Tissue‐resident memory T cells are epigenetically cytotoxic with signs of exhaustion in human urinary bladder cancer

Summary

Tissue‐resident memory T (T_RM) cells are CD8⁺ T lymphocytes that reside in the tissues, including tumours. This T cell subset possesses a magnitude of cytotoxicity, but its epigenetic regulation has not been studied. Here, we investigate the impact of perforin DNA methylation in T_RM cells and correlate it with their functional potential. Fifty‐three urothelial urinary bladder cancer (UBC) patients were recruited prospectively. The DNA methylation status of the perforin gene (PRF1) locus in T_RM cells was investigated by pyrosequencing. Flow cytometry with ViSNE analysis and in‐vitro stimulation were used to evaluate T_RM cell phenotypes. We discovered that tumour T_RM cells have low DNA methylation in the PRF1 locus (32·9% methylation), which corresponds to increased numbers of perforin‐expressing T_RM cells. Surprisingly, programmed cell death 1 (PD‐1) expression is high in tumour T_RM cells, suggesting exhaustion. Following interleukin‐15 and T cell receptor stimulation, perforin and T‐bet expressions are enhanced, indicating that T_RM cells from tumours are not terminally exhausted. Moreover, a high number of T_RM cells infiltrating the tumours corresponds to lower tumour stage in patients. In conclusion, T_RM cells from UBC tumours are epigenetically cytotoxic with signs of exhaustion. This finding identifies T_RM cells as potential new targets for cancer immunotherapy.

Introduction

Tissue‐resident memory T (T_RM) cells are defined as a memory CD8⁺ T cell subset that does not recirculate. Consequently, T_RM cells have a preferential localization in tissues, which has been demonstrated in cancer, infection, allergy and autoimmunity [1]. T_RM cells are identified by their surface expression of CD103 (αEβ7 integrin), that binds to E‐cadherin expressed on epithelia, and CD69, which antagonizes sphingosine 1‐phosphate receptor 1 (S1PR1)‐mediated egress from tissues [2]. Both CD103 and CD69 contribute to anchoring of T_RM cells in non‐lymphoid peripheral tissues [3]. Additionally, CD49a expression has been demonstrated to characterize T_RM cells in the human epidermis [4].

The microenvironment in the tissues has been reported to induce the development of T_RM cells. One signal reported to play a role is transforming growth factor (TGF)‐β1. TGF‐β1 induces CD103 expression through nuclear factor of activated T cells (NFAT)‐1 and suppressor of mothers against decapentaplegic (Smad)2/3 transcription factors binding to the promoter region of the ITGAE gene, encoding for CD103 [5]. Correspondingly, a similar effect has been demonstrated in TGF‐β‐rich tumours, in which T_RM cells are consequently generated [6].

Reports have demonstrated that a high number of tumour‐infiltrating T_RM cells is linked to improved prognosis in patients with urinary bladder, lung, cervical and ovarian cancers [7, 8, 9, 10]. The improved prognosis is due to the capacity of T_RM cells to produce IFN‐γ upon in‐situ antigen recognition [11], resulting in the recruitment of circulating T cells from the blood [12]. Furthermore, the cytotoxic capacity of T_RM cells against tumour cells is demonstrated by their production of the cytotoxic mediators, perforin, granzyme A, and granzyme B upon CD103/E‐cadherin ligation and T cell receptor (TCR) activation [8]. Notably, the cytotoxic activity of CD8⁺ T cells is initiated by perforin forming the pore that allows granzymes to enter and induce apoptosis of the tumour cells [13]. In contrast, T_RM cells residing in the tumour tissue have been demonstrated to express programmed cell death 1 (PD‐1), suggesting exhaustion [14]. In this context, the phenotype of T_RM cells displaying simultaneous cytotoxicity and exhaustion markers needs further study.

Here, we investigate the properties of perforin expression in T_RM cells, starting at the epigenetic level by means of DNA methylation. DNA methylation is defined as methylation of the cytosine residue in a cytosine–phosphate–guanine (CpG) site [15]. It has been linked to gene silencing due to the inability of transcription factors to bind to the promoter of a particular gene [16]. Moreover, study of the DNA methylation profile from the enhancer region of a gene is more critical. This region contains abundant binding sites for key lineage transcription factors, and enhancer‐initiated transcription occurs as the most rapid transcriptional change when cells commence state differentiation [17]. Additionally, epigenetic modification of the lineage‐specific enhancer corresponds to lineage‐specific cell transcription [18].

In this study we focus upon urothelial urinary bladder cancer (UBC), which in the western world is the fourth most common cancer in men and ninth in women [19]. The non‐muscle‐invasive UBC is treated routinely using Bacillus Calmette–Guerin (BCG) instillation in the bladder, indicating a substantial level of immunogenicity in this cancer [20]. In this study we investigated the DNA methylation profile of the perforin gene enhancer on T_RM cells lodged in the tumour from UBC patients. The study of the DNA methylation profile of this cell subset was accompanied by analysis of the phenotypical properties ex vivo and after in‐vitro stimulation.

