Immune checkpoint expression in mucosal-associated invariant T cells is stimulus-dependent

Abstract

Mucosal-associated invariant T (MAIT) cells are unique unconventional T cells with diverse roles in immunity, yet how their context informs their function is not well known. This contextual regulation is particularly relevant in cancer, where MAIT cells have an enigmatic role. We performed a systematic review and meta-analysis to identify MAIT cell transcriptomic signatures under different environmental conditions. We identified 4 bulk-RNA-sequencing studies that compared multiple activation stimuli. We found a stimulus-specific transcriptional signature for immune checkpoint genes, that we confirmed in a single-cell RNA-sequencing dataset. We used flow cytometry to examine in vitro human MAIT cells across 4 activation stimuli and confirmed that stimulus drives unique checkpoint signatures upon MAIT activation for Lag3, PD-L1, PD-1, NKG2A, and Tigit. Strikingly, PD-L1 was more highly induced in vitro than PD-1 in MAIT cells up to 72 h. Our data suggest that human MAIT cell regulation is context-dependent. These findings have the potential to critically inform efforts targeting MAIT cells for cancer immunotherapy.

Human MAIT cells upregulate Lag3 and PD-L1 to the cell surface upon activation, as well as PD-1, Tigit, and NKG2A in a stimulus- and kinetics-dependent manner.

1. Introduction

Mucosal-associated invariant T (MAIT) cells are unique unconventional T cells that exhibit both innate and adaptive functions. They are abundant in mucosal sites in humans, such as the skin, intestine, and lungs, as well as metabolically relevant tissues such as the liver, where they account for as many as 45% of T cells^1–3. They respond to a small subset of B-vitamin antigens presented by the major histocompatibility complex, class I-related (MR1) receptor, predominantly microbial metabolites of riboflavin (vitamin B2). In vivo, they can also be activated by the innate cytokines IL-12 and IL-18 independently of receptor engagement.⁴ Through these mechanisms, MAIT cells are primed to be responsive to either microbial or inflammatory signals. Microbial signals, however, are necessary for MAIT cell maturation, as they do not expand meaningfully in germ-free animals.⁵

MAIT cells take part in a broad variety of functions throughout the body, raising questions about how their function is informed by their context. In humans, they express a strong Th17 signature while simultaneously expressing traditional Th1 cytokines such as interferon gamma (IFNγ).^1,6 They are capable of cytotoxic functions, such as the deployment of perforin and granzyme, but are also implicated in wound repair and tissue maintenance^6–8. The diversity of their capabilities enables them to be adept responders for a wide variety of functions including pathogen clearance, maintenance of tissue homeostasis, and tumor immunity. However, how the surrounding environment dictates MAIT cell function remains incompletely understood.

Despite the promising capacity of MAIT cells to provide cancer immunity due to their rapid cytotoxic activity and lack of risk for an allogeneic mismatch,⁹ studies of the response of MAIT cells in cancer have shown conflicting results. Studies differ on whether the presence of MAIT cells is favorable or disadvantageous for patient prognosis.^10,11 A meta-analysis of over 18,000 individual tumors from 39 different cancer types showed that CD161, a marker of MAIT, NK, and iNKT cells, was consistently associated with better patient outcomes. However, tumor-associated MAIT cells from patients with hepatocellular carcinoma show an exhausted phenotype and are associated with poor prognosis.¹²

MAIT cells are nonetheless regarded as potential therapeutic targets in light of their expression of immune checkpoint genes and ligands. They have been shown to express PD-1, CTLA-4, and TIM-3 in various malignancies,^12–14 and have been shown to positively respond to PD-1 blockade.^15,16 However, the contextual factors that impact MAIT cell checkpoint expression have not been deeply explored. Although MAIT cells share many expression patterns with conventional lymphocytes, particularly CD8+ T cells, they are distinct in their innateness and semi-invariant phenotype. Their existence at the interface of innate and adaptive immunity may underpin a unique checkpoint signature as well.

In this study, our primary objective was to comprehensively compare MAIT cell transcriptional activity during varying MAIT cell activation contexts. To this end, we used a systematic survey of the literature to identify bulk RNA-seq studies with unique activation stimuli, from which we identified a common transcriptional program for activated MAIT cells, as well as stimulus-specific upregulation of checkpoint genes. We validated context-dependent variations in MAIT cell checkpoint gene expression at the single-cell level, by tissue compartment, and finally in vitro at the cell surface.

