INTRODUCTION

Cholangiocarcinoma (CCA) is a rare primary malignancy of the bile ducts, accounting for 3% of all gastrointestinal tumors.[1] The pathogenesis is insufficiently known. CCA can be divided into three subtypes depending on anatomical origin: intrahepatic (iCCA), perihilar (pCCA), and distal (dCCA).[2] The three subtypes represent distinct entities with specific risk factors, clinical presentation, management, and prognosis.[2] In general, chronic inflammation is an established risk factor given that various inflammatory diseases associate with the development of CCA. As such, primary sclerosing cholangitis (PSC) is a predisposing condition with a lifetime CCA prevalence of 5%–10%.[1, 3] The lack of early clinical symptoms and good biomarkers makes CCA diagnosis challenging, and thus the overall 5-year survival rate is ~10%. Even when resection surgery or liver transplantation are performed with a curative intent, recurrence rates are high.[1] It has been shown that immunotherapy-based treatments are successful for some cancers, but not yet for CCA. One reason for this is the limited knowledge about the composition of the intratumoral immune system in CCA.[4]

Mucosal-associated invariant T (MAIT) cells are enriched at mucosal sites, particularly in the liver where they locate to portal tracts in close proximity to intrahepatic bile ducts.[5, 6] As innate-like T lymphocytes, they have been shown to perform rapid effector functions, including the direct killing of target cells by perforin and granzyme B and the production of cytokines, such as interferon (IFN)γ, TNF, and IL-17,[7-10] thereby potentially contributing to anticancer immunity. Unlike conventional T cells, MAIT cells express a semi-invariant T-cell receptor (TCR) containing the Vα7.2 segment. The MAIT cell TCR recognizes microbial-derived riboflavin (vitamin B2) biosynthesis intermediates, such as 5-(2-oxopropylideneamino)-6-D-ribitylaminouracil (5-OP-RU), presented by the nonpolymorphic major histocompatibility complex (MHC) class I–related protein 1 (MR1).[9, 11, 12] Accordingly, MAIT cells can be identified using a fluorochrome-coupled MR1 tetramer loaded with 5-OP-RU.[13]

In addition to their antibacterial function, MAIT cells have recently been investigated in several types of cancers, although with a disputed role.[14-21] Assessment of MAIT cells in cancers at mucosal sites has shown reduced numbers of these cells in circulation, but that they are present in tumor tissue where they may impact disease progression.[14-17, 21] In primary liver tumors, however, the opposite was recently reported, where MAIT cells were decreased in HCC.[20] Some studies have reported retained functional capacity of MAIT cells in tumors, including IFNγ production and cytotoxic potential,[14, 19] whereas others have found impaired function with a shift toward IL-17[16, 17] and IL-8 production.[20] Interestingly, a recent study in mice suggested that MAIT cells may promote tumor growth and metastases through MR1, expressed by tumor cells.[18] Along these lines, infiltration of MAIT cells indicates a poor outcome in HCC.[20]

Because of a high abundance of MAIT cells in the human liver and the fact that cholangiocytes can activate MAIT cells in an MR1-dependent manner,[5, 6] we have performed a comprehensive analysis of MAIT cells from tumor and surrounding tissue of patients with iCCA and pCCA. The results are discussed in relation to current knowledge of the immunopathogenesis of CCA and immunity in the tumor microenvironment.

PATIENTS AND METHODS Study cohort

A total of 70 patients were included in this study. The study was approved by the Regional Ethics Committee of Stockholm and written informed consent was obtained from all patients. Liver tissue specimens were obtained from patients admitted for liver resection at Karolinska University Hospital under the suspicion of CCA. The diagnosis of CCA was confirmed by routine clinical pathology performed by a specialized liver pathologist and defined during fresh gross morphological assessment of the resected specimen. Tumor and surrounding tissue (and in the case of flow cytometry analysis, also tumor margin) was collected from all patients. On average, the distance for sampling between tumor and tumor margin was 21 mm (range, 13–47) and between tumor margin and the sample taken from the periphery 38 mm (range, 10–120). See Table 1 for detailed clinical characteristics and the Supporting Information for more information on the study cohort.

