PD1hi CD200hi CD4+ exhausted T cell increase immunotherapy resistance and tumour progression by promoting epithelial–mesenchymal transition in bladder cancer

Abstract Background Bladder cancer (BLCA) is one of the most diagnosed cancers in humans worldwide. Recently, immunotherapy has become a main treatment option for BC. However, most BLCA patients do not respond to immune checkpoint inhibitors or relapse after immunotherapy. Therefore, it is very important to identify novel biomarkers for the prediction of immunotherapy response in B patients. Methods Pancancer single‐cell RNA sequencing (scRNA‐seq) data were used to identify the clusters of CD4+ T cells in the tumour microenvironment (TME). The clinical significance of key CD4+ T‐cell clusters was evaluated based on the survival data of two independent immunotherapy bladder cancer (BLCA) cohorts. We also investigated the function of key clusters of CD4+ T cell in the TME of BC cells in vitro. Results This study identified two novel exhausted CD4+ T‐cell subpopulations with the expression of PD1hi CD200hi or PD1hi CD200low in BC patients. Moreover, BLCA patients with a high level of PD1hi CD200hi CD4+ exhausted T cell showed immunotherapy resistance. Cell function analysis demonstrated that PD1hi CD200hi CD4+ exhausted T cell can promote epithelial–mesenchymal transition (EMT) and angiogenesis in BLCA cells. In addition, PD1hi CD200hi CD4+ exhausted T cells were shown to communicate with malignant BLCA cells through the GAS6–AXL axis. Finally, we also found that GAS6 expression is upregulated in B cells by METTL3‐mediated m6A modification. Conclusions PD1hi CD200hi CD4+ exhausted T cell may serve as a novel biomarker for poor prognosis and immunotherapy resistance in B. Targeted inhibitors of PD1hi CD200hi CD4+ exhausted T cells may help improve the efficacy of immunotherapy.

Xuefei Liu, School of Medicine, Southern University of Science and Technology, Shenzhen 518055, P. R. China. Email: 12231372@mail.sustech.edu.cn might offer a promising therapeutic option for bladder cancer patients with high proportions of PD1hi CD200hi CD4 + exhausted T cells.
Results: This study identified two novel exhausted CD4 + T-cell subpopulations with the expression of PD1 hi CD200 hi or PD1 hi CD200 low in BC patients. Moreover, BLCA patients with a high level of PD1 hi CD200 hi CD4 + exhausted T cell showed immunotherapy resistance. Cell function analysis demonstrated that PD1 hi CD200 hi CD4 + exhausted T cell can promote epithelial-mesenchymal transition (EMT) and angiogenesis in BLCA cells. In addition, PD1 hi CD200 hi CD4 + exhausted T cells were shown to communicate with malignant BLCA cells through the GAS6-AXL axis. Finally, we also found that GAS6 expression is upregulated in B cells by METTL3-mediated m6A modification.
Conclusions: PD1 hi CD200 hi CD4 + exhausted T cell may serve as a novel biomarker for poor prognosis and immunotherapy resistance in B. Targeted inhibitors of PD1 hi CD200 hi CD4 + exhausted T cells may help improve the efficacy of immunotherapy.

K E Y W O R D S
CD200, CD4 exhausted T cells, GAS6, N6-methyladenosine

BACKGROUND
An estimated 81 180 new cases of urinary bladder cancer (61 700 men and 19 480 women) will be diagnosed in the United States in 2022, with approximately 17 100 deaths (12 120 men and 4 980 women) occurring during this same period. 1 Bladder cancer has become an increasingly prominent public health issue due to its high metastatic propensity and increased immunotolerance. 2,3 In the treatment of bladder cancer, immune checkpoint inhibitors (ICIs) targeting CTLA-4 and PD-1 have proven to be an effective strategy. [4][5][6] At present, the mechanism of antitumour immunity induced by immunotherapy mainly focuses on CD8 + exhausted T cells. 7 However, ICIs monotherapy has only an efficacy of only 20% in bladder cancer patients. 6,8,9 An increasing number of studies have shown that CD4 + T cells are susceptible to exhaustion and may contribute to the reactivation of the antitumour immune response after ICI treatment in multiple cancer types, including bladder cancer. 10,11 Therefore, it is necessary to explore the role of CD4 + exhausted T cells in bladder cancer. Upon antigen-driven activation, naive CD4 + T cells expand and differentiate into a broad spectrum of functional subsets, including T-helper 1 (Th1), T-helper 2 (Th2), T-helper 9 (Th9), T-helper 17 (Th17), regulatory T cells (Tregs) and follicular T helper (TFH) cells. 12,13 Activation of different T helper cell subtypes may exert contradictory effects on tumour immune response. Th1 cells can not only assist CD8 + T cells but also promote tumour vessel normalization by secreting IFN-γ, to inhibit cancer progression. 14,15 In contrast, regulatory CD4 + T cells (Tregs) can promote cancer development and progression by inhibiting the immune response against tumours. In addition, Tregs participate in the epithelial-mesenchymal transition (EMT) of cancer cells and angiogenesis by secreting growth factors 15,16 and support the growth of stromal cells (such as fibroblasts and endothelial cells). 17,18 Moreover, CD4 + T cells seem to be highly plastic under the influence of specific cytokines, allowing them to switch between subsets, thereby extending their functional range. 19 In this study, we identified PD1 hi CD200 hi CD4 + exhausted T cells in bladder cancer, which are closely associated with poor prognosis and immunotherapy resistance. Moreover, PD1 hi CD200 hi CD4 + exhausted T cells can promote EMT and angiogenesis in bladder cancer cells in vitro. Finally, we observed that malignant cells can recruit PD1 hi CD200 hi CD4 + exhausted T cells to induce EMT in bladder cancer cells by releasing m6A-mediated GAS6.

2.1
Single-cell transcriptome sequencing collection and data preprocessing 2.1. 1 Pan-CD4 single-cell transcriptome sequencing Single-cell transcriptome sequencing (scRNA-seq) data have been downloaded or requested from the authors of published studies. The details of each dataset are shown in Tables S1 and S2. For each sample, we used the Cell Ranger Single-Cell toolkit to align reads for each sample based on the human reference genome GRCh38 (https://cf.10xgenomics.com/supp/cell-exp/refdata-gex-GRCh38-2020-A.tar.gz).
The Seurat (v3.1.3) R toolkit was used to analyse scRNAseq data for each cancer type. 20 First, cells with more than 20% mitochondrial RNA were removed, as well as cells with UMI less than 200 or UMI greater than 6000. 'DoubletFinder' 21 was used to predict and remove doublets. For each patient, we considered the rate of doublet cells to be filtered to be 4%, using five principal components and a default value of 20% for pN.
Then, the scRNA-seq data for each cancer type were normalized for each sample using the 'NormalizeData' function in Seurat, and only the highly variable genes were retained using the 'FindVariableFeatures' function in Seurat. Next, the 'Runharmony' function in the harmony package was used to integrate the gene expression matrix of all samples, which adjusts for batch effects between samples. 22 The 'RunPCA' function was used to perform principal component analysis (PCA), and the 'FindNeighbors' function is used to construct a K-nearest neighbour graph. The most representative principal components (PCs) selected based on PCA were used for cluster analysis with the 'FindCluster' function to identify different cell types. Subclusters with high expression of CD3E, CD3G and CD3D were annotated as T cells. Separation of CD4 + T lymphocyte subclusters from T lymphocytes based on CD4 expression. We combined the expression matrix of CD4 + T lymphocytes of all cancer types and used the above clustering approach to obtain a pancancer CD4 + T lymphocyte map. In this step, we also used the Runharmony function to extract batch differences for each patient in each dataset.
Differentially expressed genes in each subcluster were identified based on the 'wilcoxauc' function in the presto package. The log-transformed fold change (log2FC), which was used to determine the magnitude of the difference, was taken into account in our single-cell RNA sequencing analysis. We considered differentially expressed genes with log 2 FC > .3 and p-values <.05 to be potentially biologically significant.