2. Materials and methods

2.1. Study design

We selected studies in a 4-tier process of initial search, abstract review, full text review and final selection (Fig. S1). We initially identified eligible studies as transcriptomic studies of MAIT cells from mice and humans using either bulk or single-cell RNA-sequencing. We excluded studies if no RNA-seq of whole MAIT cells was performed, if sequencing was not performed on an isolated population of MAIT cells, if MAIT cells were genetically modified or treated with drugs other than activation stimuli, and if raw data were not publicly available. We did not consider studies that only included the MAIT cell receptor repertoire to be reflective of whole-cell transcriptomics and excluded them. Upon collection of an initial pool of papers, we scrutinized studies for comparable variables across species, tissue location, and method (bulk vs single-cell). We made final selections based on the comparability of 2 or more studies across a given variable from bulk RNA-sequencing, which has greater depth than single-cell sequencing. We selected a single-cell study from eligible candidates that addressed both activation stimulus and tissue localization to validate findings from the bulk RNA-seq analysis. Due to complications with the formatting of publicly submitted single-cell RNA-seq data, we did not combine single-cell studies on account of batch effects, and analyzed some data from submitted count matrices rather than raw fastq files. Together, this curated dataset, compiled from multiple independent studies, enhances the robustness of our interpretations and provides a foundation for studying the transcriptional programs underlying MAIT cell identity and activation.

2.2. Bulk RNA-seq analysis

We generated count matrices within the University of Utah Center for High Performance Computing (CHPC) cluster environment. We downloaded raw fastq files for bulk RNA-seq studies from the NIH Gene Expression Omnibus (GEO) using SRA Toolkit,¹⁷ after which they were trimmed and quality filtered using Trim Galore.¹⁸ We set minimum transcript lengths to 10 base pairs to accommodate the possibility of short transcripts such as microRNAs. We then aligned transcripts to the genome index using STAR.¹⁹ All bulk RNA-seq studies were performed in humans and aligned to the Homo sapiens GRCh38 genome index. We generated final count matrices from the STAR alignment using featureCounts.²⁰ From there, count matrices were input into R (version 4.4.1) and analyzed downstream with DESeq2,²¹ with analysis and visualization support from clusterProfiler,²² pheatmap,²³ paletteer,²⁴ ggVennDiagram,²⁵ EnhancedVolcano,²⁶ RColorBrewer,²⁷ and tidyverse.²⁸

We first calculated differential gene expression using the DESeq2 pipeline within individual studies, followed by a meta-analysis of significantly expressed genes across studies (adjusted P-value < 0.05). For comparisons between stimuli, significant genes had to be significant in both studies for a given stimulus in order to be included. We assessed shared and unique genes and gene ontology pathways from DESeq2 output.

To identify common gene expression patterns between stimuli, we first analyzed each study individually for significantly differentially expressed genes (DEGs) relative to the unstimulated condition (unstim), and then compiled all of the DEGs that were significant (adjusted P-value < 0.05) for each stimulus in both studies.

For specific gene comparisons, including checkpoints, we normalized raw counts against the unstimulated condition within each study prior to combining into a joint analysis. Both raw fold change and log₂-transformed fold change are shown, and are labeled as such.

2.3. Single-cell RNA-seq analysis

We obtained single-cell RNA-seq raw data from the NIH GEO beginning from either publicly available matrices or from raw fastq files. Inconsistencies in the format of uploaded data limited the number of studies employed. We uploaded matrix data from relevant samples for analysis with the R package Seurat (version 5).²⁹ We quality filtered cells that had a minimum of 200 counts per cell, less than 9,000 features per cell, and less than 5% mitochondrial DNA per cell. We then normalized, scaled, and subjected the data to dimensionality reduction via principal coordinate analysis. Thereafter, we used the Jack straw and elbow plot tools to optimize the most appropriate clustering. We generated UMAPs on the selected dimensionality and identified markers for each cluster. We explored data visually using tools from within the Seurat and Nebulosa³⁰ packages, including heatmaps, dotplots, violin plots, and density plots.

2.4. In vitro MAIT cell activation

We obtained blood from healthy human donors from the ARUP blood bank at the University of Utah. All donors gave informed consent for their samples to be used for research, and their samples were fully deidentified prior to receipt in the lab. Peripheral blood mononuclear cells (PBMCs) were isolated using a Ficoll gradient and frozen at −80 °C until use. All cells were used within a month of storage. For activation, we thawed PBMCs, washed them several times in RPMI 1640 medium with 10% FBS and 1% penicillin/streptomycin (R10), and counted the cells. We stimulated 1 million cells per well for 6, 24, 48, or 72 h with 1 of 4 stimulations: 2 µM 5-amino-6-D-ribitylaminouracil (5-ARU) and 50 µM methylglyoxal, fixed E. coli BL21 at an MOI of 10 (1 × 10⁷ CFU per well), 50 ng/mL of IL-12 and 50 ng/mL of IL-18, or 1 µL/mL ImmunoCult CD3/CD28 T cell activator (Table S2). Stimulations were selected to recapitulate what had been shown in our in silico analysis, with the addition of 2 longer time points. We prepared all stimulations in R10 medium and incubated at 37 °C + 5% CO₂ for the duration of activation.

To prepare fixed E. coli, we grew a single colony isolated from a streak plate overnight in LB broth shaking at 37 °C. After adjusting to an OD₆₀₀ of 0.1 and adapting the volume to a precalculated MOI of 10, we fixed bacteria in 1% paraformaldehyde for 10 min at room temperature and washed several times in sterile PBS prior to use.