Materials and methods

Patient characteristics and tissue collection

Fifty‐three patients diagnosed with UBC were recruited prospectively from nine hospitals in Sweden (Umeå University Hospital, Sundsvall‐Härnösand County Hospital, Västerås County Hospital, Jönköping/Ryhov County Hospital, Vrinnevi County Hospital in Norrköping, Linköping University Hospital, Östersund County Hospital, Uppsala University Hospital and Skellefteå County Hospital) between 2016 and 2018 (Supporting information, Table S1). All patients underwent diagnostic transurethral resection of the bladder (TUR‐B) and patients with muscle invasive tumours proceeded to radical cystectomy, with or without preceding neoadjuvant chemotherapy. Peripheral blood (PB) and tumour tissues were acquired at the time of TUR‐B, whereas sentinel nodes (SN) and PB were obtained at radical cystectomy. PB was collected in sodium heparin tubes (Greiner Bio‐One, Kremsmünster, Austria). SNs were identified perioperatively with detailed mapping, as described previously [21]. Briefly, radioactive tracer ^99mTechnetium was injected into the area surrounding the tumour or tumour scar during the cystectomy operation. Lymphadenectomies were then carried out and the nodes radioactivity was measured by handheld Geiger counter. Tumour tissue and SN were divided into two parts: one part for routine pathological examination and one part for immunological analysis in this study. The tumour and SN for this study were immersed immediately in RPMI‐1640 medium (Sigma Aldrich, St Louis, MO, USA) in 4–8^oC and all the samples were processed within 24 h. Buffy coat samples from healthy donors were acquired from Karolinska University Hospital blood bank, Sweden. Informed consent from all patients recruited in this study was obtained in accordance with the Declaration of Helsinki. The study was approved by the regional ethical board (EPN‐Stockholm, registration number: 2007/71‐31 with amendment 2017/190‐32).

Processing of samples

Peripheral blood mononuclear cells (PBMC) were obtained from whole blood samples of patients and healthy donors by density gradient separation using Ficoll‐Paque Plus (GE Healthcare, Chicago, IL, USA). Isolated PBMC were suspended in AIM‐V medium (Life Technologies, Paisley, UK).

Tumour‐infiltrating lymphocytes (TIL) were isolated from the tumour by slicing the collected tissue into small pieces. The small tissues were then transferred into the C‐tubes (Miltenyi Biotech, Bergisch Gladbach, Germany) containing 10 ml of AIM‐V medium with 1% DNAse I (Sigma Aldrich) and 1% collagenase/hyaluronidase (StemCell Technologies, Cambridge, UK). Tissues inside the C‐tubes were processed in GentleMACS (Miltenyi), according to the manufacturer’s protocol. The processed tissue was filtered through a 40‐μm cell strainer, washed and suspended in AIM‐V medium.

Lymphocytes from SN were isolated by gentle pressure homogenization through a 40‐μm cell strainer in AIM‐V medium. The cell suspension was washed and suspended in AIM‐V medium.

Cell sorting

For the initial DNA methylation site screening, perforin⁺ and perforin^– CD8⁺ T cells were isolated from PBMC of healthy donors. First, bulk CD8⁺ T cells were sorted using EasySep Human CD8 Positive Selection Kit (StemCell Technologies). The selected CD8⁺ T cells were labelled with anti‐CD8 (clone RPA‐T8; BD Biosciences, San Jose, CA, USA) and anti‐CD56 (clone B159; BD Biosciences) antibodies, followed by fixation in HOPE I Fixation solution (DCS Innovative Diagnostik‐Systeme, Hamburg, Germany) overnight in 0–4^oC. The next day, the cells were permeabilized with 0·1% saponin [w/v in phosphate‐buffered saline (PBS)] (Sigma Aldrich) and stained intracellularly with anti‐perforin antibody (clone δG9; BD Biosciences) for sorting by flow cytometry.

Bulk CD8⁺ T cells were presorted from patients’ PBMC, SN and tumour using the EasySep Human CD3 Positive Selection Kit (StemCell Technologies). The selected cells were subsequently sorted by flow cytometry after cell labelling with anti‐CD8 (clone RPA‐T8; BD Biosciences) and anti‐CD56 (clone B159; BD Biosciences) antibodies. Cells to be sorted were defined as CD8⁺CD56^–.

For sorting of different memory CD8⁺ T cell subsets, CD8⁺ T cells were isolated initially from PBMC and tumour of UBC patients using the EasySep Human CD8 Positive Selection Kit. The cells were then labelled with anti‐CD8 (clone RPA‐T8; BD Biosciences), anti‐CD56 (clone B159; BD Biosciences), anti‐CD103 (clone Ber‐ACT8; BioLegend, San Diego, CA, USA), anti‐CCR7 (clone 150503; BD Biosciences) and anti‐CD45RA (clone HI100; BD Biosciences) antibodies and sorted by flow cytometry.

Cell sorting by flow cytometry was performed using the fluorescence‐activated cell sorter (FACS) Aria I instrument (BD Biosciences) and FACS Diva software (BD Biosciences).

DNA extraction and bisulfite conversion

DNA was extracted and bisulfite converted from frozen pellets of sorted cells using the EZ DNA Methylation Direct Kit (Zymo Research, Irvine, CA, USA), according to the manufacturer’s protocol.

