Single-Cell Transcriptional Profiling Reveals Signatures of Helper, Effector, and Regulatory MAIT Cells during Homeostasis and Activation

Mucosal-associated invariant T (MAIT) cells are a subset of recirculating innate-like lymphocytes enriched in human liver, lung, and gut with highly variable abundance in peripheral blood in healthy donors (<1–18% in humans) (1–3). Most MAIT cells express an evolutionarily conserved TCR (TRAV1-2 in humans) that can recognize microbially derived, non-peptide small molecule metabolites of the vitamin B pathway (4, 5) presented by the oligomorphic MHC class Ib–related molecule, MR1 (5, 6). In contrast to conventional T cells that exit the thymus in a “naive” state and require peptide-specific Ag priming to proliferate, most mature MAIT cells enter the circulation as effector cells and are activated within hours of non-peptide vitamin B metabolite recognition (7, 8). This innate-like property confers their capacity to be among the first responders to bacterial and fungal pathogens (3, 9–11) as well as to viral infections through TCR-independent cytokine stimulation (12–14).

Despite a growing body of literature implicating MAIT cells in both homeostasis and disease (1), their specialized functional states in humans are poorly characterized relative to conventional T cells. For example, whereas conventional T cells are broadly divided into two lineages defined by CD4 or CD8 coreceptor expression and stable functional phenotypes, whether these same lineages can be defined in MAIT cells remains unclear. CD4⁺ MAIT cells are reported to express increased CD25 and IL-2 along with decreased cytotoxic markers relative to CD8⁺ MAIT cells (3, 12, 15, 16). Whereas CD8⁺ and CD4⁻CD8⁻ [double-negative (DN)] MAIT cells have similar cytotoxic phenotypes (16), the DN subset is reported to have depressed function with increased propensity to activation-induced apoptosis (17). Although prior studies have suggested that MAIT cell functional states may depend on the quality and duration of activating stimuli (12, 18), the early and late effects of TCR stimulation on MAIT cell state remain largely unknown.

Although prior studies have described MAIT cell heterogeneity using flow cytometry or bulk RNA sequencing (9, 17, 19), such approaches are limited in resolution and likely do not capture the full extent of MAIT cell diversity in humans. Although some of these other studies have characterized MAIT cells together with other innate T cell populations at a modest scale with single-cell transcriptomics (19–22), a dedicated atlas of both CD4⁺ and CD8⁺ mature MAIT cells under various activation conditions in healthy human donors remains lacking. Indeed, recent reports of the importance of MAIT cells in cancer (23) and COVID-19 pathogenesis (24, 25) necessitate a large-scale single-cell atlas of MAIT cell heterogeneity stimulated with their cognate ligands.

In this study, we present a single-cell transcriptomic atlas of human MAIT cells, comprising >76,000 cells from peripheral blood under early and prolonged activation with 5-OP-RU or anti-CD3/CD28 in the presence of IL-2 or IL-2/TGF-β. Our study elucidates distinct transcriptional programs of CD4⁺ and CD8⁺ MAIT cells and identifies novel MAIT cell functional clusters that we validate in human subjects recently exposed to Mycobacterium tuberculosis. Our results demonstrate broad functional heterogeneity of MAIT cells and suggest that these innate T cells can participate in complex immune responses in addition to their role in direct killing of pathogens.

Healthy donor PBMCs were obtained with written informed consent at Sloan Kettering Institute, Memorial Sloan-Kettering Cancer Center, Rockefeller University, or ATCC. De-identified blood samples were processed according to the protocols approved by the Institutional Review Board of Memorial Sloan-Kettering Cancer Center and Rockefeller University. As PBMC samples were de-identified, no demographic or clinical information is available. Tuberculosis (TB) household contacts and unexposed community controls were recruited with written informed consent at the GHESKIO centers, Port-au-Prince, Haiti (U01AI069421). Demographic and clinical variables of this cohort were previously described (3). Sample sizes for each experiment are described in the main text and figure legends. PBMCs were isolated from peripheral blood using Ficoll-Prep (GE Healthcare) or cell preservation tubes (Becton Dickinson) according to the manufacturer’s instructions and cryopreserved using 90% FBS (Life Technologies)/10% DMSO (Sigma-Aldrich) and stored in liquid N₂ prior to thawing for in vitro assays.

PBMCs were cultured for 15 h or 7 d in RPMI 1640 (Life Technologies) supplemented with 10% (v/v) heat-inactivated FBS, penicillin/streptomycin (100 U/ml), l-glutamine (2 mM), sodium pyruvate (1 mM), nonessential amino acids (0.1 mM), HEPES buffer (10 mM), and 2-ME (50 μM) (from Life Technologies or Sigma-Aldrich) at 37°C, 5% CO₂ in 96-well plates in the presence of IL-2 [Life Technologies (100 IU/ml) or Roche (1 ng/ml)]. Additional conditions included 1 ng/ml TGF-β (Life Technologies). Two micromolar 5-amino-6-d-ribitylaminouracil (5-A-RU) was added directly to 50 μM methylglyoxal to form 5-OP-RU on day 1 for in vitro stimulation assays. 5-A-RU was synthesized as previously described (2), resuspended in 200 μM aliquots, cryopreserved, and thawed as needed. Anti-CD3/CD28 beads (Life Technologies, 11161D) were incubated with PBMCs at a 1:2 bead/cell ratio for single-cell RNA sequencing (scRNA-seq) and flow cytometric validation assays. Refer to the Table I for a complete list of cell culture reagents.