2.5. Flow cytometry and staining

At the end of the activation time, we washed cells in sterile PBS and stained with 1:2,000 Zombie Aqua viability dye for 20 min at 4 °C in the dark. We then washed them in FACS buffer (PBS + 2% FBS) and stained with surface stain antibodies for 40 min in the dark at room temperature (Table S3). Cells were washed again in FACS and fixed for 45 min in the dark at room temperature. We washed the fixed cells, resuspended them in FACS, and stored overnight in the dark at 4 °C until data acquisition on the Cytek Aurora spectral analyzer provided by the University of Utah Flow Cytometry Core. We analyzed flow cytometry data in FlowJo (version 10.10.0), with quantitative analysis completed in Excel and GraphPad Prism (version 10.4.1). MAIT cells were defined as Lymphocytes/Single Cells/Live Cells/CD3⁺MR1tetramer⁺ cells, with flow gates selected from the use of fluorescence minus one (FMO) controls and a 6-formylpterin (6-FP) control for the MR1 tetramer (Fig. S2).

2.6. Statistical analysis

We assessed significance for bulk RNA-seq analyses performed in DESeq2 using the Wald test with the Benjamini and Hochberg method to adjust for multiple testing, as included in the pipeline. The P-adjusted value (P_adj) was used for significance cutoffs (P_adj < 0.05). For investigation of individual genes from raw and normalized counts, we employed a nonparametric ANOVA (Kruskal–Wallis test) with multiple comparisons to compare fold-change values from all stimulus conditions. We also used the Kruskal–Wallis test to assess flow cytometry data to compare stimulated conditions to the unstimulated condition. To evaluate the differences between PD-L1 and PD-1 expression, we used a Wilcoxon matched-pairs rank sum test at each time point. In all cases, we set the significance threshold to P < 0.05. Asterisks are as follows: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

3. Results

3.1. Differentially stimulated MAIT cells enact a shared core transcriptional program

To examine the response of MAIT cells under different contextual stimuli, we undertook a systematic review of the literature that yielded 4 bulk RNA-seq studies of MAIT cells isolated from human PBMCs with comparable activation parameters. Across all 4 studies, there were 4 activation methods that were shared with at least 1 other study: (i) the MAIT ligand 5-(2-oxopropylideneamino)-6-D-ribitylaminouracil (5-OP-RU), generated by combining its precursor 5-ARU with methylglyoxal (n = 13), (ii) the cytokines IL-12 and IL-18 (n = 10), (iii) E. coli exposure (n = 12), or (iv) anti-CD3 anti-CD28^31–34 (n = 8) (Fig. S3A, Table S1). Each stimulus was represented by 2 studies with 3 to 7 samples per group in each study (Table S1). All studies also had an unstimulated control, which enabled baseline fold change calculations for each stimulus.

We found 245 genes that were significantly differentially expressed across all stimuli and studies, representing a core transcriptional activation signature (Fig. 1A and B, Table S4). Though expression patterns were broadly similar, the canonical MAIT ligand 5-OP-RU induced a higher fold change for many genes (Fig. 1B). We also found stimuli-specific DEGs that were unique to each condition, meaning that they were not significant for both studies of any other stimulus: 1,016 for 5-OP-RU, 1,002 for CD3-CD28, 495 for E. coli, and 141 for cytokine (Fig. 1A).

Fig. 1.

MAIT cells share a core transcriptional activation program that varies in magnitude by stimulus. Significant differentially expressed genes (DEGs) from the unstimulated condition in both studies were compiled for each stimulus (A). Of those, 245 genes were significantly differentially expressed for all studies and stimuli as a core transcriptional activation program, here shown as a heatmap of log₂ fold change (B). A subset of these genes varied substantially in magnitude by 5-OP-RU activation, and are visualized here as those genes for which 5-OP-RU prompted a log₂ fold change of ≥6 (C). Top upregulated genes that were significantly differentially expressed in all conditions are shown stratified by significance and log₂ fold change (D). Gene ontology predictably identified top upregulated pathways in the core transcriptional program to be dominated by an inflammatory immune response, along with signals of cell death (E).

Among the 245 shared significant DEGs, we found that Csf2 (Gm-csf), Fosl1, Ccl20, Ccl3 (Mip-1α), microRNA-155 (miR-155), Tnfrsf9, C2cd4a, Tnf, Mfsd2a, C2cd4b, and Nipal4 had a log₂ fold change of at least 6 for 5-OP-RU (Fig. 1C). Among all shared genes, the most significantly induced irrespective of log₂ fold change were Lta, Ccl20, Sema7a, Tnf, Ndfip2, and Mthfd2 (Fig. 1D). Other highly upregulated genes included Batf3, Slc1a5, Pdcd1 (PD-1), and Nfkb2.