TA cloning and bisulfite sequencing

Bisulfite‐converted DNA was amplified by polymerase chain reaction (PCR) using Platinum Taq DNA Polymerase High Fidelity (ThermoFisher, Fremont, CA, USA) in a T100 Thermal Cycler (Bio‐Rad). The primers used (Eurofins Genomics, Ebersberg, Germany) are listed in Supporting information, Table S2. PCR amplicons were gel‐purified using the QIAquick Gel Extraction Kit (Qiagen, Valencia, CA, USA). Next, the PCR amplicons were TA‐cloned into pCR4‐TOPO vector (ThermoFisher) and transformed into TOP10 Escherichia coli (ThermoFisher) by heat shock, according to the manufacturer’s protocol. Transformed cells were cultured on lysogeny broth (LB) plates with 50 μg/ml Kanamycin (Substrate Unit, Karolinska University Hospital) and incubated overnight at 37^oC. The following day, 20 colonies were picked from each plate and cultured overnight in LB medium containing 50 μg/ml Kanamycin (Substrate Unit, Karolinska University Hospital) at 37^oC at 210 rpm. Plasmid DNA was extracted using PureLink Quick Plasmid Miniprep Kit (ThermoFisher) and Sanger sequenced using T3 sequencing primer (Supporting information, Table S2) to establish CpG methylation profile on a single cell level.

Pyrosequencing

DNA amplification of the region containing PRF1 reporter CpG site [–1053 base pairs (bp) upstream of the transcription start site (TSS)] in the enhancer was performed with PCR using Jumpstart REDTaq DNA polymerase (Sigma Aldrich) in T100 Thermal Cycler (Bio‐Rad). One of the primers used for amplification was biotinylated (Biomers) (Supporting information, Table S2). Biotinylated PCR amplicons were purified using the Pyromark Q96 Vacuum workstation (Qiagen), followed by pyrosequencing using the Pyromark Q96 ID instrument (Qiagen) with Pyromark Gold Q96 reagent (Qiagen), according to the manufacturer’s protocol. Analysis of the data was performed using Pyromark Q96 software (Qiagen) to determine the CpG methylation profile.

Cell culture

For the 5‐azacytidine (5‐aza) experiment, CD8⁺ T cells from PB of healthy donors were cultured initially in vitro in AIM‐V medium supplemented with 200 IU/ml recombinant human interleukin (IL)‐2, 5 μg/ml plate‐coated anti‐CD3 and 1 μg/ml anti‐CD28 stimulating antibodies. After 48 h, 5‐aza was added to the culture at a final concentration of 2·5 μM/ml for 48 h. The cells were washed and fresh AIM‐V medium was added for another 48 h. The cells were kept at 37^oC in 5% CO₂ during the culture. At the end of culture, the cells were analysed by flow cytometry.

For the T_RM cells activation assay, T_RM cells were isolated from PB and tumour of patients. The cells were stimulated in vitro in AIM‐V medium supplemented with 20 ng/ml recombinant human IL‐15, 5 μg/ml plate‐coated anti‐CD3 and 1 μg/ml anti‐CD28 stimulating antibodies. The cells were incubated for 48 h at 37^oC in 5% CO₂, after which the cells were harvested and analysed by flow cytometry and pyrosequencing.

Flow cytometry

Single cell suspensions isolated from PB, SN and tumour were stained for surface and intracellular markers for flow cytometry analysis. Briefly, cells were stained with fixable live/dead dye (Life Technologies), followed by surface marker staining using anti‐CD8 (clone RPA‐T8; BD Biosciences), anti‐CD56 (clone B159; BD Biosciences), anti‐CD103 (clone Ber‐ACT8; BioLegend), anti‐CCR7 (clone 150503; BD Biosciences), anti‐CD45RA (clone HI100; BD Biosciences) and anti‐PD‐1 (clone EH12.2H7; BioLegend) antibodies. The cells were then fixated and permeabilized using the forkhead box P3 (FOXP3) transcription factor kit (eBioscience, San Diego, CA, USA). Next, the cells were stained for intracellular marker using: anti‐perforin (clone δG9; BD Biosciences), anti‐granzyme B (clone GB11; BD Biosciences), anti‐T‐bet (clone 4B10; BioLegend) and anti‐Ki‐67 (clone 20Raj1; eBioscience) antibodies. Isotype control was used to ascertain the correct gating for the following markers: perforin, granzyme B, T‐bet, Ki‐67 and PD‐1. Flow cytometry data were acquired using an LSR Fortessa instrument (BD Biosciences) and analysed with FlowJo version 10 (TreeStar, Inc., Ashland, OR, USA).

ViSNE

ViSNE was used as a bioinformatic tool to analyse high‐dimensional single cell data from flow cytometry. The Barnes–Hut Stochastic Neighbor Embedding (bh‐SNE) algorithm was used by the ViSNE tool to map the events. Samples utilized in the analysis were first gated manually using FlowJo version 10 for CD8⁺ T cells and the data were exported as flow cytometry standard (FCS) files with compensated parameters. The measured fluorescence intensity of the parameters included in the analysis was transformed using the arcsin function with a co‐factor of 150. Subsampling was generated randomly from each sample with a total number of cells varying between 2000 and 3000 cells/sample in order to adjust all groups of samples with the same comparable number of events. ViSNE analysis was performed using all parameters that were not subjected to manual gating of CD8⁺ T cell population. Analysis was perfomed using the tool CYT on the MATLAB (Mathworks) platform [22].

Statistical analysis

For comparison of the two independent groups, a two‐tailed independent t‐test was used for parametric data and the Mann–Whitney U‐test was used for non‐parametric data. One‐way analysis of variance (anova) was used to compare the means among three groups or more with parametric data, meanwhile the Kruskal–Wallis test was used when comparing non‐parametric data. Normality test was performed using the Kolmogorov–Smirnov test on all the data in this study; spss version 23 software (IBM, Armonk, NY, USA) was used for all the statistical analyses.