Cells were washed once in FACS buffer then incubated with Fc receptor binder inhibitor polyclonal Abs (eBioscience, 14-9161-73) for 15 min at room temperature prior to staining. Extracellular staining was performed at room temperature for 15 min in 50 μl of FACS buffer (eBioscience, 00-4222-26), whereas intracellular staining was performed after 2 h of incubation in brefeldin A at 37°C followed by 45 min of incubation in 50 μl of permeabilization/fixation buffer (eBioscience; Foxp3/transcription factor staining buffer kit, 00-5523-00) followed by incubation for 1 h in 50 μl permeabilization buffer at 4°C. Refer to Table I for a complete list of Abs and flow cytometry reagents.

The FACS-sorted live-cell suspensions were encapsulated in microfluidic droplets using a Chromium instrument (10× Genomics) and reagents provided in a Single Cell 3′ Reagent kit (v3). The barcoded cDNA libraries were prepared according to the manual (CG00183 rev. B). The viability of cells [stained with 0.2% (w/v) trypan blue] prior to loading onto the Chromium instrument was 80–98%. Each sample was encapsulated at a final dilution of ∼100 cells/µl. Following the reverse transcription step, the emulsion droplets were broken and barcoded cDNA was purified with Dynabeads and amplified by 12 cycles of PCR: 98°C for 180 s, 12 times (98°C for 15 s, 67°C for 20 s, 72°C for 60 s), and 72°C for 60 s. The 50 ng of PCR-amplified barcoded cDNA was fragmented with the reagents provided in the kit, purified with SPRI beads, and resulting DNA library was ligated to the sequencing adapter followed by indexing PCR: 98°C for 45 s, 14 times (98°C for 20 s, 54°C for 30 s, 72°C for 20 s), and 72°C for 60 s. The final DNA library was double-size purified (0.6–0.8×) with SPRI beads and sequenced on the Illumina Nova-Seq platform (R1, 26 cycles; i7, 8 cycles; R2, 70 or more cycles) at a depth of 220 million reads per sample. BAM files and raw and normalized counts for all single-cell RNA sequencing data presented in this article have been submitted to Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE182239) under accession number GSE182239.

The flow cytometric gating strategy to identify MAIT cells using MR1-5-OP-RU tetramers generated by the National Institutes of Health tetramer core facility (2, 26) is summarized in Supplemental Fig. 1. The percentage of MAIT cells of total CD3⁺ cells and the absolute number of cells sorted per condition are presented in Supplemental Table I. Stained cells were sorted on a BD FACSAria sorter or analyzed on a BD LSRFortessa analyzer. Flow cytometry plots and analysis were performed using FCS Express version 7 (De Novo Software, Pasadena, CA). Graphs and statistical analysis of flow cytometry data were generated using Prism version 11 software (GraphPad Software, San Diego, CA). Absolute numbers were analyzed for experiments when the initial cell number per well was identical for each donor (200,000 PBMCs/well). Otherwise, the relative frequencies (%) are reported. Paired t tests, unpaired t tests, and Mann–Whitney tests were used to determine statistical significance between incubation conditions. Level of significance was p < 0.05. Adjusted p values for multiple t test comparisons were calculated by the Holm–Sidak method.

Preprocessing of scRNA-seq fastq files was conducted using the seqc pipeline as previously described (27). Following count matrix generation, cells with >20% of transcripts derived from mitochondrial genes were considered apoptotic and were thus excluded, consistent with prior work (27). Following this step, all mitochondrial genes were filtered out of the count matrix. Ribosomal genes and the noncoding RNAs NEAT1 and MALAT1 were excluded due to prior reports of strong influence on downstream clustering. Genes with a mean raw count <3.0 were removed from the analysis, and we restricted to KLRB1⁺ cells, which are a marker of MAIT cell lineage, and defined our FACS-sorting strategy for additional quality control, resulting in a final count matrix of 76,945 cells and 12,499 genes for downstream analysis. We used Seurat v2.3.4 to perform standard library size and log normalization. The mean library size was 6680 transcripts per cell.

To examine potential batch effects in the data (28), we computed the prevalence of each cluster out of all cells in each sample, followed by a centered log-ratio transformation for compositional data and principal-component analysis. This analysis demonstrated that samples clustered by time point, not batch (i.e., donor) (Supplemental Fig. 1E). Furthermore, pairwise Euclidean distances between samples from day 1 and day 7 were higher than pairwise distances within each time point (Supplemental Fig. 1F). When attempting batch correction, we detected no differences between cluster phenotypes at day 1 and day 7, a result inconsistent with prior studies showing differences between MAIT cells in response to stimulation at early and late time points. We further found that “overcorrection” of the data introduced spurious batch effects, for example, clustering of donor rather than time point. Therefore, we proceeded to analyze the data without batch correction.

Differential expression analyses were conducted using the FindMarkers() function in Seurat using a Wilcoxon test. Specifically, we compared each cluster to all other clusters (one versus all), and genes were considered signature genes for a cluster when the log fold change was >0 and the false discovery rate (FDR)–adjusted p was ≤0.05.

Signatures for conventional immune cells were derived from Bindea et al. (29). Signatures were also manually curated from the Kyoto Encyclopedia of Genes and Genomes and Reactome databases. For analyses in Supplemental Fig. 2, significantly differentially expressed genes for each cluster were used as input to the fgsea() package in Bioconductor. Associations were considered significant at a FDR-adjusted p of ≤0.05.

