly less detectable HIV Gag p24 compared to Fiebig III-V Tx (P<0.0001), late Tx (P<0.0001) and unTx (P<0.0001) donors, though there was considerable overlap (Figs. 1D-E, S1F-G). This result was also consistent in a subset of donors Tx beyond 1 year (Fig. S1H). Quantitative image analysis of all treated LNs revealed greater area percent of Gag p24 staining in GCs compared to extrafollicular areas of the tissue (P=0.04, Fig. 1F).
Together, these data demonstrate that early ART initiation in Fiebig stage I/II limits the magnitude of HIV Gag p24 antigen in LNs, but that Gag p24 can persist predominantly in follicular areas even after 4.5 years of fully suppressive treatment.

Lymph nodes of Fiebig I/II treated individuals harbor HIV-1 RNA
To determine if viral RNA transcription was occurring, which is required to produce infectious virions, we used an in-situ hybridization (ISH) assay called RNAscope 27 to probe for HIV-1 gag-pol RNA within LN sections. 12 Fiebig I/II Tx, 4 Fiebig III-V Tx, 2 late Tx, 4 unTx and 3 HIVneg LN samples were analyzed based on sample availability. Viral RNA was detected as punctate dots in LNs from all HIV infected persons and there were no signals in the HIVneg controls (Figs. 2A, 2B and S2). Productively infected viral RNA + cells were identi ed as a dense spherical signal, whereas follicular dendritic cell (FDC)-bound virus particles were de ned by a diffuse lattice-like pattern consistent with previous reports 28

. Combined
RNAscope â ISH gag-pol staining with IF staining for CD4 + T cells con rmed viral RNA (green) within CD4 + T cells (red, Fig. 2C). RNAscope staining was quanti ed using Fiji 29 . 10 of 12 Fiebig I/II Tx donors had detectable but signi cantly lower amounts of HIV RNA compared to late Tx (P=0.04) and unTx (P=0.001) donors (Fig. 2D). However, there was no difference in RNA density between Fiebig I/II Tx and Fiebig III-V Tx donors. Notably, there was a positive correlation between Gag p24 density measured by IF and gag-pol RNA measured by in situ hybridization in Fiebig I/II Tx donors (P=0.008; r=0.8, Fig. 2E). The results are consistent with the persistence of viral RNA despite very early ART initiation in hyperacute infection and durable plasma virus suppression.

Discordant HIV-1 RNA loads in plasma and lymph nodes
To better de ne active virus transcription within LN mononuclear (LNMCs) cells and to determine the viral loads in the LNs of aviremic individuals initiated on treatment either very early or later in infection. We measured cell-associated viral loads in LNMCs using a commercial viral load assay Cobas ® Ampliprep HIV-1 test. We found a hierarchy of LNMC viral loads with the values lowest in patients that initiated therapy in Fiebig stages I/II (Fig. 2F). Interestingly, neither the peak plasma viral load (Fig. 2G), treatment duration before LN excision (Fig. 2H), nor the time to suppression (Fig. 2I) impacted viral RNA persistence in the LN. Overall, quanti able amounts HIV RNA persists in the LNs of most Fiebig I/II Tx individuals and the magnitude of LN viral loads was not dependent on the duration of treatment.