2.1.2
Endothelial and epithelial cell single-cell transcriptome sequencing of BLCA datasets We used the same approach as pan-CD4 scRNA-seq to analyse scRNA-seq data from two BLCA datasets and identified eight major clusters and annotated them according to the expression of typical gene markers, including endothelial, epithelial, fibroblast, myeloid, mast, plasma, B and T cells. CD4, CD8A and CD8B gene expressions were used to define CD4 + and CD8 + T cells. Endothelial cell and epithelial cell subclusters were named using the first marker gene.

Tissue preference statistic
In this paper, we use Ro/e values to estimate the tissue preference of each subcluster, with Ro/e denoting the ratio of the number of cells observed in a cluster to the expected number of cells. The expected number of cells per cell cluster in each tissue was obtained from the χ 2 test. Ro/e > 1 for a subcluster in a given tissue indicates that the cluster is enriched in that tissue.

Bulk RNA sequencing data collection and processing
Bulk RNA sequencing transcriptome data of patients with BLCA and the corresponding clinical data and mutation profiles were downloaded from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/, accessed on 27 October 2022). Fragments per kilobase million values were converted to transcripts per kilobase million values, and the Ensembl gene IDs of the RNA-seq data were converted to gene symbols concerning the annotation file. By using the median value of PDCD1 and CD200 gene expression of each patient, we first divided the patients into two groups, the PDCD1 hi group and the PDCD1 low group, and then divided the PDCD1 hi group into two groups according to the expression of CD200.
We included two immunotherapeutic BLCA cohorts and obtained the relative transcriptomic and clinical data from the online Supporting Information section appended to the published studies. We downloaded RNA sequencing data from the immunotherapy BLCA cohort (IMvigor210) from the Mvigor210CoreBiologies package. 9 Another immunotherapy BLCA cohort (GSE176307) was downloaded from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/, accessed on 27 October 2022) and included for analysis. 23 In addition, we included five additional immunotherapeutic cohorts with different types of cancer and obtained relevant transcriptomic and clinical data from online Supporting Information section from published studies. RNA sequencing data from an immunotherapy melanoma cohort (SKCM, GSE91061) 24 and non-small cell lung cancer cohort (NSCLC, GSE135222) 25 were downloaded from GEO. The raw data were downloaded as microarray data. Another immunotherapy melanoma cohort (SKCM, phs00052) was included in the analysis and downloaded from the Database of Genotypes and Phenotypes (dbGaP, https://dbgap.ncbi.nlm.nih.gov/, accessed on 20 November 2022). An immunotherapy cohort of stomach adenocarcinoma (STAD, PRJEB25780) was included for analysis and downloaded from the European Nucleotide Archive dataset (ENA, https://www.ebi.ac.uk/ena/browser/home, accessed on 20 November 2022). Finally, the data from an immunotherapy renal cell carcinoma cohort (RCC, Braun_2020) were included and are shown in Table S5.

2.4
Pathway enrichment analysis of bulk RNA-seq and scRNA-seq The identification of differentially expressed genes between the CD4_ex1 and CD4_ex2 subclusters was carried out using the Limma R package. 26 To explore the phenotype-specific signalling pathways of the tumour microenvironment (TME) in the CD4_ex1 subcluster, the gene set enrichment analysis (GSEA) approach was used with an adjusted p < .05 using the 'clusterProfiler' R package. 27 Next, we also used the Limma R package to identify differentially expressed genes between the PDCD1 hi CD200 hi group and the other two groups. We also used the 'clusterProfiler' R package to probe the hallmark signalling pathway in the PDCD1 hi CD200 hi group with an adjusted p < .05.

2.5
Immunological characteristics of the TME in BLCA cohort The proportions of immune cell types evaluated for immune cell infiltration in each sample (with immune infiltration scores) were computed using the CIBERSORT (https://cibersort.stanford.edu/) and the Xcell algorithm. 28

Cell-cell communication analysis
Cell-cell communication between different cell subclusters was analysed with CellChat (Version 1.1.3, R package) following the standard protocol. 29 We used the CellChat package to quantitatively characterize and comparatively infer the probability of cell-cell communication between the CD4_ex2 subcluster and all subclusters of epithelial cells, as well as the ligands and receptors for intercellular communication.

Cell developmental trajectory analysis
RNA velocity analysis was conducted by using velocyto and scVelo. 30 First, we used the 10× velocyto pipeline to count spliced and unspliced reads for each sample from cellranger-generated BAM files. To predict the root and terminal states of the underlying Markov process, the respective scVelo functions were applied. We also used the Python package PAGA to verify the pseudotime between each epithelial cluster. 31 Pseudotime trajectory analysis was constructed based on Monocle2 32 (version 2.18.0, R package) following the tutorial to order the three epithelial cell subclusters (Epi_CXCL1, Epi_OLFM4 and Epi_COL1A2) in BLCA data. To avoid omitting some important genes, we selected all genes for completely unsupervised trajectory analysis. For the differentiation trajectory analysis, we selected the top 50 pseudotime-related differentially expressed genes for these 3 clusters for further analysis.

Metabolic cell-cell communication
We used MEBOCOST which is a Python-based computational tool to infer metabolite, mediated cell-cell communication events. The cut-off parameter was set to . 25. Other parameters were set as default.

Patients and tissue samples
Peripheral blood and tumour samples were obtained from the Sun Yat-sen University Cancer Center, Sun Yat-sen University, Guangzhou, China. The patients included in this study were as follows: 18 BLCA patients, 10 non-small cell lung carcinoma (NSCLC) patients and 8 esophageal squamous cell cancer (ESCC) patients. All patients were pathologically diagnosed and collected with informed consent. This study was performed with the approval of the Ethics Committee of Sun Yat-sen University Cancer Center (GZR2018-053 and GZR2018-120).