To simulate early and prolonged MAIT cell activation under Ag-specific and nonspecific TCR stimulation, we employed an in vitro activation assay using human PBMCs incubated with MR1 ligand (5-OP-RU) or anti-CD3/CD28 beads for 15 h (day 1; early) or 7 d (day 7; prolonged) compared with unstimulated control cells. As MR1-expressing cells such as B cells and monocytes are abundant within PBMCs, they can directly present 5-OP-RU to MAIT cells (2, 30, 31). A subset of conditions at day 7 also included TGF-β to further assess potential regulatory phenotypes (32, 33). Cells were harvested at each time point for flow cytometric assays and FACS using MR1 tetramers (Fig. 1A, Supplemental Fig. 1A–C, Supplemental Table I). To define the transcriptional landscape of MAIT cell states after early and prolonged activation, we performed droplet microfluidics scRNA-seq of these sorted MAIT cells from the peripheral blood of three donors in the activation conditions defined above. In total, 76,945 MAIT cells passed quality control and were used for downstream analysis (Supplemental Table I).

Although prior studies have demonstrated that MAIT cells can express CD4 or CD8 coreceptors (3, 15, 16), it is unclear whether these markers define functionally distinct MAIT cell populations. To address this question, we sequenced all MAIT cells per condition by sorting total CD3⁺MR1-5-OP-RU tetramer⁺CD161⁺⁺ populations (Supplemental Fig. 1A–C) and included incubation conditions favoring MAIT cell expansion in vitro (Supplemental Fig. 1B), so as to enrich for rarer MAIT cell subsets (e.g., CD4⁺ subset). Representative staining of CD4⁺/CD8⁺ coexpressing MAIT cell subsets are included in Supplemental Fig. 1D.

To test the hypothesis that CD4⁺ and CD8⁺ MAIT cells are composed of distinct transcriptional lineages, we first performed a differential expression analysis of our single-cell data set to compare CD4⁺CD8⁻ and CD4⁻CD8⁺ MAIT cells at day 1 (Fig. 1B, 1C) or day 7 (Fig. 1D, 1E) in the absence of TCR stimulation (Supplemental Data Set 1). At day 1 (Fig. 1C) CD4⁺ MAIT cells highly express genes associated with conventional naive and central memory T cells, including LEF1, SELL, and CCR7, but also markers of activation and costimulation such as CTLA4, IL2RA, TNFRSF4, and TNFRSF18 (Fig. 1B). In contrast, CD8⁺ MAIT cells express markers of cytotoxic activity such as NKG7, KLRD1, EOMES, GNLY, and GZMB as well as the transcription factor RORA. We confirmed the differential expression of CD25, OX40, GZMB, and EOMES at the protein level within CD4⁺, CD8⁺, DN, and CD4⁺CD8⁺ [double-positive (DP)] MAIT cell subsets using flow cytometry (Fig. 1C). At day 7, CD4⁺ MAIT cells upregulated CTLA4, TNFRSF4, SELL, IL2RA, and CCR7, whereas CD8⁺ MAIT cells upregulated NKG7, GZMA, GZMK, PRF1, and SLAMF7 as well as several genes involved in type 1 IFN signaling (IFI44L, IFITM3, IFI6, IFITM1, and ISG15) (Fig. 1D). We also confirmed the differential expression of OX40, CTLA4, GZMA, and GZMK in each subset at the protein level by flow cytometry (Fig. 1E). Our analysis demonstrates that CD4⁺ MAIT cells are associated with expression of costimulatory receptors, IL-2 signaling, and memory markers, whereas CD8⁺ MAIT cells are defined by granzyme-mediated cytotoxicity and type 1 IFN signaling. These results suggest that CD4 and CD8 expression on MAIT cells may define distinct functional subpopulations.

To fully characterize the phenotypic states of MR1-5-OP-RU tetramer⁺CD161⁺⁺ cells, we performed dimensionality reduction by principal-component analysis followed by Louvain clustering on cells from all donors, time points, and stimuli (Fig. 2A, Supplemental Data Set 2). Integration of all samples across all time points revealed that clusters were broadly partitioned by time point rather than donor (Supplemental Fig. 1E), consistent with earlier studies of both conventional T cells and MAIT cells demonstrating transcriptional changes over time in response to stimulation (1, 14, 18, 34–38). Accordingly, distances between samples within a time point were far lower than distances between samples from different time points, lending confidence to our integration approach (Supplemental Fig. 1F). In addition, samples from each donor were largely overlapping, with more heterogeneity observed by stimulus (Supplemental Fig. 1G) or time point (Supplemental Fig. 1H, 1I). Visualization with uniform manifold approximation and projection (UMAP) together with differential expression analysis displayed 12 clusters that could be distinguished by homeostatic, cytotoxic, Th1 helper–associated cytokine release, proliferating, tissue-infiltrating, immune regulatory, and exhausted states (Fig. 2B, 2C, Supplemental Data Set 2). Clusters (C)1–C3 were enriched at day 1 (early activation; (Fig. 2D, Supplemental Data Set 2). C1 had high expression of IFNG, GZMB, TNF, TBX21, and notably the transcription factor FOXP3. Because this cluster expressed genes associated with granzymes and Th1 helper–associated cytokines, it was termed “MAIT1 cytotoxic effector,” consistent with a similar cluster detected in Koay et al. (21). C2 demonstrated increased CD69 and pronounced expression of heat shock proteins, NR4A1, and the AP-1 subunits JUN and FOS and was named “early activated.” C3 expressed effector/central memory markers CCR7 and C62L, markers of MAIT cell intrathymic development CD4 and LEF1 (21), as well as RUNX3. This cluster also upregulated type 1 IFN signaling genes, e.g., IFIT3 and ISG15. Thus, C3 was named “ISG⁺ memory” [after IFN-stimulated genes [ISGs]). Collectively, these clusters define the landscape of MAIT cells during homeostasis and early activation. The rapid induction of cytotoxic effector programs at this time point suggests that MAIT cells may be poised to respond earlier than conventional T cells.