Expansion of GCTfh cells in early ART-treated individuals
Identifying the cellular phenotypes of persistent HIV-1 protein and transcripts during therapy will be critical for future anti-HIV interventions. While follicular T helper (Tfh) cells are a key component of the adaptive immune response to HIV-1 infection and provide cognate help to B cells 30,31 and CD8 + T cells 32,33 , these cells also serve as a major HIV reservoir 13,34 . Moreover, HIV antigen can be trapped in the follicular dendritic lymphoreticular network within LNs and persist for years 35 , thus we interrogated persistence within these cell subsets.
Given that Tfh are major targets of HIV infection, we rst sought to determine the extent to which early ART mitigates HIV induced Tfh expansion. We de ned GCTfh as CD4 + CD45RA -CXCR5 hi PD-1 hi and nonGCTfh cells as CD4 + CD45RA -CXCR5 + PD-1 + in LNMCs (Fig. 3A) consistent with previous Tfh studies 36 . HIV negatives (n=9) had very low frequencies of GCTfh cells (median 1,3%, IQR; 0,6% to 1,5%) of antigen experienced (CD45RA -) CD4 + T cells whereas nonGCTfh cells were 11% (IQR; 8,5% to 13%, Fig.  3B). HIV infection resulted in signi cant expansion of GCTfh (Fig. 3C). Timing of treatment initiation impacted the extent of GCTfh expansion. Immediate therapy was associated with signi cant diminution of GCTfh expansion (Fiebig I/II Tx vs healthy controls P=0.07), whereas a slight delay in treatment initiation (Fiebig III or later) was associated with signi cant GCTfh expansion comparable to untreated HIV infection (Fiebig III-V Tx vs healthy controls P=0.03) (Fig. 3C). Notably, HIV induced Tfh expansion was restricted to GCTfh, as no signi cant expansion of nonGCTfh cells were observed between the groups (Fig. 3D). To verify these observations, we quanti ed the area densities of GCTfh in situ using FFPE LNs. Consistent with ow cytometry data GCTfh cell densities were signi cantly greater in delayed therapy and untreated infection compared to Fiebig I/II Tx and HIV negative controls (Fig. 3E). Together, these data show that early treatment initiated in Fiebig I/II mitigates HIV-induced GCTfh expansion. Reduced HIV targets in GCs might partly explain reduced HIV persistence in LN of individuals who initiate therapy very early.
To gain more insight on cellular targets of HIV infection in LNs, next, we investigated if there was a particular subcellular GCTfh that was selectively expanded. We quanti ed previously described 37,38 GCTfh subsets namely; GCTfh1 de ned as CXCR3 + CCR6 -, GCTfh2 as CXCR3 -CCR6 -, GCTfh1-17 as CXCR3 + CCR6 + and GCTfh17 de ned as CXCR3 -CCR6 + (Fig. 3F) among our study groups (Fig. 3G) and determined their relationship with HIV Gag p24 densities. While subsets had varying frequencies (Fig. 3G), within the early Tx donors, higher frequency of GCTfh1 was associated with greater degrees of Gag p24 positivity (Fig. 3H), whereas GCTfh17 displayed a reverse trend (Fig. 3I). Overall, these results show that while early treatment mitigates GCTfh responses, subset distribution of Tfh cells might impact virus persistence in early treated LNs.
Lastly, given the notion that HIV-speci c CD4 + T cells might be more susceptible to infection and contribute to viral persistence 39 , we used class II tetramers to characterize HIV-speci c Tfh responses in LN tissues. DRB1*11:01 and DRB1*13:01 class II tetramers previously described 38,40 were used to identify HIV-speci c CD4 + T cells (Fig. 3J). Summary data from the analysis of 3 Fiebig I/II Tx, 2 late Tx and 4 unTx donors within our cohort expressing the class II DRB1*11:01 and DRB1*13:01 alleles, revealed that tetramer-speci c GCTfh and nonGCTfh cells were detected in our chronic unTx and Fiebig I/II Tx participants at similar frequencies ( Since image analysis for Gag p24 indicated that most of the HIV antigen was con ned within discrete regions of GCs, and ow data showed differential correlation between levels of Gag p24 and GCTfh subsets, we next stained for other markers shown to be highly expressed on human Tfh 38 . We also used FDC markers to identify residual Gag p24 that has been reported to persist on FDCs 35 . Immuno uorescence imaging of serial sections stained with different combinations of antibodies and detected with Opal uorophores revealed that Gag p24 co-localized with several phenotypic markers (Figs. 4A and S3), including PD1 (Fig. 4A, panel i), CD4 (panel ii), CXCR3 (panel iii-iv), CCR6 (panel iv) and FDC (panel v). To more de nitively identify the Tfh subset that habored the most HIV infection burden, we quanti ed HIV RNA in LNMCs isolated from LN tissue of 3 Fiebig I/II Tx, 2 Fiebig III-V Tx and 3 late Tx donors and FACS-sorted into the 4 different Tfh subsets (Fig. 4B). HIV mRNA was detectable using digital droplet PCR in all the subsets (Fig. 4C). Importantly, when we analyzed the cells based on expression of chemokine receptors, CXCR3 and CCR6, we found that CXCR3 + Tfh subsets harbored signi cantly greater amounts of HIV RNA than other subsets (P=0.005, Fig. 4C).
