NK-like CD8+ γδ T cells are expanded in persistent Mycobacterium tuberculosis infection and chronic inflammation

Abstract

Gamma delta (γδ) T cell responses differ during the acute versus chronic phases of infection, but the mechanisms underlying these differences are unclear. γδ T cells play an important role during the acute phase of Mycobacterium tuberculosis (Mtb) infection, however their response during persistent Mtb infection is not well understood, despite most infections with Mtb manifesting as a chronic, clinically asymptomatic state. Here, we analyze peripheral blood γδ T cells from a South African adolescent cohort and identify that a unique CD8+ γδ T cell subset with features of “memory inflation” is expanded in chronic Mtb infection. These cells are hyporesponsive to T cell receptor (TCR)-mediated signaling, but like NK cells, can mount robust CD16-mediated cytotoxic responses. These CD8+ γδ T cells comprise a highly focused TCR repertoire, with clonotypes that are Mycobacterium-specific but not phosphoantigen-reactive. Using multiparametric single cell pseudotime trajectory analysis, we identify the differentiation paths that these CD8+ γδ T cells follow to develop into effectors in this infection state. Finally, we find that circulating CD8+ γδ T cells also expand in other chronic inflammatory conditions, including cardiovascular disease and cancer, suggesting that persistent antigenic exposure may drive similar γδ T cell effector programs and differentiation fates.

One sentence summary:

Antigen-driven expansion of CD8+ γδ T cells mediate NK-like functions in persistent Mycobacterium tuberculosis infection.

Introduction

The contribution of γδ T cells to host immune defense differs from αβ T cells, where γδ T cells are earlier responders in acute infections and after vaccination. In individuals with active tuberculosis (TB) and acute bacterial and parasitic infections, γδ T cell frequencies in peripheral blood can increase from <5% in healthy subjects to >45% in some patients (1). These cells predominantly express Vγ9Vδ2 TCRs and respond to small phosphorylated nonpeptide antigens, called phosphoantigens (pAgs), which are produced by bacteria and parasites via the isoprenoid pathway. Activated Vγ9Vδ2 T cells can kill infected cells in vitro by mechanisms including TCR-dependent degranulation and antibody-dependent phagocytosis (2). Additionally, increases in the frequency of circulating IL-17-producing Vγ9Vδ2 T cells have been observed in children with acute bacterial meningitis (3) and in patients with active pulmonary TB (4). These observations are consistent with findings in mouse models of infection, where γδ T cells are the major initial IL-17 producers in acute infection that initiate the inflammatory response (5). γδ T cells have also been implicated in the control of chronic viral infections. Selective γδ T cell populations with an effector/memory phenotype are expanded in peripheral blood of patients infected with HIV, CMV, EBV, or HSV (6–8). These γδ T cell responses are distinct from αβ T cell responses in most chronic infections and cancer; the latter become exhausted, decreasing in numbers and functionality (9). They are also different from the γδ T cells that respond in acute infections in that these cells primarily express TCRs encoded by Vδ1 and Vδ3 genes that pair with different Vγ chains. Some of these TCRs show reactivity to virus-infected cells (10–12).

While these studies highlight the ability of γδ T cells to respond to pathogenic challenges in a context-dependent manner, it is unclear whether the difference in these responses reflect the chronicity of the infection or the type of pathogenic challenge (bacteria or parasites vs. viruses). This question is particularly relevant to TB, since most Mtb infections manifest as a clinically asymptomatic state, presumably held in check, or cleared by the host immune system. This phase of controlled Mtb infection, commonly referred to as latent TB infection, may be best defined as a state of persistent immune response to Mtb antigens detected by an interferon gamma (IFNγ)-release assay, but without signs or symptoms of active disease. It is estimated that up to a quarter of the global population is latently infected with Mtb, and less than 10% of infected individuals ultimately progress to active TB, indicating effective infection control.

Previously, using mass cytometry analysis of peripheral blood mononuclear cells (PBMCs) from a South African adolescent cohort, we found evidence of an enhanced cytotoxic response, primarily mediated through CD16 (FcγRIIIa) and NK cells, coupled with continued inflammation and deviations in all compartments of the adaptive immune system in controlled Mtb infection (13). With additional analysis using cell-type deconvolution of transcriptomic data sets from cohorts of different ages, genetic backgrounds, geographic locations, infection stages, and treatment response, we found that cytolytic functions mediated through Fc receptors associate with distinct Mtb infection outcomes. Intriguingly, the CD16⁺GZMB^hi cytolytic cells in this cohort of controlled Mtb infection were not only present in the NK cell compartment, but also in the γδ T cells, suggesting that they also contribute to Mtb infection control.

In this study, we perform an unsupervised analysis to identify γδ T cell subsets that associate with this state of controlled Mtb infection from the same cohort and follow this with an investigation of their function and antigen specificities. We then extend the study to construct cellular progression trajectories from single cell protein and gene expression data to understand the development of these γδ T cells in this infection state and test the generality of our finding in other acute and chronic infectious diseases. Our results underscore the importance of antigen recognition in mounting a context-dependent γδ T cell response and suggest that the NK-like CD8+ γδ T cell population induced in controlled Mtb infection may also respond in other persistent inflammatory conditions.

Results

A CD8+ γδ T cell population with features of memory inflation is associated with controlled Mtb infection

To analyze γδ T cell responses during chronic Mtb infection, we followed up on our prior study of a South African adolescent cohort (13). One limitation of studying this phase of Mtb infection is that it can represent a spectrum between complete bacterial clearance and active control of subclinical disease (14, 15). However, an advantage of our study design is that this cohort is from a highly endemic area and yet has a lower rate of active tuberculosis (TB) than is seen in young children and adults (16), indicating a well-controlled Mtb infection.

Our previous CyTOF (cytometry by time-of-flight) analysis of PBMCs from 24 uninfected donors and 24 donors in this cohort with controlled Mtb infection found no significant difference in total γδ T cell frequencies between the two groups (13). To test whether changes in specific γδ T cell populations associate with controlled Mtb infection, we analyzed total γδ T cells from the same CyTOF study using Citrus (cluster identification, characterization, and regression) (17), which performs unsupervised hierarchical clustering based on surface marker and intracellular cytokine/cytolytic effector expression. We found that several γδ T cell clusters (Clusters 1–5) (Fig. S1, A to E) with shared features, specifically the expression of CD8, were present at significantly higher frequencies (False Discovery Rate (FDR) < 0.01) in individuals with controlled infection compared to uninfected controls (N=14/group). This increase in CD8+ γδ T cells was accompanied by a corresponding decrease in the frequency of the CD8− γδ T cell subset (Cluster 6). To confirm this observation, we performed flow cytometry on an additional 36 PBMC samples (from 17 uninfected donors and 19 individuals with controlled Mtb infection) and found that an increased frequency of circulating CD8+ γδ T cell was indeed a feature of these subjects (Fig. 1, A). Although most γδ T cells are double-negative for the expression of CD4 and CD8 coreceptors, γδ T cells expressing CD8αα+ homodimers (~20% of total γδ T cells) and CD8αβ+ heterodimers (~5% of total γδ T cells) are present in the peripheral blood of healthy adults (18). To determine if the CD8+ γδ T cells found in our cohort were CD8αα+ or CD8αβ+, we analyzed a small number of samples (N=3) by flow cytometry. We found that the γδ T cells in two of the three individuals predominantly consisted of CD8αβ+ cells, while the third one included both CD8αα+ and CD8αβ+ cells (Fig. S1, F).

