Research ArticleHIV

Elite control of HIV is associated with distinct functional and transcriptional signatures in lymphoid tissue CD8+ T cells

See allHide authors and affiliations

Science Translational Medicine  18 Dec 2019:
Vol. 11, Issue 523, eaax4077
DOI: 10.1126/scitranslmed.aax4077

The secret lives of CD8+s

CD8+ T cells are often referred to as cytotoxic lymphocytes, but their functions extend beyond lysis of targets. Moreover, resident CD8+ T cells are not identical to their better studied circulating counterparts. To better understand functions of highly effective lymphoid CD8+ T cells, Nguyen et al. sampled lymph nodes from HIV elite controllers. Compared with cells from people with progressive disease, the elite controller cells had a distinct transcriptional profile and were able to suppress viral replication in the absence of cytolysis. CD8+ T cells from elite controllers translated proteins more efficiently, which could contribute to viral control. These results elucidate natural mechanisms of HIV control that could be informative for cure efforts or vaccine design.

Abstract

The functional properties of circulating CD8+ T cells have been associated with immune control of HIV. However, viral replication occurs predominantly in secondary lymphoid tissues, such as lymph nodes (LNs). We used an integrated single-cell approach to characterize effective HIV-specific CD8+ T cell responses in the LNs of elite controllers (ECs), defined as individuals who suppress viral replication in the absence of antiretroviral therapy (ART). Higher frequencies of total memory and follicle-homing HIV-specific CD8+ T cells were detected in the LNs of ECs compared with the LNs of chronic progressors (CPs) who were not receiving ART. Moreover, HIV-specific CD8+ T cells potently suppressed viral replication without demonstrable cytolytic activity in the LNs of ECs, which harbored substantially lower amounts of CD4+ T cell–associated HIV DNA and RNA compared with the LNs of CPs. Single-cell RNA sequencing analyses further revealed a distinct transcriptional signature among HIV-specific CD8+ T cells from the LNs of ECs, typified by the down-regulation of inhibitory receptors and cytolytic molecules and the up-regulation of multiple cytokines, predicted secreted factors, and components of the protein translation machinery. Collectively, these results provide a mechanistic framework to expedite the identification of novel antiviral factors, highlighting a potential role for the localized deployment of noncytolytic functions as a determinant of immune efficacy against HIV.

INTRODUCTION

HIV/AIDS is a persistent global health issue with no existing vaccine or cure. Most individuals infected with HIV experience high amounts of ongoing viral replication in the absence of ART, leading to a progressive loss of CD4+ T cells and the eventual onset of disease. However, a small subset of HIV-infected individuals (<1%), termed ECs, spontaneously control viral replication below the limit of detection and generally do not progress to AIDS. It is established that virus-specific CD8+ T cells are critical determinants of the EC phenotype in humans and rhesus macaques (1, 2). In addition, HIV-specific CD8+ T cells in ECs are qualitatively distinct from HIV-specific CD8+ T cells in CPs, typically displaying enhanced polyfunctionality (3, 4), cytolytic activity (57), and proliferative capacity (5, 8), as well as a more differentiated memory phenotype and a distinct immunodominance profile (4, 911). These attributes have been documented primarily among circulating lymphocytes whereas HIV replication occurs predominantly in lymphoid tissues (1215).

Lymphoid tissues are major reservoir sites for HIV. Recent studies have further demonstrated that almost 99% of viral RNA–positive (vRNA+) cells in simian immunodeficiency virus (SIV)–infected rhesus macaques occur in lymphoid tissues (16), reinforcing the need to understand anatomically colocalized mechanisms of immune control. It has long been known that circulating CD8+ T cells are more cytolytic than CD8+ T cells in the lymphoid tissues of donors infected with HIV (17). Moreover, a state of immunoprivilege exists in lymphoid tissues, which limits immunosurveillance by cytolytic HIV-specific CD4+ and CD8+ T cells (18, 19). In conjunction with the identification of distinct lymphoid tissue–resident memory CD8+ T cell subsets (2022), these observations suggest that HIV-specific CD8+ T cells limit viral replication in lymphoid tissues via effector mechanisms that differ from those used by circulating HIV-specific CD8+ T cells (22). It also seems reasonable to propose that noncytolytic suppression rather than cytolytic eradication dictates effective immune control of HIV, given reports of ongoing viral evolution (23, 24) and the presence of replication-competent viral strains in ECs (25). However, this proposition remains unproven to date, because previous studies have not defined the antiviral efficacy and functional characteristics of HIV-specific CD8+ T cells in the lymphoid tissues of ECs.

In this study, we used a variety of methodological approaches, including polychromatic flow cytometry and single-cell RNA sequencing (scRNAseq) analyses, to compare the functional and transcriptional properties of HIV-specific CD8+ T cells in the peripheral blood and LNs of ECs and CPs. Our findings demonstrate that immune control of viral replication is associated with the occurrence of polyfunctional HIV-specific memory CD8+ T cells that exhibit a weak cytolytic signature and preferentially home to B cell follicles in the LNs of ECs.

RESULTS

CD8+ T cells actively suppress HIV replication in the LNs of ECs

To define the nature of protective CD8+ T cell responses in LNs, where HIV replicates in vivo, we obtained tissue biopsies (cervical, iliac, inguinal, mesenteric, pelvic, or peribronchial LNs) and fine-needle aspirates (inguinal LNs) from HIV+ individuals on ART and untreated HIV+ individuals categorized as acute seroconverters, ECs, or CPs (table S1). Extremely low numbers of vRNA+ cells were detected in the LNs of ECs compared with the LNs of CPs (P = 0.0174; Fig. 1A). In line with previous findings (26), vRNA+ cells were visualized in the B cell follicles and in the paracortical region (T cell zone) (fig. S1). Total CD4+ T cell–associated HIV DNA and RNA measurements were also lower in the LNs of ECs compared with the LNs of CPs (P = 0.0215 and P = 0.0339, respectively; Fig. 1B), and CD8+ T cells from the LNs of ECs displayed greater efficacy in a modified viral suppression assay (27) compared with CD8+ T cells from the LNs of CPs (P = 0.0224) and CD8+ T cells from the LNs of HIV+ individuals on ART (P = 0.0323; Fig. 1C). These results suggest that CD8+ T cells maintain active control of HIV replication in the LNs of ECs.

