ARID1A Loss plus CD8+ T-Cell Infiltration Associate with Favorable Clinical Outcomes in Urothelial Carcinoma.

In the ZSHS cohort, 25.2% of patients (34/135) were categorized as ARID1A-loss patients based on IHC evaluation. In the FUSCC cohort, 28.8% of patients (34/118) harbored ARID1A loss. Baseline characteristics of the ZSHS and FUSCC cohorts are displayed in Table 1 . Patients with ARID1A loss were associated with high PIK3CA expression and low KDM6A expression ( Table 1 , PIK3CA, P = 0.038; KDM6A, P = 0.054). In addition, ARID1A-loss patients were more frequently associated with deficient mismatch repair (dMMR), as evidenced by a significantly higher incidence of MMR protein (MLH1, MSH2, MSH6, and PMS2) deficiency ( Table 1 , P = 0.045). Consistent with these findings, analyses of the Shanghai-sequencing and IMvigor210 cohorts revealed that ARID1A loss was also associated with higher TMB (Supplementary Fig. S3). In both ZSHS (discovery) and FUSCC (validation) cohorts, Kaplan–Meier analysis revealed no significant association between ARID1A loss and OS in urothelial carcinoma. In the ZSHS cohort, the median OS was 94.0 months for patients with ARID1A loss and 75.0 months for non-loss individuals ( Fig. 1A , log-rank P = 0.463). Similarly, in the FUSCC cohort, the median OS was not reached in ARID1A-loss patients, whereas it was 68.5 months in the non-loss subgroup ( Fig. 1B , log-rank P = 0.298). We observed that high infiltration of CD8 + T cells correlated with improved OS in both cohorts. In the ZSHS cohort, patients with high CD8 + T-cell infiltration had a median OS of 154.0 months, compared with 70.0 months for moderate infiltration and 80.0 months for low infiltration ( Fig. 1C , log-rank P = 0.050). In the FUSCC cohort, the median OS was not reached for patients with high CD8 + T-cell infiltration, whereas those with moderate and low infiltrations had median OS of 42.4 and 14.9 months, respectively ( Fig. 1D , log-rank P = 0.004). Notably, ARID1A loss did not significantly affect CD8 + T-cell infiltration levels in urothelial carcinoma, as determined by either IHC ( Fig. 1E and F ) or transcriptomic profiling (Supplementary Fig. S4). Association between ARID1A status and CD8 + T-cell infiltration in patients with UC. A and B, Kaplan–Meier analysis of OS according to ARID1A status in the ZSHS ( A ) and FUSCC ( B ) cohorts. C and D, Kaplan–Meier analysis of OS according to the infiltration level of CD8 + T cells in the ZSHS ( C ) and FUSCC ( D ) cohorts. E and F, Box plot and pie chart depicting the correlation between CD8 + T-cell infiltration and ARID1A status in the ZSHS ( E ) and FUSCC ( F ) cohorts. CI, confidence interval; HPF, high-power field; n.s., not significant; UC, urothelial carcinoma. We further investigated the prognostic significance of CD8 + T cells based on ARID1A status. In patients with ARID1A non-loss across ZSHS and FUSCC cohorts, no substantial differences in survival were observed based on infiltration levels of CD8 + T cells ( Fig. 2A and B , ZSHS cohort, log-rank P = 0.638; FUSCC cohort, log-rank P = 0.082). In contrast, high CD8 + T-cell infiltration conferred a clear survival advantage in ARID1A-loss patients. In the ZSHS cohort, patients with ARID1A loss and high infiltration of CD8 + T cells (ARID1A loss CD8 high ) did not reach the median OS, contrasting with 63.0 months for patients with ARID1A loss and moderate/low infiltration of CD8 + T cells (ARID1A loss CD8 mod/low ; Fig. 2C , log-rank P = 0.010). Similarly, in the FUSCC cohort, ARID1A loss CD8 high patients did not reach the median OS, whereas the median OS for ARID1A loss CD8 mod/low patients was 30.0 months ( Fig. 2D , log-rank P = 0.015). Multivariable Cox regression analyses further confirmed that high infiltration of CD8 + T cells indicated favorable prognosis independent of age, gender, and