Figure 1. A CD8+ γδ T cell population with features of memory inflation is associated with controlled Mtb infection.

(A) Percentage of circulating CD8+ γδ T cells in 17 uninfected donors and 19 donors with controlled Mtb infection from a South African adolescent cohort. P-value was derived using the Mann-Whitney test. Error bars represent mean and 95% confidence intervals. (B) Percentages of CD11c+, CD70+, CD16+, GZMB+, PRF1+, IFNγ+, and polyfunctional (GZMB+PRF1+IFNγ+) cells within CD8− and CD8+ γδ T cell subsets. P-values were derived using Wilcoxon matched-pairs signed rank test. Error bars represent mean and 95% confidence intervals. (C) Percentages of naïve, central memory (CM), effector memory (EM), and effector memory cells that re-express CD45RA (EMRA) within CD8− and CD8+ γδ T cell subsets. P-values were derived using multiple Wilcoxon’s test followed by Holm-Sidak correction. (D) Percentages of CD8+ γδ T cells in a longitudinal cohort (N=7) of South African adolescents. (E) Volcano plot showing differential gene expression between bulk sorted CD8− and CD8+ γδ T cells derived from peripheral blood samples of 5 donors with controlled Mtb infection. (F) Heatmaps showing the expression patterns of selected genes significantly differentially expressed (adjusted P <0.05) between CD8− and CD8+ γδ T cells. (G) KEGG Pathway analysis of transcripts differentially upregulated in CD8+ γδ T cells (red bars) and CD8− γδ T cells (blue bars) analyzed by the ShinyGO gene-set enrichment tool, which calculates enrichment based on hypergeometric distribution followed by FDR correction. The pathways listed have enrichment FDR<0.05.

Notably, these CD8+ γδ T cells, from both the CyTOF and FACS analyses, displayed significantly lower cell surface CD3 expression compared to CD8− γδ T cells (Fig. S1, E and G). This suggests persistent exposure to antigenic stimulation, which is known to trigger TCR internalization. Consistent with this observation, CD8+ γδ T cells also had increased expression of two classical myeloid cell markers, CD11c and CD70 (Fig. 1, B and Fig. S2, A), whose expression on αβ T cells has been linked to antigen-driven responses and chronic immune activation (19, 20).

The CD8+ γδ T cells primarily displayed a terminally differentiated effector memory phenotype (CD45RA+ CD45RO− CCR7− CD27−) (Fig. 1, C and Fig. S2, A and B) and expressed significantly higher levels of CD16, granzyme B (GZMB), and perforin (PRF1) compared to the CD8− γδ T cells (Fig. 1, B and Fig. S2, A and B), indicating enhanced cytotoxic function, potentially mediated through CD16. Furthermore, these cells were polyfunctional, showing robust co-expression of GZMB, PRF1, and IFNγ (Fig. 1, B). Taken together, the γδ T cell response in controlled Mtb infection is similar to the “memory inflation” phenomenon noted in CMV and some other viral infections, where effector memory CD8+ αβ T cells and γδ T cells with robust effector functions and clonal expansions persist even after the major T cell response has occurred (6, 21). A key aspect of this phenotype is stability in cell numbers (22). We found that the increased frequencies of CD8+ γδ T cells were stable for at least 6 months to a year in a longitudinal cohort (N=7) of South African adolescents with controlled Mtb infection (Fig. 1, D).

To broaden this analysis, we also performed bulk RNA sequencing (bulk-RNA-seq) on FACS sorted CD8+ and CD8− γδ T cells from peripheral blood samples of five of these donors and identified 1,273 significantly (Adjusted P<0.05) differentially expressed genes (Fig. 1, E and Table S1). Consistent with the expression of effector-memory markers, these CD8+ γδ T cells showed significantly lower expression of CCR7, CD27, CD28, CD62L, and CD127 relative to CD8− γδ T cells (Fig. 1, F and Table S1). But they showed a preferential upregulation of NK-cell-associated inhibitory and activating receptor genes, including KLRG1 (also a marker for inflationary CD8+ αβ T cells), KLRD1 (CD94), KLRC3 (NKG2E), KLRC2 (NKG2C), KLRC4 (NKG2F), NCR1 (NKp46), CD244 (2B4), and the CD158 family of KIR receptors: KIR2DS4, KIR2DL1, KIR2DL3, KIR3DL1, KIR3DL2, and KIR3DL3 (Fig. 1, F and Table S2). Furthermore, consistent with the analysis described above, they displayed significantly increased expression of RNA for cytolytic molecules (GZMA, GZMH, GNLY, PRF1) and NKG7 (a regulator of exocytosis of cytotoxic granules), indicative of robust cytotoxicity. These CD8+ γδ T cells also expressed high levels of IL-32, a proinflammatory cytokine, which plays a protective role in TB (23), and chemokines including XCL1, XCL2, CSF3, CCL4L1(MIP1b), CCL4L2, CCL3L1 (MIP1AP), and CCL19 (Fig. 1, F), which target a wide range of immune cells, including lymphocytes, dendritic cells, and myeloid cells. These observations suggest that mobilizing leukocytes to the site of infection may be one of the ways in which these CD8+ γδ T cells could regulate the inflammatory response to Mtb.

Unlike exhausted αβ T cells, the CD8+ γδ T cells did not upregulate PD-1 expression but had elevated levels of TIGIT, LAG3, B3GAT1 [CD57], PRDM1 [Blimp-1], NR4A2, and TOX – genes associated with prior antigenic exposure and chronic TCR activation (Fig. 1, F). In addition, compared with the CD8− γδ T cells, the CD8+ cells showed significantly higher levels of TBX21 (Tbet), a transcription factor with a well-defined role in driving effector and inflationary CD8+ αβ T cell responses (21, 24). In contrast, the CD8− γδ T cells showed higher levels of RORC, SOX4, ID3, and LEF1 expression – transcription factors associated with IL-17-committed γδ T cells (25). Furthermore, a suite of integrin genes – ITGAD (CD11D), ITGA10, ITGAL (CD11A), ITGA4 (CD49D), ITGB1 (VLAB), were also upregulated on the CD8+ γδ T cells (Fig. 1, F), suggesting specific tissue homing properties, especially to nonlymphoid tissues (since they lack CD62L and CCR7) and extra-intestinal sites, a characteristic of inflationary CD8+ αβ T cells as well (26). The CD8− γδ T cells, in contrast, expressed higher levels of the gut-homing receptor ITGAE (αE).

To identify specific pathways that appear upregulated in these T cells, we performed KEGG (Kyoto Encyclopedia of Genes and Genomes) enrichment analysis and found that relative to CD8− γδ T cells, their CD8+ counterparts were significantly enriched for genes associated with FcγR-mediated functions, and genes encoding components of the TCR, PI3K-AKT, and Rap1 signaling machinery (Fig. 1, G). Signaling via Rap1 and PI3K, whose activation by TCR ligation and chemokines is known to regulate integrin-(specifically, ITGAL and ITGA4)-mediated cell adhesion (27, 28), may reflect the mechanistic underpinnings of the migratory behavior of peripheral CD8+ γδ T cells. In addition, these cells showed a significant enrichment for genes involved in immune response elements associated with chronic viral infections, consistent with their similarities with the CD8+ αβ T cell response in CMV infection (Fig. 1, G).