Fig. 1 CD8+ T cells from the LNs of ECs display superior antiviral efficacy.

(A) Representative RNAscope images of paraffin-embedded LN biopsies from one EC and one CP (left) and data quantification (right). TCZ, T cell zone; BCF, B cell follicle. EC, n = 7; ART, n = 10; CP, n = 11. (B) Quantification of cell-associated HIV DNA and RNA from negatively selected LN CD4+ T cells by qPCR. One data point on the DNA plot was adjusted from 0 to 1 to display on the log scale. Error bars represent median and interquartile ranges. EC, n = 5; CP, n = 8. (C) Representative flow plots from a viral suppression assay using LN-derived CD8+ T cells from an EC cocultured with autologous LN-derived CD4+ T cells infected with HIV BAL virus in vitro (left) and data quantification (right). Dotted lines represent median and interquartile ranges. EC, n = 4; ART, n = 8; CP, n = 6. Flow cytometry data were pregated on single live CD8 T cells. Significance was determined using the Kruskal-Wallis test with Dunn’s correction (A), an unpaired t test with Welch’s correction (B), or Welch’s ANOVA with Benjamini-Hochberg correction (C). *P < 0.05.

CD8+ T cells exhibit weak cytolytic activity and increased follicle-homing potential in the LNs of ECs

In peripheral blood, the frequency of cytolytic HIV-specific CD8+ T cells, defined by expression of perforin and granzyme B, correlates inversely with plasma viral load (pVL) (57). We therefore hypothesized that enhanced CD8+ T cell–mediated cytolytic activity may underlie viral control in the LNs of ECs. However, lower frequencies of perforin+, granzyme B+, and perforin+ granzyme B+ memory CD8+ T cells were detected in the LNs of ECs compared with the LNs of CPs (P = 0.0001, P = 0.0004, and P = 0.0002, respectively; Fig. 2A and fig. S2, A and B). The frequency of perforin+ granzyme B+ memory CD8+ T cells also correlated positively with pVL (r = 0.8213, P < 0.0001; Fig. 2B). Perforin and granzyme B were broadly expressed among circulating memory CD8+ T cells in donor-matched samples from ECs, indicating that the noncytolytic phenotype was restricted to LNs (fig. S2C). Immunohistochemical analyses confirmed that fewer perforin+ CD8+ T cells (P = 0.006), but not granzyme B+ CD8+ T cells, were present in the LNs of ECs compared with the LNs of CPs (Fig. 2, C and D).

Fig. 2 Cytolytic CD8+ T cells are rare in the LNs of ECs.

(A) Representative flow plots showing perforin and granzyme B expression in LN memory CD8+ T cells (left) and data quantification (right). EC, n = 13; ART, n = 11; CP, n = 13; acutely infected individuals (acute), n = 7. (B) Correlation between the frequency of perforin+ granzyme B+ LN memory CD8+ T cells and pVL. Values for pVL below the limit of detection were plotted as 40 copies/ml. (C and D) Representative immunohistochemistry images of LN biopsies (left) stained for perforin (C) or granzyme B (D) and data quantification across donor groups (right). EC, n = 5; ART, n = 9; CP, n = 14. (E) Representative flow plots showing HIV-specific tetramer staining of LN memory CD8+ T cells (left) and data quantification (right). Each data point on the quantification plot represents a distinct tetramer+ population. (F) Representative flow plots showing HIV-specific tetramer+ cells (colored contours) overlaid on total LN memory CD8+ T cells (gray background) gated to display perforin and granzyme B expression (left) and data quantification (right). Each data point on the quantification plot represents a distinct tetramer+ population. EC, n = 25; ART, n = 6; CP, n = 8. (G) Representative flow plots of redirected killing assays (left) and data quantification (right). Data were pregated on single live TFL4+ cells. (H) Representative flow plots showing CXCR5 expression on LN-derived memory CD8+ T cells (left) and data quantification across donor groups (right). EC, n = 12; ART, n = 9; CP, n = 8. (I) Quantification of CXCR5 expression on LN tetramer+ memory CD8+ T cells. Each data point on the quantification plot represents a distinct tetramer+ population. EC, n = 25; ART, n = 6; CP, n = 8. (J) Representative flow plots showing perforin and granzyme B expression in LN-derived CXCR5+ and CXCR5 memory CD8+ T cells (left) and data quantification (right). Error bars represent median and interquartile ranges. EC, n = 12; ART, n = 9; CP, n = 8. Flow cytometry data, except in (G), were pregated using scheme shown in fig. S2A. Significance was determined using the Kruskal-Wallis test with Dunn’s correction (A, C, D, E, F, H, I, and J) or the Spearman’s rank correlation (B) or a two-way ANOVA (G). *P < 0.05, **P < 0.01, and ***P < 0.001.

We then used human leukocyte antigen (HLA) class I–matched tetramers to examine HIV-specific CD8+ T cells. Higher frequencies of tetramer+ CD8+ T cells were present in the LNs of ECs compared with the LNs of CPs (P = 0.04; Fig. 2E). However, a very low frequency of HIV-specific CD8+ T cells expressed perforin and granzyme B, especially in the LNs of ECs (Fig. 2F). In response to cognate peptide stimulation, HIV-specific CD8+ T cells from the LNs of ECs also failed to up-regulate either perforin or granzyme B, unlike CD8+ T cells paired by donor and specificity from the peripheral blood of ECs (fig. S3A). Moreover, CD8+ T cells from the LNs of ECs did not exhibit discernable expression of perforin or granzyme B after 3 days in culture with HIV-infected CD4+ T cells from the same LNs (fig. S3B).

To quantify rather than infer cytolytic activity, we performed ex vivo redirected killing assays based on the detection of active caspase-3, which reliably identified nonviable target cells (fig. S4, A and B). In contrast to circulating CD8+ T cells, donor-matched CD8+ T cells from the LNs of ECs largely failed to kill P815 mastocytoma target cells precoated with a CD3-specific monoclonal antibody, which mimics signals delivered via the T cell receptor (Fig. 2G). A similar anatomical discrepancy was observed using paired samples from CPs (Fig. 2G). Moreover, no substantial increases in target cell death were observed after extended incubation (24 hours), ruling out the involvement of temporally delayed killing mechanisms (fig. S4C). Collectively, these data suggest that cytolytic activity is unlikely the major mechanism of protection by CD8+ T cells in the LNs of ECs.