TABLE 1. Clinical characteristics of study subjects Variable Cholangiocarcinoma patients (n = 70) Age, median years (range) 66 (29–83) Sex, n (%) Female 35 (50) Male 35 (50) BMI, median (range) 25 (17–37) Surgery, n (%) 70 (100) Minor resection 7 (10) Major resection 63 (90) Diagnosis, n (%) pCCA 21 (30%) iCCA 47 (67%) dCCA 1 (1%) Gallbladder cancer 1 (1%) Tumor grade, n (%) Well differentiated 6 (9) Poorly/moderately differentiated 58 (83) N.A. 6 (9) ASA, n (%) 1–2 51 (73) 3–4 19 (27) PSC, n (%) 9 (13) Viral hepatitis (HBV or HCV), n (%) 3 (4) Fibrosis, n (%) None 7 (10) Mild 24 (34) Moderate 34 (49) Severe 5 (7) CRP mg/L, median (range) 6 (1–123) Bilirubin μmol/L, median (range) 8 (3–168) Bilirubin >25 μmol/L, n (%) 5 (7) Bilirubin <25 μmol/L, n (%) 65 (93) Albumin g/L, median (range) 35 (21–43) CA19-9 Ke/L, median (range) 138 (1–40,590) Biliary drainage, n (%) No drainage 49 (70) Percutaneous transhepatic cholangiography (PTC) 1 (1) Endoscopic retrograde cholangiography (ERC) 14 (20) PTC and ERC 6 (9) Glasgow Prognostic Score, n (%) 0 27 (38) 1 18 (26) 2 18 (26) N.A. 7 (10) Disease free survival, median months (range) 10 (0–64) Overall survival time, median months (range) 22 (0–95) Abbreviations: BMI, body mass index; ASA, American Society of Anesthesiologist classification; CRP, C-reactive protein; CA19-9, cancer antigen 19-9; N.A., not applicable. Sample processing

Blood samples were collected in heparin tubes and isolated using Ficoll gradient centrifugation. Liver tissue was enzymatically digested into single cells using collagenase II. See the Supporting Information for details.

Cell lines

Cell lines derived from CCA patients (HuCCT-1 and TFK-1) were used for protein expression analysis on cholangiocytes. See the Supporting Information for details.

Flow cytometry staining and analysis

Flow cytometry stainings were performed using optimized combinations of fluorescently labeled antibodies. Standard flow cytometry analysis was performed using FlowJo software (FlowJo, LLC, Ashland, OR) and high-dimensional analysis using the uniform manifold approximation and projection (UMAP) and Phenograph plugins. See the Supporting Information for details on staining protocol, antibody panels, and analysis.

Immunohistochemistry staining and analysis

Immunohistochemistry (IHC) was performed on tissue sections from frozen liver specimens and analyzed in a blinded fashion by two different methods: acquired computerized image analysis (ACIA) and digital image analysis. See the Supporting Information for details on staining and analysis procedures.

Bacterial visualization and quantification in CCA tissue

Bacterial visualization was performed using fluorescence in situ hybridization (FISH) and with a probe against the universal bacterial gene, 16S rRNA (ribosomal RNA), and quantification performed after DNA extraction followed by real-time PCR using the Femto DNA Quantification Kit (Zymo Research, Irvine, CA, USA). See the Supporting Information for details on protocols.

Deconvolution of RNA-sequencing data sets

Publicly available bulk RNA-sequencing (RNA-seq) data from CCA tissue and matched surrounding tissue were deconvoluted for absolute immune-cell content using the absolute immune signature for RNA-seq tool, a recently published algorithm,[22] which uses RNA-seq signatures normalized by mRNA abundance. See the Supporting Information for details on data sets analyzed and the analysis procedure.

Gene expression analyses

Gene set enrichment analyses (GSEA) was performed against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database using GSEA software (version 4; Broad Institute) with default settings. In these analyses, MAIT-high (>median) iCCA was compared with MAIT-low (<median) iCCA, with statistical significance considered at p < 0.05.