CD8+ γδ T cells are hyporesponsive to TCR-mediated signaling but mount robust CD16-mediated cytolytic responses

As persistent antigen exposure is known to diminish TCR expression and TCR-mediated αβ T cell responses, we tested how these CD8+ γδ T cells could respond to TCR-triggering. We stimulated PBMCs in vitro with either anti-CD3 antibody or an Mtb whole cell lysate and measured the upregulation of CD69 as a read-out for T cell activation. We found that both the percentage of cells that upregulated CD69, and the CD69 expression level, were significantly lower for the CD8+ γδ T cells compared to the CD8− cells (Fig. 2, A). We also found that the response to the Mtb lysate was inhibited by Cyclosporine A, indicating that the activation was mediated through the TCR. These observations indicated that the CD8+ γδ T cells were hyporesponsive to antigenic challenge, in contrast to the robust TCR-mediated response reported for inflationary CD8+ αβ T cells and for memory αβ T cells (26).

Figure 2. Peripheral blood CD8+ γδ T cells are effectors with a diminished ability for TCR-mediated activation, but a robust capacity for CD16-mediated cytotoxic response.

(A) Histograms showing the expression of CD69 on CD8− and CD8+ γδ T cells under unstimulated and stimulated conditions (Mtb-lysate, Mtb-lysate with Cyclosporin A, anti-CD3) (left) and the percentage change in CD69+ cells post stimulation in CD8− and CD8+ γδ T cells from 8 donors with controlled Mtb infection, performed in four independent experiments (right). Circles represent Mtb-lysate stimulation, while squares represent anti-CD3 stimulation. P-value was derived using Wilcoxon matched-pairs signed rank test. (B) Mean mass intensities of selected phospho-signaling effectors in CD8− and CD8+ γδ T cell subsets [N=20 samples/subset (10 uninfected and 10 donors with controlled Mtb infection)] at baseline (unstimulated condition) and post PMA-ionomycin stimulation. P-values were derived using Wilcoxon matched-pairs signed rank test. Lower and upper hinges of boxes represent 25th to 75^th percentiles, the central line represents the median, the plus sign represents the mean, and the whiskers extend to the highest and lowest values. (C) ADCC response of CD8− and CD8+ γδ T cells, measured by the % increase in CD107a degranulation. Total γδ T cells from donors with controlled Mtb infection (N=5) were sorted and incubated in the presence of antibody coated or uncoated P815 target cells (data are from two independent experiments). P-value was derived using Wilcoxon matched-pairs signed rank test. (D) ADCC response of CD8+ γδ T cells and NK cells measured by incubating total PBMCs from donors with controlled Mtb infection in the presence of antibody coated or uncoated P815 target cells (N=6 donors). P-values were derived using Wilcoxon matched-pairs signed rank test. (E) Histograms showing MitoTracker Green staining of CD8− and CD8+ γδ T cells (left). The mean fluorescence intensity (middle) and the percentage of MitoTracker Green^hi cells (right) within the CD8− and CD8+ γδ T cell subsets (right) are shown (N=17 uninfected donors and 19 donors with controlled Mtb infection). Data are from three independent experiments. P-values were derived using Wilcoxon matched-pairs signed rank test.

These CD8+ γδ T cells also showed an altered signaling capacity through the mitogen-activated protein kinase (MAPK) and the AKT pathways, which regulate diverse cellular programs by relating extracellular signals to intracellular responses, including TCR-mediated signaling. In particular, phospho-CyTOF analysis indicated that these cells displayed significantly lower phosphorylation levels of MAPKs, including ERK-1/2, p38, and MAPKAPK2, a substrate for p38, at baseline (unstimulated condition) as well as after stimulation with phorbol myristate acetate (PMA) and ionomycin (Fig. 2, B and Fig. S2, C). Instead, they exhibited enhanced AKT phosphorylation, which, in cytotoxic CD8+ αβ T cells, has been shown to play an important role in driving TCR− and IL-2-induced transcriptional programs that control effector and cell fate determination (29).

From the CyTOF/FACS and gene expression analysis described above, we have noted that these CD8+ γδ T cells have the potential to mount CD16-mediated cytolytic responses. To test this ability, we measured antibody-dependent CD107a degranulation from isolated γδ T cells and from total PBMCs. We found that the CD8+ γδ T cells showed significantly higher degranulation than the CD8− γδ T cells, and their antibody-dependent cell-mediated cytotoxicity (ADCC) potential was comparable to that of NK cells (Fig. 2, C and D and Fig. S2, D).

In addition, we found that these CD8+ γδ T cells had significantly higher mitochondrial mass than the CD8− γδ T cells as assayed by the intake of MitoTracker-Green (Fig. 2, E). Different states of activated αβ T cells are known to have different patterns of metabolism, but all effector αβ T cells have higher mitochondrial mass when compared with naïve or exhausted T cells developed in chronic infections and in the tumor microenvironment (30). Taken together, these analyses indicated that the CD8+ γδ T cells in controlled Mtb infection were terminally differentiated effectors with a diminished ability for TCR-mediated activation, but a robust capacity for CD16-mediated cytotoxic response.

CD8+ γδ T cells express clonally focused TCRs and respond to Mtb whole cell lysate and tumor cell lines

To assess the antigen-specific repertoire of these CD8+ γδ T cells, we performed direct ex vivo single-cell TCR sequencing (scTCR-seq) (31) on CD8+ and CD8− γδ T cells isolated from the peripheral blood of four donors with controlled Mtb infection. We found that in all donors, the CD8+ γδ TCRs were comprised of just a few dominant expanded clonotypes (multiplets). In contrast, most of the CD8− γδ TCRs appeared only once (singletons) (Fig. 3, A). The percent clonality (calculated as the proportion that the multiplets occupy in the total repertoire) across all donors showed a significant difference between these two types of γδ T cells (P=0.0045) (Fig. 3, B). These observations indicate that the peripheral CD8+ γδ T cells consist of antigen-expanded γδ T cell clones, potentially responsive to dominant antigens. Notably, in contrast to acute Mtb infection, which is known to induce increased frequencies of circulating γδ T cells expressing Vγ9Vδ2 TCRs (32), the repertoire of peripheral CD8+ γδ T cells in controlled Mtb infection was skewed towards Vδ1 and to a lesser extent, Vδ3 gene usage, but also included Vδ2 and Vδ4 encoded TCRs. These TCRs displayed chain-pairing diversity with Vγ2/3/4 (Fig. 3, C and Table S3). Thus, the V gene usage of expanded γδ T cells in controlled Mtb infection is similar to the V gene usage described for γδ T cells that expand in chronic CMV infection (33). Nonetheless, we found that a substantial population, approximately a third of the CD8− γδ T cells in two of the donors comprised of Vδ1/3 TCRs. This suggests that the CD8+ and CD8− γδ T cells cannot be strictly defined by their V-gene usage.

Figure 3. CD8+ γδ T cells express clonally focused TCRs and respond to Mtb-lysate.

(A) Pie charts depicting the clonal expansion of circulating CD8− and CD8+ γδ T cells in 4 donors with controlled Mtb infection. The number of cells with both γ and δ chains identified is shown below each pie chart. For each TCR sequence expressed by two or more cells (clonally expanded), the absolute number of cells expressing that clone is shown by a distinct colored section. (B) Percent clonality of CD8− and CD8+ γδ T cells determined from paired single cell γ and δ chain sequencing. Clonality is defined as the total number of sequences that appear more than once relative to the total number sequenced per sample. P-value was determined using a paired t test. Error bars represent mean and 95% confidence intervals. (C) Stacked bar plots showing the Vγ/δ chain pairings of CD8− (top) and CD8+ (bottom) γδ T cells isolated from the 4 donors with controlled Mtb infection. (D) CDR3δ/γ sequences from four CD8+ γδ TCRs (GD8-1, GD8-2, GD8-3, GD8-5) and two control Vγ9Vδ2 expressing TCRs (TB5, AT1G9). Jurkat α-β-cells expressing these TCRs were tested for reactivity to cell lines, Mtb-lysate, and HMBPP, measured by the increase in CD69 expression 14 hours post stimulation. P-values were determined using one-way ANOVA followed by Dunnett’s multiple comparisons test. Error bars represent mean and standard deviations.