In contrast, C4–C12 were enriched with prolonged activation and demonstrated marked phenotypic heterogeneity (Fig. 2E, Supplemental Data Set 2). C4 and C5 (“homeostatic effectors”) displayed similar expression patterns, with high levels of IL7R, GZMM, GZMK, KLRG1, and IFI44L and decreased expression of CD69 and CD25. C4 was distinguished from C5 by expression of vascular endothelial growth factors VEGFA/B and was named “homeostatic effector VEGF.” C6 (degranulating effector”) showed high expression of LAMP1, cytokines TNF and TNFSF13B (B cell activating factor), IFNGR1, as well as the homing molecule CXCR3. C7 (“MR1⁺ MAIT17”) expressed GZMA, ENTPD1, RORA (21), as well as MR1 itself, suggesting that this population may interact with other MAIT cells. C8 (“tissue-infiltrating-like”) highly expressed ITGA1 and ITGAE and was only present in activation conditions including TGF-β. C9 was termed “innate-like” given its high expression of HOPX, KLRB1, and NCR3 together with GZMA and GZMK. C10–C12 were defined by a gradient of activation/exhaustion marker expression including CTLA4, PDCD1, and LAG3, with C12 (“late activated/exhausted”) resembling a terminally exhausted population. C11 (“proliferating”) demonstrated a strong mitotic phenotype. C10 (“cytotoxic/regulatory”) upregulated GZMA, GNLY, and IL26 as well as immune-modulating genes TGFB and LAG3. Taken together, these analyses define a highly diverse MAIT cell transcriptional landscape shaped by prolonged activation, with important parallels with conventional CD4⁺ and CD8⁺ T cells.

We next asked to what extent previously published markers associated with mouse and human MAIT cells identified in key prior studies (39–41) are represented within the heterogeneous collection of MAIT clusters we present in the current study. We thus analyzed a wide array of transcription factors, cytokines/chemokines and their receptors, effector molecules, NK receptors, and other markers previously shown to be expressed on MAIT cells, and plotted them on top of our clusters (Supplemental Fig. 2A). We detected expression of key transcription factors IKZF2, RORC, and EOMES as well as the chemokines CCR2, CCR5, and CXCR6 in clusters also expressing CD8. In contrast, CD4⁺ clusters demonstrated expression of IL2RA, CCR7, and SELL expression, consistent with our analysis directly comparing CD4⁺ with CD8⁺ MAIT cells (Fig. 1B).

We further asked whether markers typically associated with conventional CD4⁺ and CD8⁺ T cell subpopulations are expressed in MAIT cells (Supplemental Fig. 2B). We therefore manually curated markers that define key conventional T cell lineages that were not already represented in the heatmap in Supplemental Fig. 2A. This analysis revealed that FOXP3 was strongly expressed in C1, which also differentially expressed CD4. Moreover, we observed expression of key transcription factors that define the ontogeny and activation of conventional T cell lineages, for example, STAT1, RUNX3, PRDM1 (BLIMP1), and GATA3. We also found differential expression of activation/exhaustion markers, for example, HAVCR2 (TIM3), CTLA4, and PD1, in clusters where CD8 was differentially expressed. Consistent with this idea, we performed an additional pathway analysis using signatures of immune populations derived in Bindea et al. (29). This analysis revealed enrichment of NK and CD4⁺/CD8⁺ T cell signatures within our clusters, including differential enrichment of Th1 helper, Th2, central memory, and effector memory phenotypes (Supplemental Fig. 2C). We also performed pathway analyses using curated databases from Kyoto Encyclopedia of Genes and Genomes and Reactome (Supplemental Fig. 2D, 2E) that revealed enrichment of various antiviral immune response pathways (e.g., NOD-like receptor signaling), general effector pathways (e.g., TNF signaling), and pathways related to Ag presentation, for example, MHC class II protein complex binding. Given the recent association of MAIT cell responses with severity of SARS-CoV-2 infection, we also plotted markers of cytotoxicity found in patients with severe disease (25) on our clusters (Supplemental Fig. 2F). A subset of markers associated with severe COVID-19 disease were significantly upregulated in our clusters (Supplemental Fig. 2F). Specifically, GNLY (C3 and C10–C12), PRDM1 (C1, C2, C7, and C12), and IFITM1 (C3 and C12) were significantly differentially expressed in day 1 and day 7 clusters, whereas TXNIP (C4–C6) was more highly expressed in day 7 clusters. Taken together, we show that CD4⁺ and CD8⁺ MAIT cell subsets are transcriptionally distinct and can adopt similar programs to innate lymphocytes and conventional T cells.

We next asked how the prevalence of these populations varied across stimuli and donors during early and prolonged activation (Fig. 2D, 2E, Supplemental Fig. 1H, and 1I). In particular, C1 MAIT1 cytotoxic effector—which expressed high levels of IFNG, GZMB, and TNF—was enriched by TCR stimulation conditions with 5-OP-RU or anti-CD3/CD28 compared with cells incubated with IL-2 alone (Fig. 3A, 3B). To validate these early transcriptional phenotypes, we performed flow cytometric characterization of MAIT cells under the same early activation conditions, which confirmed the differential expression of common T cell surface activation markers IL-7R, CD69, and CD25 (Fig. 3C). Unexpectedly, IL-7R was significantly downregulated by 5-OP-RU, but not anti-CD3/CD28. CD69 and CD25 were significantly upregulated by both stimulation conditions, although CD69 was more highly expressed with anti-CD3/CD28. To confirm that these phenotypes were not driven by activation-induced apoptosis, we stained with early activation/apoptosis marker annexin V and late apoptosis marker, activated caspase-3/7 (Supplemental Fig. 3). Whereas annexin V staining was increased by 5-OP-RU or anti-CD3/CD28 stimulation (Supplemental Fig. 3A), no difference in caspase staining was observed between resting and 5-OP-RU conditions (Supplemental Fig. 3B–E). This suggests that the increased annexin V staining observed represents early activation rather than apoptosis (42). In sum, these results demonstrate that MAIT cells rapidly adopt distinct immune phenotypes after early Ag-specific and nonspecific TCR stimulation.