Immunogenomics analyses

The activity of different stages of the cancer-immunity cycle was estimated in iCCA samples by tumor immunophenotyping.[23] The analysis was performed on RNA-seq data (counts in reads per kilobase of transcript per million reads mapped) through the TIP server (http://biocc.hrbmu.edu.cn/TIP/) with TCGA-CHOL specified as the reference cancer type. Intratumoral MAIT content estimates were correlated with activity of each stage of the cancer-immunity cycle and visualized as a volcano plot in GraphPad Prism software (GraphPad Software Inc., La Jolla, CA).

Single-cell RNA-seq data analysis

To reanalyze the CCA single-cell RNA-seq data set (GSE138709) from the Gene Expression Omnibus,[24] data were first filtered, normalized, integrated, and scaled, cell clusters were identified, and immune and nonimmune cell types annotated. Malignant cells were identified by inferCNV. In total, 34 tumor and 1,527 peripheral MAIT cells were identified. Differentially expressed genes between tumor MAIT cells and peripheral MAIT cells were determined using DESeq2. Enriched Gene Ontology terms or gene sets were identified and significantly activated or inhibited canonical pathways were detected by Ingenuity Pathway Analysis (IPA) software. CellPhoneDB analysis[25] was performed to identify significant interactions between receptors on tumor or peripheral MAIT cells and ligands on malignant cells. See the Supporting Information for more details.

Statistical analysis

Statistical analysis was conducted using GraphPad Prism software (version 7; GraphPad Software), where data sets were first tested for normality followed by specific tests depending on data structure: *p < 0.05; **p < 0.01; ***p < 0.001. See the Supporting Information Materials and Methods for details on statistical analysis.

RESULTS Assessment of conventional T cells in CCA tumors

We set off to characterize the CCA microenvironment using IHC. As expected, the microenvironment contained a high expression of cytokeratin 19 (CK19), a marker that identifies CCA cells (Figure 1A,B). Next, for an overall understanding of the T-cell composition, IHC analysis was performed for expression of CD3, CD8, and forkhead box P3 (FoxP3) in 20 matched CCA tumor and surrounding nonaffected tissue samples. This revealed a similar presence of total T cells as well as of CD8+ T cells within the tumor microenvironment as compared to outside the tumor and an enrichment of regulatory T (Treg) cells, a finding validated by another digital image analysis approach (Figure 1A,C-E and Figure S1A). Our CCA patient cohort consisted predominantly of iCCA (almost 70%) and pCCA patients, and when we stratified our IHC data (Figure 1B-E) based on the CCA subtype, we observed similar results between iCCA and pCCA cases (Figure S1B). Several CCA patients in our IHC cohort had PSC, but there were no significant differences in immune cell presence when CCA patients were stratified according to their PSC status (Figure S1C). IHC findings were confirmed using flow cytometry, through comparing cells isolated from the tumor with cells from the tumor margin, nontumorous liver periphery, and matched peripheral blood (Figure 1F). Thus, conventional CD8+ T cells infiltrate iCCA and pCCA tumors in which Tregs were enriched.

T cells infiltrate CCA tumors. (A) Digital scans of representative IHC staining of CK19, CD3, CD8, and Foxp3 in nontumor and tumor tissues of CCA patients. (B-E) IHC quantification analysis (ACIA) of (B) CK19 (control, n = 4; nontumor, n = 22; tumor, n = 19); (C) CD3 (control, n = 5; nontumor, n = 19; tumor n = 17); (D) CD8 (control, n = 5; nontumor, n = 20; tumor, n = 19); and (E) Foxp3 (control, n = 5; nontumor, n = 20; tumor, n = 17), with a line indicating the median in all plots. (F) Flow cytometry data showing percentages of T cells out of live CD45+ CD14− CD19− cells (blood, n = 13; CCA liver, n = 11–13), CD8+ T cells (blood, n = 13; CCA liver, n = 11–13), and Foxp3+ Treg cells (blood, n = 10; CCA liver, n = 9–10) out of total T cells. Statistical significance was determined using the Kruskal-Wallis test, followed by Dunn`s multiple comparison test or one-way ANOVA, followed by Tukey`s multiple comparison test (B-F). *p < 0.05; **p < 0.01; ***p < 0.001