B cell follicles are thought to represent immunoprivileged sites that potentiate HIV persistence as a consequence of limited immunosurveillance by CD8+ T cells (2, 2831). Previous studies have also demonstrated an inverse correlation between the frequency of follicle-homing (CXCR5+) CD8+ T cells and pVL (18, 32, 33). In line with these findings and the relative absence of vRNA+ cells in the B cell follicles of ECs (fig. S1, C and D), we detected higher frequencies of CXCR5+ memory CD8+ T cells in the LNs of ECs compared with the LNs of CPs (P = 0.0288; Fig. 2H). No increased frequencies were observed for HIV-specific CXCR5+ CD8+ T cells from ECs compared to ARTs (P = 0.07) or CPs (P = 0.124; Fig. 2I). However, perforin and granzyme B were typically expressed at higher frequencies among CXCR5 CD8+ T cells compared with CXCR5+ CD8+ T cells, reaching significance in the ART cohort (P = 0.0026; Fig. 2J).

HIV-specific CD8+ T cells have a distinct transcriptional profile in the LNs of ECs

To identify potential correlates of viral control in lymphoid tissues, we analyzed the transcriptomes of HIV-specific CD8+ T cells from the LNs of ECs and CPs (Fig. 3A). The scRNAseq data were initially processed using an unsupervised kernel-based algorithm to determine cellular similarity in a global manner (see Materials and Methods). Clusters of transcriptionally distinct cells were present in each donor group (Fig. 3B). Normalized expression of PRF1 and GZMB was lower in HIV-specific CD8+ T cells from the LNs of ECs compared with HIV-specific CD8+ T cells from the LNs of CPs (Fig. 3C), and normalized expression of transcripts encoding other cytolytic molecules, including GZMA, GZMH, GZMK, GZMM, FASL, and TNFSF10 (34, 35), was either comparable between groups or enriched in HIV-specific CD8+ T cells from the LNs of CPs compared with HIV-specific CD8+ T cells from the LNs of ECs (Fig. 3C).

Fig. 3 Distinct transcriptomic signatures characterize HIV-specific CD8+ T cells from the LNs of ECs and CPs.

(A) Setup of the scRNAseq experiment, including data exclusion criteria. ERCC, External RNA Controls Consortium. (B) tSNE visualization of the unsupervised kernel-based algorithm. (C) Violin plots of cytolysis-related genes showing z-normalized expression. (D) Average percent error of k-fold cross-validation of support vector machine (SVM) models using subsets of the L0 norm–ranked gene list. Each cross-validation was reiterated 100 times. (E) Heatmap showing z-normalized expression of the 200 feature genes. (F) GO analysis of the feature gene list using topGO. The top 25 enriched terms were reported, and P values were calculated using the classic Fisher method. (G) tSNE feature plots showing the distribution of gene expression. tSNE coordinates were calculated using the first 50 principal components and iterated 1000 times. The color gradient displays relative log-normalized gene expression. Significance was determined using scDD calculations (C and G). **P < 0.01 and ***P < 0.001. The color of asterisks indicates the group with higher average expression of the indicated gene.

We then applied a supervised machine-learning algorithm to determine which transcripts best characterized HIV-specific CD8+ T cells from the LNs of ECs and HIV-specific CD8+ T cells from the LNs of CPs (see Materials and Methods). A set of 200 genes reliably distinguished each donor group (average error, <0.5%) (Fig. 3D). Hierarchical clustering of these “feature” genes clarified the key intergroup differences (Fig. 3E), most of which could be assigned to immune response pathways via Gene Ontology (GO) analysis (Fig. 3F), and t-distributed stochastic neighbor embedding (tSNE) visualization generated two distinct clusters of HIV-specific CD8+ T cells that differentiated ECs from CPs (Fig. 3G). Overlaying transcript abundance of specific feature genes onto the tSNE plot further revealed that HIV-specific CD8+ T cells from the LNs of CPs preferentially expressed inhibitory receptors, such as TIGIT, LAG3, and CD244, as well as the terminal differentiation marker KLRG1 and the transcription factor EOMES, collectively indicating an exhausted phenotype (3638), whereas HIV-specific CD8+ T cells from the LNs of ECs preferentially expressed IL7R, which is crucial for homeostasis (39), and several chemokines/cytokines, such as CCL5 and IL32, collectively indicating a highly functional memory phenotype (Fig. 3G).

To confirm and extend these findings, we reanalyzed our scRNAseq data using the reproducibility-optimized test statistic (ROTS) (40), a Bayesian model of differential distribution (scDD) (41), and Seurat (42). A total of 2264 differentially expressed genes achieved significance in at least two of these independent analytical frameworks (fig. S5A and table S2). These genes were enriched for immune-related terms, such as “immune response,” “immune system process,” and “defense response,” in GO analyses (fig. S5B). Of particular note, Ingenuity Pathway Analysis further identified a core cassette of 11 transcripts encoding predicted secreted factors that were selectively up-regulated in HIV-specific CD8+ T cells from the LNs of ECs, including TNF, CCL5, RNASE1, and IL32, which have been shown to suppress HIV replication (fig. S5C) (4349).

HIV-specific CD8+ T cells are polyfunctional and translate proteins efficiently in the LNs of ECs

To examine the functionality of HIV-specific CD8+ T cells, we stimulated LN mononuclear cells (LNMCs) with HIV peptides and measured the up-regulation of various cytokines by flow cytometry. In response to cognate peptide stimulation, higher frequencies of HIV-specific CD8+ T cells from the LNs of ECs produced tumor necrosis factor (TNF) compared with HIV-specific CD8+ T cells from the LNs of CPs (P = 0.0003) and HIV-specific CD8+ T cells from the LNs of HIV+ individuals on ART (P = 0.039) (Fig. 4, A and B). A similar difference was observed for interferon-γ (IFNγ) (P = 0.045), but the differences for macrophage inflammatory protein–1β (MIP-1β) and interleukin-2 (IL-2) did not reach statistical significance (Fig. 4B). Moreover, HIV-specific CD8+ T cells from the LNs of ECs were frequently polyfunctional, whereas HIV-specific CD8+ T cells from the LNs of CPs were predominantly monofunctional (Fig. 4, C and D).

Fig. 4 Polyfunctionality is associated with protein translation efficiency in HIV-specific CD8+ T cells from the LNs of ECs.