Cell lines
The human microvasculature endothelial cell line HMEC1 was purchased from Biospecies, Ltd. with cell authentication via the STR multiamplification method. Human bladder cancer cells UMUC-3 and T24 were obtained from Sun Yat-sen University Cancer Center. UMUC-3 and T24 cell lines were maintained in RPMI-1640 medium (Gibco, Canada) supplemented with 10% foetal bovine serum (Sigma, USA). HMEC-1 cell line specialty medium (Biospecies, Guangzhou, China). All cell lines were cultured in a humidified atmosphere of 5% CO2 at 37 • C. Indirect coculture assays were performed using .4 μm cell culture inserts (Corning, NY, USA).

Primary human T-cell isolation and culture
Human CD200 hi CD4 + T cells and CD200 low CD4 + T cells were obtained from PBMCs with a negative CD4 magnetic selection Kit (StemCell Technologies, Vancouver, CA, USA) and a human PE Position Selection Kit (StemCell Technologies, Vancouver, CA, USA). To increase the expression level of PD1, the T cells were cultured in the presence of 2 μg/mL anti-CD3 antibodies (Pepro-Tech, Rocky Hill, NJ, USA), 2 μg/mL anti-CD28 antibodies (PeproTech) and 10 ng/mL IL2 (PeproTech). Forty-eight hours later, the purity (>98%) of PD1 hi CD200 hi CD4 − T cells and PD1 hi CD200 low CD4 − T cells were confirmed by using flow cytometry.

Flow cytometry
Human PD1 hi CD200 hi CD4 + exhausted T cells and PD1 hi CD200 low CD4 + exhausted T cells were stimulated with Leukocyte Activation Cocktail

Cell invasion assays
For the in vitro invasion assays, the upper chambers of Transwells plates (Corning, NY, USA) were coated with 100 μL diluted Matrigel (Corning, NY, USA). Before the invasion assays, T24 and UMUC3 cells (1 × 10 5 ) were cocultured with PD1 hi CD200 hi CD4 + exhausted T cells/PD1 hi CD200 low CD4 + exhausted T cells (1 × 10 6 ) for 48 h in Transwell plates. Two hundred microliters of T24 and UMUC3 cells (1 × 10 5 ) in serum-free media and 1% serumcontaining media were plated in the upper and lower chambers, respectively. After 48 h of incubation, the cells in the upper chamber were removed. Invasive cells located on the lower side of the chamber were fixed with 100% methanol for 10 min and stained with crystal violet for 30 min at room temperature. Four random fields per well were observed, and cells were counted under a microscope. Experiments were performed in triplicate.

Matrigel tube formation assay
For the in vitro tube formation assays, 48-well culture plates were coated with 150 μL of Matrigel (Corning, NY, USA) per well and then allowed to polymerize for 30 min at 37 • C. Before the tube formation assays, HMEC1 cells (2 × 10 5 ) were cocultured with PD1 hi CD200 hi CD4 + exhausted T cells/PD1 hi CD200 low CD4 exhausted T cells (1 × 10 6 ) for 48 h in Transwell plates. After coculture, 200 μL HMEC1 cells (5 × 10 4 ) were seeded on polymerized Matrigel. After incubation at 37 • C for 4 h, each culture was photographed by using a confocal microscope (Leica, Wetzlar, Germany) to capture four random fields per well.

RNA interference
Small interfering RNA (siRNA) duplexes targeting METTL3 (siRNA1400 and siRNA1640) and the negative control (NC) siRNA were designed and synthesized by GenePharma (Shanghai, China). Cell transfection was achieved using Lipofectamine 3000 (Invitrogen, CA, USA) according to the manufacturer's instructions. Briefly, a 100 nM siRNA duplex was incubated with cell lines for 48 h for each transfection, after which the following experiments and assays were performed.

Western blotting
Cells were lysed in RIPA lysis buffer (Biyuntian, Hangzhou, China) containing 1% protease and phosphatase inhibitors (Thermo Fisher Scientific, MA, USA). Protein concentration was measured using a BCA Protein Assay Kit (Biyuntian, Hangzhou, China). Samples (30 μg) were separated by electrophoresis on 8%-15% SDS-PAGE gels and then transferred onto PVDF membranes (.45 μm; Merck-Millipore, Darmstadt, Germany) using a wet transfer system (Bio-Rad, Hercules, CA, USA). After blocking with blocking solution (Biyuntian, Hangzhou, China) for 15 min at room temperature, the membranes were incubated overnight with primary antibodies at 4 • C, followed by secondary antibodies for another hour.
Blots were detected by the enhanced chemiluminescence system (Millipore, MA, USA).

Methylated RIP analysis
TRIzol was used to extract total RNA from bladder cancer cells transfected with 100 nM siMETTL3 or control siRNA. Then, the Magna MeRIP m6A Kit (RiboBio, Guangzhou, China) was used for methylated RNA immunoprecipitation (RIP) according to the manufacturer's protocol. After that, an RNA purification kit (Zymo Research Corp., Irvine, CA, USA) was used to extract the enriched RNA. The enrichment of m6A-containing RNA relative to the negative (IgG) sample was analysed by qRT-PCR with the primers GAS6: 5′-GGTAGCTGAGTTTGACTTCCG-3′ (forward) and 5′-GACAGCATCCCTGTTGACCTT-3′ (reverse).

RNA immunoprecipitation analysis
RIP was performed with antibodies specific for METTL3 by using the Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Merck-Millipore, Darmstadt, Germany) according to the manufacturer's protocol. In brief, 4 × 10ˆ7 bladder cancer cells were harvested and lysed in RIP lysis buffer. After centrifugation at 4 • C, the supernatant was incubated with METTL3 antibodies and NC IgG at room temperature. Then, the bead-antibody complex was washed and incubated with Proteinase K buffer.
The input and immunoprecipitated RNAs were isolated by TRIzol reagent and reverse transcribed into cDNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, MA, USA). The fold enrichment relative to the negative (IgG) sample was measured by qRT-PCR with the primers GAS6: 5′-GGTAGCTGAGTTTGACTTCCG-3′ (forward) and 5′-GACAGCATCCCTGTTGACCTT-3′ (reverse).