Next, to determine the antigen specificity of some of the clonally expanded TCRs identified in controlled Mtb infection, we selected four clonal TCRs (clonal frequency >2%) with different γδ chain-pairings and established these as TCR transfectants in Jurkat α-β-cells. We also expressed two Vγ9Vδ2 TCRs with CDR3 sequences similar to the TCR sequences identified from CD8− γδ T cells as controls (Fig. 3, D). We found that three out of the four CD8+ γδ TCRs responded to Mtb-whole cell lysate, but none responded to (E)-4-hydroxy-3-methyl-but-2-enyl pyrophosphate (HMBPP), a pAg (Fig. 3, D). pAgs stimulate Vγ9Vδ2 T cells through the induction of a specific conformation of BTN2A1 and BTN3A1 expressed on the cell surface, which interact with the γδ TCRs (34–36). As expected, the control Vγ9Vδ2 TCRs responded to both HMBPP and the Mtb lysate, which contains pAgs as one of its components (Fig. 3D). These findings suggest that prevalent Mtb antigens present in different stages of infection preferentially trigger different subsets of γδ T cells. Further work is required to identify individual components in the Mtb lysate which stimulate CD8+ γδ T cells.

Some of the non-Vγ9Vδ2 cells induced in chronic viral infections can respond to tumor cell lines in addition to viral infected cells (10). We found that Clones GD8-1 and GD8-5 showed robust CD69 expression upon coculture with T2, but not K562 or Daudi cells. Clone GD8-3 was reactive to both T2 and Daudi, but not K562 cells, while clone GD8-2 recognized Daudi cells, but not T2 or K562 cells (Fig. 3, D). These results indicated that the CD8+ γδ T cells, as a group, could respond to antigens induced in tumor cell lines, but unlike the Vγ9Vδ2 T cells, they did not respond to antigens common to all the tumor cells. These observations underscored the difference in antigen specificity of γδ T cells in acute vs. controlled Mtb infection.

CD8+ γδ TCRs responded to T2 cells, which have a homozygous deletion in the H-2 region, including both peptide transporter genes, and all functional MHC class II genes, such that they have low/no surface MHC class I and no surface MHC class II expression (37). A CD8+ γδ TCR also showed reactivity to Daudi cells, which have a defective β₂ microglobulin gene and therefore do not express either classical or non-classical MHC class I on the cell surface. These results indicate that MHC class I and II molecules are not obligatory components of the antigens of CD8+ γδ T cells. In this context, it should be noted that CMV exposure can induce CD8αα-expressing γδ T cells (10). While the expression of CD8αα on γδ T cells was demonstrated to play a co-stimulatory role in the recognition of CMV-infected cells, it was dispensable for reactivity to tumor cell lines.

CD8+ γδ T cells primed by non-Mtb antigens can respond to Mtb-whole cell lysate

Mtb responsive γδ T cells may be induced at the site of infection and/or the draining lymph nodes. Given the difficulty in acquiring human lung and draining lymph node samples, we utilized human tonsil organoids (38) to evaluate Mtb-specific, lymphoid organ γδ T cell responses. We found that stimulation of tonsil organoid cultures with Mtb lysate resulted in increased frequencies and numbers of CD8+ γδ T cells in all three independent tonsil organoid cultures (Fig. 4, A). Ex vivo single cell TCR sequencing of CD8+ γδ T cells from Mtb lysate stimulated organoid cultures demonstrated clonal expansions, albeit to different degrees (Fig. 4, B). Nonetheless, in all three donors, we identified individual TCRs that clonally expanded (i.e., showed increased clonal frequency) upon Mtb lysate stimulation (Table S4). 70% of the expanded clonotypes expressed Vδ1 encoded TCRs. γδ T cells with other TCRs, including those encoded by Vγ9Vδ2 and with CDR3 sequences compatible for phosphoantigen recognition also showed clonal expansion in the tonsil organoid cultures (Fig. 4, C, Table S4). These observations indicate that the Mtb-whole cell lysate consists of components which stimulate γδ T cells in both the acute and the controlled stages of infection, and that Mtb-specific γδ T cell responses can be generated in lymphoid tissues regardless of the stage of infection.

Figure 4. Mtb-specific T cells can be generated in lymphoid organs and CD8+ γδ T cells primed by non-Mtb antigens can respond to Mtb-whole cell lysate.

(A) Percentage and number of CD8+ γδ T cells in unstimulated and Mtb-lysate (10μg/ml)-stimulated tonsil organoid cultures (day 7) from three children (T1-T3) with no known exposure to TB. (B) Clonal composition of CD8+ γδ T cells in unstimulated and Mtb-lysate-stimulated tonsil organoid cultures from three donors. The total number of paired γδ TCRs is shown below each pie chart. For each TCR expressed by two or more cells (clonally expanded), the absolute number of cells expressing that TCR sequence is shown by a distinct colored section. (C) Stacked bar plots showing the Vγ/δ chain pairings used by the clonally expanded tonsillar CD8+ γδ T cells in the unstimulated and Mtb-lysate stimulated cultures. (D) CDR3δ/γ sequences from two CD8+ γδ TCRs that showed clonal expansion (appeared two or more times) in both unstimulated and Mtb-stimulated organoid cultures. Jurkat α-β-cells expressing these TCRs were tested for response to Mtb lysate (10, 30, 50 and 100μg/ml) and HMBPP (5μM). P-values were determined using one-way ANOVA followed by Dunnett’s multiple comparisons test. Error bars represent mean and standard deviations.

Unexpectedly, we observed that some of the clonally expanded CD8+ γδ TCR pairs present in the unstimulated tonsil organoid cultures were also found among the expanded TCR clones in the Mtb lysate stimulated organoid cultures established from the same donors (Table S4). We tested two of these TCRs, which were not encoded by Vγ9Vδ2 sequences and confirmed that these clones could respond to Mtb lysate in a dose-dependent manner, but not to HMBPP (Fig. 4, D). Since these organoids were generated from the tonsils of children that had no prior exposure to Mtb antigens, this finding suggests that these Mtb lysate responsive CD8+ γδ T cells have proliferated in response to non-Mtb antigens. These observations suggested a way for CD8+ γδ T cells to encounter antigen repetitively in Mtb infection and to respond in Mtb unrelated pathological/physiological conditions.

CD8+ γδ T cells in controlled Mtb infection traverse distinct differentiation trajectories characterized by gradual TCR clonal focusing

To better understand the developmental and maturation sequences that program γδ T cells to the effector state in this infection stage, we employed the newly developed Cytoskel algorithm (39) to construct pseudotime trajectories from single cell targeted mRNA expression and concomitant cell surface marker expression. The single cell analysis also included concomitant TCR determination allowing the trajectory analysis to determine the developmental paths of γδ T cells with different TCRs. This approach is different from the γδ T cell trajectory analysis described in an earlier study, where the Vδ1+ and Vδ2+ peripheral blood γδ T cells were analyzed separately and only linear pseudotemporal ordering of cells was considered (40).