Results

Perforin expression is regulated by DNA methylation in the enhancer region

The enhancer region of the perforin coding gene, PRF1, was identified previously to be located upstream of the TSS (Fig. 1a), and regulates the expression of perforin [23]. This enhancer region is CpG‐rich, with 10 identified CpG sites (chr10: 72, 363, 525–72, 363, 863) (Fig. 1a). In order to investigate the evolutionary conservation of the regulatory element in this locus, we aligned human and mouse DNA sequences using the VISTA tool. We found that the enhancer region of the PRF1 was not conserved evolutionarily between the two species (< 70% conservation) (Fig. 1a). This indicated the greater importance of using a human model to explore DNA methylation of the PRF1 enhancer. Moreover, instead of using gene silencing analysis, we validated that DNA methylation regulated perforin expression, as displayed when treating CD8⁺ T cells with the demethylating agent, 5‐aza, which resulted in elevated perforin expression (Fig. 1b).

Figure 1.

DNA methylation in the enhancer region regulates perforin transcription. (a) Promoter region sequence containing enhancer in the PRF1 gene was aligned between human and mouse using the VISTA tool. The conservation degree is indicated by the peaks, with coloured peaks (blue for exon and light red for the non‐coding sequence) indicating > 70% conservation. Ten cytosine–phosphate–guanine sites (CpGs) in the enhancer are displayed upstream from the transcription start site (TSS). (b) Sorted CD8⁺ T cells from healthy donors were cultured initially in the presence of 200 IU/ml recombinant human interleukin (IL)‐2, 5 μg/ml plate‐coated anti‐CD3 and 1 μg/ml anti‐CD28 stimulating antibodies in vitro. 5‐Azacytidine (5‐aza) was added to the culture at a final concentration of 2·5 μM/ml for 48 h, followed by culture medium replacement and incubation for another 48 h. Flow cytometry data demonstrating perforin expression pre‐ and post‐treatment are shown. (c) Perforin^– and perforin⁺ CD8⁺ T cells were sorted. DNA was extracted and bisulfite‐converted from each cell subset, TA‐cloned to pCR4‐TOPO vector and transformed into TOP10 Escherichia coli. Colonies were picked from each cell subset and their DNA was sequenced to measure DNA methylation status of the 10 CpGs in the enhancer. The data display rows represent individually sequenced clone. (d) The bar graphs display total DNA methylation percentage in each CpG from clones in (c).

In order to identify CpGs responsible for regulating perforin transcription, we examined the methylation profile of the 10 CpGs identified in the enhancer region. Bisulfite‐converted DNA from perforin^– and perforin⁺ CD8⁺ T cells obtained from healthy donors was TA‐cloned and sequenced in order to acquire the methylation status from a single cell level. We discovered that in the perforin^– CD8⁺ T cells, all the CpGs were methylated completely, in contrast to perforin⁺ cells (35% total methylation) (Fig. 1c), indicating that CpG demethylation in this enhancer is crucial for perforin transcription. The three CpGs closest to the TSS (position –1053, –1012 and –994) were the least methylated in perforin⁺ CD8⁺ T cells (Fig. 1d). Correspondingly, CpG in position –1053 was chosen as the reporter for perforin expression.

CD8⁺ T cells possess distinct cytotoxicity in different tissues by epigenetic regulation

Next, we investigated the methylation profile of the PRF1 reporter CpG site (–1053) in CD8⁺ T cells isolated from UBC patients by pyrosequencing of the PCR‐amplified bisulfite‐converted target locus. We discovered no significant difference between CD8⁺ T cells from tumour (mean = 44·1%) and PBMC (mean = 46·6%) (P = 0·95) (Fig. 2a). In contrast, CD8⁺ T cells from SN had higher methylation in the PRF1 reporter CpG site (mean = 66·8%) compared to PBMC and tumour (both P < 0·05) (Fig. 2a).

Figure 2.

Distinct epigenetic regulation of perforin in CD8⁺ T cells from different tissues. (a) DNA methylation profile of PRF1 reporter cytosine–phosphate–guanine (CpG) site (–1053) was measured by pyrosequencing of polymerase chain reaction (PCR)‐amplified bisulfite‐converted DNA isolated from CD8⁺ T cells from peripheral blood mononuclear cells (PBMC), sentinel nodes (SN) and tumour (n = 15). The data are the mean with error bars indicating standard error of the mean (s.e.m.); one‐way analysis of variance (anova) was used as the statistical test. (b) The frequency of perforin⁺ CD8⁺ T cells from PBMC, SN and tumour (n = 25) was measured by flow cytometry. The data are the mean with error bars indicating s.e.m.; Kruskal–Wallis was used as the statistical test. (c) The expression of perforin in CD8⁺ T cells from different tissues (n = 25) were measured by flow cytometry. The gate was based on isotype control and the frequency of perforin⁺ cells was counted out of CD8⁺ T cells. Dot‐plots display representative data from one patient. *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.

To validate the DNA methylation status to the expression of perforin, we performed flow cytometry on CD8⁺ T cells from UBC patients. Correspondingly, the number of perforin‐expressing CD8⁺ T cells was nine times higher in tumour (P < 0·001) and 17 times higher in PBMC (P < 0·001) in comparison to SN (Fig. 2b and c). In addition, there was a significant difference between the frequency of perforin⁺ CD8⁺ T cells from tumour and PBMC (P = 0·01) (Fig. 2b). The findings indicate epigenetic regulation of the CD8⁺ T cell phenotype in different tissues with regard to the cytotoxic constituent perforin.