MAIT cells are reduced in the CCA microenvironment

Besides conventional T cells, hepatic MAIT cells have recently gained attention in the context of tumor immunology.[20] Using IHC, we stained for Vα7.2 expression as an indirect marker for MAIT cells and noted a decline within the tumors compared to surrounding tissue (Figure 2A, left and middle, and Figure S1A,B,E). Similarily, lower levels of Vα7.2 mRNA were present in tumor tissue compared to nontumorous tissue (Figure 2A, right). Given that non-MAIT cells can also express Vα7.2, we confirmed the intratumoral decline in MAIT cells with flow cytometry staining for either coexpression of Vα7.2 and CD161 or by using the MR1 5-OP-RU tetramer (Figure 2B and Figure S1D). Finally, using a recently published gene signature for MAIT cells,[22] we deconvoluted bulk RNA-seq data from matched CCA tumor and nontumor tissues from four published data sets and could validate the loss of MAIT cells in the CCA microenvironment in all four, three of which contained only iCCA cases (Figure 2C; Supporting Materials and Methods). Chronic antigen stimulation has been suggested as one cause for driving the loss of MAIT cells under chronic inflammatory conditions, including chronic liver diseases.[26-29] Given that cholangiocytes can stimulate MAIT cells through MR1,[6] we investigated the presence of MR1 in the CCA microenvironment. We observed similar levels of MR1 mRNA in CCA nontumor and tumor tissues (Figure 2D, left). A phenotypic analysis of two CCA cell lines as well as of primary cholangiocytes further showed that MR1 was expressed on the surface of these cells, but higher protein levels could be identified intracellularly (Figure 2D, right). Analysis of the same four RNA-seq data sets we looked at previously validated findings from our cohort, given that MR1 was present at similar levels in nontumorous and tumorous tissues across three data sets, and significantly more highly expressed in tumor tissues in one of the data sets (Figure 2E). We also assessed the presence of bacteria and bacterial fragments within tumors, using samples from tumors and surrounding tissue. Bacteria were indeed present within tumors (Figure 2F, left) as well as surrounding tissue (not shown), as determined by FISH. To quantify the presence of bacteria, we performed qPCR for 16S rRNA on matched tumor and nontumor tissue. All CCA tumor samples were positive and contained a high absolute concentration of bacterial DNA, which applied to only 40% of the surrounding tissue (Figure 2F, middle and right, respectively).

Heterogenous loss of MAIT cells in the CCA tumor microenvironment. (A) Left: digital scan of representative IHC staining of Vα7.2; middle: IHC quantification analysis (ACIA) of staining in (A) (control, n = 5; nontumor, n = 48; tumor, n = 44), with a line indicating the median; right: quantification of mRNA encoding Vα7.2, determined by qPCR (control, n = 5; nontumor, n = 20; tumor, n = 19), with a line indicating the median. (B) Left: representative flow cytometry plots showing the percentage of MAIT cells from peripheral blood and CCA liver; right: quantification of MAIT cells in percentage as gated in plots on the left (blood, n = 14; CCA liver, n = 12–14). (C) Quantification of the MAIT cell absolute immune signal for CCA tumor and peripheral nontumor tissue after deconvolution of four publicly available CCA RNA-seq data sets. (D) Left: quantification of MR1 mRNA determined by qPCR (control, n = 5; nontumor, n = 20; tumor, n = 19), with a line indicating the median; right: representative flow cytometry staining of the extracellular (red line) and intracellular (blue line) MR1 expression on two CCA lines and primary cholangiocytes defined as live CD45−, CD31−, EpCAM+ CK19+ (one staining out of three is shown). (E) MR1 gene expression across CCA tumor and nontumor peripheral tissues of four publicly available CCA RNA-seq data sets. (F) Left: representative FISH staining for bacterial detection in CCA livers (one positive staining and negative control are shown out of 11, scale bars indicated in the images); middle: quantification of the number of CCA tumors and matched nontumor peripheral tissue in which bacteria could be detected by qPCR; right: quantification of bacterial DNA by qPCR in matched tumor and nontumor parts of CCA livers (n = 12). A line indicates the median. Statistical significance was determined using the Kruskal-Wallis test, followed by Dunn`s multiple comparison test (A,D); one-way ANOVA followed by Tukey`s multiple comparison test (B); Wilcoxon matched-pairs signed rank test, paired t test, or Mann-Whitney U test (C); chi-square test (F, middle); and Wilcoxon matched-pairs rank test (F, right). *p < 0.05; **p < 0.01; ***p < 0.001. EpCAM, epithelial cell adhesion molecule; TCGA, The Cancer Genome Atlas