stage in patients with ARID1A loss [ Fig. 2E and F , ZSHS cohort, multivariable Cox (HR) = 0.12, P = 0.011; FUSCC cohort, multivariable Cox (HR) = 0.02, P = 0.002]. This phenomenon was not observed in patients without ARID1A loss. CD8 + T-cell infiltration yields favorable prognosis in ARID1A-loss patients with UC. A and B, Kaplan–Meier analysis of OS according to the infiltration level of CD8 + T cells in patients with ARID1A non-loss in the ZSHS ( A ) and FUSCC ( B ) cohorts. C and D, Kaplan–Meier analysis of OS according to the infiltration level of CD8 + T cells in patients with ARID1A loss in the ZSHS ( C ) and FUSCC ( D ) cohorts. E and F, Multivariable analyses incorporating clinical–pathologic characteristics and CD8 + T-cell infiltration in patients with or without ARID1A loss in the ZSHS ( E ) and FUSCC ( F ) cohorts. AJCC, American Joint Committee on Cancer; CI, confidence interval; UC, urothelial carcinoma. Given the limited number of patients with urothelial carcinoma with ARID1A loss undergoing chemotherapy, a pooled analysis encompassing the ZSHS and FUSCC cohorts was conducted. High infiltration of CD8 + T cells was not significantly associated with OS among patients with ARID1A non-loss who received ACT [ Fig. 3A (left), log-rank P = 0.249]. Conversely, we found that ARID1A loss CD8 high patients demonstrated a pattern of prolonged OS after ACT. The median OS for ARID1A loss CD8 high patients was not reached, whereas it was 58.0 months for ARID1A loss CD8 mod/low patients [ Fig. 3A (right), log-rank P = 0.015]. Additionally, the recurrence rate for patients with urothelial carcinoma receiving ACT was notably lower in ARID1A loss CD8 high patients (11.1%) compared with other subgroups (42.4%–44.1%; Fig. 3B ). CD8 + T-cell infiltration elicits clinical benefit from ACT and PD-1/PD-L1 blockade in ARID1A-loss patients with UC. A, Kaplan–Meier analysis of OS for patients with different ARID1A statuses and CD8 + T-cell infiltration levels treated with ACT in the pooled analyses of the ZSHS and FUSCC cohorts. B, Proportion of the recurrence rate for patients with different ARID1A statuses and CD8 + T-cell infiltration levels treated with ACT. C, Serial CT imaging of two ARID1A -loss patients with mUC treated with GC plus PD-1 blockade in the Shanghai-sequencing cohort. The red arrows highlight the target lesions. D, Kaplan–Meier analysis of OS for patients with different ARID1A statuses and CD8 + T-cell infiltration levels treated with atezolizumab in the IMvigor210 cohort. E, Proportion of the response rate for patients with different ARID1A statuses and CD8 + T-cell infiltration levels treated with atezolizumab. CI, confidence interval; CR, complete response; M & L, moderate and low; PD, progressive disease; PR, partial response; SD, stable disease; UC, urothelial carcinoma; UTUC, upper tract urothelial carcinoma. We further explored the association among ARID1A status, CD8 + T-cell infiltration and immunotherapeutic response in urothelial carcinoma. For example, two patients from the Shanghai-sequencing cohort, both diagnosed with mUC, received PD-1 blockade. The ARID1A loss CD8 high patient (RJP002) demonstrated significant remission (RECIST: partial response), whereas the ARID1A loss CD8 low patient (RJP007) showed disease progression (RECIST: progressive disease; Fig. 3C ). Consistently, in the IMvigor210 cohort, ARID1A loss CD8 high patients treated with atezolizumab exhibited the most favorable survival outcomes. The median OS was not reached in this group, compared with 9.2 months in ARID1A loss CD8 mod/low patients, 10.5 months in patients with ARID1A non-loss and high infiltration of CD8 + T cells ( ARID1A non-loss CD8 high ), and 7.9 months in patients with ARID1A non-loss and moderate/low infiltration of CD8 + T cells ( ARID1A non-loss CD8 mod/low ; Fig. 3D , ARID1A