Integrating molecular subtype and CD8⁺ T cells infiltration to predict treatment response and survival in muscle-invasive bladder cancer

The study utilized data from four independent patient cohorts, consisting of a total of 889 patients (Supplementary Fig. 1A). For our local Zhongshan Hospital Affiliated to Fudan University (ZSHS) cohort, with the approval of the Clinical Research Ethics Committee of Zhongshan Hospital affiliated to Fudan University, 142 patients who were diagnosed with MIBC and underwent radical cystectomy at Zhongshan Hospital from 2002 to 2014 were followed up regularly till July 2016. Among them, 7 patients were excluded from the analysis due to dot loss on tissue microarray (TMA). Of these, 64 patients received adjuvant chemotherapy to reduce the risk of disease relapse according to the guidelines and the preferences of the patients, while 71 did not receive any neoadjuvant or adjuvant therapy. In clinical settings, platinum-based chemotherapy remains the standard for treating advanced and metastatic bladder cancer [17]. Our local ZSHS cohort maintained consistency in systemic therapy over a 12-year period, with patients being administered platinum-based chemotherapy rather than immunotherapy. The Cancer Genome Atlas (TCGA) cohort encompasses 391 MIBC patients with comprehensive clinical profiles. Initially, a pool of 412 patients diagnosed with bladder cancer was sourced from the TCGA network using TCGA-Assembler 2.0.6 as of July 2021. Subsequently, 21 cases were excluded from this pool due to reasons such as non-muscle-invasive bladder cancer (NMIBC) pathologic diagnosis (n = 4), absence of survival or mRNA data (n = 7), or prior neoadjuvant therapy (n = 10). Regarding the IMvigor210 cohort, a phase II clinical trial that comprises 348 individuals diagnosed with mUC treated with the PD-L1 inhibitor atezolizumab, clinical and RNA-seq data were procured from http: //research-pub.gene.com/IMvigor210CoreBiologies utilizing the "IMvigor210CoreBiologies" R package. As for the phase II clinical trial NCT03179943 cohort, it encompassed 21 patients who underwent treatment with guadecitabine and atezolizumab. Ultimately, the mRNA data of 15 patients were incorporated into this study. Detailed clinical and pathological attributes are presented in Supplementary Tables 1–3.

For transcriptomic analyses, RNA-seq data were normalized through the formula Log₂(FPKM + 1). Immune cells infiltration signature including CD8⁺ T cells signature [18], effector CD8⁺T signature [4], exhausted CD8⁺T signature [19] and immune function signature including IFN-γ-related signature [20], pan fibroblast TGF-β signature [21] were calculated by Single sample gene set enrichment analysis (ssGSEA) algorithms based on related genes expression as previously reported (genes are listed in Supplementary Table 4). ssGSEA algorithms was accomplished by a ‘single sample’ extension of GSEA that allows one to define an enrichment score that represents the degree of absolute enrichment of a gene set in each sample within a given data set. The gene expression values for a given sample were rank-normalized, and an enrichment score was produced using the empirical cumulative distribution functions (ECDF) of the genes in the signature and the remaining genes [22].

For genomic analyses, tumor mutation burden (TMB) was defined as total non-silent somatic mutation per megabase (mut/Mb) which was acquired from https://​portal.​gdc.​cancer.​gov/​. The copy number variants (CNV) data in TCGA cohort originated from http://​www.​cbioportal.​org. Amplified or deleted regions of the genome were identified using the Genomic Identification of Significant Targets in Cancer (GISTIC) 2.0 algorithm, which divided into five discrete steps: (1) accurately defining the copy-number profile of each cancer sample; (2) identifying the somatic copy number alterations (SCNAs) that most likely gave rise to these overall profiles and estimating their background rates of formation; (3) scoring the SCNAs in each region according to their likelihood of occurring by chance; (4) defining the independent genomic regions undergoing statistically significant levels of SCNA; and (5) identifying the likely gene target(s) of each significantly altered region [23].