We analyzed FACS sorted γδ T cells (N=24,888) from six donors with controlled Mtb infection (Fig. S3) and found that ~10% of the cells had a naïve phenotype (defined as CD45RA+ CCR7+) (represented as FlowSOM Cluster 4 and X on the Cytoskel map) (Fig. 5, A). These cells navigated a trajectory with a branch point (Branch Point 1) that led to two alternative developmental fates – trajectory 1 and trajectory 2, with similar numbers of cells in each trajectory. Trajectory 1 further bifurcated (Branch point 2) into two separate paths which ended at A and B, while trajectory 2 showed an early branching event (Branch Point 3) leading to two divergent cellular paths, eventually resulting in three branch termini (identified as C, D, and E). Additionally, some of the naïve γδ T cells appeared to travel a short distinct path (identified as F) that ran parallel to paths A and B and was considered a part of trajectory 1 (Fig. 5, A). Except path F, nearly all the cells along trajectory 1 (comprised of FlowSOM Clusters 2 and 3) appeared to be programmed to cytolytic CD8+ effectors with the expression of NK cell receptors, including CD16 (Figs. 5, A and B, and Fig. S4). In contrast, ~85% of the cells in trajectory 2 appeared to be in a transitory state, trailing around Branch Point 3 (FlowSOM Cluster 8) (Fig. 5, A). Only ~10% of the cells in trajectory 2 differentiated into cytolytic effectors, that expressed CD16, but not CD8 (path C) (Fig. 5, B and Fig. S4). These divergent paths were marked by the expression of different transcription factors and related genes (Fig. S5, A), which may play instructive roles in γδ T cell effector fate decisions and maintenance.

Figure 5. CD8+ γδ T cells in controlled Mtb infection traverse distinct differentiation trajectories characterized by gradual TCR clonal focusing.

(A) Trajectory initiated UMAP plot of γδ T cells (N=24,888 from 6 donors with controlled Mtb infection) constructed using the Cytoskel package. The cells are colored based on FlowSOM clustering (left). The starting point “X”, the various branch points and branch termini of the Cytoskel map are annotated (shown in the middle). The branches that comprise two major trajectories (trajectory 1 = black and trajectory 2 = gray) are shown (right). (B) Expression of selected cell surface markers along the trajectories as determined by antibody staining. The expression intensity of each marker is indicated, independently for each marker, by the colored gradient for which the range corresponds to the arcsinh transformed expression [arcsinh (x/5), where x=counts]. (C) Pie charts showing the clonal expansion of γδ T cells along pseudotime axes. Cells along each Cytoskel branch were grouped into 4 pseudotime bins of approximately equal number of TCR sequences. For each TCR clone expressed by two or more cells (clonally expanded), the absolute number of cells expressing that clone is shown by a distinct colored section.

Analysis of the TCR paired chain sequences of the cells along each of these trajectories showed gradual increases in clonal expansion along trajectory 1 (Branch point 2 to A and B, and X to F) (Fig. 5, C and Fig. S6, A). In contrast, the cells in trajectory 2 showed limited clonal expansion, with no association between clonality and the pseudotemporal ordering of cells. In fact, BTG1 and LGALS1, two of the ten most abundantly expressed genes in trajectory 2 (FlowSOM Clusters 7, 9, and 10) (Fig. S5, B) are noted for their anti-proliferative function (41, 42). These results were consistent with the TCR analysis of CD8+ and CD8− γδ T cells from the 4 donors described above (Fig. 4, A).

Among the γδ T cells whose TCRs were successfully sequenced (N=12,706), 39%, 27%, and 26% comprised of Vδ1+, Vγ9-Vδ2+, and Vγ9+Vδ2+ TCRs, respectively (Fig. S6, B and Table S5). The Vδ1+ T cells predominately appeared in Trajectory 1 (86% of the Vδ1+ cells) and traversed distinct differentiation paths (from Branch Point 2 to A and B, and X to F, FlowSOM Cluster 3) leading to cytolytic effector cell states. In contrast, ~73% of all Vγ9+Vδ2+ T cells belonged to trajectory 2, and appeared to be in transitional states, gathered around branch points (between Branch Points 1 to 2, 1 to 3, and 3 to 4, FlowSOM Cluster 8) (Fig. S6). Notably, ~60% of the cells in path F (trajectory 1) and ~30% of the cells in path C (trajectory 2), both of which are comprised of CD8− γδ T cells, expressed Vδ1 TCRs. These results demonstrate that neither CD8 expression on γδ T cells nor the trajectory bifurcation is restricted by TCR V-gene usage. In fact, Vδ1+ γδ T cells can differentiate into diverse effector subsets. For instance, the ones along path C are distinguished by their enhanced CD158e1 expression and lack of KLRG1 expression (Fig. S4, A). Since KLRG1 expression is a defining signature of memory inflation, it stands to reason that only the CD8+ Vδ1-expressing γδ T cells participate in the inflationary response in controlled Mtb infection.

Unlike the Vγ9+Vδ2+ T cells, the Vγ9-Vδ2+ T cells were similarly abundant in trajectories 1 and 2, and both at branch points and branch termini. This is in line with reports that Vγ9-Vδ2+ and Vγ9+Vδ2+ T cells have different antigenic specificities and participate in different pathological conditions (43). Indeed, Vγ9-Vδ2+ cells comprised ~30% of cells with cytolytic effector fate in trajectory 1 (from Branch Point 2 to A and B, and X to F), and approximately one third of cells with cytolytic effector fate in trajectory 2 (from Branch Point 3 to C). It is important to note that γδ T cells with different Vγ/Vδ TCRs showed preferential, but not exclusive distribution in any of the transition nodes and effector end points. This observation is consistent with the TCR usage recently described for cell clustering analysis from scRNA-seq data of γδ T cells from two adult PBMCs and two cord blood samples (25).

Taken together, these results indicated that in chronic Mtb infection, the responding γδ T cell population (predominantly the Vδ1+ T cells) traversed distinct differentiation paths leading to cytolytic effector cell states. In contrast, most of the Vγ9+Vδ2+ T cells were transitory, poised to develop into effectors. These observations suggest that antigen encounter and environment cues determine the functional state of γδ T cells in the different stages of infection.

Increased frequency of circulating CD8+ γδ T cells is a feature of chronic inflammatory conditions of diverse etiologies

In keeping with the findings that CD8+ γδ T cells may be activated in various pathological conditions and that γδ T cells traverse distinct effector differentiation paths to provide infection stage-specific response, we asked if these CD8+ γδ T cells are a common feature of other persistent or chronic infectious and inflammatory conditions. We analyzed flow cytometry data of PBMCs from cohorts of chronic HIV (adults), chronic cardiovascular disease (older adults) and acute influenza (adults) infection as well as publicly available datasets (https://flowrepository.org) from cohorts of melanoma (chronic) [FR-FCM-ZY34] (44) and COVID-19 (acute) patients [FR-FCM-Z2XC]. We found that with striking consistency, circulating CD8+ γδ T cells increased across all cohorts presenting chronic or persistent inflammatory conditions, even though the frequency of total γδ T cells remained unchanged (Fig. 6, A). This trend was absent in donors with acute infections (Fig. 6, B). Consistent with this finding, higher levels of circulating CD8+ γδ T cells have been reported for a small group of HIV seropositive subjects (45), and in immunocompromised patients after allogeneic stem cell transplantation with CMV reactivation as well as congenitally CMV-infected newborns (10). These findings suggest that antigen persistence induces a distinct CD8+ γδ T cell effector program in different chronic conditions and shed new light on the role of antigen-specificity in γδ T cell differentiation and function in humans.