Taken together, these data show that MAIT cells are partially lost from the iCCA and pCCA microenvironment, and that these tumors contain an increased quantity of bacteria and, by extension, bacterial-derived antigens that can potentially stimulate MAIT cells.

Tumor-infiltrating MAIT cells express tissue-residency markers, but are not activated or exhausted

Next, we comprehensively characterized residual tumor-infiltrating MAIT cells, assessing expression of 14 phenotypic markers using multicolor flow cytometry (Figure 3A,B). The phenotype of tumor-infiltrating MAIT cells was compared to those found at the tumor margin as well as in the liver periphery and peripheral blood in matched samples taken from the same patient. Tumor MAIT cells expressed high levels of the tissue-residency markers, CD69 and CD103, but expression of exhaustion (programmed cell death protein 1; PD-1) and activation/cytotoxicity markers (CD25, human leukocyte antigen [HLA]-DR, and granzyme B) was not significantly changed when comparing peripheral and tumoral MAIT cells (Figure 3B). The vast majority of MAIT cells in matched periphery, margin, and tumor expressed IL-18Rα, similar to normal and PSC biliary mucosa obtained from bile ducts (Figure S2A). CD56, a receptor that was recently shown to identify MAIT cells with higher cytokine responsiveness,[30] was expressed significantly less on MAIT cells within tumors (Figure 3B). Furthermore, the chemokine receptors, C-X-C motif chemokine receptor (CXCR) 6 and C-C motif chemokine receptor (CCR) 6, known to be important for liver homing,[31, 32] were less frequently expressed on tumor MAIT cells, but we did not observe any significant differences in CXCR3 and CCR9 expression on tumoral MAIT cells compared with peripheral and marginal MAIT cells (Figure 3A,B and Supporting Figure S2A). Decreased CXCR6 expression was independent of the CCA subtype, but CCR6 expression appeared overall lower on MAIT cells across all investigated areas from pCCA (Figure S2D). Similar phenotypic alterations were found on non-MAIT CD4+ and CD8+ T cells, albeit with a more clear PD-1 up-regulation (Figure S2B,C). Nevertheless, and as compared to conventional non-MAIT T cells, both CXCR6 and CCR6 were highly expressed on hepatic MAIT cells. Furthermore, we observed that CD69 expression was particularly high in tumor MAIT cells compared to non-MAIT T cells (Figure 3B and Figure S2B,C). Finally, tumor-infiltrating MAIT cells were more proliferative than their counterparts in matched liver periphery, measured by intracellular Ki-67 expression (Figure 3B).