Tissue‐resident memory T cells in tumour are cytotoxic but exhausted

To explore further potential phenotypical differences between CD8⁺ T cells from tumour and PBMC, not revealed by DNA methylation of the PRF1 locus (Fig. 2a), we characterized memory CD8⁺ T cell subsets using flow cytometry and ViSNE analysis. Tissue‐resident memory T (T_RM) cells, based on CD103 expression (Fig. 3a), were discovered to be increased significantly in the tumour (mean = 56·8%) compared to PBMC (mean = 2·7%) (P < 0·0001) (Fig. 3b and c). Interestingly, a small number of T_RM cells could be observed both in the blood (PBMC) of UBC patients and healthy individuals (Supporting information, Fig. S1).

Figure 3.

Cytotoxic tissue‐resident memory T cells in tumour despite of exhaustion. (a) ViSNE analysis was performed from flow cytometry data of CD8⁺ T cells from peripheral blood mononuclear cells (PBMC) and tumour (n = 8). Tissue‐resident memory T (T_RM) cells are indicated in green based on CD103 expression. (b) ViSNE analysis of CD8⁺ T cells derived from PBMC and tumour (n = 8) were plotted (red  =  PBMC and yellow  =  tumour). (c) T_RM cell frequency in PBMC and tumour (n = 13) were measured by flow cytometry and counted out of total CD8⁺ T cells. The bar graph shows the mean with error bars indicating standard error of the mean (s.e.m.). Independent t‐test was used as statistical test. (d) The parameter expressions, in which the algorithm of ViSNE in (a,b) was calculated, are displayed with colour indicator representing fluorescence intensity (n = 8). (e) DNA methylation profile of PRF1 reporter cytosine–phosphate–guanine site (CpG) site (–1053) was measured by pyrosequencing of polymerase chain reaction (PCR)‐amplified bisulfite‐converted DNA isolated from T_RM cells of PBMC and tumour (n = 10). The bar graph shows the mean with error bars indicating s.e.m.; independent t‐test was used as statistical test. (f,g) Fraction of perforin⁺ T_RM cells or programmed cell death 1 (PD‐1)⁺ T_RM cells in PBMC and tumour (n = 8) were measured by flow cytometry. The bar graph shows the mean with error bars indicating s.e.m.; independent t‐test was used as statistical test. *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.

Furthermore, DNA methylation of the PRF1 reporter CpG site (–1053) in tumour‐derived T_RM cells was decreased twofold compared to PBMC (P < 0·0001) (Fig. 3e). Consequently, perforin‐expressing T_RM cells were increased in tumours (P = 0·008) (Fig. 3d and f). In addition, we found that the fraction of T_RM cells expressing granzyme B was increased in the tumours compared to PBMC (P < 0·001) (Fig. 3d, Supporting information, Fig. S2a).

The tumour contained a significantly higher number of T_RM cells that expressed the exhaustion marker PD‐1 compared to PBMC (mean tumour = 79% versus mean PBMC = 21·9%) (P < 0·0001) (Fig. 3d and g). Additionally, T‐bet expression was low in tumours (Fig. 3d, Supporting information, Fig. S2b), suggesting exhaustion of tumour‐derived T_RM cells.

Tissue‐resident memory T cells can be activated despite signs of exhaustion

In order to investigate the functional capacity of T_RM cells despite the signs of exhaustion (high PD‐1 and low T‐bet expression), we activated T_RM cells from tumour and PBMC with IL‐15 in the presence of anti‐CD3 and anti‐CD28 stimulating antibodies. Following 48 h of stimulation, the number of perforin‐expressing T_RM cells from the tumour increased twofold compared to PBMC (P = 0·008) (Fig. 4a and b), with no difference in the proportion of granzyme B⁺, PD‐1⁺ and T‐bet⁺ cells (Supporting information, Figure S3). No prominent change in the DNA methylation of PRF1 signature locus was observed (Supporting information, Fig. S1).

Figure 4.

Tissue‐resident memory T cells in tumour are not terminally exhausted. (a) Tissue‐resident memory T (T_RM) cells from peripheral blood mononuclear cells (PBMC) and tumour (n = 3) were activated using 20 ng/ml recombinant human interleukin (IL)‐15, 5 μg/ml plate‐coated anti‐CD3 and 1 μg/ml anti‐CD28 stimulating antibodies for 48 h. The expression of perforin post‐stimulation was measured by flow cytometry. The gate was based on isotype control and the frequency of perforin⁺ cells was counted out of T_RM cells. Dot‐plots display representative data from one patient. (b) The frequency of perforin⁺ T_RM cells in PBMC and tumour (n = 3) from (a) is shown. The bar graph demonstrate the mean with error bars indicating standard error of the mean (s.e.m.); independent t‐test was used as statistical test. (c) The expression of T‐bet and programmed cell death 1 (PD‐1) in T_RM cells from tumour (n = 3) were measured by flow cytometry post‐culture, comparing unstimulated and stimulated groups. The gate was based on isotype control. Dot‐plots display representative data from one patient. (d) The frequency of PD‐1⁺ T‐bet⁺ T_RM cells in unstimulated and stimulated groups (n = 3) from (c) is shown. The bar graph shows the mean with error bars indicating s.e.m.; independent t‐test was used as statistical test. (e) The same as in (c), but the dot‐plots show the expression of Ki‐67 in T_RM cells (n = 3). (f) Same as in (d), but the graph shows the frequency of Ki‐67⁺ T_RM cells from (e) (n =3). *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.