To further interrogate preferential infection of CXCR3 + Tfh cells, we used a broadly neutralizing antibody (bNAb) called 3BNC117 to stain HIV infected cells while simultaneously staining for CXCR3. 3BNC117 targets the CD4 binding site on the surface of HIV-1 Envelope (Env) glycoprotein 41 . LNMC and paired PBMC samples obtained from 7 Fiebig I/II treated donors were analyzed without prior manipulation. Representative ow plots for one donor and aggregate data for 7 donors showed detection of HIV-1 Env (3BNC117) positive LNMCs at signi cantly greater frequency compared to paired PBMC samples (P=0.03, Fig. 4D). To con rm detection of low frequency HIV-1 positive cells ex vivo, we intracellularly stained aliquots of the same samples with anti-Gag p24 antibody. Similarly, Gag p24 + CD4 + T cells were readily detectable in LNMCs compared to PBMCs (P=0.01, Fig. 4E). We phenotyped infected cells by dual staining of 3BNC117 and CXCR3 and observed a trend towards more Env + CD4 + T cells co-expressing CXCR3 (P=0.06, Fig. 4F) than those not expressing CXCR3. Together, these data suggest that CD4 + CXCR3 + expressing Tfh cells may be preferentially infected in vivo compared to other subsets.
Impact of HIV-speci c CD4 + and CD8 + T cell responses on HIV persistence in the lymph node during ART We previously showed that immediate ART initiation augments HIV-speci c T cell function in peripheral blood 24 . To investigate the effects of early ART on LN responses, we begun by investigating if there were compartimental differences in the frequency of HIV-speci c responses between LN and PB. We used intracellular cytokine staining (ICS) to measure the proportions of HIV-speci c CD4 + and CD8 + T cells in LNs and paired blood samples using 15 fully suppressed Fiebig I/II Tx donors on uninterrupted therapy for greater than a year. Representative ow plots for one donor and aggregate data show signi cantly higher frequency of Gag-speci c CD8 + T cells (P=0.05) compared to PB responses (Fig. 5A). HIV-speci c CD4 + T cell frequencies also trended towards greater frequencies in LN relative to PB (P=0.06; Fig. 5B).
Next, we investigated whether HIV-speci c CD8 + T cell responses limit HIV persistence in the LN, and found a negative correlation between the frequency of HIV-speci c CD8 + T cell responses and HIV Gag p24 density (r=-0.7, P= 0.02; Fig. 5C). No correlation was observed between peripheral responses and the amount of persistant Gagp24 antigen in the LN, indicating the peripheral responses do not accurately depict HIV persistence in LNs. Notably, there was no correlation between LN or peripheral CD4 + T cell responses and perisistant HIV Gag p24 in the LN (Fig. 5D).
Considering that proliferative CD8 + T cell responses are often associated with protection 42,43 , we next measured virus-speci c responses by carboxy uorescein succinimidyl ester (CFSE) dilution. Representative ow plots for a donor with low and high Gagp24 density are shown (Fig. 5E). Agregate data show proliferative Gag-speci c CD8 + T cell responses negatively correlated with HIV Gag p24 burden (r=-0.7, P=0.04; Fig. 5F). A similar results was obtained for Gag-speci c CD4 + T cell responses (r= -0.7, P=0.04: Fig. 5G). Together, these data show an association between maintenance of functional cellular responses and reduced HIV viral antigens in LNs.
Given most of the residual virus was concentrated within GCs, we next assessed the capacity of HIVspeci c T cell responses to tra c into the GCs by enumerating the frequencies of CXCR5 + HIV-speci c responses in LN, which denote capacity to migrate into GCs. CXCR5 + HIV-speci c CD8 + T cells were signi cantly lower compared to CXCR5 + CD8 -T cells (P=0.001), suggesting reduced capacity to migrate into GCs, which may partly explain the observed greater HIV antigen burden in GCs relative to extrafollicular areas (Fig 5H). Moreover, SEB stimulation showed signi cantly lower proportion of IFN-Y + CXCR5 + CD8 + T cells relative to CXCR5 -CD8 + T cells (Fig 5I), indicating an inherent de ciency of CXCR5 + CD8 + T cells to secrete cytokines. Additionally, there was no correlation between plasma CXCL-13, (part of the CXCR5-CXCL-19 axis crucial for recruitment of immune cells into GCs) and density of HIV antigens in LNs (Fig. S4). Together, these data show that reduced functional HIV-speci c CD8 + T cell responses within GCs might contribute to HIV persistence in this tissue microenvironment. Most donors exhibited persistent HIV antigens in LN despite prompt blunting of initial peak viremia and sustained plasma viral suppression for as long as 55 months, suggesting that early therapy initiation may not fully eradicate persistent virus in lymphoid tissue sites. Immediate therapy reduced GCTfh expansion which is typically associated with dysregulation of B cell responses due to excessive GC reactions in untreated HIV infection 36,44 . Moreover, mitigated GCTfh responses decreased the number of cellular targets of HIV infection. Importantly, the association between functional immune responses and reduced viral burden in LNs indicates that T cell responses contribute towards elimination of infected cells during therapy. Combined, these data highlight the need to prioritize elimination of active HIV persistence in LNs as a critical step to achieving a cure or prolonged HIV remission off therapy.