As GZMB, IFNG, and TNF were significantly upregulated in early activation C1 and C2, we next asked whether these markers were coexpressed at the protein level. A representative gating strategy of intracellular costaining for GZMB, IFNG, and TNF is presented alongside cumulative data from 30 healthy donors (Fig. 3D). These analyses confirmed the presence of a MAIT cell population with a combined granzyme and Th1 helper–associated cytokine phenotype enriched with early activation, consistent with MAIT1 cytotoxic C1 enriched in the anti-CD3/CD28 condition in the scRNA-seq analysis. The primary early activation response in MAIT cells was upregulation of granzyme B alone, which was increased in anti-CD3/CD28 relative to 5-OP-RU conditions (Fig. 3D). 5-OP-RU and anti-CD3/CD28 both induced significant coexpression of GZMB and TNF, although increased coexpression was observed with anti-CD3/CD28 (Fig. 3D). Only anti-CD3/CD28 induced IFNG and TNF coexpression (Fig. 3D). These results demonstrate increased expression of IFNG and TNF with nonspecific TCR stimulation, consistent with previous reports (3, 12, 43).

We next performed a differential expression analysis between MAIT cells stimulated with 5-OP-RU and anti-CD3/CD28 at day 1 to define the gene expression programs associated with Ag-specific or nonspecific TCR stimulation (Fig. 3E, Supplemental Data Set 3). Although both early TCR stimulation conditions induced similar transcriptional states (Fig. 2D), we did detect stimulus-specific gene expression patterns, with anti-CD3/CD28 robustly upregulating GZMB, PRF1, IFNG, IL2RA, BATF3, CTLA4, and LAG3. In contrast, 5-OP-RU induced a strong type 1 IFN signaling signature with upregulation of IFIT1, ISG15, IFI44L, IFIT3, IFITM1, and IFI6. These findings are consistent with distinct MAIT cell transcriptional programs driven by early Ag-specific signals versus nonspecific T cell stimulation.

Given detection of FOXP3 transcripts across clusters and significant differential expression in C1, we further characterized the transcriptional programs associated with MAIT cell FOXP3 expression using our single-cell data. We first performed a differential expression analysis between FOXP3⁺ and FOXP3⁻ MAIT cells in the resting condition at day 1 (Supplemental Fig. 4A, Supplemental Data Set 4) that revealed FOXP3⁺ MAIT cells differentially express CD4 in addition to markers of activation/exhaustion including TIGIT, ENTPD1, and ICOS, together with IL-2RA, a hallmark of the conventional regulatory T cell (Treg) lineage (44).

We next hypothesized that FOXP3 expression in MAIT cells was associated with acquisition of a transcriptome that might resemble conventional Tregs. To evaluate this, we assembled a list of four conventional Treg gene signatures published by Miragaia et al. (45). In this paper, the authors constructed gene signatures of four Treg populations spanning brachial and mesenteric lymph nodes, colon, and spleen, and established broad conservation of Treg identity between mice and humans. Using these gene signatures, we performed gene set enrichment analysis using the differentially expressed genes from Supplemental Fig. 4A and observed positive enrichment (normalized enrichment score > 0) in all four cases (Supplemental Fig. 4B, 4C, Supplemental Data Set 4). These analyses reveal that resting FOXP3⁺ MAIT cells differentially express CD4 and are transcriptionally similar to conventional CD4⁺ Tregs.

We then repeated these analyses using FOXP3⁺ MAIT cells in the 5-OP-RU condition at day 1. In this case, the cells expressed markers associated with activation and cytotoxicity, for example, CTLA4 and LTB, but were associated with neither CD4 nor CD8 (Supplemental Fig. 4D, Supplemental Data Set 4). In the anti-CD3/CD28 condition, FOXP3⁺ MAITs also differentially expressed CD4 as well as effector markers similar to the 5-OP-RU–activated condition (Supplemental Fig. 4E, Supplemental Data Set 4). In all conditions, the gene signatures of FOXP3⁺ MAITs overlapped with signatures of conventional Tregs from the four tissues evaluated in Miragaia et al. (45).

We then applied these analyses to day 7 data. FOXP3⁺ MAIT cells in the resting condition again demonstrated differential expression of CD4 (Supplemental Fig. 4F, Supplemental Data Set 4) and also expressed activation/effector markers such as GZMB, MKI67, HLA-DR, and TNFRSF18. In contrast, 5-OP-RU–activated FOXP3⁺ MAITs differentially expressed CD8 together with GZMB, TIGIT, NKG7, and GNLY (Supplemental Fig. 4G, Supplemental Data Set 4). FOXP3⁺ MAITs in the anti-CD3/CD28 condition were not associated with CD4 or CD8 expression (Supplemental Fig. 4H, Supplemental Data Set 4). As at day 1, gene signatures for FOXP3⁺ MAITs at day 7 were shared with the Miragaia et al. (45) signatures for conventional Tregs. These results demonstrate that FOXP3 expression on CD4⁺ or CD8⁺ MAIT cells differs based on antigenic versus nonspecific TCR stimuli. Our data also show that FOXP3⁺ MAITs at day 7 demonstrate a pronounced cytotoxic effector phenotype, similar to one recent report (23).