Tumor-infiltrating MAIT cells express tissue-residency markers and display a nonactivated phenotype. (A) Representative histograms showing the expression of 11 phenotypical markers expressed on MAIT cells on/in MAIT cells from blood (dark gray), nontumor liver periphery (green), tumor margin (blue), and tumor (red). FMO control is depicted in light gray. (B) Summary of marker expression as percentage (%) or median fluorescence intensity (MFI) on MAIT cells from different tissue origins determined by flow cytometry (blood, n = 7–14; CCA liver, n = 3–12). A line indicates the mean. Statistical significance was determined using the Kruskal-Wallis test, followed by Dunn`s multiple comparison test. (C) UMAP plots displaying MAIT cells from a different origin within the map (n = 4). MAIT cells were defined by flow cytometry as live CD45+, CD3+, TCR-Vα7.2+ MR1 5-OP-RU+ cells. (D) Intensity of marker expression of 11 phenotypic markers that served as parameters to run UMAP (n = 4). *p < 0.05; **p < 0.01; ***p < 0.001. Bcl-2, B-cell lymphoma 2; FMO, fluorescence minus one; MAR, tumor margin; PBMC, peripheral blood mononuclear cells; PER, periphery (nontumor); TUM, tumor

In summary, tumor-infiltrating MAIT cells express high levels of the tissue-residency markers, CD69 and CD103, but do not exhibit an exhausted or activated phenotype.

High-dimensional data analysis reveals specific clusters of tumor-infiltrating MAIT cells

Conventional single-parameter flow cytometry provides an overall representation of phenotypic alterations, but it does not show whether these changes apply to the same cells. To address this, we performed a high-dimensional data UMAP analysis that reveals multivariate relationships between phenotypic markers in an unsupervised way. This algorithm can be applied to multiple donors simultaneously and is therefore possible to visualize clusters of cells presenting phenotypical similarities and differences in a map.[33] To this end, MAIT cells from four matched sites (tumor, margin, peripheral liver tissue, and peripheral blood) were electronically barcoded, concatenated, and analyzed (Figures 3C,D and 4A). We observed that peripheral blood MAIT cells localized separately from liver MAIT cells. Interestingly, within the liver, tumoral MAIT cells also separated clearly compared with MAIT cells localized at the tumor margin or in surrounding nontumor liver tissue (Figure 3C), meaning that tumor-infiltrating MAIT cells were phenotypically distinct whereas tumor margin- and periphery-derived MAIT cells were similar. High-dimensional UMAP analysis confirmed the shift in phenotype to higher expression of the tissue-residency markers, CD69 and CD103, whereas PD-1 was found on both nontumor and tumor-infiltrating MAIT cells (Figure 3D). As opposed to MAIT cells, the UMAP analysis for non-MAIT T cells revealed no clearly distinct clusters for T cells originating from the tumor and mostly coexpressed CD69 and CXCR6 (Figure S3A,B). Unlike intratumoral conventional CD8+ cells, tumor-infiltrating MAIT cells expressed higher levels of CCR6, CD56, and CD69, but lower levels of PD-1 and HLA-DR (Figure S3C).

High-dimensional data analysis reveals specific clusters of tumor-infiltrating MAIT cells. (A) UMAP analysis showing clusters of MAIT cells from different origin and the clusters identified by unsupervised PhenoGraph clustering. Clusters are based on 11 phenotypic markers used as clustering parameters and are indicated in different colors for each subset, followed by a subdivision for clusters that are most abundant for cells of different origin. (B) Percentage of cells of different origin within each PhenoGraph cluster. (C) Heatmap showing the expression of 11 phenotypical markers on MAIT cells from each cluster identified by PhenoGraph. (D) The scRNA-seq GSE138709 data set from Zhang et al. was reanalyzed in Seurat 4.0. UMAP shows an annotated data set, including a cluster of MAIT cells. (E) Putative receptor-ligand interactions between tumoral or nontumoral MAIT cells and malignant (CCA tumor) cells, obtained from the CellPhoneDB analysis. Receptors and ligands expressed in p 0.25 were kept. Dashed black line represents an adjusted p value of 0.01 (-log10[FDR] of 2). AKT, protein kinase B; ARP, acidic ribosomal phosphoprotein P0; BAG2, Bcl2-associated athanogene 2; FDR, false discovery rate; iNOS, inducible nitric oxide syntahse; JAK, Janus kinase; MIF, macrophage migration inhibitory factor; MSP, macrophage stimulating protein; PI3K, phosphoinositide 3-kinase; RANK, receptor activator of NF-κB; RON, recepteur d’origine nantais; STAT, signal transducer and activator of transcription; TNFR2, TNF receptor 2; WASP, Wiskott-Aldrich syndrome protein