non-loss, log-rank P = 0.019; ARID1A loss, log-rank P = 0.020). Furthermore, the objective response rate (complete response/partial response) of immunotherapy was significantly higher in ARID1A loss CD8 high patients (52.6%) compared with other subgroups (17.7%–25.6%; Fig. 3E ). We further observed functional phenotypes of CD8 + T cells in the context of ARID1A loss. We found no difference in the distribution of CD8 + T-cell subsets, including CD103 + CD8 + T cells, CXCR5 + CD8 + T cells, and TIGIT + CD8 + T cells, between ARID1A loss CD8 high and ARID1A non-loss CD8 high patients ( Fig. 4A ). However, the infiltration number of CD8 + T cells was positively correlated with the number of effector molecules IFN-γ and PRF1 in ARID1A-loss patients ( Fig. 4B , IFN-γ, r = 0.476, P = 0.004; PRF1, r = 0.400, P = 0.019), which was not observed in patients with ARID1A non-loss. In the IMvigor210 cohort, differential gene expression showed that TLS signature genes ( CCL5 , CCL21 , CXCL13 , PTGDS , RBP5 , and CETP ; refs. 32 – 34 ), activated CD8 + T-cell signature genes ( PRF1 , GZMK , CD69 , CR2 , and IL7R ; refs. 35 – 37 ), and immunostimulators (KLRC2, CXCL12, CXCL14, and CCL14) were upregulated in ARID1A loss CD8 high patients compared with ARID1A non-loss CD8 high patients ( Fig. 4C ). Meanwhile, in the ZSHS cohort, we found that ARID1A loss CD8 high patients had increased infiltration level of CXCL13 + cells, B cells, DCs and Th17 cells, which were associated with TLS formation ( Fig. 4D ; ref. 32 ). Furthermore, IHC assays disclosed the most prevalent presence of TLS in ARID1A loss CD8 high patients (78.6%), compared with 25.0% in ARID1A loss CD8 mod/low patients, 51.6% in ARID1A non-loss CD8 high patients, and 12.9% in ARID1A non-loss CD8 mod/low patients ( Fig. 4E ). From an immunogenomic perspective, ARID1A loss CD8 high patients in the IMvigor210 cohort exhibited significantly higher TMB and neoantigen burden [ Fig. 4F (left)]. In addition, patients with ARID1A loss consistently exhibited higher cGAS-STING signature. Patients with high infiltration of CD8 + T cells also exhibited higher MHC I signature, irrespective of ARID1A status [ Fig. 4F (right)]. CD8 + T-cell infiltration confers increased TLSs in ARID1A-loss patients with UC. A, Radar chart demonstrating proportion of CD103 + CD8 + T cells, CXCR5 + CD8 + T cells, and TIGIT + CD8 + T cells between ARID1A loss CD8 high and ARID1A non-loss CD8 high patients in the ZSHS cohort. B, Scatter plot depicting the correlation between IFN-γ + cells (left), PRF1 + cells (right), and CD8 + T cells in the ZSHS cohort stratified by ARID1A status. C, Volcano plot showing differential gene mRNA expression between ARID1A loss CD8 high and ARID1A non-loss CD8 high patients in the IMvigor210 cohort. D, Box plot illustrating different infiltration number of CXCL13 + cells, CD19 + B cells, DCs, and Th17 cells in patients with different ARID1A statuses and CD8 + T-cell infiltration levels in the ZSHS cohort. E, Sankey plot displaying the relationship among ARID1A status, CD8 + T-cell infiltration level, and TLSs in the ZSHS cohort. F, Box plot illustrating different levels of TMB, neoantigen burden, cGAS-STING signature, and MHC I signature in patients with different ARID1A statuses and CD8 + T-cell infiltration levels in the IMvigor210 cohort. Mod, moderate; HPF, high-power field; UC, urothelial carcinoma. We proceeded to investigate the immune composition of different subgroups based on ARID1A status and CD8 + T-cell infiltration. In the IMvigor210 cohort, ARID1A l oss CD8 high patients displayed a more immunologically active profile, with enrichment in antitumor cytokine signatures and reduced expression of immune suppression by myeloid signature, myeloid cells traffic signature, CAF signature, EMT signature, and immune suppression signature, compared with ARID1A