To validate the FOXP3⁺ MAIT cell phenotypes revealed by scRNA-seq, we performed intracellular staining for FOXP3 and flow cytometric analysis stratified by CD4 or CD8 coexpression (Fig. 4) compared with conventional Tregs. As FOXP3 is a transcriptional regulator of conventional CD4⁺ Tregs and is often coexpressed with CD25 (46, 47), we hypothesized that MAIT cells could share similar staining characteristics. We found that FOXP3 protein was significantly upregulated with Ag-specific stimulation (Fig. 4A, 4B). When we stratified MAIT cells by CD4 or CD8 coexpression at day 1, we observed that CD4⁺ MAIT cells express more FOXP3 at baseline whereas CD8⁺ MAIT cells significantly upregulated FOXP3 protein after 5-OP-RU stimulation (Fig. 4C).

As FOXP3 may be transiently upregulated after activation, we also profiled FOXP3 protein expression at day 7 (Fig. 4D–H). Whereas CD4⁺ MAIT cells had increased FOXP3 expression in the resting condition, consistent with day 1, 5-OP-RU–activated CD8⁺ MAIT cells robustly upregulated FOXP3 (Fig. 4D). We next compared FOXP3/CD25 coexpression in MAIT cells compared with Tregs and found that 5-OP-RU–activated MAIT cells strongly upregulated these markers of the conventional Treg lineage (Fig. 4E). Direct comparison of CD4 and CD8 expression in FOXP3⁺CD25⁺ by the stimulation condition revealed that coreceptor expression was dependent on the stimulus used (Fig. 4F), in agreement with the single-cell analysis (Supplemental Fig. 4). CD4⁺ MAIT cells expressed increased FOXP3 and CD25 in the resting condition relative to CD8⁺ MAIT cells but did not significantly upregulate these markers after prolonged activation (Fig. 4F). In contrast to CD4⁺ MAIT and conventional Tregs, FOXP3 and CD25 were robustly upregulated in the CD8⁺ MAIT cell subset after 5-OP-RU activation (Fig. 4F). The addition of TGF-β did not significantly affect MAIT cell FOXP3 expression. As previously reported (48), anti-CD3/CD28 stimulation alone resulted in significant Treg expansion (Fig. 4F). Analysis of absolute numbers of FOXP3⁺CD25⁺ MAIT cells at day 7 demonstrated that FOXP3⁺CD25⁺ MAIT cells were most prevalent within the CD8⁺ MAIT cell subset (Fig. 4G) and that the vast majority of FOXP3-expressing cells were detected among CD25⁺IL-7R⁻ cells and rarely among CD25⁻ cells (Fig. 4H). We also found that FOXP3 expression in CD8⁺ MAIT cells of some donors is sustained in some donors for 2 wk following a single Ag-specific stimulation, a phenotype that was dependent on MR1 (Fig. 4I). Taken together, these results demonstrate that FOXP3 is expressed in CD4⁺ and CD8⁺ MAIT cells and is associated with distinct stimulatory conditions for each subset. In our attempts to document the functional properties of FOXP3⁺ MAIT cells, we observed significant interindividual variability in suppression assays (data not shown), and thus we cannot draw any firm conclusions whether FOXP3 expression by MAIT cells is accompanied by suppressive capacity.

We next examined clusters enriched during prolonged activation (Fig. 5A). These clusters were broadly associated with expression of MKI67, GZMA, CTLA4, ICOS, LAG3, and PDCD1 (Figs. 2, 5A, 5B). Despite largely overlapping transcriptional states observed between prolonged 5-OP-RU and anti-CD3/CD28 conditions (Fig. 2E), 5-OP-RU significantly upregulated genes associated with lymphoid development (LTB), self-renewal (LST1), maturation (CD52), motility and proliferation (STMN1, ACTG1), and recruitment of immune cells (CCL3, CCL4). In contrast, anti-CD3/CD28 upregulated genes were associated with cytotoxicity (GZMB, GZMK), cell stress (HILPDA), and exhaustion (CTLA4, LAG3). This granzyme/exhaustion signature observed was largely driven by C12 defined by cells from donor 1 (Supplemental Fig. 1I).

The transcriptional differences between stimuli were further elucidated at the protein level, where proliferation marker Ki67 was significantly upregulated in the 5-OP-RU but not the anti-CD3/CD28 condition (Fig. 5C). This was consistent with robust Ag-specific MAIT cell expansion not observed with prolonged anti-CD3/CD28 (Fig. 5D, Supplemental Fig. 1B). In contrast, sorted MAIT cells alone stimulated with 5-OP-RU do not significantly expand (49), demonstrating the requirement for accessory cells for proliferation, similar to observations made in γδ T cells (50).

Similar to early activation, we again detected significant downregulation of IL-7R at the protein level only in the 5-OP-RU condition, which was associated with significant upregulation of CD25 not observed with prolonged anti-CD3/CD28 stimulation (Fig. 5E). Similar to early activation, GZMB was the dominant MAIT cell effector protein (Fig. 5F). Activation/exhaustion markers CTLA4, PD1, FAS, ICOS, and LAG3 expression levels were also enhanced with prolonged 5-OP-RU stimulation at the protein level (Fig. 5G).

To determine whether MAIT cell subtypes revealed by scRNA-seq are present during human infection, we assessed functional markers associated with CD4⁺ or CD8⁺ MAIT cells during human exposure to Mycobacterium tuberculosis. To this end, we used flow cytometry to analyze MAIT cell phenotypes in PBMCs collected from household contacts of active TB patients in Port-au-Prince, Haiti compared with healthy unexposed donors from the same community (3).