Figure 6. Increased frequency of circulating CD8+ γδ T cells associates with chronic inflammatory conditions of diverse etiologies.

(A, B) Percentages of total γδ T cells and CD8+ γδ T cells in different cohorts of chronic (A) and acute (B) inflammatory conditions compared to healthy controls. P-values were determined using the Mann-Whitney test. Lower and upper hinges of boxes represent 25th to 75^th percentiles, the central line represents the median, and the whiskers extend to the highest and lowest values.

Discussion

A better understanding of how lymphocytes control both the acute and chronic phases of the same infection is key to the understanding of how they provide immune protection overall and will aid the development of new intervention strategies and improved vaccines. γδ T cell response in active TB was one of the earliest studies documenting γδ T cell responses in a human disease (46). Here, we show that chronic Mtb infection leads to the expansion of a unique type of CD8+ γδ T cells and skews them towards an effector-memory phenotype with enhanced functionality, such as cytokine release and cytolytic response. These features share characteristics of the “memory inflation” described for the effector memory CD8+ αβ T cells developed in chronic CMV and other chronic infections (21). However, unlike those T cells that show a strong TCR-mediated response, the CD8+ γδ T cells identified here have attenuated TCR mediated responses, but express CD16 and have a robust ADCC response. This feature would allow antibodies, in addition to T cells, to contribute to the antigen specific response. In this context, it was reported that antibodies from individuals with persistent Mtb infection have unique Fc functional profiles that promote selective binding to CD16 and can effectively drive intracellular Mtb killing (47). CD16+ γδ T cells with ADCC potential are also induced in CMV infection and can cooperate with anti-CMV antibodies to drive anti-viral immunity (48). Additionally, CD16 expression is upregulated on cytotoxic Vδ1 but not Vγ9Vδ2 effector γδ T cells after repeated Plasmodium falciparum infection (49). These observations indicate that the development of CD16+ terminally differentiated cytolytic effector T cell response may be a general strategy for the control of chronic infections. Chronic, untreated HIV-1 infection has been associated with elevated numbers of CD45RA+CD57+ terminal effector CD8+ αβ T cells expressing CD16, and these CD16+ CD8+ αβ T cells mediate HIV-specific ADCC activity at levels comparable with NK cells on a per cell basis (50).

In fact, the NK-like functions adopted by the CD8+ γδ T cells in chronic or persistent infections may be driven by sustained antigenic stimulation. It has been reported that in vitro culture of PBMCs in the presence of IL-2 or IL-15 induces the expression of natural cytotoxicity receptors (NCRs) on Vδ1+ γδ T cells. These γδ T cells express high levels of granzyme B and show cytotoxicity against leukemic and other types of neoplastic cells. Importantly, this process requires prolonged stimulation through the TCR and a functional phosphatidylinositol 3-kinase (PI-3K)/AKT signaling pathway (51). We found that in comparison to CD8− γδ T cells, CD8+ γδ T cells showed enhanced AKT phosphorylation and expressed higher levels of genes enriched in the (PI-3K)/AKT pathway and the IL-2R/IL-15R common β chain (Fig. 2, B and Fig. S4). A recent study has shown that the leukemia reactivity of such in vitro expanded Vδ1+ γδ T cells is significantly more reliant on the NCR NKp30 than the TCR, further strengthening the point that γδ T cells adapt NK-like functions upon prolonged antigenic stimulation (52). In addition, a population of γδ T cells with NK cell functions has also been described in mice (53). These cells are characterized by very low cell surface TCR expression, but strong intracellular CD3 staining, suggesting ligand-induced TCR internalization. These observations indicate the importance of antigenic induction of γδ T cell functionality, even when the response is triggered through receptors other than the TCR.

Indeed, our study underscores the importance of antigen recognition in mounting a context-dependent γδ T cell response. We demonstrate that γδ T cells elicited in persistent or controlled Mtb infection are responsive to Mtb lysate but are distinct from the widely studied pAg-reactive Vγ9Vδ2 T cells induced in acute Mtb infection. Like Mtb, P. falciparum can also induce pAg (e.g., merozoite-derived HMBPP)-reactive Vγ9Vδ2 T cells upon initial infection. However, repeated infection induces clonally expanded, parasite- (but not pAg) reactive Vδ1+ effector T cells (49). These observations suggest that distinct γδ T cell subsets operate sequentially and mount stage-specific responses. Interestingly, both sets of γδ T cells show broad reactivities, but most likely through different mechanisms. Vγ9Vδ2 T cells can respond to pAgs, such as HMBPP, produced by cellular pathogens via the isoprenoid pathway. They can also respond to pAgs, such as IPP (isopentenyl pyrophosphate) and DMAPP (dimethylallyl pyrophosphate), produced in eukaryotic cells via the mevalonate pathway. This ability to respond to metabolites generated in proliferating cells allows Vγ9Vδ2 T cells to be activated by different tumor cell lines. The CD8+ γδ T cells also show reactivities to tumor cell lines, but each individual TCR responds to different sets of tumor cell lines. Our study also found that clonally expanded CD8+ γδ T cells present in children with no history of TB, or Mtb antigen exposure can respond to Mtb lysate, but not HMBPP. Given the similarity between γδ TCRs and immunoglobulins in antigen recognition and antigen specific repertoires (54), it is possible that some of these γδ TCRs are cross-reactive, and like cross-reactive antibodies, they may respond to a broad range of antigens. In this respect, we found increased circulating CD8+ γδ T cells to be a common feature of persistent or chronic infectious and inflammatory conditions. Defining what triggers this subset of γδ T cells, not only in chronic infection but also in atherosclerosis and cancer will be an important task going forward.

Methods

Study design

The overall objective of this study was to determine the role of γδ T cells in persistent Mtb infection. To accomplish this, peripheral blood samples from an adolescent South African cohort were used. PBMCs from individuals who were uninfected or with controlled Mtb infection, aged 13–18 years, were used for CyTOF, flow-cytometry, scTCR-seq, bulk-RNA-seq, and targeted scRNA-seq analyses. No sample size calculations were performed. Sample sizes were determined based on sample availability and were randomly assigned for the various experiments. Additionally, the same sample could not be assayed multiple times (i.e., replication) due to the limited number of cells available from each sample. The study was unblinded.

Cohort description

South African adolescent cohort study (ACS)

As previously described (13), all eligible adolescent donors themselves gave written informed assent while their parents/legal guardians gave written, informed consent. The study protocols were approved by the Human Research Ethics Committee of the University of Cape Town. Individuals were classified as Mtb-infected based on a positive QuantiFERON TB Gold In-tube assay (Qiagen; >0·35 IU/mL). All participants were healthy without signs or symptoms of active disease. Only adolescents who remained disease free for two years from the time of enrolment were included in the analysis. For some donors, PBMCs were collected at ~6 month intervals for a period of two years and were included as the longitudinal cohort. None of the donors had any other known chronic illness.