To interrogate differences among MAIT cells in even greater depth, we subjected the UMAP data to Phenograph clustering, which revealed subclusters of phenotypic similarities[34] (Figure 4A). This analysis displayed 25 clusters based on the MAIT cell phenotype, which were represented by a different proportion of cells depending on their origin (Figure 4B). In more detail, six clusters were almost exclusively composed of MAIT cells originating from peripheral blood, five clusters largely represented intratumoral MAIT cells, whereas the remaining 14 clusters consisted of MAIT cells from the tumor margin and surrounding liver tissue, which overlapped to a large extent (Figure 4A,B). Intratumoral MAIT cell clusters were characterized by high expression of CD69 and CD103, lower levels of CXCR6 and CCR6, and low expression of CD25, HLA-DR, and CD57 as well as of the effector molecules, perforin and granzyme B (Figures 3D and 4C and Figure S4A-C). Of note, only one of the five tumor-specific MAIT cell clusters was high in PD-1.

Finally, publicly available single-cell RNA-seq (scRNA-seq) data of nontumor and tumor tissue from CCA patients were analyzed to determine the single-cell transcriptome and interactome of tumor MAIT cells[24] (Figure 4D). Using CellPhoneDB for determining tentative receptor-ligand interactions between MAIT cells and CCA tumor cells revealed that the majority of the interactome was similar for nontumor and tumor MAIT cells. However, CD44, CD47, and CCR6, interacting with galectin 9 and C-C motif chemokine ligand 20, were significantly enriched interactions for tumor MAIT cells (Figure 4E). Pathway analysis of differentially expressed genes in tumor MAIT cells revealed a down-regulated T helper 1 (Th1) pathway, but enrichment for the PD-1/PD-L1 (programmed death ligand 1) pathway (Figure 4F).

Taken together, as compared to MAIT cells from surrounding tissue, this analysis revealed that CCA-infiltrating MAIT cells represents a specific cluster of cells with a distinct transcriptome and interactome.

MAIT cell levels predict survival independent of clinical prognostic factors

Given that all patients in our study underwent liver surgery with a curative intent, we had the possibility to assess any relationship between MAIT cells, other immune cells, and survival. In a first analysis of 46 CCA patients, we divided the cohort into those who had high- and low-MAIT cells. Interestingly, CCA patients with high numbers of MAIT cells in surrounding liver tissue had a significantly longer survival time (59.5 vs. 25 months median overall survival; p = 0.01) compared to patients with low MAIT cell numbers (Figure 5A, left). To validate this finding, we used the previously described transcriptome signature for MAIT cells, deconvoluted bulk RNA-seq data available from 111 iCCAs, and compared tumors with the presence of either a high or low MAIT cell level.[35] A similar association between MAIT cell presence and longer survival was also noted in this second cohort (p = 0.02; Figure 5A, right). Strikingly, when combining MAIT cell presence with these significant clinical parameters, which are prognostic markers for survival after resection surgery, only the presence of MAIT cells remained significant in a multivariate Cox analysis (HR, 0.75; 95% CI, 0.58–0.96; Figure 5B).

High MAIT cell numbers are associated with better overall survival in CCA. Survival analysis showing the Kaplan-Meier curve for (panel A, left, IHC, n = 46) MAIT cells from nontumor CCA tissue or from tumorous iCCA tissue (panel A, right, RNA-seq, n = 111) with high and low MAIT cell division based on median values. (B) Forest plot showing the results of multivariate Cox proportional hazards modeling of tumoral MAIT cells and clinical surivival predictors, taking into account tumor differentiation (Mod. = moderate), presence of lymph node metastasis (pN = pathological lymph nodes, lymph node metastasis), and tumor stage (stage; n = 108). (C) Correlogram showing the correlation analysis of immune cell subsets after deconvolution (n = 119). (D) Forest plot showing the results of multivariate Cox proportional hazards modeling of clinical surivival predictors shown in (B) and immune cell subsets (n = 111). (E) GSEA peformed against the KEGG database for high- and low-MAIT cells in iCCA based on the median (n = 119). (F) Volcano plot showing the correlation of intratumoral MAIT cell content with different stages of the cancer-immunity cycle in relation to high or low MAIT cell content based on the median (n = 119); *p < 0.05. mDCs, myeloid DCs; pDCs, phosphorylated DCs; Th17, T helper 17; Th22, T helper 22