Discussion
Our unique ability to obtain excisional lymph node biopsies in the FRESH cohort allowed for characterization of sites of virus persistence within the LN architecture in persons in whom peak viremia is blunted. The topological analysis of persistent HIV antigens within intact LN tissues identi ed greater HIV protein antigen burden within B cell follicles. Notably, onset and duration on therapy did not signi cantly affect the amount of detectable Gag p24 protein, consistent with the notion of rapid HIV reservoir establishment followed by very slow decay rate 45 . Importantly, our data highlight LN GCs as major sites of HIV persistence, with the potential to be a major source of rebound viremia upon treatment interruption.
Using a highly speci c in-situ hybridization assay, RNAscope 46,47 , which, in addition to HIV antigen detection provided further evidence in support of persistent HIV transcription in LNs in the face of ART in 86% of very early treated donors. We identi ed densely spherical signals by RNAscope staining in some early treated individuals suggestive of productively infected viral RNA + cells, consistent with a previous report in which active HIV RNA transcription was detected in the LNs of patients who initiated therapy in the chronic phase of illness 13 . Moreover, our data reveal heterogeneity in the amount of persistent HIV transcripts in very early treated aviremic individuals despite similar levels of peak viremia and rapid plasma viral suppression kinetics following ART initiation. Viral antigen persistence and ongoing transcription could indicate that HIV continues to cause immune damage in anatomical sites despite full suppression in peripheral blood, but it may also suggest that even in early treated individuals, priming, stimulation and harnessing of HIV-speci c immunity for curative strategies will not be insurmountable because functional HIV-speci c immunity is preserved.
Identifying cellular reservoirs of HIV in tissues has been a major area of research, (reviewed in 48 ). These studies describe GCTfh subset compositions anatomically and phenotypically during HIV infection and their contributions to persistent virus. The observed positive correlation between the proportions of GCTfh1 cells (which are CXCR3 + CCR6 -GCTfh) and detection of greater amounts of HIV RNA relative to CXCR3 + Tfh cells, indicates that CXCR3 may yet be another phenotypic marker of Tfh cells that have greater HIV transcription activity on ART. These data are consistent with a study that reported greater amounts of SIV DNA in CXCR3 + GCTfh compared to CXCR3 -GCTfh in macaques 49 . It is reasonable to attribute increased HIV burden in the CXCR3 + Tfh subset to CXCR3 being used as an alternative coreceptor for HIV entry, which has previously been reported 50 . However, con rmatory work is needed. In any case, we have identi ed a marker for HIV infected cells during ART that could be targeted for elimination as part of an HIV eradication strategy. Whether or not the CXCR3 + Tfh population identi ed in this study represents the same population as the PD-1 + subset that was recently implicated in HIV persistence is intriguing and warrants further investigation 13,51 .
Low CD8 + T cell density in GCs is thought to be a major reason for persistently high HIV antigen burden in this anatomical niche 52,53 . Interestingly, we detected signi cantly greater proliferative responses in individuals with little to no detectable HIV antigens compared to those with greater LN HIV antigen burden. These ndings are consistent with our recent study showing that very early ART is associated with functionally superior cellular responses 24 . Together, these data suggest that HIV-speci c T cell responses contribute to HIV suppression in LN during therapy 54 . However, longitudinal studies using serial biopsies from the same donor are needed to con rm these ndings.
A notable limitation of the study is that we could only obtain one LN sample per study participant, thus we could not conduct intra-individual longitudinal HIV decay kinetics, or longitudinally de ne the immune responses in tissues associated with control. This limitation was partly overcome by sampling LNs from many participants over a very wide time range. Serial LN studies have been reported in the absence of complications 55 . Alternatively, future studies could attempt serial ne needle aspirates, though cell yields may be limiting for some of the studies described here and the architectural structure would be disrupted. An additional limitation is the lack of gender diversity. Although we studied only women, this is the group at disproportionate risk of infection and women are underrepresented in most studies to date 56 . These issues can and should be addressed in future studies focused by our ndings.