U.S. adult cohorts

(a) Chronic cardiovascular disease: As previously described (55), peripheral blood samples were collected from older patients (aged above 55 years) who underwent heart transplantation, and age-matched healthy controls. All eligible subjects gave their informed and written consent. The study protocol was approved by the Stanford Institutional Review Board. All donors were tested for a series of infections, including HIV and TB. None of the donors tested positive for HIV. A few donors had a history of TB, but these cases were excluded from this study. (b) Influenza virus infection: The cohort were adult patients attending the emergency department or the express outpatient clinic at Stanford with flu-like symptoms. Influenza infection was confirmed by nasal swab. Healthy controls were adult blood donors from the Stanford Blood Bank. The study protocol was approved by the Stanford Institutional Review Board. (c) HIV infection: All HIV-1-infected adults provided written informed consent before participation in the study (www.clinicaltrials.gov;NCT02018510), which was conducted in accordance with Good Clinical Practice. Peripheral blood samples at baseline (i.e., before drug injection) were analyzed for this study. The protocol was approved by the Federal Drug Administration in the USA, the Paul Ehrlich Institute in Germany, and the Institutional Review Boards at the Rockefeller University and the University of Cologne. (d) Healthy tonsils for organoid cultures: Tonsils from children undergoing surgery for obstructive sleep apnea or hypertrophy were collected after written informed consent was obtained from parents/legal guardians. Ethics approval was granted by the Stanford University Institutional Review Board.

Flow cytometry measurements and analysis

For all flow cytometry experiments, PBMCs were blocked with Human TruStain FcX (Biolegend) for 20 minutes on ice and then stained on ice for 30 minutes with different antibody cocktails, all of which included anti-CD3 (UCHT1), anti-TCRδ (5A6.E9), anti-CD8α (RPA-T8), and anti-CD19 (HIB19) antibodies. Anti-CD8β (2ST8.5H7) antibody was used to measure CD8αα+ or CD8αβ+ γδ T cells. Cells were washed with FACS buffer (PBS+ 1% FBS) and then analyzed on the LSRII or sorted using the BD Aria machines. Dead cells were defined as aqua-positive and were excluded from the analysis. All antibodies were validated by the manufacturers for flow cytometry application, as indicated on the manufacturer’s website. Data were analyzed using FlowJo version 10.2.

Bulk-RNA-sequencing and analysis

CD8+ and CD8− γδ T cells from previously cryopreserved PBMC samples were directly FACS sorted into TRIZOL reagent from 5 donors with controlled Mtb infection (1000–5000 cells/sample) for RNA extraction. cDNA was generated using the SMART-Seq v4 Ultra Low Input RNA Kit (Takara). The amplified cDNA was fragmented, and libraries were made using the Illumina Nextera-XT kit, following manufacturer’s instructions. Raw Illumina sequences were checked for quality using FastQC version 0.11.8 (Babraham Bioinformatics) and aligned to the GRCh38/hg38 Human genome (UCSC Genome Browser) using star RNASeq aligner version 2.5.4b. Resulting alignments were processed with the feature Counts software version 2.0.0 (subread) to obtain raw counts for each gene. The raw counts were then analyzed using DESeq2 version 3.10 (Bioconductor) to get differential expression data as well as normalized counts.

Mitochondrial mass detection

For the assessment of mitochondrial mass, we purchased the MitoTracker Green reagent from Invitrogen. The samples were stained according to manufacturer’s instructions. All cells were subsequently analyzed by flow cytometry.

In vitro stimulation of PBMCs with anti-CD3 antibody and an Mtb whole cell lysate

PBMCs were thawed in complete RPMI 1640 medium at 2×10⁶ cells per ml and recovered 12 hours before stimulation. PBMCs were stimulated with an Mtb-lysate (10μg/ml) (obtained from BEI Resources) or plate-bound anti-CD3 antibody (10μg/ml) for 14 hours, and then stained with anti-CD3 (UCHT1), anti-TCRδ (5A6.E9), anti-CD8 (RPA-T8), anti-CD19 (HIB19), and anti-CD69 (FN50) antibodies. All antibodies were purchased from BioLegend. Dead cells were stained using the LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (ThermoFisher Scientific).

FcγR-mediated antibody-dependent cell-mediated cytotoxicity (ADCC)

ADCC was carried out as previously described (56), with some modifications. Briefly, P815 cells (a mouse leukemic cell line) were incubated with 10μg/ml concentration of P815-specific monoclonal antibody, 2.4G2. Coated and uncoated P815 cells were then cocultured with FACS sorted and rested (24 hours) γδ T cells, or total PBMCs at an effector: target ratio of 1:10 in the presence of anti-CD107a (H4A3) antibody. After incubation, the cells were washed, stained, and CD107a degranulation was measured by cell acquisition on the LSRII machine.

Direct ex vivo single cell γδ TCR determination

Barcode-enabled direct ex vivo single cell TCR determination was carried out and analyzed as described (31). Briefly, the method involved single cell sorting of CD8+ and CD8− γδ T cells followed by reverse transcription and 3 progressive nested PCR reactions to amplify the CDR3 regions of both gamma and delta chains. The last PCR added barcodes to allow the identification of sequences derived from each individual cell studied. Amplified products from barcoded individual cells were combined and sequenced with the Illumina™ MiSeq™ platform. The resulting sequences were analyzed using VDJFasta, and the CDR3 nucleotide sequences were then extracted and translated. Human TCR sequencing primers are listed in Table S6.

CyTOF measurements and analysis

The antibody panels, staining protocols, and analysis methods used here have been thoroughly described in a previous study (13). Briefly, PBMCs from the South African adolescent cohort were stained with two panels – one measuring 25 surface markers and 12 cytokines/effector molecules (cytokine panel) (N=14 uninfected and 14 controlled Mtb infection), and the other measuring 27 surface markers and 13 signaling effectors (phospho panel) (N=10 uninfected and 10 controlled Mtb infection) (Table S7). All antibodies were validated by the manufacturers for mass cytometry applications (as indicated on the manufacturer’s datasheet, available online) and were conjugated using MAXPAR reagents (Fluidigm Inc.). Cisplatin and iridium intercalators were used to identify live and dead cells. We used palladium barcoding (Fluidigm Inc.) according to the manufacturer’s instructions. Cells were stained (1h; room temperature) and then fixed in 1.6% paraformaldehyde. For intracellular staining, cells were permeabilized in MeOH at −80°C overnight and then stained (1h; room temperature). Cells were acquired at approximately 500 events s⁻¹ on a CyTOF 2 instrument. Data analysis was done on the Cytobank website (http://www.cytobank.org). First, total γδ T cells were manually gated for each sample and then Citrus analysis was performed to identify stratifying subsets between the infected and control groups. To determine differences in cell subset abundances, we used the SAM algorithm in Citrus, which assesses the false discovery rate (FDR) by permutations. We also used the dimensionality reduction technique – viSNE for the visualization of similarities and heterogeneity across individual cells. Manually gated γδ T cells from all samples were first concatenated and then visualized using viSNE. The analysis was performed on Cytobank.

Cell surface expression of γδ TCRs by lentiviral transduction and stimulation assays

Lentiviral transduction was performed as previously described (57). Briefly, TCR γ and δ chain gene fragments were cloned into lentiviral constructs (nLV Dual Promoter EF-1a-MCS-PGK-Puro). For TCR expression, the TCR γ and δ chain constructs were transfected into 293X cells separately. The virus was collected after 72 hours of transfection and transduced into Jurkat α-/β-cells, which were selected for highest TCR expression by FACS sorting. 100μl TCR transduced Jurkat α-/β-cells (10⁶ per ml) were co-cultured with 100μl T2, K562, and Daudi cells (10⁶ per ml) in a 96-well plate. Plate-bound anti-CD3 stimulation was used as positive control. TCR transfectants in media only was used as negative control. HMBPP and Mtb lysate were used at 5μM and 10μg/ml concentrations, respectively. After 14-hour incubation, cells were collected and CD69 expression was measured with anti-CD69 antibody (FN50) using flow cytometry.