To assess broader signaling pathways shared by all stimuli, we performed a gene ontology analysis on the pool of 245 shared significant genes from all stimuli (Fig. 1E). Pathways shared by all stimuli prominently featured cellular mechanics associated with rapid proliferation, such as transcription, translation, and protein folding. We observed some variation in the average adjusted P-value for significant genes by study, possibly influenced by biological or technical variation, or the number of samples included (Fig. S3B, Table S1). The distribution of log₂ fold change for each study was comparable (Fig. S3C).

Together, these data show that there are many genes core to the activation program of MAIT cells regardless of the activation stimuli. Many of these genes and their associated pathways are relevant for rapid proliferation, as expected for activated immune cells. The magnitude of gene expression is influenced by the specific activation stimulus.

3.2. Differential activation stimuli induce unique transcriptional programs in MAIT cells

In order to better understand the transcriptional programs for each activation stimulus, we first identified the DE genes between activation and unstimulated MAIT cells for each stimulus from each study with adjusted P-values in the lowest fifth percentile, and then combined genes for each stimulus that met this criterion for each study of a given stimulus. This gave us the most highly significant DE genes for each stimulus represented by more than 1 study, amounting to 54 genes for 5-OP-RU, 30 genes for cytokine stimulation, 45 genes for E. coli, and 61 genes for antiCD3-CD28 (Fig. 2, Table S5).

Fig. 2.

MAIT cells show stimulus-specific transcriptional programs. Studies were compiled by activation stimulus to include only genes that were significant (adjusted P-value < 0.05) for both studies per stimulus. Volcano plots (A to D) are representative of all shared significant genes for each stimulus. DEGs were then further filtered on the most highly significant genes by selecting the bottom fifth percentile of adjusted P-value per stimulus. Heatmaps (A to D) show up to the top 15 log₂ fold change DEGs from this list, along with any strongly downregulated genes from the same list. For the full table of most significant DE genes per stimulus, see Table S5.

The MAIT ligand 5-OP-RU strongly upregulated transcription signatures of both Il17a and Ifng, as well as multiple chemokines such as Ccl4l2, Ccl3, Ccl3l1, and Ccl4, and amino acid transporters Slc7a5 and Slc1a5 (Fig. 2, Table S5). Cytokine stimulation with IL-12 and IL-18, however, favored an Ifng response, with additional upregulation of granzyme (Gzmb), Il26, and the chemokine Ccl20 but not Il17a. Cytokine stimulation also promoted a checkpoint signature including strong upregulation of both Ctla4 and Cd274 (PD-L1). E. coli also strongly induced Ifng and Gzmb expression, as well as interferon regulatory factors Irf4 and Irf8, NF-kB-regulatory genes, and Cd274. Anti-CD3-CD28 stimulation promoted a variety of genes, including Csf2 (Gm-csf), Irf4, chemokines Ccl4 and Ccl20, and others.

To narrow in on genes that were unique to each stimulus, without overlapping expression among other stimuli, we compiled all shared genes that were significant for each stimulus, and then pared down to genes that were only represented by 1 stimulus (Fig. S4, Table S6). This included over 1,000 genes for 5-OP-RU and anti-CD3-CD28 activation, 495 for E. coli, and 141 for cytokine exposure, although base expression of many genes was quite low. The unique gene expression profiles for each stimulus underline both the potential for functional diversity as well as functional homology. While each unique transcriptional signature contained genes associated with proliferation, cell division, and metabolic function, it is noteworthy that stimuli can activate unique genes within these pathways. For example, both E. coli and anti-CD3-CD28 significantly induced the intrinsic apoptosis pathway, but from 14 to 17 nonoverlapping genes, respectively (data not shown). Similarly, both cytokine and anti-CD3-CD28 induced mitochondrial pathways, though different ones (Fig. S4). At the same time, some functions did not overlap with other stimuli, such as Fc receptor expression induced by cytokine exposure (Fcmr), amphiregulin (Areg) signaling for E. coli, or succinate metabolism for 5-OP-RU. Once again, a unique checkpoint signature arose with Havcr2 (Tim3) significantly induced by cytokine exposure alone.

To look more broadly at cell signaling pathways affiliated with unique gene signatures, we performed a gene ontology analysis on the significant gene sets that were unique to each stimulus (Fig. S4). Despite 1,016 unique genes for 5-OP-RU activation, our analysis identified only 6 significant pathways, the most significant of which was the caspase complex. Cytokine exposure had 8 unique pathways, many of which were largely focused on mitochondrial dynamics and cell cycle. E. coli exposure prompted many pathogen response gene pathways, including those belonging to viral response gene sets. Anti-CD3-CD28 exposure prompted a number of ribosomal metabolism pathways, as well as multiple mitochondrial-associated pathways.

3.3. MAIT cells express unique checkpoint gene signatures depending on their activation context

Given the prominence of checkpoint genes and ligands among DEGs, we hypothesized that MAIT cells express unique checkpoint gene signatures depending on their activation context. We thus focused on examination of stimulus-specific gene expression of the checkpoint genes Ctla4, Pdcd1 (PD-1), Havcr2 (Tim3), Lag3, Tigit, Klrc1 (NKG2A), Vsir (Vista), and the ligand Cd274 (PD-L1).