2.20
Immunofluorescence assay and spatial analysis Multiplex immunofluorescence staining was performed using the PANO 7-plex IHC kit (Panovue, Beijing, China). In detail, the slides were incubated for 2 h at 65 • C, deparaffinized with xylene and ethanol diluted in a graduated series (100%, 95%, 70%, 50%) and then fixed in 10% neutral buffered formalin for 30 min. After that, microwave antigen retrieval was performed using EDTA antigen repair buffer (pH = 9.0, ZSGB-Bio, Beijing, China), followed by blocking. Following those procedures, horseradish peroxidase (HRP)-conjugated secondary antibodies were incubated, and tyramide signal amplification was carried out after sequentially applying different pri-  Table  S14. Subsequently, sections were incubated using biotinylated rabbit polyclonal anti-rabbit and rabbit anti-mouse antibodies as secondary antibodies followed by HRPconjugated streptavidin treatment according to the manufacturer's instructions (Panovue, Peking, China). Biotinylated secondary antibodies were used for streptavidinlinked alkaline phosphatase-dependent chromogen reactions for streptavidin-linked fluorophores for IF (excitation wavelengths of 480, 520, 570, 690 or 780 nm, Table  S14). Multispectral images were acquired using the Vectra Polaris Automated Quantitative Pathology Imaging System (Akoya Biosciences, State of Delaware, USA).
To calculate the spatial relationship between different cells, inferences are made about the interactions between cells. Proximity analysis in HALO software (Indica Labs, Corrales, NM, USA) was performed to assess the microenvironment around PD1 hi CD200 hi CD4 + exhausted T cells in bladder cancer. The PD1 hi CD200 hi CD4 + exhausted T cells were registered and plotted together as a single figure. The spatial analysis tool was used to calculate the average distance between PD1hi CD200hi CD4 exhausted T cells and CD31 + , IFNG + and GAS6 + cells. The 50 μm cutoff was selected based on analysis from previous studies. 33 Spatial data for all analyses were exported for statistical analysis.

2.21
The construction of the mouse model The subcutaneous syngeneic BLCA mouse models were generated by injecting 5 × 105 MB49 cells in 100 μL PBS into 5-6-week-old male C57BL/6J mice (n = 15). Tumour sizes were measured using callipers every day in an unblinded manner, and tumour volumes were calculated (V = .5 × L × W2). Once tumours reached 50 mm 3 , animals were randomly assigned to receive anti-PD1 (antimurine PD-1 mAb clone, BioXCell BE0273) or PBS. The MB49 model was used as validation, and all animals were sacrificed 8 days after drug administration. Tumours were collected and processed for haematoxylin-eosin staining and mIHC assays. All animal studies were approved by the laboratory animal ethics committee of Sun Yat-sen University Cancer Center.

Statistical analysis
Using the Kaplan-Meier method to estimate overall survival (OS), FPI or DSS, the Kaplan-Meier curves were compared using the log-rank test. A two-sided p-value of less than .05 was regarded as significant. Moreover, all of the sample sizes were large enough to enable proper statistical analysis. Spearman correlation analysis was applied in all of the correlation analyses. GraphPad Prism (https:// www.graphpad.com/, accessed on 25 November 2022) was also used to perform statistical analyses. p-Values less than .05 were deemed to be statistically significant. All of the t tests were two-sided t tests (paired or unpaired, depending on the experiments).
Differential gene expression analysis revealed that no immuno-effecting properties were associated with the CD4 Tex2 cluster. Moreover, CD4 Tex2 cells have significantly higher expression of TNFRSF4 and TCF7 than that of CD4 Tex1 cells, indicating their Treg-like activity and a certain naive state, respectively. [34][35][36] In contrast, GZMA and CST7 were specifically expressed in the CD4 Tex1 cluster, suggesting that the subcluster has an immunological effect to a certain extent ( Figure 1E; Table S4). Although the CD4 Tex1 cluster was considered to contain cells in an exhausted state, it still secreted certain factors that promote its antitumour function, including IFNG ( Figure 1F). GO biological pathway enrichment analysis also revealed that the CD4 Tex1 cluster was predominantly involved in T cell activation ( Figure 1G). Additionally, based on the differential gene expression profiles, CD4_Tex1-expressing genes were related to those of Th17 cells, including IL17A, IL17F, RORC, RORA and IL22, 37,38 whereas CD4_Tex2expressing genes exhibited features similar to those of Tfh cells, including BCL6, CXCR5, ICA1, IL6ST and MAGEH1 39-41 ( Figure S1G).
Interestingly, the CD4 Tex2 cluster is characterized by high levels of CD200 expression. CD200 was specifically highly expressed in the CD4 Tex2 cluster compared with CD4 Tex1 and all the other subclusters of CD4 T cells ( Figure 1H; Figure S1H). Therefore, the CD4 Tex1 cluster was defined as PD1 hi CD200 low CD4 exhausted T cells and the CD4 Tex2 cluster was defined as PD1 hi CD200 hi CD4 exhausted T cells. In addition, we discovered that CD200 expression was significantly and positively correlated with the exhaustion-associated molecules ENTPD1, HAVCR2 and TOX in TCGA ( Figure 1I). Finally, F I G U R E 1 Two subclusters of CD4 + exhausted T cells were defined in the tumour microenvironment using the pancancer CD4 + T cell atlas: (A) the UMAP plot of the subclusters of CD4 + T cells integrated from nine tumour types. Each dot indicated a single cell. Colour-coded for the cell type; (B) the dot plot showing the particular genes for each subcluster of CD4 + T cells; (C) tissue prevalence estimated by Ro/e multiplex immunofluorescence was performed to determine the infiltration of PD1 hi CD200 hi CD4 exhausted T cells in bladder cancer tissue, which showed that PD1 hi CD200 hi CD4 exhausted T cells were commonly present in the tumour-infiltrating CD4 + T cells in the bladder cancer tissues ( Figure 1J). In addition, PD1 hi CD200 low CD4 and PD1 hi CD200 hi CD4 exhausted T-cell subsets were also isolated from the blood of 18 bladder cancer patients, 10 non-small cell lung carcinoma patients and 8 esophageal squamous cell carcinoma patients by magnetic sorting ( Figure S1I).
In summary, we identified two novel subclusters of CD4 + exhaustion T cells with substantially different properties. The CD4 + Tex1 cluster with a PD1 hi CD200 low signature had a certain immuno-effecting function similar to that previously reported, 42 whereas the CD4 + Tex2 cluster, marked by high expression of PD1 and CD200, was in a complete exhaustion state.