Moreover, the frequency of T_RM cells expressing PD‐1⁺ T‐bet⁺ increased 21 times following stimulation (P = 0·002) (Fig. 4c and d). Additionally, Ki‐67 expression was elevated significantly in stimulated T_RM cells, demonstrated by a 6·6 times increase after IL‐15 and T cell receptor (TCR)‐stimulation (P = 0·007) (Fig. 4e and f). Taken together, PD‐1‐expressing T_RM cells from the tumour possessed great cytotoxic and proliferative potential following activation of the cells, demonstrating that they are not terminally exhausted.

Tissue‐resident memory T cells are the most dominant subset in the tumour

As we have demonstrated the cytotoxic and functional capacity of T_RM cells when they resided in the tumour, the next step was to compare the phenotype of T_RM cells to other memory cell subsets present. ViSNE analysis was performed on total CD8⁺ T cells isolated from the tumour. As mentioned earlier, T_RM cells co‐expressed CD49a and CD69 alongside CD103 and constituted the dominant memory cell subset in the tumour (P < 0·001) (Fig. 5a and b). However, other memory cell subsets were also present, such as central memory T (T_CM) cells (CD103^–CD45RA^–CCR7⁺, mean = 4·25%), effector memory T (T_EM) cells (CD103^–CD45RA^–CCR7^–, mean = 27·3%) and effector memory T cells with CD45RA up‐regulation (T_EMRA) (CD103^–CD45RA⁺CCR7^–, mean = 8·7%) (Fig 5a and b).

Figure 5.

Tumour tissue is highly infiltrated by tissue‐resident memory T cells. (a) ViSNE analysis was performed on tumour CD8⁺ T cells data measured by flow cytometry (n = 4). Distinct CD8⁺ T cell memory subsets were plotted [red = tissue‐resident memory T (T_RM), green = T_CM, blue = T_EM and yellow = T_EMRA]. The parameter expressions, in which the algorithm of ViSNE was calculated, are displayed with colour indicator representing fluorescence intensity. (b) The frequency of memory CD8⁺ T cell subsets were measured by flow cytometry and counted out of total CD8⁺ T cells (n = 13). The bar graph shows the mean with error bars indicating standard error of the mean (s.e.m.); one‐way analysis of variance (anova) was used as statistical test. (c) DNA methylation profile of PRF1 reporter cytosine–phosphate–guanine (CpG site) (–1053) was measured by pyrosequencing of polymerase chain reaction (PCR)‐amplified bisulfite‐converted DNA isolated from different tumour memory CD8⁺ T cell subsets (n = 10). The bar graph shows the mean with error bars indicating s.e.m.; one‐way anova was used as statistical test. (d,e) Fraction of perforin⁺, programmed cell death 1 (PD‐1)⁺ or T‐bet⁺ in different memory CD8⁺ T cell subsets were measured by flow cytometry (n = 8). The bar graph shows the mean with error bars indicating s.e.m.; one‐way anova was used as statistical test. *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.

When investigating the DNA methylation profile of the PRF1 reporter CpG site (–1053) in each cell subset, T_RM cells and T_EM cells demonstrated intermediate methylation (mean = 32·9 and 42·6%, respectively) with significant lower to T_CM cells, but higher when compared to T_EMRA cells (Fig. 5c). This corresponded inversely with the number of perforin⁺ cells among T_RM and T_EM cells, which was significantly higher compared to T_CM cells (P < 0·01) and lower than terminally differentiated T_EMRA cells (P < 0·05) (Fig. 5d). This expression pattern was also observed for granzyme B (Supporting information, Fig. S5).

The fraction of PD‐1‐expressing cells was the highest among T_RM cells when compared to other cell subsets (P < 0·001) (Fig. 5e, left panel), which resulted consequently in a significantly lower expression of T‐bet (P = 0·014 T_RM versus T_EMRA) (Fig. 5e, right panel). We conclude that the T_RM cells are dominant in the tumours demonstrating higher cytotoxic quality compared to the other cell subsets.

High infiltration of tumour tissue‐resident memory T cells corresponds with lower tumour stage

Because T_RM cells were demonstrated to have positive cytotoxic properties, we correlated the proportion of different memory CD8⁺ T cell subsets infiltrating the tumour to clinical tumour–node–metastasis (TNM) stage in UBC patients. We observed that in the patients with stage I (T1 N0 M0), there was a trend towards dominant T_RM cell infiltration in the tumour (Fig. 6a). Notably, patient 48 had a lower T_RM cell number, similar to high‐stage tumour. Hypothetically, it was due to the high‐risk stage I tumour (cT1 + carcinoma in situ) in this patient. In contrast, in patients with muscle invasive tumour, i.e. stage II (T2 N0 M0) and stage III (T3 N0 M0), the trend demonstrated a lower fraction of T_RM cells in the tumour (Fig. 6a).

Figure 6.