(A) Representative flow plots showing the production of TNF and IL-2 by HIV-specific CD8+ T cells from the LNs of one EC and one CP in response to cognate peptide stimulation. Plots were pregated on total memory CD8+ T cells using the scheme shown in fig. S2A. (B) Quantification of cytokine production by HIV-specific CD8+ T cells from LNs. The total response for each readout was determined by summing the corresponding frequencies after background subtraction of all Boolean gates using permutations of CD107a, IFNγ, TNF, MIP-1β, and IL-2. Error bars represent mean and SEM. (C) Polyfunctionality plot of HIV-specific CD8+ T cells from LNs. Only combinations of four and five functions are shown. Error bars represent mean and SEM. EC, n = 49; ART, n = 10; CP, n = 27. (D) Pie charts summarizing all combinations of functions. (E) Heat map showing z-normalized expression of genes from the feature list identified by GO protein translation terms. (F) GSEA of the scRNAseq data with reference to the Reactome Pathway Database, reporting the top 10 enriched pathways in ECs and CPs. UTR, untranslated region; NES, normalized enrichment score. (G) Quantification of protein translation efficiency in LN-derived CD8+ T cells. LNMCs were incubated with HPG in methionine-free medium for 6 hours in the presence or absence of HIV-derived peptides or anti-CD3. Responding cells were defined using Boolean gating for IFNγ+ and/or TNF+ events. CM, central memory; EM, effector memory. EC, n = 3; ART, n = 3; CP, n = 4. Error bars represent mean and SEM. Significance was determined using a one-way ANOVA (D and G) or a two-way ANOVA (B). *P < 0.05, **P < 0.01, and ***P < 0.001. MFI, mean fluorescence intensity. SEM, standard errors of the mean.

To seek an explanation for these findings, we returned to our GO analysis of the 200 feature genes. In addition to immune-related terms, several protein translation–related terms were also highlighted, including “signal recognition particle (SRP)–dependent cotranslational protein targeting to membrane,” “protein targeting to endoplasmic reticulum,” and “translation initiation” (Fig. 3F). Most of these genes encoded ribosomal protein subunits that were preferentially expressed in HIV-specific CD8+ T cells from the LNs of ECs (Fig. 4E). Gene set enrichment analysis (GSEA) similarly identified “translation,” “SRP-dependent cotranslational protein targeting to membrane,” and “3′ untranslated region–mediated translation regulation” as the most highly enriched gene sets in the EC group (Fig. 4F).

We then performed in vitro translation experiments that measure the uptake of the fluorescently labeled amino acid analog l-homopropargylglycine (HPG) in resting and activated CD8+ T cells from the LNs of ECs, CPs, and HIV+ individuals on ART. Higher frequencies of resting memory CD8+ T cells from the LNs of ECs incorporated HPG into newly synthesized proteins compared with resting memory CD8+ T cells from the LNs of CPs (P = 0.003) (Fig. 4G). These differences emanated primarily from the central memory compartment (Fig. 4G). After stimulation with anti-CD3 or HIV-derived peptides, functionally responsive CD8+ T cells from the LNs of ECs also incorporated more HPG than functionally responsive CD8+ T cells from the LNs of CPs (P = 0.008 and P = 0.0004, respectively) and functionally responsive CD8+ T cells from the LNs of HIV+ individuals on ART (P = 0.04 and P = 0.004, respectively) (Fig. 4G). These results suggest that efficient protein translation can explain, at least in part, the antiviral efficacy and polyfunctional nature of HIV-specific CD8+ T cells in ECs.

DISCUSSION

In this study, we found that weakly cytolytic CD8+ T cells homed to B cell follicles, which constitute a major viral reservoir in vivo (2, 1215), and potently suppressed HIV replication in the lymphoid tissues of ECs. Moreover, these LN-derived CD8+ T cells down-regulated inhibitory receptors, up-regulated multiple soluble factors and cytokines, and displayed efficient protein translation. Although cause-and-effect arguments can always be made in the context of human studies, these features likely identify a protective immune signature because CD8+ T cells from the LNs of HIV+ individuals on ART failed to acquire the characteristics of CD8+ T cells from the LNs of ECs even in the absence of viremia.

Ultrasensitive assays have revealed ongoing viral replication and evolution in the plasma of ECs (2325). In line with the notion of anatomically localized immune control, however, we detected very few vRNA+ cells and very little HIV DNA and RNA in the LNs of ECs. We also found that CD8+ T cells from the LNs of ECs suppressed HIV replication in autologous CD4+ T cells far more efficiently than CD8+ T cells from the LNs of CPs and HIV+ individuals on ART. Depletion studies in nonhuman primates have shown that CD8+ T cells are required to maintain effective control of SIV (2, 50). Moreover, we and others have demonstrated that circulating CD8+ T cells exhibit superior cytolytic activity in ECs (6, 7). It was therefore unexpected to find that very few CD8+ T cells from the LNs of ECs expressed cytolytic molecules compared with CD8+ T cells from the LNs of CPs. This particular finding suggests that limited immunosurveillance by cytolytic T cells is a general phenomenon in HIV-infected lymphoid tissues (18). Although we cannot exclude a role for cytolytic mechanisms, especially during the early stages of infection, our data show that a potent cytolytic response is not a prerequisite for durable immune-mediated suppression of HIV. These data are consistent with the observation that terminally differentiated effector memory T cells, which are highly cytolytic, primarily circulate in the blood, and patrol highly vascularized organs rather than other tissue sites (51).

The general exclusion of cytolytic CD4+ and CD8+ T cells from lymphoid tissues (18, 19) is likely a conserved feature of the immune system that has evolved to limit the destruction of antigen-presenting cells and stromal tissues under conditions of persistent stimulation. Noncytolytic effector functions may therefore be necessary for effective immune control of viruses and other pathogens that specifically target these sites. An equivalent scenario has been described in the liver, where hepatitis B virus–specific CD8+ T cells eliminate the infection without killing hepatocytes via the secretion of IFNγ and TNF (52). Similarly, noncytolytic mechanisms of immune control have been described in the context of infection with herpes simplex virus, measles virus, and vaccinia virus (53, 54), as well as HIV and SIV infection (5557). Our results add to this body of literature and further suggest that spatial distribution and the efficient translation of antiviral proteins are key features of a protective HIV-specific CD8+ T cell response in lymphoid tissues.