Quantitative Image analysis
Quantitative image analysis of Gag p24 in IF images of whole tissue section scans was conducted with TissueQuest software (TissueGnostics). Two independent experiments of total area measurements and nuclear segmentation analyses were performed on each whole tissue scan. The numerical data generated from the analyses are displayed in scattergrams. Greyscale images were analysed and each channel was processed separately by the software using DAPI as a master marker. In cases where images were stained with another nuclear marker such as BCL-6, then the FITC channel was used as a virtual channel for nuclei identi cation. Negative control slides were used to set the threshold values in the scattergrams and to distinguish speci c staining signals from non-speci c or background uorescence signals. Although HIV Gag p24 staining was generally intense, there was no notable spillover of the signal to other channels (Fig. S1A-B). Also, p24 co-staining was only observed with FDCs and CD4 markers but not CD8 cells (Fig. S1C-D).
Analysis of gag-pol RNA signals was done using Fiji, an open-source software based on ImageJ which is optimized for biological image analysis 29 . Brie y, images were segmented using the color segmentation plugin with the algorithm for Hidden Markov Model. Thresholding was applied to the segmented image and the total area of brown RNA signals was measured and recorded. Five images were analysed per sample and averaged. Pixel measurements were converted to mm using the scale bar.

Digital droplet PCR
Total RNA was extracted from FACS-sorted LNMC Tfh subsets using Qiagen RNeasy kit (Qiagen) after lysing cells with QIAzol lysis reagent (Qiagen, Hilden, Germany) according to manufacturer's instructions, and used for cDNA synthesis using the iScript cDNA synthesis kit (Bio-Rad, Hercules, CA, USA). The cDNA was used as a template for HIV gag mRNA quanti cation by TaqMan digital droplet PCR assay using custom probes (Assay ID: APCE4R6, Thermo Fisher Scienti c) in a two-step digital droplet PCR reaction.
PCR thermal cycling was conducted following optimized cycling conditions: an initial denaturation at 95°C for 10 min, 40 cycles of 30 seconds at 94°C, 1 min at 60°C, followed by a nal incubation at 98°C for 10 min and holding at 4°C until reading time. After PCR ampli cation, droplets were measured in the QX200 ddPCR Droplet Reader (Bio-Rad), and target gene copy number was analyzed using QuantaSoft analysis software (Bio-Rad) and recorded as mRNA copies/20μL. Absolute gag mRNA counts were normalized to the expression of the housekeeping gene b2M.

Statistical analyses
All statistical analyses were conducted with GraphPad Prism 7.0 (GraphPad Software, La Jolla, California, USA) and P values were considered signi cant if less than 0.05. Speci cally, the Mann-Whitney U and Kruskal-Wallis H tests were used for group comparisons. Additional post hoc analyses were performed using the Dunn's multiple comparisons test. Correlations between variables were de ned by the Spearman's rank correlation test.

Declarations
Con ict of interest statement The authors have no con icting nancial interests.
Online supplemental material Fig. S1 shows that there is no co-localization of Gag p24 signals with non-infectious cell subsets like CD8 + T cells and validates the image quantitative analyses using a different approach of nuclear segmentation. Fig. S2 displays representative images of HIV RNA detected in lymph node sections by in situ hybridization. Fig. S3 demonstrates the co-localization of Gag p24 antigens with cells expressing PD1 + , CD4 + , CXCR3 + and CCR6 + surface markers. Fig. S4 shows that the low antigen persistence within lymph nodes is not associated with plasma CXCL-13 levels.   Single RNA transcripts are seen as punctate dots; clusters of transcripts are also observed. Red