non-loss CD8 high patients ( Fig. 5A ). Corroborating these findings, IHC analysis in the ZSHS cohort revealed that ARID1A loss CD8 high patients had a higher M1/M2 ratio and lower DC-SIGN + TAM infiltration than other subgroups. Moreover, ARID1A loss CD8 high patients exhibited decreased infiltration of PDPN + cells and TGF-β + cells, which were considered markers of CAFs and EMT, respectively ( Fig. 5B ; ref. 38 ). IHC assays also disclosed a reduced expression level of B7-H3 and B7-H4 in ARID1A loss CD8 high patients compared with others, whereas no significant differences were observed in the expression level of other immune checkpoints ( Fig. 5C ). ARID1A status plus CD8 + T-cell infiltration stratifies tumor immune contexture in UC. A, Radar chart demonstrating different levels of signatures related to tumor immune characteristics between ARID1A loss CD8 high and ARID1A non-loss CD8 high patients in the IMvigir210 cohort. B, Box plot illustrating different levels of M1/M2 ratio, DC-SIGN + TAMs, PDPN + cells, and TGF-β + cells in patients with different ARID1A statuses and CD8 + T-cell infiltration levels in the ZSHS cohort. C, Heatmap illustrating different levels of immune checkpoints in patients with different ARID1A statuses and CD8 + T-cell infiltration levels in the ZSHS cohort. HPF, high-power field; Mod, moderate; M & L, moderate and low; UC, urothelial carcinoma.

The chromatin remodeling pathway is frequently altered in urothelial carcinoma. Our previous study demonstrated that gender disparities were associated with chromatin remodeling gene KDM6A status on urothelial carcinoma’s diverse clinical outcomes and therapeutic responsiveness ( 39 ). Building on this, the present study further investigates the clinical relevance and immunologic profiles of ARID1A loss and CD8 + T-cell infiltration in urothelial carcinoma. We found that in patients with urothelial carcinoma with ARID1A loss, high CD8 + T-cell infiltration is associated with a favorable prognosis and heightened sensitivity to both chemotherapy and immunotherapy. This study revealed the potential of combining ARID1A loss and CD8 + T-cell infiltration as biomarkers to inform treatment decisions in urothelial carcinoma. Our study confirmed that ARID1A loss had no impact on OS in urothelial carcinoma. This result was consistent with findings from endometriosis-associated ovarian carcinoma ( 12 , 13 ), colorectal cancer ( 14 ), and gastric cancer ( 15 ), although this was in contrast with other smaller-scale studies suggesting uncertain impact of ARID1A loss in urothelial carcinoma ( 40 , 41 ). ARID1A mutations have been shown to be correlated with immunotherapeutic benefits in urothelial carcinoma, as evidenced in previous studies of the IMvigor210 and CheckMate275 cohorts ( 19 , 23 ). Sarfaty and colleagues validated that ARID1A loss is consistently associated with better outcomes in ICB ( 20 ). ARID1A loss increases DNA mutability and ICB response by disrupting the cBAF complex’s interaction with the MMR complex ( 42 ). In line with these findings, we observed that ARID1A loss was linked to higher TMB and a greater proportion of dMMR in patients with urothelial carcinoma. Additionally, KDM6A loss, which we previously identified as associated with PD-1/PD-L1 blockade sensitivity ( 39 ), frequently co-occurred with ARID1A loss. Maxwell and colleagues ( 16 ) demonstrated that the ARID1A loss amplified antitumor immune response via the cGAS-STING pathway. We similarly found that patients with ARID1A loss exhibited higher cGAS-STING signature, irrespective of CD8 + T-cell infiltration. These observations collectively suggested that ARID1A mutations could alter the tumor immune microenvironment through various mechanisms, potentially enhancing immunotherapy effectiveness. The relationship between ARID1A status and CD8 + T-cell abundance