A limitation of our study was the small sample size, which was most apparent in the scRNAseq experiment. Ideally, we would validate signatures identified in this dataset with an independent cohort. However, such cohorts are not easily accessible, given the extremely low frequency of ECs (<1% of all infected individuals) and the increasingly early initiation of ART. Several of the key gene signatures were nonetheless confirmed in terms of functionality and protein expression, indicating that these signatures are generalizable. The interpretation of the scRNAseq data was potentially confounded by antigen load because the comparison was limited to only between ECs and CPs. Technical difficulties and the low frequency of HIV-specific CD8+ T cells in LNs of ART-treated individuals did not permit the inclusion of this group in our scRNAseq experiment. To alleviate this caveat, we included the ART-treated group in most of our functional experiments. Because CD8+ T cells from LNs of ECs were more polyfunctional, suppressed viral replication more potently, and translated proteins more efficiently than CD8+ T cells from LNs of both CPs and ART-treated individuals, these major signatures could not be explained by antigen load alone.

In summary, we have shown that noncytolytic CD8+ T cells with distinct functional, phenotypic, and transcriptional signatures, including the up-regulation of known or predicted soluble factors, efficient protein translation, and a high ribosome biogenesis, suppress viral replication in the LNs of ECs. These observations are highly pertinent in light of current “kill” strategies designed to achieve a functional cure for HIV (58). Collectively, our data support a model in which viral suppression rather than viral eradication dictates immune efficacy, without excluding a role for cytolytic mechanisms, and provides fresh impetus in the search for novel antiviral factors that may complement existing therapies for HIV (59).

MATERIALS AND METHODS

Study design

Peripheral blood samples, tissue biopsies (cervical, iliac, mesenteric, pelvic, peribronchial, or inguinal LNs), and fine-needle aspirates (inguinal LNs) were obtained from HIV+ individuals on ART (n = 14) and untreated HIV+ individuals categorized as acute seroconverters (Fiebig stages IV to VI; n = 7), ECs (n = 12), or CPs (n = 18). Donors were recruited at three different sites: Instituto Nacional de Enfermedades Respiratorias (INER)–Centro de Investigación en Enfermedades Infecciosas (CIENI), Mexico City, Mexico; the University of California, San Francisco, CA, USA; and the University of Pennsylvania, Philadelphia, PA, USA. Clinical characteristics are summarized in table S1. Paraffin-embedded LN slides from CPs (n = 7) for microscopy experiments were obtained from the Pathology Core at the Hospital of the University of Pennsylvania (Philadelphia). All donors provided informed consent in line with the protocols approved by the INER-CIENI Ethics Committee, the Federal Commission for the Protection against Sanitary Risk (COFEPRIS), and the Institutional Review Boards of the University of California (San Francisco) and the University of Pennsylvania (Philadelphia). Sample sizes were not predetermined by power calculations but instead were dependent on sample availability. Investigators were not blinded to group identity during the course of the study. Primary data are reported in data file S1.

Sample processing and storage

Peripheral blood mononuclear cells (PBMCs) were isolated using standard density gradient centrifugation, and LNMCs were collected from whole biopsies via mechanical homogenization. Fine-needle aspirates were processed for immediate experimentation. PBMCs and LNMCs were cryopreserved at −140°C.

RNAscope

Next-generation in situ hybridization was performed on LN biopsies as described previously (60).

qPCR quantification of cellular HIV-1 DNA and RNA

Cell-associated DNA and RNA were purified using the AllPrep DNA/RNA Mini Kit (Qiagen), concentrated in a SpeedVac System (Thermo Fisher Scientific), and normalized to cell equivalents via quantitative polymerase chain reaction (qPCR) with reference to genomic telomerase reverse transcriptase for DNA and expressed ribosomal protein lateral stalk subunit P0 for RNA (Thermo Fisher Scientific). Total cell-associated HIV DNA and RNA were quantified via qPCR using the QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems) with the long terminal repeat–specific primers F522–43 (5′-GCCTCAATAAAGCTTGCCTTGA-3′; HXB2 522–543) and R626–43 (5′-GGGCGCCACTGCTAGAGA-3′; HXB2 626–643), and a 5’-Fluorescein-CE Phosphoramidite-Black Quencher (FAM-BQ) probe (5′-CCAGAGTCACACAACAGACGGGCACA-3′). Cell-associated HIV DNA copy number was determined in a final reaction volume of 20 μl incorporating 4 pmol of each primer, 4 pmol of probe, 5 μl of DNA, and 10 μl of 2× TaqMan Universal Master Mix II, with uracil N-glycosylase (Thermo Fisher Scientific). Thermal parameters were as follows: 50°C for 2 min (1 cycle), 95°C for 10 min (1 cycle), and 95°C for 15 s, followed by 59°C for 1 min (60 cycles). Cell-associated HIV RNA copy number was determined in a final reaction volume of 20 μl incorporating 4 pmol of each primer, 4 pmol of probe, 5 μl of RNA, 0.5 μl of reverse transcriptase, and 10 μl of 2× TaqMan RNA-to-CT 1-Step (Thermo Fisher Scientific). Thermal parameters were as follows: 48°C for 20 min (1 cycle), 95°C for 10 min (1 cycle), and 95°C for 15 s, followed by 59°C for 1 min (60 cycles). For HIV DNA measurements, external quantification standards were prepared from the ACH-2 cell line in a background of HIV human cellular DNA, calibrated against the Division of AIDS Virology Quality Assurance Program HIV-1 DNA Quantification Standard (National Institutes of Health AIDS Reagent Program). For HIV RNA measurements, external quantification standards were prepared from full-length NL4-3 virion RNA, and copy number was determined using a RealTime HIV-1 Viral Load Assay (Abbott Molecular), calibrated against the Division of AIDS Virology Quality Assurance Program HIV-1 RNA Quantification Standard (National Institutes of Health AIDS Reagent Program). Copy number for all samples was determined in triplicate by extrapolation against a 7-point standard curve (1 to 10,000 cps).