First, we focused our analysis on a subset of markers that characterized CD4⁺ MAIT cells in our single-cell analyses. We observed a similar differential expression pattern of CD25, CCR7, and OX40 at the protein level in CD4⁺ MAIT cells relative to other subsets (Fig. 6A–C). Moreover, each of these markers was significantly upregulated in contacts, indicating that the CD4⁺ MAIT cell gene expression pattern observed in our clusters in vitro corresponds to in vivo MAIT cell subsets in the context of recent M. tuberculosis exposure. Consistent with our previous work (3), these findings indicate that CD4⁺ MAIT cells respond to early M. tuberculosis exposure.

We next investigated MAIT cell FOXP3 expression, which was predominant in the CD8⁺ subset after in vitro prolonged antigenic stimulation (Fig. 4). We found that recently exposed contacts expressed increased FOXP3 relative to healthy controls (Fig. 6D). Consistent with our flow cytometry results at day 7 (Fig. 4G), FOXP3⁺CD25⁺ MAIT cells were most prevalent in the CD8⁺ subset, and FOXP3⁺CD25⁺CD8⁺ MAIT cells were significantly enriched in contacts (Fig. 6E). When stratifying by M. tuberculosis–specific IFN-γ release assay (IGRA) status, which distinguishes contacts who were exposed and latently infected (IGRA⁺) from those who remained uninfected (IGRA⁻), we observed that FOXP3⁺CD25⁺CD8⁺ MAIT cells were enriched in latently infected contacts (Fig. 6F), demonstrating that this MAIT cell subtype is responding to M. tuberculosis infection. This MAIT cell subpopulation detected in TB contacts closely resembles the CD8⁺FOXP3⁺CD25⁺ MAIT cells detected at the transcript and protein level at day 7 following Ag-specific stimulation (Fig. 4, Supplemental Fig. 4) and indicates that these cell states defined by single-cell transcriptomics are also present during early human M. tuberculosis exposure and infection.

MAIT cells have been well characterized as innate-like, rapid responders to microbial Ags with the capacity to mount cytotoxic responses and secrete Th1 helper/Th17-associated cytokines within hours of Ag recognition (3, 9, 39). However, less is known about the transcriptional programs that define MAIT cell adaptation during early and prolonged activation states. In this study, we provide a single-cell transcriptional atlas of >76,000 human MAIT cells in homeostatic and early/prolonged activation states coupled with flow cytometric and functional characterization. Our analysis demonstrates distinct transcriptional programs defined by incubation period and quality of stimulus, revealing novel MAIT cell subpopulations not yet described by bulk RNA-seq or traditional flow cytometric approaches.

First, our work contributes further evidence to support divergent transcriptional and functional phenotypes associated with human MAIT cell CD4 or CD8 coexpression. Several prior studies have used CD8 as a prerequisite for defining MAIT cells or did not distinguish between subsets (21, 22, 51). Differential expression analysis revealed that CD4⁺ MAIT cells upregulated IL7R, CD25, CCR7, and SELL along with TNF superfamily receptors TNFRSF4 and TNFRSF18, suggesting that CD4⁺ MAIT cells respond to distinct costimulatory signals from the CD8⁺ subset. We hypothesize that this CD4⁺ MAIT cell signature may represent unique tissue-homing functions through CCR7 and SELL. Future studies will help to evaluate whether this CD4⁺ MAIT cell signature confers the capacity to exit the circulation and migrate into tissues.

Recent work in mice and humans shows that CD4 and the transcription factor LEF1 are coexpressed during MAIT cell intrathymic development with reciprocal downregulation of CD319 (SLAMF7) (21). CD4 is subsequently downregulated as MAIT cells mature with reciprocal upregulation of CD8 and CD319 during thymic egress. In a parallel observation in human peripheral blood MAIT cells, we observed that CD4⁺ MAIT cells significantly upregulated LEF1, whereas SLAMF7 was significantly differentially expressed in CD8⁺ MAIT cells. Our findings motivate additional study of how MAIT cell activation conditions can select for CD4 or CD8 coexpression in peripheral circulating populations, and, further, how these subsets differentially respond to major pathogens. Indeed, our TB contact data demonstrate that CD25, CCR7, and OX40 are significantly upregulated in CD4⁺ MAIT cells and FOXP3 in CD8⁺ MAIT cells of TB contacts, suggesting that these subsets respond to early M. tuberculosis exposure.

The second aim of our study was to distinguish between transcriptional phenotypes induced during early and prolonged activation and to characterize functional programs associated with MR1 ligand versus nonspecific T cell stimulation. We observed that early activated MAIT cells with either stimulus were driven toward similar states defined by a rapid upregulation of GZMB, IFN-γ, and TNF-α. At both the transcript and protein levels, this signature was more potently induced by anti-CD3/CD28, which is possibly explained by nonspecific stimulation of all T cells within PBMCs triggering cytokine release by MAIT and non-MAIT T cells alike (52). Moreover, anti-CD3/CD28 may mimic other nonspecific T cell stimuli derived from whole bacteria or cytokines, which have previously been shown to significantly upregulate GZMB and IFN-γ in MAIT cells relative to 5-OP-RU alone (12, 43). We acknowledge that our study design could not identify the additional inflammatory signals induced by anti-CD3/CD28 stimulation compared with MR1 ligand (22) nor quantify MR1 expression by specific immune subsets within the PBMC aliquot. Future studies will identify which accessory cell types (53) and costimulatory conditions (38, 54) drive activation of specific MAIT cell subpopulations in healthy and disease states. In contrast to nonspecific TCR stimulation, early activation with 5-OP-RU induced a strong type 1 IFN–driven gene expression phenotype along with T cell maturation marker CD52. Costimulation of MAIT cells with 5-OP-RU and type 1 IFNs have previously been shown to enhance GZMB and IFN-γ release (54). Our results demonstrate that early Ag stimulation alone also enhances MAIT cell type 1 IFN signaling as detected at the transcript level in our scRNA-seq dataset, which may promote survival and proliferation observed during prolonged stimulation (55) as well as modulate immunity against bacterial or viral pathogens (56).