3.2
The PD1 hi CD200 hi CD4 + exhausted T cells correlate with response to ICI It is well known that rejuvenating exhausted T cells is an important research direction of cancer immunotherapy. 43 In light of this, we investigated whether the PDCD1 hi CD200 hi group is related to the immunotherapeutic response in bladder cancer. Compared with the PDCD1 hi CD200 hi and PDCD1 low groups, the PDCD1 hi CD200 low group had the best survival rate and was most sensitive to immunotherapy based on the two immunotherapy cohorts GSE176307 and IMVigor210 (Figure 2A,B; Table  S5; Figure S2A,C). Similar results were also obtained in the extended dataset of NSCLC (GSE135222), gastric cancer (PRJEB25780) and melanoma (GSE91062, phs000452) cohorts ( Figure S2D-H). There was a strong indication from the above results that patients in the PDCD1 hi CD200 low group could benefit from immunotherapy, whereas those in the PDCD1 hi CD200 hi or PDCD1 low group appeared to be resistant to immunotherapy.
In terms of immunocyte infiltration, more CD8 + T cell infiltration was found in the PDCD1 hi CD200 low group than in the PDCD1 hi CD200 hi and PDCD1 low groups by employing the CIBERSORT algorithms. Moreover, more infiltration of M1 macrophage cells ( Figure 2C, Table S6) and cytotoxic lymphocyte infiltration ( Figure S3A) were found in the PDCD1 hi CD200 hi and PDCD1 hi CD200 low groups compared with those in the PDCD1 low group. In addition, a striking finding in the single-cell RNA sequencing data was that the effector T-cell molecules IFNG, GZMA, and GZMB were substantially expressed in the PDCD1 hi CD200 low group ( Figure S3B-D). To further confirm these findings, flow cytometry was used to estimate IFNG expression in three bladder cancer samples, and the results revealed that PD1 hi CD200 low CD4 + exhausted T cells showed significantly higher expression of IFNG ( Figure 2D). Immunohistochemistry staining in bladder cancer tissues further validated that strong staining of IFNG was observed near the tumour borders rather than around PD1 hi CD200 hi CD4 + exhausted T cells ( Figure 2E; Figure S3E). Among the CD4 + T-cell clusters, Th1 cells are characterized by the secretion of IFNG, which activates macrophages and CD8 + T cells. 44,45 Overall, our findings suggest that compared with PD1 hi CD200 low CD4 exhausted T cells, PD1 hi CD200 hi CD4 exhausted T cells may contribute to immune tolerance, and patients with high proportions of this subcluster often fail to respond to immunotherapy.

PD1 hi CD200 hi CD4 + exhausted T cells promote angiogenesis and recruit tip cells through the uridine triphosphate (UTP)/P2RY6 axis
To explore the mechanism by which the PDCD1 hi CD200 hi Group T cells reduce immunotherapy efficacy and patient survival, xCell algorithm analysis was performed using the TCGA dataset of 406 BLCA patients. The results revealed a significant increase in the number of endothelial cells in the PDCD1 hi CD200 hi group, indicating its role in promoting angiogenic activity ( Figure 3A; Table  S7). As angiogenesis is one of the ways that tumours are able to evade detection by the immune system, 46 we hypothesized that PD1 hi CD200 hi CD4 + exhausted T cells might contribute to immunotherapy resistance by promoting angiogenesis. In line with expectations, differential gene expression profiles across the PDCD1 hi CD200 low , (J) multiplex immunofluorescence staining was performed for DAPI (blue), CD4 (white), PD1 (red) and CD200 (green); each stain was shown separately and merged. The circles represent the PD1 hi CD200 hi CD4 + exhausted T cells. The field of view in the square is the reduced field of view.

F I G U R E 2
The PD-1 hi CD200 hi CD4 + exhausted T cells predict a poor response to immunotherapy: (A) overall survival of patients with immunotherapy in PDCD1 low , PDCD1 hi CD200 low and PDCD1 hi CD200 hi groups in the GSE176307 and IMvigor210 cohorts; (B) barplots showing the proportion of responders among PDCD1 low , PDCD1 hi CD200 low and PDCD1 hi CD200 hi groups in the GSE176307 and IMvigor210 PDCD1 hi CD200 hi and PDCD1 low groups indicated that the PDCD1 hi CD200 hi group exhibited the highest levels of angiogenesis-related genes ( Figure 3B; Figure S4A; Table S8). GSEA revealed that PDCD1 hi CD200 hi group was enriched in the angiogenesis pathway ( Figure 3C; Table S9). GO biological pathway enrichment analysis also indicated that the PDCD1 hi CD200 hi group was predominantly related to the regulation of angiogenesis ( Figure  S4B). Next, we evaluated the proangiogenic potency of the PDCD1 hi CD200 hi group in vitro. HMEC-1 cells (human microvascular endothelial cells) were cocultured with PD1 hi CD200 low T cells or PD1 hi CD200 hi T cells for 48 h and then used to assess tube formation in vitro. As revealed by fluorescence microscopy images of the tube formation assays, cocultivation with the PD1 hi CD200 hi group increased the capillary length and number of branch points of HMEC1 cells ( Figure 3D,E). Next, we used multiplex immunofluorescence (CD4, PD1, CD200, CD31 and DAPI) to explore the spatial distance between the PD1 hi CD200 hi CD4 exhausted T cells and CD31, a marker for specific vascular endothelial cells ( Figure 3F). Proximity analysis revealed that CD31 + endothelial cells were more abundant around PD1 hi CD200 hi CD4 + exhausted T cells than around PD1 hi CD200 low CD4 + exhausted T cells, indicating their proangiogenic potential ( Figure 3G; Figure S4C). Given the above results, we speculated that PD1 hi CD200 hi CD4 + exhausted T cells can induce angiogenesis.
To gain insight into the underlying molecular mechanisms of the proangiogenic effect of PD1 hi CD200 hi CD4 + exhausted T cells, we created an endothelial cell atlas with single-cell RNA-Seq data on 9801 endothelial cells derived from tumour tissue and adjacent normal tissues of nine bladder cancer patients (Table S10). Endothelial cells were mainly found in tumour tissues, showing little heterogeneity between patients ( Figure S4D,E). Ten endothelial cell clusters with distinct gene expression patterns were identified in the dataset ( Figure 3H; Figure S4F; Table  S11). Among the clusters, Tip_COL4A1 showed the highest expression levels of the angiogenesis-promoting genes FLT1, KDR, CD93 and NRP1 ( Figure 3I; Figure S4G). FLT1 and KDR are receptors of VEGF, which is the most important factor in promoting angiogenesis in tumours. 47,48 In addition, COL4A1 was also shown to play a major role in promoting neovascular sprouting. 49 As shown above, Tip_COL4A1 seemed to have potent angiogenic activity. Notably, CellChat analysis revealed that the Tip_COL4A1 cluster communicates most frequently with PD1 hi CD200 hi CD4 exhausted T cells among the 10 endothelial cell clusters ( Figure 3J). When applying the MEBOCOST algorithm on the CD4 Tex2 cluster as well as endothelial cells, we found that the Tip_COL4A1 cluster had the largest number of communications (133 events) compared to that of other endothelial cells as receiver cells ( Figure S4H). In addition, among all metabolite-sensor partners, uridine triphosphate (UTP) and its transporter P2RY6 were found to mediate communication between CD4 Tex2 cells and Tip_COL4A1 endothelial cells ( Figure 3K). Furthermore, a higher expression of P2RY6 predicted worse survival in patients with bladder cancer in the TCGA database ( Figure  S4I). Thus, we suggest that PD1 hi CD200 hi CD4 exhausted T cells might induce angiogenesis and recruit Tip_COL4A1 endothelial cells through the UTP/P2RY6 axis.