Peptide stimulation assay

Cryopreserved LNMCs and PBMCs were thawed and rested overnight at 37°C in RPMI supplemented with 10% fetal bovine serum, 1% l-glutamine, and 1% penicillin/streptomycin (R10). Cells were then washed in R10 and resuspended at 2 × 106 cells/ml. About 0.5 × 106 to 1 × 106 cells were used per condition, with lower bounds defined by cell availability in each sample. All stimulation conditions included anti-CD28 and anti-CD49d (each at 1 μg/ml; BD Biosciences), GolgiStop (0.7 μl/ml; BD Biosciences), and brefeldin A (1 μg/ml; Sigma-Aldrich). Cells were stimulated for 6 hours at 37°C with peptides matching known HLA-specific optimal HIV epitopes (each at a final concentration of 1 μg/ml; New England Biolabs). Positive controls incorporated staphylococcal enterotoxin B (1 μg/ml; Sigma-Aldrich). Degranulation was detected via the addition of anti-CD107a–phycoerythrin (PE)–Cy5 (eBioH4A3, eBioscience) at the start of the assay to capture surface-mobilized events in real time (61).

Flow cytometry

LNMCs and PBMCs were stained with anti-CCR7–allophycocyanin (APC)–Cy7 (G043H7, BioLegend) for 10 min at 37°C. Cells were then labeled with LIVE/DEAD Fixable Aqua (Thermo Fisher Scientific) for 10 min at room temperature to identify nonviable events, stained with a cocktail of directly conjugated monoclonal antibodies for 20 min at room temperature to detect surface markers, fixed/permeabilized using the Cytofix/Cytoperm kit (BD Biosciences), stained with another cocktail of directly conjugated monoclonal antibodies for 1 hour at room temperature to detect intracellular markers, and fixed in 1% paraformaldehyde (Sigma-Aldrich). Data were acquired using an LSR II flow cytometer (BD Biosciences) and analyzed with FlowJo software v9.9.4 (Tree Star Inc.).

Antibodies

The following directly conjugated reagents were used in flow cytometry experiments: tetramer panel: anti-CCR7–APC-Cy7 (G043H7, BioLegend), anti–CD14-BV510 (M5E2, BioLegend), anti–CD19-BV510 (HIB19, BioLegend), anti–CD3-BV711 (UCHT1, BioLegend), anti-CD4–PE-Cy5.5 (S3.5, Thermo Fisher Scientific), anti–CD8-BV570 (RPA-T8, BioLegend), anti–CD27-BV785 (O323, BioLegend), anti–CD45RO-BV650 (UCHL1, BioLegend), anti-CD69–PE-Cy5 (TP1.55.3, Beckman Coulter), anti–CD103-BV605 (2E7, BioLegend), anti–CXCR5-AF488 (RF8B2, BD Biosciences), anti-perforin–PE-Cy7 (dG9, eBioscience), anti–granzyme B–AF700 (GB11, BD Biosciences), anti-Tbet–PE-Dazzle (4B10, BioLegend), and anti-Eomes–AF647 (WD1928, eBioscience); peptide stimulation panel: anti-CD107a–PE-Cy5 (eBioH4A3, eBioscience), anti-CCR7–APC-Cy7 (G043H7, BioLegend), anti–CD14-BV510 (M5E2, BioLegend), anti–CD19-BV510 (HIB19, BioLegend), anti–CD3-BV711 (UCHT1, BioLegend), anti-CD4–PE-Cy5.5 (S3.5, Thermo Fisher Scientific), anti–CD8-BV570 (RPA-T8, BioLegend), anti–CD27-BV785 (O323, BioLegend), anti–CD45RO-BV650 (UCHL1, BioLegend), anti–CXCR5-AF647 (RF8B2, BD Biosciences), anti-IFNγ–fluorescein isothiocyanate (FITC) (B27, BD Biosciences), anti–TNF-BV605 (MAb11, BioLegend), anti–IL-2–APC-R700 (MQ1-17H12, BD Biosciences), anti–MIP-1β–PE-Cy7 (D21-1351, BD Biosciences), anti–perforin-BV421 (B-D48, BioLegend), anti–granzyme B–PE–Texas Red (GB11, Thermo Fisher Scientific), and anti–Tbet-PE (4B10, eBioscience); redirected killing assay panel: anti–active caspase-3–FITC (C92-605, BD Biosciences), anti–CD3-PE (SK7, BD Biosciences), and anti-perforin–PE-Cy7 (dG9, eBioscience); suppression assay panel: anti–p24-FITC (KC57, Beckman Coulter), anti–CD14-BV510 (M5E2, BioLegend), anti–CD19-BV510 (HIB19, BioLegend), anti–CD3-BV711 (UCHT1, BioLegend), anti-CD4–PE-Cy7 (RPA-T4, BioLegend), anti–CD8-BV570 (RPA-T8, BioLegend), anti–CD25-Tricolor (CD25-3G10, Thermo Fisher Scientific), anti–perforin-BV421 (B-D48, BioLegend), and anti–granzyme B–PE–Texas Red (GB11, Thermo Fisher Scientific); translation assay panel: anti-CCR7–APC-Cy7 (G043H7, BioLegend), anti–CD14-BV510 (M5E2, BioLegend), anti–CD19-BV510 (HIB19, BioLegend), anti–CD3-BV711 (UCHT1, BioLegend), anti–CD4-APC (S3.5, Thermo Fisher Scientific), anti–CD8-BV570 (RPA-T8, BioLegend), anti–CD45RO-BV650 (UCHL1, BioLegend), anti–azide-AF488 (A10266, Thermo Fisher Scientific), anti-TNF–PE-Cy7 (MAb11, Thermo Fisher Scientific), and anti–IFNγ-AF700 (B27, BD Biosciences); and viral quantification panel: anti-CCR7–APC-Cy7 (G043H7, BioLegend), anti-CD3–APC-R700 (UCHT1, BD Biosciences), anti-CD4–PE-Cy7 (RPA-T4, BioLegend), anti-CD8–PE-Cy5.5 (RPA-T8, eBioscience), anti–CD14-BV510 (M5E2, BioLegend), anti–CD19-BV510 (HIB19, BioLegend), anti-CD45RA–PE-CF594 (HI100, BD Biosciences), anti–CXCR5-AF647 (RF8B2, BD Biosciences), and anti–PD-1–BV421 (EH12.2H7, BioLegend).