Tonsil organoid culture experiments

Tonsil organoids were established as previously described (38). Briefly, whole tonsils (overall healthy, without obvious signs of inflammation) were collected in saline after surgery and then immersed in an antimicrobial bath of Ham’s F12 medium (Gibco) containing Normocin (InvivoGen), penicillin and streptomycin for 1 hour at 4 °C for decontamination of the tissue. Tonsils were then briefly rinsed with PBS and manually disrupted into a suspension by processing through a 100-μm strainer with a syringe plunger and cryopreserved. Frozen cells were thawed, washed, enumerated, and then plated (6X10⁶ cells in 100 μl per well) into permeable (0.4-μm pore size) membranes placed in standard 12-well tissue-culture plates. Mtb-lysate (10μg/ml) was then directly added to the cells and cultured for 7 days. Organoids were harvested 7 days post stimulation, cells were washed and then used for flow cytometry analysis or single-cell sorting for TCR determination. The tonsils used to generate these organoid cultures were from children who were healthy, did not have TB, but underwent tonsillectomy for sleep apnea at the Stanford Hospital, CA, USA. These children had the standard vaccination regimen given in the US, which does not include BCG, and are therefore unlikely to have been exposed to Mtb antigens.

BD Rhapsody single-cell analysis

γδ T cells were FACS-sorted from six donors with controlled Mtb infection. Samples were stained with oligonucleotide-conjugated Sample Tags from the BD Human Single-Cell Multiplexing Kit in BD staining buffer following the manufacturer’s protocol. Barcoded samples were then washed and spun down at 350xg for 10 minutes and pooled. Pooled sample was then stained concurrently with a panel of 12 oligonucleotide-conjugated antibodies: anti-CD3 (SK7), anti-CD8 (RPA-T8), anti-CD16 (3G8), anti-CCR7 (2-L1-A), anti-CD62L (DREG-56), anti-CD45RA (HI100), anti-CD45RO (UCHL1), anti-CD158e1/KIR-NKB1 (DX9), anti-CD335 (9E2), anti-CD337/NKp30 (P30-15), anti-KIR-NKAT2 (DX27), and anti-NKp44 (P44-8) from BD. Staining was performed using the BD staining buffer for 30 minutes on ice, samples were then spun down at 350xg for 10 minutes and then washed three times. Pellet was resuspended in Rhapsody buffer for sort and capture. Cell capture and library preparation were completed using the BD Rhapsody Targeted mRNA and AbSeq Reagent kits. Briefly, cells were captured with beads in a microwell plate, followed by cell lysis, bead retrieval, cDNA synthesis, template switching, Klenow extension, and library preparation following the BD Rhapsody protocol in the Stanford Human Immune Monitoring Center. Libraries were prepared for T cell receptor, sample tags, targeted mRNA using the BD standard Immune Response panel, and AbSeq. Sequencing was completed on NovaSeq (Illumina, San Diego, CA) in the Stanford Genome Sequencing Service Center, and at Novogene (US Davis, CA).

Data were processed using the Seven Bridges Genomics online platform (San Francisco, CA) and BD Rhapsody Targeted Analysis Pipeline with V(D)J processing incorporated. After processing, data were imported into SeqGeq version 1.6.0 (BD, Ashland, OR). The import included a CSV file of all the data, and CSV files identifying the Sample Tag and V(D)J calls. Then the plug-in Lex BDSMK was run to separate out the Sample Tags, and the VDJ Explorer to identify clones. Data was further processed in Seurat pipeline(58) to annotate cell subsets by clustering algorithm, and remove contaminating B cells, alpha beta T cells, and some myeloid cells. The data was further cleaned of various outlier cells – these consisted of cells with CD22 mRNA counts greater than 1 (6 cells), and cells which had a gene with an mRNA count greater than 500 and were identified as outliers (204 cells). The final data set consisted of 24,888 cells with each cell having counts for 12 proteins and 399 transcripts.

Pseudotime Trajectories analysis

The data from Rhapsody single-cell analysis was transformed by replacing each count x by arcsinh(x/5). We refer to the D = 411 transformed coordinates (12 cell surface markers and 399 transcripts) for each cell as the feature coordinates and the D dimensional space as feature space. The data cells form a cloud of points in the feature space. Branching trajectories were then constructed using the Cytoskel package. Briefly Cytoskel constructs a k-nearest neighbor graph (k-NN graph) with edges connecting cells labeled by distances - dissimilarities between data points. This step captures the local and global structure of the cell cloud. Distances are simply the Euclidean distances between cells in the feature space. For the current data set we used k = 30 neighbors. Cytoskel then constructs a minimum spanning tree (MST) of the k-NN graph. The MST is the set of cell connections with the shortest total length which connect all cells. In the MST, each cell is only connected to the minimum set of cells with greatest similarity to itself such that all cells are connected.

Branch or trajectory construction then proceeds by first finding the two cells which are furthest apart as measured along the edges of the MST and constructing the graph path joining them. The next path segment is set of cells and edges from this first existing path to the cell furthest from the existing path. The process is then repeated a specified number of times. The result is a trajectory tree graph linking some subset of the data cells. An averaging step is carried out. The data cells are duplicated, and the coordinates of each duplicated cell are replaced by an average over near neighbors out to some distance in the MST. The averaged cells are referred to as pseudo-cells. The pseudo-cells which are part of the found trajectories are also referred to as trajectory cells. In the case of scRNA-seq data, this averaging is similar to imputation as done in MAGIC (59) and kNN smoothing (60).

The trajectories are represented as linked pseudo-cells. No assumption is made in the algorithm about the starting point of the trajectories. After trajectory construction the user can specify the starting branch. Given the starting branch, pseudo-time can be assigned to the trajectory pseudo-cells. Original data cells are associated with their closest pseudo-cell. Two dimensional plots of the pseudo-cell trajectories were constructed using metric multidimensional scaling (MDS) based on distances between trajectory cells in the original N dimensional feature space. This method creates a two-dimensional layout of the trajectory data points while trying to preserve as closely as possible the N dimensional distances between the data points. The cells in the plot are then colored by feature values of interest. This method clearly shows relationships between the trajectory branches and progression of features along the trajectories.

FlowSOM Clustering

We performed FlowSOM (61) meta-clustering on the data cells. FlowSOM first constructs a self-organizing map (SOM) which is a set of cell groups arranged on a (for example) 2D 10 by 10 grid such that the cells in each group are similar to each other and nearby groups are more similar than distant groups. The groups are then joined by a minimum spanning tree between groups. Finally, higher level clustering is performed on the groups forming meta-clusters each of which contains one or more low level groups. In the following we refer to the meta-clusters simply as clusters. The data cells were clustered into 10 meta-clusters using the FlowSOM R package. The clustering used the same distance calculation as the trajectory algorithm. For each cluster, the cluster mean was calculated for each gene. For the subset of markers of interest, a dataframe was constructed with each row containing the average values for each marker of interest for a given cluster. This dataframe was passed to the clustermap function of the Python seaborn plotting package with scaling set so that the maximum value in each column was scaled to a value of 1.0 to make the gene expression differences and similarities between the clusters clearer.

Statistical analysis

Data were analyzed by either Student’s t tests, one-way analyses of variance (ANOVA), significance analysis of microarray (SAM), Mann-Whitney U, or Wilcoxon ranked-sign tests, as indicated in the figure legends. For all comparisons, P values are shown in the figures. Error bars represent mean and 95% confidence intervals.

Data and materials availability

The bulk-RNA-seq and single-cell RNA-seq data have been deposited in the GEO repository under accession codes GSE216652 and GSE216654. The source code for Cytoskel analysis is available at https://zenodo.org/record/4818819#.Y_el9i-B3wY. All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.