Crosstalk between PD1 hi CD200 hi CD4 + exhausted T cells and tumour cells through AXL-GAS6 promotes EMT in bladder cancer
The xCell algorithms on BLCA data in the TCGA database demonstrated not only a decrease in epithelial cells but also an increase in fibroblasts in the PDCD1 hi CD200 hi group, indicating a possible EMT phenotype ( Figure 4A). EMT has been associated with cancer invasion and immune resistance. 50 Consequently, we hypothesized that PD1 hi CD200 hi CD4 + exhausted T cells may promote EMT in bladder cancer cells, thus contributing to immunotherapy resistance. Consistent with our hypothesis, a higher expression of the EMT-related genes CDH11 and DCN was found in the PDCD1 hi CD200 hi group than in the PDCD1 hi CD200 low group and other CD4 + T-cell subclusters ( Figure 4B; Figure S5A). We also found that CD200 was positively correlated with the EMT-related transcription factors ZEB1, ZEB2, TWIST1 and SNAI1 ( Figure 4C). GSEA showed that the EMT and myogenic pathways were enriched in the PDCD1 hi CD200 hi group ( Figure 4D; Figure S5B). Furthermore, an invasion experiment was performed using the bladder cancer cell Lines T24 and UMUC3 cells cocultured with exhausted T cells. The results showed that the invasive capacity of T24 and UMUC3 cells significantly increased after coincubation with PD1 hi CD200 hi CD4 + exhausted T cells ( Figure 4E).

F I G U R E 3
The PD-1 hi CD200 hi CD4 exhausted T cells recruit tip cells to promote angiogenesis: (A) boxplot showing that estimation of the abundance of endothelial cells in the PDCD1 low , PDCD1 hi CD200 low and PDCD1 hi CD200 hi groups using xCell_counter algorithm; (B) volcano map showing differentially expressed genes between PDCD1 hi CD200 hi and PDCD1 hi CD200 low groups (PDCD1 hi CD200 hi : The above results indicated that PD1 hi CD200 hi CD4 + exhausted T cells induce EMT and enhance the invasion of bladder cancer cells. Next, we created an epithelial cell atlas using 24 172 epithelial cells derived from tumour tissue and adjacent normal tissues from 8 BLCA patients (Table S12). The identified epithelial cells were mainly derived from tumour tissues ( Figure S5C). Nine different epithelial cell clusters with unique gene expression patterns were identified using unsupervised clustering ( Figure 4F). High expression of epithelial-related genes, such as KRT8, KRT18 and EPCAM, was found in all clusters ( Figure S5D; Table S13), proving that all cells can be identified as epithelial cells. Surprisingly, high levels of COL1A1, MYL9, COL3A1 and vimentin were found in the clusters Epi_OLFM4 and Epi_CXCL1, indicating that the cells might be fibroblast-like epithelial cells ( Figure 4G). Additionally, the clusters Epi_OLFM4 and Epi_CXCL1 expressed high levels of CD44, CD55, KLF4 and ALDH1A1, suggesting that they are stem-like cells ( Figure S5E). The above results revealed that the epithelial cell clusters Epi_OLFM4 and Epi_CXCL1 may have undergone EMT and transitioned into invasive mesenchymal cells. The results from trajectory analysis, RNA velocity analysis and PAGA analysis showed that Epi_COL1A2 developed from Epi_OLFM4 and Epi_CXCL1 was the intermediate transition stage ( Figure 4H,I; Figure S5F). Next, CellChat analysis revealed that the receptor-ligand pair GAS6-AXL was significantly enriched in CD4 ex2-Epi_OLFM4, CD4 ex2-Epi_CXCL1 and CD4 ex2-Epi_COL1A2 ( Figure 4J). Moreover, multiplex immunofluorescence (CD4, PD1, CD200, GAS6 and DAPI) was performed to investigate the spatial distance between the PD1 hi CD200 hi CD4 exhausted T cells and GAS6. The proximity analysis showed plentiful GAS6 in the vicinity of the PD1 hi CD200 hi CD4 + exhausted T cells compared to the PD1 hi CD200 low CD4 + exhausted T cells ( Figure 4K,L; Figure S6A). Thus, we proposed that PD1 hi CD200 hi CD4 exhausted T cells and malignant cells may communicate via AXL-GAS6 signalling in bladder cancer. The Kaplan-Meier survival analysis showed that the higher expression of GAS6 was associated with worse OS, progression-free survival and platinum-free interval in 406 BLCA patients in the TCGA dataset ( Figure 4M; Figure  S7A). A similar trend was found in the Kaplan-Meier survival analysis for BLCA patients in the TCGA dataset with high levels of AXL ( Figure 4M; Figure S7B). High GAS6 expression was also associated with more advanced N stage ( Figure S7C). In addition, the correlation analysis showed a significant and positive correlation between CD200 and GAS6 (R = 0.43, Figure 4N). The PDCD1 hi CD200 hi group exhibited a higher level of GAS6 expression than the PDCD1 low and PDCD1 hi CD200 low groups ( Figure S7D). Furthermore, we also found that GAS6 was positively related to the EMT-related transcription factors ZEB1, ZEB2, TWIST1, SNAI1 and vimentin ( Figure 4O). GSEA results also revealed that GAS6 expression significantly and positively correlated with the EMT signatures ( Figure S7E,F). The results above indicated that the GAS6-AXL axis might interact with the PD1 hi CD200 hi CD4 exhausted T cells, trigger EMT and promote bladder cancer progression.

PD1 hi CD200 hi CD4 + exhausted T cells negatively correlate with the response to ICIs and positively recruit vascular cells in vivo
To verify the function of PD1 hi CD200 hi CD4 + exhausted T cells in vivo, we subcutaneously implanted MB49 bladder cancer cells into immunocompetent mice and treated mice every 2 days with anti-PD1 antibody when tumours were 50 mm 3 in size ( Figure 5A,B). According to the tumour growth curve, mice were divided into anti-PD1responsive and nonresponsive groups ( Figure 5C,D). Multiplex immunofluorescence was performed to determine the infiltration of PD1 hi CD200 hi CD4 + exhausted T cells in mice divided into anti-PD1-responsive and nonresponsive groups, which showed that the nonresponsive groups logFC >0.3 and log p-value >2; PDCD1 hi CD200 low : logFC < −0.3 and log p-value >2). Angiogenesis pathway-related genes were highlighted; (C) gene set enrichment analysis (GSEA) showing that the angiogenesis pathway was overrepresented in PDCD1hi CD200hi groups; (D and E) fluorescence micrograph of tube formation by HMEC1 (top) and network mask (bottom). Branch points and capillary length were analysed to evaluate angiogenic activity. The bars represent mean ± SD (n = 3), **p <0.01, ***p < 0.001; (F) multiplex immunofluorescence staining was performed for DAPI (blue), CD4 (white), PD1 (red), CD200 (green) and CD31 (orange). The circles represent the PD1 hi CD200 hi CD4 + exhausted T cells. The arrows represent the CD31; (G) spatial distribution of CD31 + cells around PD-1 hi CD200 hi CD4 − T cells.