Although previous studies have used prolonged 5-OP-RU costimulation to expand MAIT cells for downstream use during in vitro functional assays (49, 57), to our knowledge, this is the first investigation of the transcriptional landscape of prolonged Ag-specific activation during this proliferative phase. Although we observed largely overlapping transcriptional signatures between 5-OP-RU and anti-CD3/CD28 activation conditions, 5-OP-RU significantly upregulated genes associated with T cell development, self-renewal, proliferation, and recruitment of inflammatory cells. In contrast, prolonged anti-CD3/CD28 activation upregulated genes associated with cytotoxicity, stress, and exhaustion. Taken together, these results demonstrate that Ag-specific stimulation is required for MAIT cell survival, expansion, and sustained effector function during the prolonged phase of stimulation.

Moreover, we describe a FOXP3⁺ peripheral blood MAIT cell subset enriched in early activation C1 with abundant FOXP3 detected at the protein level during prolonged 5-OP-RU stimulation. We observed that CD4⁺ MAIT cells expressed increased FOXP3 in the resting state relative to the CD8⁺ subset, similar to a recent study that detected tumor-associated FOXP3⁺CD4⁺ MAIT cells (23). We also demonstrate that peripheral FOXP3⁺ induction was detected almost exclusively in the CD8⁺CD25⁺IL-7R⁻ subset after Ag-specific induction. This surface coexpression of FOXP3 and CD25 in MAIT cells parallels that of CD4⁺ Tregs and raises the question of whether FOXP3 expression in MAIT cells may confer a similar immune regulatory function (32). We observed considerable interindividual variability in our MAIT cell suppression assays and thus cannot draw firm conclusions regarding MAIT cell regulatory function at this time. Further study is required to investigate the suppressive capacity of FOXP3⁺ MAIT cells. Importantly, we extend our findings of MAIT cell FOXP3 induction in vitro to immune responses of donors recently exposed to M. tuberculosis. We found that CD8⁺FOXP3⁺CD25⁺ MAIT cells expanded after early M. tuberculosis exposure in healthy household TB contacts and were significantly more abundant in recently infected donors (IGRA⁺ contacts). Our results suggest that CD8⁺FOXP3⁺CD25⁺ MAIT cells expand during early M. tuberculosis infection.

Our study also highlights that MAIT cell activation rapidly drives expression of phenotypic exhaustion markers. This was evident in both Ag and nonspecific stimulation conditions at the transcriptional and protein levels. These findings mirror evidence from chronic disease states such as TB, viral hepatitis, and malignancy where increased MAIT cell PD1 expression is associated with peripheral MAIT cell depletion and dysfunction and directly correlates with severity of disease in some studies (11, 58–64). Although immune checkpoint molecules can be transiently upregulated after T cell activation, we observed sustained expression in prolonged activation clusters. At the protein level, these exhaustion markers were significantly upregulated in the 5-OP-RU condition, suggesting that prolonged Ag-specific activation strongly induces negative feedback mechanisms to suppress further activation and proliferation. Taken together, these data imply that checkpoint blockade may enhance MAIT cell activity against infectious or malignant disease (57, 65, 66) or contribute to autoimmune complications of immunotherapy (67).

Given the recently described importance of MAIT cells in SARS-CoV-2 pathogenesis (24, 25, 68), we also show that MAIT cell transcriptional markers in severe COVID-19 disease are expressed in distinct MAIT cell clusters defined in the present study (Supplemental Fig. 2F). The MAIT cell subtypes identified in this study may provide additional refinement into the pathologic role of MAIT cell subpopulations in COVID-19 disease severity.

We acknowledge certain limitations to our study. The foremost of these is the small number of individuals in our single-cell cohort and donor-to-donor variability, some of which is inherent in human studies from genetically diverse subjects. However, the functional flow cytometric validation in multiple donors, including a genetically distinct population from Haiti, strengthens our confidence that the MAIT cell states observed can be applied to other populations. Future studies will test the reproducibility of this transcriptional atlas in additional healthy donors or in specific disease states.

In sum, we present an integrated transcriptional and immune landscape of human MAIT cell activation at single-cell resolution. Our results provide a detailed analysis of novel MAIT cell clusters defined by early and prolonged stimulation, including distinct CD4⁺ and CD8⁺ MAIT cell signatures. We also show that these phenotypes are recapitulated in the setting of early M. tuberculosis exposure and infection. This functional map of MAIT cell activation biology will help inform future investigations into how MAIT cells specialize in healthy and disease states.

We thank Ojasvi Chaudhury for assistance in preparation of single-cell RNA-seq libraries. We acknowledge staff at the Immune Monitoring, Flow Cytometry, and Integrated Genomics Operation Core Facilities at the Sloan Kettering Institute, Memorial Sloan-Kettering Cancer (New York, NY) for expert consultation and services rendered during this project. We also acknowledge the National Institutes of Health Tetramer Core (Emory University, Atlanta, GA) for providing MR1 tetramers.

M.S.G. reports consulting fees and equity in Vedanta Biosciences, Inc., is a consultant for Fimbrion, and is on the Scientific Advisory Board of PRL-NYC. The other authors have no financial conflicts of interest.