At the transcriptional level, we observed clear differences for each checkpoint gene by activation stimulus (Fig. 3A). The MAIT ligand 5-OP-RU significantly upregulated Pdcd1, Cd274, Vsir, and the NK-associated checkpoint Klrc1 (Fig. 3B to E). By contrast, MAIT receptor-independent stimulation with cytokine significantly upregulated Ctla4, Pdcd1, Cd274, Havcr2, and Lag3 (Fig. 3B, C, F, G, and H). Stimulation with E. coli appeared to provide an intermediate transcriptional signature, with significant upregulation of Ctla4, Pdcd1, Cd274, Havcr2, Lag3, and Klrc1 with a nonsignificant trend for Vsir. This shared phenotype may reflect that E. coli stimulation presumably involves both stimulation of the MAIT ligand via antigen presentation and cytokine stimulation via inflammation in the milieu. Anti-CD3-CD28 stimulation generally had a lower magnitude of upregulation for most checkpoint genes studied, but was nonetheless significantly upregulated for all genes except for Klrc1, Vsir, and Cd274, and was the only stimulus for which Tigit was transcriptionally upregulated (Fig. 3I).

Fig. 3.

MAIT cells express unique checkpoint signatures dependent upon activation stimuli. A) Checkpoint gene expression log₂ fold change was averaged for each stimulus and compiled into a heatmap for visual comparison. B to I) Individual gene counts for each checkpoint gene and sample across all studies were normalized against the unstimulated condition to calculate fold change and compared in aggregate. Each stimulus is represented by 2 studies of 3 to 7 donors each. Significance is calculated by nonparametric ANOVA, Kruskal–Wallace with multiple comparisons. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Taken together, our data underline the heterogeneity of MAIT cell checkpoint gene expression, which may be responsive to contextual microenvironmental cues.

3.4. Stimulus-specific impact of checkpoint gene expression validated at the single-cell level

Having identified a stimulus-specific checkpoint gene signature at the bulk RNA-seq level of sorted MAIT cells, we sought to validate these findings at the single-cell transcriptomic level. While bulk RNA-seq has greater coverage than single-cell RNA-seq, single-cell analyses can resolve heterogeneity among similar cell types. We analyzed our checkpoint genes of interest using publicly available single-cell data from sorted MAIT cells that had been activated with 2 different stimulations³⁵ (the MAIT ligand 5-OP-RU and cytokines IL-12 and IL18).

At the single-cell level, MAIT cells showed a clear separation of broad transcriptomic signatures by stimulation (Fig. 4A and B). The density of checkpoint gene expression varied by the gene and stimulation, with overall density of expression being lowest for Ctla4 and Havcr2, intermediate for Pdcd1 and Klrc1, and highest for Cd274, Tigit, Vsir, and Lag3, depending on the stimulation (Fig. 4C). MAIT cells most broadly expressed Lag3 upon stimulation from both 5-OP-RU and cytokine stimulation, with approximately 60% of activated MAIT cells expressing it in either case. In terms of magnitude of expression, MAIT cells most strongly expressed Cd274 and Tigit, although in a lower percentage of cells (Fig. 4C). Cd274 was upregulated under cytokine stimulation and Tigit was upregulated under MAIT ligand expression (Fig. 4C).

Fig. 4.

Single-cell transcriptomics confirm a role for stimulus specific checkpoint gene expression. Single-cell RNA-sequencing from sorted MAIT cells activated with either the MAIT ligand 5-OP-RU or cytokines IL-12 and IL-18. Sequencing revealed 12 clusters (A), which clearly diverge from the unstimulated condition into either ligand or cytokine activation (B). Specific analysis of checkpoint gene expression by stimulus shows differences in expression intensity and breadth of expression (C). This is additionally visualized by a density plot showing some genes that show broad expression in activated MAIT cells, some that favor cytokine stimulation and some that favor 5-OP-RU (D).

In many cases, the stimulus-specific nature of checkpoint gene expression from the single-cell study matched the pattern observed in the bulk RNA-seq studies. MAIT ligand stimulation with 5-OP-RU favored Havcr2, Vsir, and Klrc1 expression, while cytokine stimulation prompted a greater upregulation of Ctla4 and Cd274 (Fig. 4C and D). Pdcd1 was upregulated by both stimulations, but more highly by cytokine stimulation than ligand stimulation. Tigit, which had only been upregulated by anti-CD3-CD28 in the bulk RNA-seq study, was strongly upregulated by 5-OP-RU in the single-cell study. In all cases, MAIT cell activation promoted an upregulation of checkpoint gene expression by at least 1 stimulus. Taken together, our analysis of single-cell RNA-seq datasets confirm, and demonstrate with greater detail, the nature by which activated MAIT cells upregulate checkpoint gene expression across a broad array of checkpoints, and in a stimulus-specific manner.