F I G U R E 4
The PD1 hi CD200 hi CD4 exhausted T cells promote epithelial-mesenchymal transition (EMT): (A) boxplot showing that estimation of the abundance of epithelial and fibroblast cells in the PDCD1 low , PDCD1 hi CD200 low and PDCD1 hi CD200 hi groups using xCel algorithm; (B) volcano map showing differentially expressed genes between PDCD1 hi CD200 hi and PDCD1 hi CD200 low groups (PDCD1 hi exhibited higher levels of Pd1 hi Cd200 hi Cd4 + exhausted T cells than the anti-PD1-responsive groups ( Figure 5E). Next, multiplex immunofluorescence revealed that Pd1 hi Cd200 hi Cd4 + exhausted T cells secreted less Ifng than Pd1 hi Cd200 low Cd4 + exhausted T cells, and the nonresponsive groups had higher levels of Ifng ( Figure 5F). Next, as a result of multiplex immunofluorescence, Pd1 hi Cd200 hi Cd4 + exhausted T cells were found to produce more Cd31 than Pd1 hi Cd200 low Cd4 + exhausted T cells, and Cd31 levels were higher in the nonresponsive group than in the anti-PD1 group ( Figure 5G). In addition, multiplex immunofluorescence revealed that Pd1 hi Cd200 hi Cd4 + exhausted T cells produced more Gas6, and Gas6 levels were higher in the nonresponsive group than in the anti-PD1 group ( Figure 5H).

METTL3-mediated m6A modification enhances GAS6 expression in bladder cancer cells
The most abundant base modification in eukaryotic mRNA, N6-methyladenosine (m6A), is known to modulate gene expression in various cancer types. 51,52 Based on the SRAMP (sequence-based m6A modification site predictor) algorithm, 53 we identified six m6A sites with high scores in the 3′UTR of GAS6 mRNA ( Figure 6A). We also discovered that the knock-down of METTL3 in MOLM13, HeLa and Caco2 cell lines can downregulate the mRNA expression of GAS6 using three GSE datasets ( Figure 6B). Based on the above results, we proposed that METTL3 may affect the expression of GAS6 by regulating the level of m6A modification. To validate this hypothesis, we first performed gene-specific m6A methylated RIP qRT-PCR analysis in bladder cancer cells transfected with or without METTL3 siRNA. As expected, METTL3 knock-down significantly reduced the m6A enrichment of GAS6 ( Figure 6C). The findings suggested that METTL3 could directly regulate the m6A methylation of GAS6 mRNA. Next, GAS6 expression at both RNA and protein levels was reduced after METTL3 knock-down in the bladder cancer cell Lines T24 and UMUC3 ( Figure 6D,E). In addition, RIP assays demonstrated direct interactions between GAS6 mRNA and METTL3 ( Figure 6F,G). In conclusion, these findings suggested that METTL3 could regulate GAS6 expression in an m6A-dependent manner in bladder cancer cells.