Thus, in two independent data sets using two separate methods to enumerate MAIT cells, the high presence of these cells within CCA livers predicted better long-term survival and this appeared to be independent of known clinical prognostic factors.

Intratumoral MAIT cell presence is associated with a favorable antitumoral immune signature in iCCA

We next sought to determine whether the association between high MAIT cell presence and survival could be linked to the immune status of the tumor microenvironment. As a first step, we analyzed bulk RNA-seq data from the 119 iCCAs described above. We deconvoluted an additional 17 immune cell populations from the tumor microenvironment and assessed their relationship to intratumoral MAIT cells. When performing pair-wise correlations between these signatures, MAIT cell presence within the tumor microenvironment correlated positively with signatures of primarily other innate immune cells, such as monocytes, neutrophils, γδ T cells, myeloid dendritic cells (DCs), and plasmacytoid DCs, among others (Figure 5C), whereas no correlation was present between MAIT cells and memory CD4 or CD8 T cells. Based on this inter-relatedness between MAIT cells and other immune cells, we next performed a new multivariate Cox regression including both immune signatures and clinical prognostic parameters. Strikingly, in this analysis, intratumoral MAIT cells still remained positively associated with increased survival whereas Vδ2 γδ T cells were the only other significant factor in the multivariate analysis, but negatively associated with survival (Figure 5D).

To dissect the significance of having a high presence of MAIT cells within CCA tumors, we performed pathway analysis comparing CCA tumors with high versus low MAIT cell presence (Figure 5E). This revealed that tumors with a high presence of MAIT cells were further enriched for a large number of immune-related pathways, including hematopoietic cell linage, cytokine-cytokine receptor interaction, and antigen processing and presentation, whereas no such enrichments could be found in tumors with low MAIT cell presence (Figure 5E). Finally, to understand how a high presence of MAIT cells associated with antitumor immunity, we used the entire cohort of 119 CCAs and, on a sample-by-sample basis, compared MAIT cell presence within the tumors with seven defined steps of the tumor-immunity cycle, which are based on defined immune-related signatures.[23] Somewhat surprisingly, this analysis revealed a negative association between MAIT cells and the killing of cancer cells (Figure 5F). Nevertheless, a high presence of MAIT cells strongly correlated with a number of defined steps of the tumor-immunity cycle, including cancer antigen presentation, priming and activation of immune cells, and T-cell recognition of cancer cells, but especially with immune cell recruitment of a variety of immune cells (Figure 5F).

Taken together, this suggests that elevated intratumoral MAIT cell presence, independent of other immune cells, associates with long-term survival and that this is linked to an overall favorable antitumoral immune signature within the tumor microenvironment.

DISCUSSION

CCA, the second-most frequent primary tumor of the liver, is a major clinical challenge and often diagnosed when the tumor has already spread or is at an advanced stage.[36] Immunotherapy efforts are still scarce, which might be attributable to the fact that the immune environment of CCA is less well understood.[4, 37] MAIT cells localize to bile ducts and have the capacity to release cytokines and cytotoxic granules.[5, 6] In this context, we set out to comprehensively characterize intratumoral MAIT cells in a large cohort of patients with iCCA or pCCA. We found that the tumor microenvironment was characterized by the presence of CD8+ T cells and high numbers of Treg cells. However, we observed a heterogeneous decline or loss of MAIT cells in the tumor as compared to surrounding tissue. The CCA tumor microenvironment was further characterized by maintained MR1 expression, but a higher presence of bacteria,