was inconsistent ( 13 , 15 , 43 ). In our study, we observed no significant difference in CD8 + T-cell infiltration between ARID1A-loss and ARID1A non-loss patients, suggesting that ARID1A loss and CD8 + T-cell infiltration are not directly linked but rather may work in concert to influence the tumor-immune microenvironment. Furthermore, we found that only in ARID1A-loss patients, high infiltration of CD8 + T cells was associated with a favorable prognosis and indicated enhanced sensitivity to chemotherapy and immunotherapy. Our previous research identified three distinct CD8 + T-cell subsets in urothelial carcinoma: CD103 + CD8 + T cells (tissue-resident memory; ref. 44 ), CXCR5 + CD8 + T cells (follicular cytotoxic T cells; ref. 45 ), and TIGIT + CD8 + T cells (exhausted CD8 cells; ref. 46 ). However, we found no significant difference in the distribution of these subsets between ARID1A loss CD8 high and ARID1A non-loss CD8 high patients. TLSs have been associated with CD8 + T-cell activity, favorable outcomes, and response to immunotherapy in solid tumors, including urothelial carcinoma ( 29 , 32 , 47 ). In our study, TLSs and key players in their formation, including CXCL13 + cells, B cells, DCs, and Th17 cells, were significantly enriched in ARID1A loss CD8 high patients. We hypothesize that the presence of TLS enhances the effector functions of infiltrating CD8 + T cells in ARID1A-loss patients, which warrants evaluation in future studies. In addition, the immunogenomic background, including TMB, neoantigen, and MHC I–mediated antigen-presenting machinery, had also contributed to the cytotoxic capacity of CD8 + T cells in ARID1A loss CD8 high patients. We found that immunosuppressive cells, including DC-SIGN + TAMs ( 48 ), PDPN + cells ( 49 ), and TGF-β + cells ( 50 ), were decreased in ARID1A loss CD8 high patients. We also found that immune checkpoints such as B7-H3 and B7-H4 ( 28 ) were downregulated in these patients. These factors may account for the best clinical outcomes and the most sensitivity to chemotherapy and immunotherapy in ARID1A loss CD8 high patients. Previous studies found that ARID1A loss might be predictive of EZH2 inhibitor sensitivity ( 51 , 52 ). For ARID1A loss CD8 mod/low patients, the use of EZH2 inhibitors could be a promising therapeutic consideration. Study limitations included the retrospective study design, varied treatment regimens, and modest local cohort sizes, challenging the comparability of patient profiles across different urothelial carcinoma cohorts. ARID1A status was assessed via IHC in the ZSHS and FUSCC cohorts but by sequencing in the IMvigor210 cohort, reflecting cohort-specific data constraints. Nevertheless, both methods target biologically congruent populations given ARID1A’s predominant loss-of-function mutation pattern. Additionally, transcriptomic proxies for CD8 + T cells, although well-validated, cannot resolve spatial heterogeneity compared with IHC. However, the consistent prognostic trends observed across these orthogonal methods strengthen the robustness of our findings. As a chromatin remodeler, ARID1A interacted with a variety of proteins to modulate the tumor-immune microenvironment. The underlying molecular mechanisms of how ARID1A loss influenced TLS formation and CD8 + T-cell function should be evaluated in future studies. In conclusion, our research found that ARID1A-loss patients with urothelial carcinoma with high CD8 + T-cell infiltration exhibited a favorable prognosis and increased sensitivity to both chemotherapy and immunotherapy. The enhanced antitumor function of CD8 + T cells in urothelial carcinoma with ARID1A loss might be affected by TLSs. By accurately determining the ARID1A status and CD8 + T-cell infiltration level, we can enhance risk stratification and provide personalized therapy for urothelial carcinoma.