DISCUSSION
CD4 + T-cell subsets with different functions are found in a wide range of cancers. [54][55][56] To date, a pancancer Tcell atlas has been established, and two types of CD8 + exhausted T cells were identified. 57 However, knowledge on CD4 + exhausted T cells is still scarce. In the pancancer CD4 + T-cell atlas, we found two different clusters of CD4 + exhausted T cells, CD4_Tex1 and CD4_Tex2, defined according to PD-1 and CD200 expressions. In addition, our findings revealed that CD4_Tex1 cells have Th17-like gene expression, whereas CD4_Tex2 cells have Tfh-like gene expression. CD200 (also known as OX2), a transmembrane glycoprotein of the immunoglobulin supergene family, is mainly expressed in several types of cancer cells, endothelial cells and activated immune cells, whereas its receptor (CD200R) is mostly expressed in monocytes/myeloid cells and T lymphocytes. [58][59][60] As a newly emerging immune checkpoint molecule, CD200 mediates immunoregulatory signals by binding to CD200Rs to suppress the antitumour immune response and regulate immune tolerance, adhesion, differentiation and chemotaxis. 61,62 In CD4 + T cells, the CD200-CD200R pathway is important for regulating their differentiation. The imbalance of Tregs/Th17 cells can be caused by the CD200-CD200R signalling pathway, which induces Treg generation and prevents Th17 differentiation, thus resulting in immune tolerance. 63,64 Moreover, the activation of the CD200/CD200R signalling pathway is also a key contributor to the Treg phenotype. 65,66 In light of the above studies, we hypothesized that PD1 hi CD200 hi CD4 + exhausted T cells may have the traits of Tregs. Next, our results indicated that patients with high proportions of PD1 hi CD200 low CD4 + exhausted T cells had an excellent prognosis following immunotherapy, whereas those with high proportions of PD1 hi CD200 hi CD4 + exhausted T cells demonstrated immune tolerance. Furthermore, in addition to the high infiltration of CD8 T cells and multiplex immunofluorescence staining was performed for DAPI (blue), Cd4 (white), Pd1 (red) and Cd200 (green); each stain was shown separately and merged. Histograms show the percentages of Pd1hi Cd200hi Cd4 + exhausted T cells in anti-PD1-responsive and non-responsive groups; (F) multiplex immunofluorescence staining was performed for DAPI (blue), Cd4 (white), Pd1 (red) and Cd200 (green); each stain was shown separately and merged. Histograms show the percentages of Pd1hi Cd200hi Cd4 + exhausted T cells in anti-PD1-responsive and non-responsive groups; (G) multiplex immunofluorescence staining was performed for DAPI (blue), Cd4 (white), Pd1 (red), Cd200 (green) and Ifng (yellow); each stain was shown separately and merged. Histograms show the percentages of Ifng in anti-PD1-responsive and non-responsive groups; (G) multiplex immunofluorescence staining was performed for DAPI (blue), Cd4 (white), Pd1 (red), Cd200 (green) and Cd31 (yellow); each stain was shown separately and merged. Histograms show the percentages of Cd31 in anti-PD1-responsive and non-responsive groups; (H) multiplex immunofluorescence staining was performed for DAPI (blue), Cd4 (white), Pd1 (red), Cd200 (green) and Gas6 (yellow); each stain was shown separately and merged. Histograms show the percentages of Gas6 in anti-PD1-responsive and non-responsive groups. cytotoxic lymphocytes, PD1 hi CD200 low CD4 + exhausted T cells are also shown to consistently produce IFNG, GZMA and GZMB despite being exhausted. 67 Therefore, we proposed that PD1 hi CD200 low CD4 + exhausted T cells in the TME might induce the expansion and activation of intratumoural CD8 + T cells, thus leading to better immunological responses in bladder cancers patients. In contrast, Tregs are known to represent the majority of CD4 + T cells that inhibit immunity. 68,69 As the CD200-CD200R signalling pathway can contribute to the generation of Tregs, we speculate that the characteristics of PD1 hi CD200 hi CD4 + T cells may be similar to those of Tregs. Further exploration is necessary on this point. As determined by xCell algorithms, the PDCD1 hi CD200 hi group exhibited an increased number of endothelial cells, which is associated with angiogenesis. The expression of the angiogenesis-related genes KCNJ8 and VCAN was also higher in the PDCD1 hi CD200 hi group than in the other CD4 + T-cell subclusters. Using in vitro experiments, we further proved that PD1 hi CD200 hi CD4 + exhausted T cells contribute to stimulating angiogenesis in bladder cancer. In the TME, angiogenesis plays a key role in facilitating immune tolerance by blocking cytotoxic T lymphocytes from entering. 70 Thus, one of the reasons why PD1 hi CD200 hi CD4 + exhausted T cells are resistant to immunotherapy may lie in their ability to induce angiogenesis. Numerous studies have shown that Tregs promote angiogenesis. [71][72][73][74][75][76][77] In detail, Tregs have been shown to increase VEGFA levels directly, thereby stimulating angiogenesis. 78 Through their release of paracrine factors (e.g. IL-10 and amphiregulin) and modulation of macrophage function, Tregs can indirectly influence angiogenesis beyond their VEGF-mediated effects. 79,80 The results of the above research confirmed our conjecture that PD1 hi CD200 hi CD4 + T cells and Tregs may form a transformation relationship in some contexts. In the next step, we employed scRNA-seq to comprehensively reveal novel cellular interactions between CD4 + exhausted T cells and endothelial cells at single-cell resolution. Of all endothelial cell clusters, Tip_COL4A1 most frequently communicated with PD1 hi CD200 hi CD4 + exhausted T cells. Known as sprouting endothelial cells, Tip cells play a significant role in neovascularization, which is strongly affected by the tumour metabolic microenvironment. [81][82][83] In light of this, we discovered that the metabolite-sensor partner UTP-P2YR6 was activated specifically in the cell interaction between PD1 hi CD200 hi CD4 + exhausted T cells and the Tip_COL4A1 subcluster. When stressed or inflamed, cells will increase the exocytotic release of UTP and then bind with P2YR6 to activate a series of downstream pathways, including inducing chemotaxis and migration and regulating vascular remodelling. [84][85][86][87][88] Overall, we conclude that paracrine secretion of UTP by PD1 hi CD200 hi CD4 exhausted T cells can recruit the sprouting endothelial cells Tip_COL4A1 to initiate angiogenesis.
There is widespread recognition that EMT plays a critical role in immune resistance. 89 An EMT phenotype with decreased epithelial cells and increased fibroblasts was demonstrated in the PDCD1 hi CD200 hi group by using xCell algorithms in this study. Moreover, the expression of the EMT transcription factors CDH11 and DCN was higher in the PDCD1 hi CD200 hi group than in the other CD4 + T-cell subclusters. A significant and positive correlation was found between CD200 and the EMT-related transcription factors ZEB1, ZEB2, TWIST1 and SNAI1 in TCGA data. The EMT pathway was enriched in the PDCD1 hi CD200 hi group by GSEA. Therefore, driving EMT in bladder cancer might be another mechanism by which PD1 hi CD200 hi CD4 + exhausted T cells promote immune escape in bladder cancer, in addition to inducing angiogenesis. Recent studies have indicated that Tregs can facilitate EMT by promoting the expression of β-Catenin and TGF-β in epithelial cells. [90][91][92] Nonetheless, studies providing evidence of how other CD4 + T-cell clusters regulate EMT are scarce. To discover the underlying mechanism, an atlas of epithelial cells was constructed using bladder cancer scRNA-seq, and nine distinct epithelial cell subclusters were identified. Among them, the Epi_OLFM4 and Epi_CXCL1 subclusters showed EMT transition with a high expression of mesenchymal biomarkers. Interestingly, the GAS6-AXL axis was specifically enriched in the cell interaction between PD1 hi CD200 hi CD4 + exhausted T cells and Epi_OLFM4, Epi_CXCL1 and Epi_COL1A2. The GAS6/AXL signalling pathway has been shown to promote tumour cell proliferation and survival as well as EMT and immune evasion. 93 As reported, AXL has major roles in the migration of cells during development and regulates cell-cell communication among cancer cells, endothelial cells and immune cells by binding to its ligand GAS6. 94,95 In addition, GAS6 has a direct positive effect on the functions of Tregs mainly through its interaction with the AXL receptor. 96 Evidence exists that GAS6 can induce AXL-mediated chemotaxis. 97 As expected, using six-colour mHIC, we validated the result that GAS6 is capable of inducing the chemotaxis of PD1 hi CD200 hi CD4 + exhausted T cells. Considering the above studies, we hypothesized that malignant cells in bladder cancer may recruit PD1 hi CD200 hi CD4+ exhausted T cells via the GAS6/AXL axis, and that PD1 hi CD200 hi CD4 + exhausted T cells may promote EMT in bladder cancer cells. The mechanisms by which bladder cancer cells upregulate GAS6 expression are also discussed. m6A represents the most prevalent epigenetic modification of RNAs. 98 There is evidence pointing to a regulatory role for m6A modification in GAS6 expression. 99,100 Additionally, we verified that m6A can regulate GAS6 expression through METTL3 in bladder cancer cells using immunoprecipitation experiments and in vitro experiments. Overall, the PD1 hi CD200 hi CD4 + T cells might be recruited and contribute to the EMT of bladder cancer cells through m6A-mediated GAS6.

CONCLUSION
In this study, we constructed a pancancer CD4 + T-cell atlas based on single-cell RNA-Seq data from patients with multiple cancers types and identified two clusters of PD1 hi CD200 low and PD1 hi CD200 hi CD4 + exhausted T cells in bladder cancer patients. More importantly, bladder cancer patients with high proportions of PD1 hi CD200 hi CD4 + exhausted T cells showed a worse response to immunotherapy than those with PD1 hi CD200 low CD4 + exhausted T cells. We created a single-cell atlas of endothelial cells and revealed that PD1 hi CD200 hi CD4 + exhausted T cells might recruit tip endothelial cells and initiate angiogenesis through UTP and its transporter P2RY6. In addition, we also found that bladder cancer cells can recruit PD1 hi CD200 hi CD4 + exhausted T cells to promote EMT by enhancing m6A-mediated GAS6. Finally, our findings suggest that immunotherapy combined with anti-CD200 drugs might be a promising therapeutic option for bladder cancer patients with high proportions of PD1 hi CD200 hi CD4 + exhausted T cells. Therefore, this research provides a valuable insight into CD4 + exhausted T cells and their influence on immunotherapy. However, there are several limitations to our study. Further validation will be conducted in a prospective and multicentre randomized trial in the future (Graphical Abstract).

A C K N O W L E D G E M E N T S
We acknowledge TCGA and GEO databases for providing their platforms and contributors for uploading their meaningful datasets. This work was supported by the National Natural Science Foundation of China (NOs. 81772884, 82273051, 81972640 and 81871986).

C O N F L I C T O F I N T E R E S T S TAT E M E N T
The authors declare that they have no conflict of interest.

D ATA AVA I L A B I L I T Y S TAT E M E N T
The data supporting the findings of this study are accessible within the article and its supplementary materials.