The increased number of tissue‐resident memory T (T_RM) cells in tumours correlates with lower tumour stage. (a) The proportion of distinct memory CD8⁺ T cell subsets in the tumour was measured by flow cytometry and displayed from each urinary bladder cancer (UBC) patient (n = 10). Patients were grouped based on the tumour–node–metastasis (TNM) tumour stage judged clinically. Stage I = T1 N0 M0, stage II = T2 N0 M0 and stage III = T3 N0 M0. (b) DNA methylation profile of PRF1 reporter cytosine–phosphate–guanine (CpG) site (–1053) was measured by pyrosequencing of polymerase chain reaction (PCR)‐amplified bisulfite‐converted DNA isolated from tumour T_RM cells and compared between patients with stage I and ≥ stage II (n = 10). The bar graph shows the mean with error bars indicating standard error of the mean (s.e.m.); Mann–Whitney was used as statistical test. (c) The frequency of programmed cell death 1 (PD‐1)⁺ T_RM cells from tumour was measured by flow cytometry and compared between patients with stage I and ≥ stage II (n = 10). The bar graph shows the mean with error bars indicating s.e.m.; Mann–Whitney was used as statistical test.

Furthermore, when comparing DNA methylation in the PRF1 locus, there was no significant difference between patients with non‐muscle invasive (stage I) and muscle‐invasive (≥ stage II) tumours (P = 0·571) (Fig. 6b). In addition, the PD‐1‐expressing T_RM cell number was not seen to be different between the two stage groups (P = 1·0) (Fig. 6c). This suggests that the tumour T_RM cells have similar phenotypes among tumour stages.

Discussion

Here, for the first time to our knowledge, we demonstrated that phenotypically exhausted T_RM cells are epigenetically available for the transcription of perforin and harbour effector functions, which may correlate with a less advanced tumour stage. Our findings suggest that T_RM cells residing in the tumour tissue of urothelial UBC have low DNA methylation in the reporter CpG site located in the enhancer of the PRF1 locus (Fig. 3e). This discovery has a solid correlation with the perforin protein expression inside the cells (Fig. 3f) and renders the possibility to activate tumour T_RM cells (Fig. 4). Moreover, a high number of T_RM cells infiltrating the tumour corresponds to a lower tumour stage in UBC patients (Fig. 6a). Epigenetic profiling of T_RM cells may help us in revealing mechanisms that can be targeted to reverse exhaustion and enhance cytotoxic potency in future cancer immunotherapy regimens.

The rationale for investigating the enhancer region of the PRF1 gene relies on the fact that the regulation by enhancers are proven to be highly dynamic upon CD8⁺ T cell activation and differentiation [24]. Furthermore, the transcription of PRF1 is regulated mainly by the enhancer located at position –1kb from the TSS, which has binding sites for transcription factors related to T cell activation, such as signal transducer and activator of transcription (STAT)‐5, nuclear factor kappa B (NF‐κB) and activator protein 1 (AP‐1) [25]. Hence, it is compelling to examine the DNA methylation status of the CpG signature site in the enhancer region, where we found a good correlation with perforin expression (Fig. 2). However, we observed that PRF1 methylation was equivalent in PBMC and tumour CD8⁺ T cells, but the perforin protein expression is lower in the tumour (Fig. 2). This discrepancy may be due to additional mechanisms that regulate perforin expression, such as microRNAs [26] or histone modifications.

It is widely acknowledged that DNA methylation has a critical role in gene expression and specific cell lineage maintenance. Therefore, DNA methylation establishes a stable epigenetic mark which can be inherited through successive cell divisions [27]. In the context of CD8⁺ T cell differentiation, a shift in DNA methylation is not a straightforward phenomenon once CD8⁺ T cells have retained their specific memory subset. Accordingly, effector‐associated loci, such as PRF1, have a stably unchanged DNA methylation in peripheral blood‐derived naive, T_CM and T_EM CD8⁺ T cells, even after three rounds of cell division by ex‐vivo stimulation using homeostatic cytokines IL‐7 and IL‐15 [28].

These insights prompted us to study DNA methylation of the PRF1 locus in tissue‐resident memory T (T_RM) cells, which have unique transcriptional signatures compared to T_CM and T_EM cells [3]. T_RM cells are defined as memory CD8⁺ T cells that reside in the non‐lymphoid tissues and have a distinct functional profile [29]. Here, we demonstrated that the PRF1 locus of T_RM cells was hypomethylated compared to T_CM and T_EM cells from the tumour (Fig. 5c), which corresponded to an increased fraction of perforin‐expressing T_RM cells (Fig. 5d). Although we could see that terminally differentiated T_EMRA cells displayed the lowest DNA methylation in the PRF1 locus, leading to consequently higher perforin expression compared to T_RM cells, we know that T_EMRA cells are short‐lived in contrast to T_RM cells [30]. Moreover, long‐lived T_RM cells are demonstrated to elicit more potent local immune protection in the tissue against infection [31].

In cancer, T_RM cells are reported to delay tumour growth and prolong survival [32]. Additionally, T_RM cells are necessary for the efficacy of cancer vaccine [11, 33]. The protective effect of T_RM cells in cancer is linked to their high degree of cytotoxicity, which is correlated with the magnitude of their perforin and granzyme expression [14]. In spite of expressing cytotoxic mediators, T_RM cells in the tumour express the exhaustion markers PD‐1, T cell immunoglobulin and mucin domain 3 (TIM‐3) and lymphocyte‐activation gene 3 (LAG‐3) [34]. In accordance with these previous findings, in this study we have demonstrated a high number of PD‐1⁺ T_RM cells from the tumour (Fig. 3g) and a low fraction of T‐bet⁺ cells in this subset (Supporting information, Fig. S2b). However, when we activated T_RM cells from tumours using the homeostatic cytokine IL‐15 and T cell receptor (TCR) stimulation in vitro, we observed that the dominant PD‐1⁺ T_RM cells began to gain T‐bet expression (Fig. 4c and d). Consequently, this resulted in elevated perforin and granzyme B expression in T_RM cells from tumour and increased proliferation, i.e. elevated Ki‐67 (Fig. 4e and f). Taken together, this indicates that the tumour‐derived T_RM cells are not terminally exhausted. Consistent with our data, skin T_RM cells have been described to have high expression of PD‐1 and other co‐inhibitory molecules after viral infection, thus suggesting exhaustion. Surprisingly, following antigenic challenge, in‐situ proliferation of T_RM cells, marked by up‐regulation of Ki‐67, occurs [35]. Therefore, we may speculate that the T_RM cell is a memory CD8⁺ T cell subset with an outstanding potential for cytotoxicity and is not terminally exhausted.