Tetramers

HLA class I tetramers conjugated to BV421 or PE were produced as described previously (62). The following specificities were used to detect HIV-specific CD8+ T cells: A*0201-IV9 (ILKEPVHGV), A*0201-SL9 (SLYNTVATL), A*0201-FK10 (FLGKIWPSHK), A*0201-TV9 (TLNAWVKVV), A*2402-RW8 (RYPLTFGW), A*2402-KW9 (KYKLKHIVW), A*2402-RL9 (RPMTYKGAL), B*0702-GL9 (GPGHKARVL), B*0702-HI10 (HPRVSSEVHI), B*0702-SM9 (SPAIFQSSF), B*2705-KK10 (KRWIILGLNK), B*3501-VY10 (VPLDEDFRKY), B*3501-NY9 (NSSKVSQNY), B*5701-KF11 (KAFSPEVIPMF), B*5701-TW10 (TSTLQEQIGW), B*5701-ISW9 (ISPRTLNAW), and B*5701-QW9 (QASQEVKNW).

Immunohistochemistry

LN biopsy material was cut at a thickness of 5 μm and processed as described previously (63). Briefly, tissue sections were heated in 0.01% citraconic anhydride containing 0.05% Tween 20 and incubated overnight at 4°C with monoclonal or polyclonal antibodies specific for perforin (1:100; 5B10/VP-P967, Vector Laboratories Inc.) or granzyme B (1:200; HPA003418, Sigma-Aldrich). Slides were then washed in tris-buffered saline containing 0.05% Tween 20, and endogenous peroxidases were blocked using 1.5% (v/v) H2O2 in tris-buffered saline (pH 7.4). Antigens were revealed using mouse Polink-1 or rabbit Polink-2 horseradish peroxidase in conjunction with ImmPACT DAB (Vector Laboratories Inc.) and Warp Red (Biocare Medical Inc.). Slides were then washed in H2O, counterstained with hematoxylin, mounted in Permount (Thermo Fisher Scientific), and scanned at ×200 magnification using a ScanScope CS System (Aperio Technologies). Representative regions of interest (0.4 mm2) were identified visually, and high-resolution images were extracted from the whole-tissue scans. The percent area positive for CD4+ T cells was quantified using CellProfiler v3.1.5 (64).

Single-cell RNA sequencing

The experimental setup was described previously by Buggert et al. (22). Briefly, single HIV tetramer+ cells were index-sorted directly into 96-well microtiter plates containing lysis buffer using a FACSAria II flow cytometer (BD Biosciences). Cellular nucleic acids were recovered using the RNEasy Plus Micro Kit (Qiagen) and RNAClean Solid Phase Reversible Immobilization Beads (Beckman Coulter). Performance was assessed using the External RNA Controls Consortium RNA Spike-In Mix (Ambion). Reverse transcription was performed using oligo-dT primers, and complementary DNA (cDNA) was amplified over 22 to 24 PCR cycles using universal primers and the Kapa HiFi HotStart ReadyMix Kit (Kapa Biosystems). After cleanup, amplified cDNA was barcoded using Illumina Nextera libraries and sequenced to a depth of about 2 million 150–base pair paired-end reads per cell on a HiSeq 4000 (Illumina). Whole transcriptome data were obtained from 221 cells (n = 552 sorted cells).

scRNAseq analyses

A kernel method learning framework was used to assess the pairwise similarity between cells and compute the distance metric that best fitted the structure of the data (https://arxiv.org/pdf/1808.02061.pdf). Cell-cell similarity was learned using a rank-based, nonparametric function, and tSNE kernelization allowed the use of data from all available features. Dimensionality reduction was performed using kernel tSNE, wherein the pairwise similarity between cells was computed using the Semblance kernel as a distance measure. In determination of the cell-cell similarity matrix, the Gini index was used to account for the distribution of each gene and weigh it appropriately in the kernel calculation. The Gini index enabled the prioritization of genes with high variance as the genes that were more likely to provide useful features for niche group detection. Single cells were then projected onto a two-dimensional space using kernel tSNE, enabling intuitive visualization of hidden structures within the data. The supporting code is freely available online and can be implemented via the R package Semblance (https://cran.rproject.org/web/packages/Semblance/index.html) (65).

For SVM analyses, genes were selected using a t score cutoff of P < 0.01. Selected genes were then ranked using L0 norm regularization. The L0 norm–ranked gene lists were fed into the SVM algorithm for cross-validation (k-fold = 10) in subsets (increment = 100 genes). The average prediction error for each cross-validation was used to determine how well each subset of genes classified the desired labels. All steps were performed in MATLAB (MathWorks). L0 norm regularization was implemented using the MATLAB Feature Selection Library (MathWorks).

Differentially expressed genes were identified using three approaches: ROTS (40), which optimizes the t statistics based on the inherent characteristics of the data; scDD (41), which is a differential expression analysis method that accounts for the possibility of multimodally distributed gene expression; and Seurat (42), which uses the nonparametric Mann-Whitney U test to assess the null hypothesis that a randomly selected mean expression value for a given gene in one group will have an equal chance of being less than or greater than a randomly selected mean expression value for the same gene in a second group. As different genes in the dataset exhibited different distribution properties, these approaches in combination allowed us to winnow the list of differentially expressed genes, specifying a cutoff of P < 0.05 in at least two of the three outputs. The scDD P values are reported for simplicity.

The topGO package Bioconductor v3.8 was used for GO analyses (https://bioconductor.org/packages/release/bioc/html/topGO.html). GO terms associated with each gene were obtained using the August 2017 Ensembl database. Significance was determined using the classic Fisher method. GSEAs were performed using the software developed by the Broad Institute (66, 67).

Redirected killing assay

P815 mastocytoma target cells were labeled with LIVE/DEAD Violet (Thermo Fisher Scientific) and TFL4 (OncoImmunin), washed twice in phosphate-buffered saline, and incubated for 30 min at room temperature with anti-CD3 (5 μg/ml; UCHT1, Bio-Rad). CD8+ T cells were negatively selected from LNMCs or PBMCs using the CD8+ T Cell Enrichment Kit (STEMCELL Technologies). Isolated CD8+ T cells were rested in R10 for at least 45 min at 37°C and then incubated with anti-CD3–coated P815 cells at different effector-to-target ratios in a 96-well V-bottom plate for 4 hours at 37°C. Cells were then stained as described above (see “Antibodies” section) and acquired using an LSR II flow cytometer (BD Biosciences). Killing capacity was calculated by subtracting the frequency of active caspase-3+ TFL4+ LIVE/DEAD P815 cells in target-only wells from the frequency of active caspase-3+ TFL4+ LIVE/DEAD P815 cells in wells containing effector cells.