Urothelial carcinoma ranks as the tenth most common malignant tumor worldwide ( 1 ). Among its defining molecular features, alterations in chromatin remodeling pathways are particularly prevalent, with up to 70% of urothelial carcinoma tumors harboring inactivating mutations in at least one chromatin remodeling gene ( 2 , 3 ). Notably, these mutations are also frequently detected in normal urothelial tissue, suggesting that they may confer substantial clonal advantages that promote malignant transformation ( 4 , 5 ). Despite their high prevalence, the clinical implications and accompanying changes in the tumor microenvironment resulting from chromatin remodeling gene alterations in urothelial carcinoma remain poorly understood. The switch/sucrose nonfermentable complex is a multisubunit chromatin remodeling complex that utilizes ATP hydrolysis to regulate chromatin accessibility, transcription, and DNA repair ( 6 ). ARID1A encodes the cBAF subunit of switch/sucrose nonfermentable complexes and is one of the most frequently mutated genes in urothelial carcinoma, with mutation rates ranging from 21% to 36% ( 2 , 7 – 9 ). Notably, ARID1A is the only chromatin remodeling gene more commonly mutated in high-grade invasive and metastatic urothelial carcinoma (mUC; ref. 10 ). The majority of ARID1A alterations are nonsense truncating point mutations that lead to loss of protein expression ( 11 , 12 ). Beyond urothelial carcinoma, ARID1A mutations are also prevalent in 30% to 60% of endometriosis-associated ovarian carcinoma, colorectal cancer, and gastric cancer. Although ARID1A plays critical roles in regulating the cell cycle, DNA damage response, p53 targets, and telomerase activity, its loss has not consistently shown prognostic significance in these malignancies ( 12 – 15 ), raising important questions about its biological and clinical relevance. In addition to its role in cell-autonomous cancer progression, ARID1A loss is sufficient to induce antitumor immunity. Animal models of melanoma and colon cancer showed that ARID1A loss enhanced inflamed tumor microenvironment by triggering a type I IFN response through the cGAS-STING pathway, thereby promoting T-cell infiltration and cytotoxicity ( 16 ). Moreover, ARID1A loss has been reported to endow immunotherapeutic benefits across various solid tumor types, including urothelial carcinoma ( 17 , 18 ). Goswami and colleagues ( 19 ) demonstrated that ARID1A mutation plus CXCL13 expression levels could predict immune checkpoint blockade (ICB) response in mUC. Sarfaty and colleagues ( 20 ) reinforced prior ARID1A findings and developed a novel genetic classifier for ICB sensitivity based on ARID1A mutations. Despite these promising findings, the lack of a consistent association between ARID1A loss and overall prognosis remains puzzling. One possible explanation is the immunologic heterogeneity among patients, particularly with respect to CD8 + T-cell infiltration. In this study, we explored the non–cell-autonomous effects of ARID1A loss in urothelial carcinoma, focusing on its impacts on CD8 + T cells. We further investigated the prognostic significance, predictive value for treatment response, and functional phenotype of CD8 + T cells in the context of ARID1A loss. These findings may inform the development of novel therapeutic strategies and refined patient stratification approaches in urothelial carcinoma.

This study primarily involved two institutional cohorts: the Zhongshan Hospital, Fudan University (ZSHS) cohort and the Fudan University Shanghai Cancer Center (FUSCC) cohort. The study followed the Declaration of Helsinki and was approved by the Clinical Research Ethics Committee of Zhongshan Hospital and Fudan University (No. B2015-030). Written informed consent was obtained from each patient. The ZSHS cohort comprised 135 patients who were diagnosed with muscle-invasive bladder cancer (MIBC) and underwent radical cystectomy between 2002 and 2014. The FUSCC cohort included 118 patients with MIBC who underwent RC between 2008 and 2012. All patients from the two cohorts were classified as stage II or III according to the American Joint Committee on Cancer eighth edition staging criteria. One hundred and fifteen patients from these two cohorts received adjuvant chemotherapy (ACT) and lasted at least one therapeutic cycle. The follow-up schedule follows the MIBC guidelines ( 21 , 22 ). Detailed