3.5. Tissue-based variation in checkpoint gene expression

Given that MAIT cells are known to have unique characteristics depending on their location,^6,36,37 we used single-cell RNA-seq data from the same study to examine differences in checkpoint gene expression between MAIT cells from human blood and liver.³⁵

We found a clear signature of tissue vs peripheral MAIT cell checkpoint expression (Fig. S5A to D). Checkpoint gene expression overall was generally lower, as these MAIT cells were not specifically stimulated. The exception was Vsir, which is known to be expressed in naïve-like cells.³⁸ Despite low overall expression, we observed the highest checkpoint expression in the liver for Havcr2, Pdcd1, Cd274, Lag3, and Klrc1. By contrast, peripheral MAIT cells expressed higher levels of Ctla4, Tigit, and Vsir (Fig. S5C).

These data suggest an additional layer of contextual influence on checkpoint gene regulation in MAIT cells, with the tissue showing a higher prevalence of most checkpoint genes studied. This may reflect that tissue-based MAIT cells are more likely to have encountered antigen or activating cytokines relative to circulating MAIT cells.

3.6. Immune checkpoint genes are upregulated in tumor MAIT cells

Having identified a transcriptional upregulation of checkpoint genes tissue MAIT cells, we also asked whether this would be relevant to tumor-associated MAIT cells. We identified a single-cell RNA-seq study of colorectal cancer in humans that had data for tumor-associated MAIT cells and lymphocytes from healthy tissue.³⁹ There was no isolated MAIT healthy control. However, our analysis of checkpoint genes in tumor-associated MAIT cells shows a clear upregulation of checkpoint genes relative to healthy lymphocytes for all genes except for Vsir (Fig. S6A to C).

Most checkpoint genes, with the exception of Cd274 and Klrc1, were expressed in approximately 30% to 40% of tumor MAIT cells (Fig. S6C). This is a greater expression level than seen in raw steady-state tissue (Fig. S5C). This potentially reflects a higher or more chronic activation state in the tumor than seen in unstimulated tissue, and underlines the relevance of MAIT cell checkpoint expression in a malignant context.

3.7. In vitro activated MAIT cells upregulate Lag3 and PD-L1 at the cell surface

To confirm our findings from transcriptomic data, we stimulated peripheral human MAIT cells from healthy donors in vitro with either 5-ARU or methylglyoxal to generate 5-OP-RU, E. coli, the cytokines IL-12 and IL-18, or anti-CD3-CD28 and measured surface protein expression by flow cytometry. We stimulated MAIT cells at 4 different time points: 6 and 24 h to reflect the time points seen in our transcriptomic studies, and 48 and 72 h to better reflect chronic stimulations that may drive checkpoint expression. MAIT cells from 14 to 19 donors total were analyzed across 3 independent experiments.

All stimuli induced a highly significant increase in Lag3 expression on the cell surface by 24 h (Fig. 5A, B, and D). This reflected our findings from the single-cell data, in which Lag3 was the most broadly upregulated gene by percent of cells regardless of the type of stimulus. Several other checkpoint genes saw significant changes to checkpoint surface expression by the 24 h time point. PD-L1 was also significantly upregulated at the cell surface following all stimulations by 24 h (Fig. 5C and E). Tigit was significantly upregulated when exposed to 5-OP-RU, E. coli, and anti-CD3/CD28, but not the receptor-independent cytokine activation (Fig. 5A). NKG2A (Klrc1) was significantly upregulated by E. coli exposure, but not until 48 h (Fig. 5A). PD-1 was also significantly upregulated, but not until 48 h (Fig. 5A).

Fig. 5.

MAIT cells differentially express checkpoint genes at the cell surface. MAIT cells from human PBMCs were activated by 4 stimulation conditions for 6, 24, 48, or 72 h. The heatmap represents flow cytometry data for all stimulations, checkpoints and time points (A). B and C) Lag3 and PD-L1 are significantly upregulated beginning at 24-h stimulation of MAIT cells and remain elevated for the duration of the experiment at 72 h (B and C). The 24-h time point of Lag3 and PD-L1 demonstrates the significant upregulation across all groups in greater clarity, along with representative flow cytometry histograms (D and E). Lag3 and PD-L1 show a higher expression in MAIT cells than in non-MAIT CD3⁺ cells (F and G). MAIT cells and non-MAIT CD3⁺ cells are distinguished by presence or absence of MR-1 tetramer binding. PD-L1 is more significantly induced in MAIT cells than PD-1 across all stimulations at nearly all time points (H to K). A to G were assessed with nonparametric ANOVA (Kruskal–Wallace with multiple comparisons). H to K were assessed with a Wilcoxon matched-pairs rank sum test at each time point. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

We observed higher Lag3 and PD-L1 expression in MAIT cells than in non-MAIT CD3⁺ cells (CD3⁺MR1tetramer⁻), even for the anti-CD3/CD28 stimulation, which would activate conventional T cells (Fig. 5F and G).