The mechanism behind the co‐expression of exhaustion markers and cytotoxic mediators seen in T_RM cells may be explained as part of T cell activation [36]. We know that PD‐1 expression in CD8⁺ T cells results from chronic antigenic exposure in the tumour tissue [37]. However, recent evidence describes that in the case of chronic infection, permanent PD‐1 expression in T_RM cells is imprinted epigenetically and is antigen‐independent [38, 39]. Additionally, PD‐1 signalling has been demonstrated to be important for antigen‐independent maintenance of memory CD8⁺ T cells [40]. Therefore, we may speculate that PD‐1 expression is a mere T_RM cell marker that does not affect cell functionality, and may be seen as an engagement marker with tumour cells. Accordingly, it was seen in our data that PD‐1 expression was similar at different stages of the tumour (Fig. 6c).

Notably, our finding indicated that T_RM cells present in the peripheral circulation had the capacity to increase their expression of granzyme B, PD‐1 and T‐bet, following IL‐15 and TCR stimulation. The results were comparable to tumour T_RM cells (Supporting information, Fig. S3). It is important to note that in this context PD‐1 acts as an indicator of T cell activation instead of exhaustion [41]. Therefore, T_RM cells from the circulation may have a promising feature of activation following TCR engagement and possibly CD103/E‐cadherin ligation once they lodge in the tumour tissue [6, 42, 43]. However, their specificity towards distinct tumour antigens needs to be investigated further.

The limitation of this study was that we correlated T_RM cell numbers with the tumour stage in only 10 patients. However, from this observation we could recapitulate the trend in which high T_RM cell infiltration in the tumours corresponded to less advanced tumour stage, suggesting the potential anti‐tumour capacity of T_RM cells. Therefore, further correlation between patients’ clinical prognosis with T_RM cell numbers is required in a larger cohort.

In conclusion, epigenetic profiling adds to our understanding about T_RM cell phenotypes and functionality. These findings help us to recognize the cytotoxic potential of T_RM cells and therefore provide us with new strategies for cancer immunotherapy.

Disclosure

None.

Author contributions

C. A. H., A. S. and O. W. designed the study, developed the methodology, wrote and reviewed the manuscript. C. A. H, E. A. B., A. B. and S. B. performed the experiments. M. J., F. A., T. J., Y. H., F. A., K. P., B. H., K. R. and A. S. provided clinical samples and patients’ clinical information. A. S. assembled the clinical information database of all patients. C. A. H. and S. B. performed data analysis. H. G. supervised the study.

Supporting information

Fig S1. A small fraction of T_RM cells is present in the blood of UBC patients and healthy individuals. ViSNE analysis was performed on PBMC CD8⁺ T cells data measured by flow cytometry (n = 7). Tissue‐resident memory T (T_RM) cells are indicated in green based on CD103 expression. PBMC CD8⁺ T cells from healthy donors and UBC patients were plotted (red = healthy donors and yellow = UBC patients).

Fig S2. Cytotoxic tissue‐resident memory T cells in tumour despite of exhaustion. (a) The frequency of granzyme B⁺ T_RM cells in PBMC and tumour (n = 8) were measured by flow cytometry and counted out of T_RM cells. The bar graph shows means with error bars indicating SEM. Independent t‐test was used as statistical test. (b) Same as in (a), but the data show T‐bet⁺ T_RM cell frequency (n = 8). Mann‐Whitney was used as statistical test. *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.

Fig S3. T_RM cells from blood can potentially be activated. (a) The frequency of granzyme B⁺ T_RM cells from PBMC and tumour (n = 3) in unstimulated and stimulated condition with IL‐15, anti‐CD3 and anti‐CD28 stimulating antibodies were measured by flow cytometry and counted out of T_RM cells. The bar graph shows means with error bars indicating SEM. Independent t‐test was used as statistical test. (b)(c) Same as in (a), but the data show the frequency of PD‐1⁺ T_RM cells or T‐bet⁺ T_RM cells (n = 3). *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.

Fig S4. DNA methylation of PRF1 locus does not change following activation. DNA methylation profile of PRF1 reporter CpG site (‐1053) was measured by pyrosequencing of PCR‐amplified bisulfite‐converted DNA isolated from pre‐culture, unstimulated and stimulated (IL‐15 + αCD3 + αCD28) tumour T_RM cells. The data was acquired from one representative patient.

Fig S5. T_RM cells are cytotoxic in the tumour. The fraction of tumour‐derived memory CD8⁺ T cell subsets expressing granzyme B were measured by flow cytometry (n = 8). The bar graph shows means with error bars indicating SEM. One‐way ANOVA was used as statistical test. *P < 0·05, **P < 0·01, ***P < 0·001, ****P < 0·0001.