clinicopathologic features in ZSHS and FUSCC cohorts have been shown in Table 1 . Additionally, data from another 134 patients at FUSCC and Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, referred to as the Shanghai-sequencing cohort, with matched genomic data, were analyzed to characterize the genomic features associated with ARID1A loss. The IMvigor210 cohort included 274 patients with mUC treated with atezolizumab ( 23 – 26 ). Clinical, genomic, and transcriptomic data from the IMvigor210 cohort were obtained using the R package downloaded from https://researchpub.gene.com/IMvigor210CoreBiologies/ . We investigated the clinical outcomes and treatment sensitivity of ARID1A loss and CD8 + T-cell infiltration in the ZSHS, FUSCC, and IMvigor210 cohorts. In the ZSHS and IMvigor210 cohorts, we further examined the impact of ARID1A loss and CD8 + T-cell infiltration on immune microenvironment characteristics. The Shanghai-sequencing cohort primarily served for analyzing genomic characteristics of ARID1A -loss patients with urothelial carcinoma. In total, this study included 661 patients with urothelial carcinoma from three local and one public cohorts. The specific inclusion criteria for three local cohorts are presented in Supplementary Fig. S1. Clinical characteristics of patients with urothelial carcinoma according to ARID1A IHC staining status in ZSHS and FUSCC cohorts. Abbreviations: AJCC, American Joint Committee on Cancer; N, node; no., number; pMMR, MMR-proficient; T, tumor; UC, urothelial carcinoma; UNC, University of North Carolina. P value < 0.05 is marked in bold. Statistical P values were computed using the log-rank test. PD-L1 IC data were absent for one patient because of dot loss. KDM6A data were absent for six patients because of dot loss. PIK3CA data were absent for one patient because of dot loss. The primary outcome of survival analysis was overall survival (OS), representing the time from surgery to death or last follow-up. The therapeutic response to ICB was assessed using RECIST v1.1 for the Shanghai-sequencing and IMvigor210 cohorts. For the ZSHS and FUSCC cohorts, the expression status of ARID1A was evaluated using IHC for ARID1A (Abcam, ab182560). ARID1A loss was assessed for the absence of nuclear staining in the tumor epithelium, with retained stromal nuclear staining serving as an obligate internal control ( 13 ). Representative images of ARID1A are available in Supplementary Fig. S2A. For the Shanghai-sequencing and IMvigor210 cohorts, ARID1A loss referred to point mutations of ARID1A according to original genomic data, considering that point mutations of ARID1A significantly diminished its transcriptomic expression (Supplementary Fig. S2B). For the ZSHS, FUSCC, and Shanghai-sequencing cohorts, the infiltration level of CD8 + T cells was evaluated using IHC for CD8A (Abcam, ab17147). We divided patients with urothelial carcinoma into high, moderate, and low infiltration of CD8 + T cells based on tertiles in the ZSHS cohort. The cutoff values were 39 cells/high power field and 17 cells/high power field in the ZSHS cohort and were used in the FUSCC and Shanghai-sequencing cohorts as well. Representative images of CD8 + T cells are available in Supplementary Fig. S2C. For the IMvigor210 cohort, we used MCP-counter algorithm to evaluate the infiltration level of CD8 + T cells based on transcriptomic data ( 27 ). The patients were divided into three subgroups based on tertiles of the MCP-counter score of CD8 + T cells, which were 0.117 and 0.043. Genomic sequencing was conducted on patients’ fresh or formalin-fixed, paraffin-embedded tumor tissues in the Shanghai-sequencing cohort. Blood samples were collected and processed in Streck tubes within 72 hours. The quality of the sample must be equivalent to 100 ng tissue DNA. Tissue DNA was sheared to 200 bp fragments using a Covaris E210 system. Library preparation was done using the AccelNGS 2S DNA Library Kit and xGen Lockdown Probes Kit, with custom probes for gene exons and intronic regions. Libraries were quantified using a Qubit 3.0 Fluorometer and analyzed with an Agilent 2100 Bioanalyzer. Paired-end sequencing was performed on an Illumina NovaSeq 6000 with 2 × 150 bp reads. Sequencing data were aligned to the hg19 genome with the Burrows–Wheeler Aligner, followed by duplicate removal and local realignment. Single-nucleotide variants and indels were identified