Viral suppression assay

The viral suppression assay was modified from Sáez-Cirion et al. (27). Briefly, CD4+ T cells were positively selected from LNMCs or PBMCs using the CD4+ T Cell Enrichment Kit (STEMCELL Technologies) and activated using a cocktail of IL-2 (100 U/ml; Chiron), anti-CD3 (1 μg/ml; UCHT1, Bio-Rad), anti-CD28 (1 μg/ml; L293, BD Biosciences), and anti-CD49d (1 μg/ml; L25, BD Biosciences). Concurrently, autologous CD8+ T cells were negatively selected from LNMCs or PBMCs using the CD8+ T Cell Enrichment Kit (STEMCELL Technologies) and rested in R10. After 2 days, CD4+ T cells were infected with HIV-1 BAL (University of Pennsylvania Center for AIDS Research Virology Core) by spinoculation and incubated with or without autologous CD8+ T cells in the absence of exogenous IL-2. Cells were harvested after a further 3 days, stained as described above (see “Antibodies” section), and acquired using an LSR II flow cytometer (BD Biosciences). Suppression capacity was calculated by dividing the frequency of p24+ CD4+ T cells in wells containing autologous CD8+ T cells by the frequency of p24+ CD4+ T cells in wells lacking autologous CD8+ T cells.

HPG translation assay

The protein translation assay was adapted from Araki et al. (68). Briefly, LNMCs were rested overnight and incubated for 30 min in methionine-free R10. The cultures were then supplemented with Click-iT HPG (100 μM; Thermo Fisher Scientific). Cells were stimulated with PepMix HIV (GAG) Ultra (2 μg/ml per peptide; JPT Peptide Technologies), PepMix HIV (NEF) Ultra (2 μg/ml per peptide; JPT Peptide Technologies), or anti-CD3 (5 μg/ml; UCHT1, Bio-Rad) in the presence of GolgiStop (0.7 μl/ml; BD Biosciences) and brefeldin A (1 μg/ml; Sigma-Aldrich). After 6 hours, cells were stained as described above (see “Antibodies” section) and acquired using an LSR II flow cytometer (BD Biosciences), following the instructions in the Click-iT Plus Alexa Fluor Picolyl Azide Toolkit (Thermo Fisher Scientific).

Statistics for nonsequencing data

Data were checked for normality using the Shapiro-Wilk normality test. Parametric tests were used if the data passed the normality test, and nonparametric tests were used if the data failed the normality test. Multiple corrections for two-way analysis of variance (ANOVA) were performed using the two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli. Specific tests are indicated in the relevant figure legends. Analyses were performed using R Studio or Prism v7.0 (GraphPad).

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/11/523/eaax4077/DC1

Fig. S1. HIV RNA is present in B cell follicles and the T cell zone in LNs.

Fig. S2. Relatively few CD8+ T cells from the LNs of ECs express perforin and granzyme B.

Fig. S3. CD8+ T cells from the LNs of ECs do not up-regulate cytolytic molecules in response to cognate antigen exposure.

Fig. S4. Active caspase-3 detection effectively captures killed target cells.

Fig. S5. Differential expression analysis of the scRNAseq data.

Table S1. Clinical characteristics of the donors included in this study.

Table S2. Complete list of 2264 genes that were differentially expressed between HIV-specific CD8+ T cells from the LNs of ECs and HIV-specific CD8+ T cells from the LNs of CPs.

Data file S1. Primary data.

REFERENCES AND NOTES

Acknowledgments: We thank the Virus and Reservoirs Core at the Penn Center for AIDS Research for assistance with viral quantification, the Pathology Core at the Hospital of the University of Pennsylvania for access to biopsy materials, K. Noyan-Gertler for assistance with viral suppression assays, and M. Li and Y. Che for initial contributions to the scRNAseq analyses. This study was funded, in part, by the Oregon National Primate Research Center NIH grant P51OD011092 and by federal funds from the NCI under contract HHSN261200800001E. S.D., A.R., and D.C.D. were supported by the Intramural Research Program of the National Institute of Allergy and Infectious Diseases at the NIH. M.A.-M., J.A.H., and M.R.B. were supported by the Penn Center for AIDS Research (P30 AI045008). M.A.-M. was further supported by the NIH R21 grant AI129636, the Campbell Foundation, and a grant from the W.W. Smith Charitable Trust (A1701). M.R.B. was further supported by the NIH R01 grants AI076066 and AI118694 and the BEAT-HIV Delaney Collaboratory (UM1AI126620). D.A.P. was supported by a Wellcome Trust Senior Investigator Award (100326/Z/12/Z). M.B. was supported through the Swedish Research Council (VR), Karolinska Institutet, Swedish Society for Medical Research (SSMF), Jeansson Stiftelser, Ake Wibergs Stiftelse, The Swedish Society of Medicine (SLS), Magnus Bergvalls Stiftelse, Lars Hiertas Stiftelse. The SCOPE cohort was supported by the University of California, San Francisco (UCSF)–Gladstone Institute of Virology and Immunology Center for AIDS Research (P30 AI027763), the Delaney AIDS Research Enterprise (AI127966), and the amfAR Institute for HIV Cure Research (109301). The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services (DHHS) nor does the mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government. Author contributions: S.N., D.A.P., D.C.D., M.B., and M.R.B. conceptualized and designed the experiments and wrote the manuscript. M.B. and M.R.B. supervised the study. S.N., A.S.J., and M.B. performed and analyzed the flow cytometry experiments. C.D. and J.D.E. performed and analyzed the RNAscope and immunohistochemistry experiments. E.G. and D.A.P. generated bespoke HLA class I tetramers. S.D., A.R., and D.C.D. performed the scRNAseq experiments. S.N., S.D., D.P.T., D.A., N.R.Z., and V.H.W. analyzed the scRNAseq data. L.K.-C. and M.A.-M. performed the viral quantification experiments. J.A.H. provided reagents for the study. P.M.D.R.E., Y.A.-T., A.N., G.R.-T., and S.G.D. provided the samples for the study. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. The sequencing data reported in this paper have been deposited in the Gene Expression Omnibus database with the accession number GSE110684.

Stay Connected to Science Translational Medicine

Navigate This Article