PD-1, not its ligand PD-L1, is traditionally more associated with lymphocytes and has historically been the focus for MAIT checkpoint studies to date. PD-L1, by contrast, is associated with antigen-presenting cells, regulatory T cells, and tumors. However, in our study with MAIT cells, PD-L1 was significantly more broadly induced by all stimuli than PD-1 up to 72 h (Fig. 5H to K). These protein expression data are consistent with our transcriptomic findings that MAIT cell activation transcriptionally upregulates Cd274 (PD-L1), and suggests that MAIT cells may have a role in providing tolerance signals to the broader milieu and dampening activation signals around them. Expression of checkpoint ligands like PD-L1 may provide clues to their inconsistent role in cancer immunity.

4. Discussion

MAIT cells have capacities for rapid degranulation and cytotocity,⁷ as well as tissue repair and homeostasis.^6,8 In this study, we showed that contextual cues from the environment, such as activation stimulus, can shape their regulatory architecture via checkpoint gene expression.

We used a systematic review and meta-analysis approach to identify 4 different publicly available bulk transcriptomic datasets with 4 conditions of MAIT cell activation to examine how activation stimulus impacts MAIT cell transcriptional programming. We found a shared transcriptional program across all stimuli that underlines rapid proliferation, while identifying several differences among stimuli. We found that cytokine stimulation prompted a more robust cytotoxic response, while stimulation with the MAIT ligand 5-OP-RU promoted a mixed IL-17 and IFNγ phenotype. In transcriptomic studies, we found key checkpoint genes and ligands to be differentially upregulated. At the protein level, we used in vitro assays to confirm that MAIT cells significantly upregulate Lag3 and PD-L1 across all stimuli beginning at 24 h, Tigit under receptor-associated stimuli, NKG2A with E. coli exposure, and PD-1 by 5-OP-RU, cytokine, and CD3-CD28, though the kinetics for the latter 3 are variable.

Our findings have important implications for the role of the local microenvironment on MAIT cell tolerance. Being primarily activated through either microbial exposure or inflammatory cytokines, the location of MAIT cell activation may influence the tolerance mechanisms employed. MAIT cells at mucosal sites are more likely to encounter microbial antigens as an activating stimulus (ie 5-OP-RU), which could have implications for the checkpoint genes that are upregulated in those environments. By contrast, MAIT cell activation in nonmucosal sites with low microbial exposure may be more likely to be activated through inflammatory signals (ie cytokines). If these stimuli promote unique tolerance cascades, this may contribute to the variability observed in MAIT cell response to malignancy, or even variations observed within mouse models of inflammatory disease.

We found that Pdcd1 and Klrc1 are most transcriptionally upregulated by 5-OP-RU (Fig. 3D and H) and also upregulated and expressed in a greater number of MAIT cells in the liver compared to the blood (Fig. S5C). Although not directly a mucosal organ, the liver receives signals directly from the gut via the hepatic portal vein.⁴⁰ MAIT cell involvement in mucosal cancers has shown a poorer prognosis than some other cancers,¹¹ and it has been shown that some microbial species in colorectal cancer are directly associated with receptor-dependent exhaustion of tumor MAIT cells.¹³ The observation that 5-OP-RU promotes IL-17 expression while cytokine exposure promotes a cytotoxic program (Fig. 2) supports the hypothesis that 5-OP-RU-activated MAIT cells, potentially more prolific at mucosal sites, may be less fit for cancer-killing than MAIT cells at other tissue locations.

Many MAIT checkpoint studies so far have focused on PD-1, a target for currently available clinical immunotherapeutic agents^15,41–43. However, our findings may open the door into exploration of targeting additional checkpoints that may be more relevant to MAIT cell biology. We demonstrate Lag3 to be broadly and significantly upregulated to the cell surface in all stimulated conditions by 24 h and out to 72 h. Lag3 is particularly interesting as a finding for MAIT cells, as it engages with the major histocompatibility complex II (MHC-II) on antigen-presenting cells (APCs), which MAIT cells do not interact with. Lag3 has also been found on other immune cell types such as dendritic cells, B cells, γδ T cells, and NK cells, all of which also do not engage MHC-II.⁴⁴ In dendritic cells, Lag3 is expressed at higher rates than in T cells, and is thought to contribute to bidirectional homeostasis between dendritic cells and their cognate T cells.⁴⁴ For MAIT cells, it's possible that Lag3 expression is more relevant to modification of the surrounding lymphocytic milieu than to intrinsic regulation.

We found PD-L1 to be surprisingly more highly expressed than PD-1 both transcriptionally and at the cell surface. Although conventional T cells are known to express PD-L1, they generally produce higher amounts of PD-1, and receive signals from its ligand PD-L1 from surrounding APCs or tumor cells.⁴⁵ Studies support a role for PD-1 blockade in improving the function and cytotoxic capacity of MAIT cells in cancer therapy, suggesting that PD-1 still plays an important role in MAIT cell function.^15,42,43 Our study shows that MAIT cells do upregulate PD-1 across several stimuli, but not until 48 h. However, our observation that MAIT cells more significantly upregulate PD-L1 than PD-1 following all stimulations up to 72 h suggests that MAIT cells may also have a prominent role in communicating tolerance signals to conventional T cells around them. Further study over longer time courses using in vivo models is needed to determine whether this is true in an active disease model, not just in the context of MAIT cell activation.