using Genome Analysis Toolkit and LoFreq, annotated with ANNOVAR, and compiled into a mutation annotation file document. IHC staining was performed on fresh or formalin-fixed, paraffin-embedded tissue microarrays as described previously ( 28 ). Briefly, slides were heated at 60°C for 6 hours, deparaffinized in xylene (3 × 15 minutes), and rehydrated through a graded ethanol series. Antigen retrieval was performed by immersing slides in 0.01 mol/L sodium citrate buffer (pH 6.0) at 95°C to 100°C for 20 minutes, followed by endogenous peroxidase blocking with 3% H 2 O 2 in methanol at 37°C for 30 minutes. For single IHC staining, slides were incubated with primary antibodies at 4°C overnight and visualized using a 3,3′-diaminobenzidine chromogen system. For double IHC staining, after completing the single IHC–3,3′-diaminobenzidine protocol, slides were further incubated with secondary primary antibodies at 4°C for 2 hours, followed by application of the Vector Blue Alkaline Phosphatase Substrate Kit (Vector Laboratories). All slides were independently scored by two blinded pathologists from separate institutions with no access to clinicopathologic data. The mean cell count (immune cells and stroma cells) and IHC score (tumor cells) of their evaluation were adopted. In instances of marked discrepancies in IHC scoring results, the findings were independently reassessed by two pathologists. The antibodies for IHC staining are detailed in Supplementary Table S1. Thirteen groups of double staining were performed: CD68 and HLA-DR for M1 macrophages, CD68 and CD206 for M2 macrophages, CD68 and DC-SIGN for DC-SIGN + tumor-associated macrophages (TAM), CD11c and HLA-DR for dendritic cells (DC), CD8 and CD103 for tissue-resident memory T cells, CD8 and CXCR5 for follicular cytotoxic CD8 + T cells, CD8 and TIGIT for TIGIT + CD8 + T cells. Single IHC staining was conducted for the rest of the immune cells, including CD8 + T cells, CD3 + T cells, B cells (CD19), Th17 cells (IL17A), and immune checkpoint and immunoregulatory molecules like IFN-γ, PRF1, CXCL13, PDPN, TGF-β, PD-L1, B7-H3, B7-H4, VISTA, PD-1, TIGIT, LAG-3, CTLA-4, TIM-3, NKG2A, and CD39. Expression levels of TGF-β and PDPN were interpreted as microenvironmental indices, not cellular identity determinants. The existence of intratumoral tertiary lymphoid structures (TLS) was assessed via both hematoxylin and eosin and CD3/CD20 double staining using a previously published scale ( 29 ). Tumors with at least one intratumoral TLS were defined as TLS-positive. For genomic analysis, tumor mutation burden (TMB) was identified as the number of somatic synonymous and nonsynonymous mutations (base substitutions and indels) per megabase of analyzed DNA in the Shanghai-sequencing and IMvigor210 cohorts. For transcriptomic analysis, RNA sequencing data were mined along with the process of acquiring clinical information and normalized through the formula log 2 (FPKM+1). The score of the cGAS-STING pathway, MHC I signature, antitumor cytokine signature, immune suppression signature, epithelial–mesenchymal transition (EMT) signature, cancer-associated fibroblast (CAF) signature, myeloid cells traffic signature, and immune suppression by myeloid signature were calculated by the single sample gene set enrichment analysis algorithms based on related gene expression as previously reported ( 30 , 31 ). The sources and details of these gene signatures are detailed in Supplementary Table S2. The Kaplan–Meier approach combined with the log-rank test and multivariable Cox regression analysis were used for survival analysis. Analyses of correlations between categorical variables were performed using the χ 2 test or Fisher exact test. For continuous variables, the Student t test and one-way ANOVA test were used to analyze continuous variables among two or more subgroups. All data analysis was based on SPSS Statistics v27.0 and R v4.3.2 ( http://www.rproject.org/ ), and P < 0.05 was considered statistically significant.

Enrollment, clinicopathological features and inclusion criteria for ZSHS, FUSCC, and Shanghai-sequencing cohorts. Representative IHC images of ARID1A and CD8+ T cells in UC. Association between ARID1A status and TMB in UC. Association between ARID1A status and CD8+ T cell signature in the IMvigor210 cohort. Details of immunohistochemistry antibodies. Sources and details of the utilized gene signatures.