We found that NKG2A was transcriptionally upregulated upon 5-OP-RU and E. coli stimulation among the studies included in our review, which was confirmed for E. coli in vitro, but not until 48 h. NKG2A is an inhibitory NK cell receptor that has also been found on Vδ2 T cells and a small proportion of MAIT cells^46–48. The role of NKG2A in MAIT cell function is unclear but may point to a noncanonical inhibitory signal for MAIT cells.

Finally, we observed upregulation of Tigit in response to CD3-CD28 stimulation at the transcriptional level, confirmed at the protein level with other receptor-associated stimuli 5-OP-RU and E. coli stimulation in vitro. Others have reported higher Tigit expression in MAIT cells from colonic tumors, but failed to find a statistically significant association between co-expression of Tigit with PD-1 and Tim-3.⁴⁹ Our data suggest that this may stem from nuances in the unique drivers for each checkpoint gene. Notably, one of the ligands of Tigit is the bacterial antigen Fap2, which can dampen both T and NK cell responses.⁵⁰ Given MAIT cells' connection to the microbiome, expression of Tigit may provide an additional layer of host–microbe co-regulation.

Together, our data suggest that in addition to receiving intrinsic checkpoint regulation, MAIT cell expression of Lag3 and PD-L1 may make them guardians of the environment around them. Checkpoint blockade treatments aimed at MAIT cells ought to consider their secondary impact on other cells.

Though findings from our in silico analyses suggest relevance to both tissue and tumor sites, our in vitro study is limited to stimulated MAIT cells from peripheral human blood and may not represent tissue-resident MAIT cells. Many checkpoints studied were transcriptionally upregulated in the liver compared to the blood, but also only expressed in a small subset of steady-state MAIT cells (with the exception of Vsir). It is notable that the MAIT cells from the tissue analysis were not activated prior to sequencing, so low levels of the checkpoint genes may simply reflect a lack of stimulation. In addition, we did not confirm the functional impact of checkpoint genes like PD-L1 or Lag3 to MAIT cells or surrounding cells in the milieu.

In the tumor, many checkpoints were expressed in approximately 30% to 40% of MAIT cells, which is comparable to what we see in vitro, but the level of Cd274 (PD-L1) expression was lower than indicated by the in vitro studies. In both steady-state liver and in colorectal tumor-associated MAIT cells, Pdcd1 is more highly expressed than Cd274. This may suggest that PD-L1 is more relevant to an acute activation context than the chronic stimulation of the tumor microenvironment, or else that there are tissue-inherent signatures that promote PD-1 in lieu of PD-L1. Future studies into tissue-resident MAIT cells are needed to better understand their potential for tumor immunotherapy.

Despite these limitations, we have identified core and unique transcriptional signatures for activated MAIT cells, as well as context-dependent checkpoint gene expression at the gene and protein level. We identify an elevated expression of Lag3 and PD-L1 in acutely activated MAIT cells. Our findings may contribute to the development of better therapeutics in checkpoint-relevant diseases, such as cancer.

Author contributions

C.H.C. and D.L.: conceptualization; C.H.C.: data curation; C.H.C.: formal analysis; C.H.C., A.R., and K.J.: investigation; J.A.: resources; D.T.L., K.C., and R.M.O.: supervision; C.H.C. and D.L.: writing—original draft; C.H.C., A.R., J.A., K.C., R.M.O., and D.T.L.: writing—review and editing.

Supplementary material

Supplementary material is available at Journal of Leukocyte Biology online.

Funding

Funding for this study was provided by the National Institute of Diabetes and Digestive and Kidney Diseases Ruth L. Kirschstein National Research Service Award in Metabolism (T32DK091317, awarded to C.H.C.) and the National Cancer Institute (R21CA280224, awarded to D.T.L.).

Data availability

No new transcriptomic data were produced from this study. Publicly available data of sorted and activated MAIT cells for bulk RNA-seq was obtained from the NIH Gene Expression Omnibus (GEO) with accession numbers GSE158439, GSE129906, GSE123805, and PRJNA559574. Single-cell data from sorted MAIT cells either activated in vitro or assessed at steady state from blood and liver were acquired from GSE194189. Code used for the meta-analysis of bulk RNA-seq studies will be made publicly available.

Ethics statement

For our in vitro work on human PBMCs, blood samples were collected from healthy human donors who gave full informed consent for their samples to be used for research purposes. All samples collected at the University of Utah were fully deidentified. This protocol has been reviewed and approved by the University of Utah Internal Review Board (IRB_00097389).