Serial tissue sections fixed in formalin and embedded in paraffin were used to construct TMA and the antibodies used for immunohistochemistry (IHC) staining are summarized in Supplementary Table 5. IHC staining for CD8⁺ T cells and other molecules were carried out according to previous protocols [14, 24, 25] (representative IHC images, Supplementary Fig. 1B). As before, the infiltration density of immune cells was evaluated as the mean value of cells/HPF under 3 randomized high-power magnification filed (HPF, × 200 magnification) independently by two pathologists who were blind to the clinical information with NanoZoomer-XR (Hamamatsu) and Image Pro plus 6.0 digitally. The cut-off values of CD8⁺ T cells were determined by the median value in each study cohort. In TCGA cohort, IMvigor210 cohort and NCT03179943 cohort, the cut-off points were 0.910, 1.035 and 0.530, respectively, while the cut-off value of was set at 25 cells/HPF in ZSHS cohort.

The molecular subtype of each patient was identified through ‘BLCAsubtyping’ R package from https://​github.​com/​cit-bioinfo/​BLCAsubtyping in TCGA, IMvigor210 and NCT03179943 cohorts. In ZSHS cohort, we performed IHC single staining of GATA3 and KRT5/6 to classify patients into either Luminal or Basal subtype according to an IHC subtyping method reported in previous articles [25, 26]. GATA3 is predominantly expressed in the Luminal subtype, whereas KRT5/6 is regarded as a Basal marker. IHC analysis determined whether KRT5/6 and GATA3 expression was positive or negative, with the cut-off value being 20% tumor tissue positivity under a 200 × high power field. KRT5/6⁺GATA3⁻ tumors were classified as Basal subtype, while the rest were classified as Luminal subtype.

Kaplan–Meier analysis and log-rank test were performed to conduct survival analyses. Univariate analysis was conducted using the Cox proportional hazards regression model. Pearson's Chi-square test and Fisher’s exact test were applied to detect categorical variables. Statistical P values were computed using Kruskal–Wallis test, and detailed statistical tests were described in corresponding figure legends. A P value less than 0.05 was considered statistically significant. All statistical analyses were conducted using IBM SPSS Statistics 26.0 and R software 4.1.2.

We first explored the association between OS and the four molecular subtype/CD8⁺ T cells-stratified subgroups. The results demonstrated that patients classified under the Basal subtype with low CD8⁺ T cells infiltration (median OS: 17.7 months) exhibited the poorest prognosis, in contrast to the more favorable outcomes observed for the Basal-CD8⁺T^high (median OS: 32.0 months), Luminal-CD8⁺T^high (median OS: 44.9 months), and Luminal-CD8⁺T^low (median OS: 44.3 months) subgroups within the TCGA cohort (log-rank P = 0.019, Fig. 1a). Importantly, a coherent pattern emerged from the analysis of the ZSHS cohort, consistently indicating that the Basal-CD8⁺T^high subgroup was linked to comparatively inferior OS outcomes (log-rank P = 0.025, Fig. 1c). Moreover, univariate analyses confirmed the independent correlation between the Basal-CD8⁺T^high subgroup and decreased mortality when contrasted against the Basal-CD8⁺T^high (HR: 0.571, 95% CI: 0.347–0.940, P = 0.027 and HR: 0.348, 95% CI: 0.143–0.848, P = 0.020 in respective cohorts), Luminal-CD8⁺T^low (HR: 0.487, 95% CI: 0.297–0.799, P = 0.004 and HR: 0.475, 95% CI: 0.229–0.985, P = 0.046 in respective cohorts) and Luminal-CD8⁺T^high (HR: 0.454, 95% CI: 0.258–0.796, P = 0.006 and HR: 0.323, 95% CI: 0.147–0.711, P = 0.005 in respective cohorts) subgroups, consistently across both the TCGA and ZSHS cohorts (Fig. 1b, d).

The potential of adjuvant chemotherapy (ACT) to extend OS among patients with the Basal subtype was observed (Supplementary Fig. 2A). To delve deeper into the interplay between molecular subtype, CD8⁺ T cells infiltration, and the effectiveness of ACT in MIBC, we conducted further exploration. Our analysis unveiled a noteworthy pattern that only patients characterized by the Basal subtype and high levels of CD8⁺ T cells infiltration experienced discernible clinical advantages from platinum-based chemotherapy. This pattern was consistent across both the TCGA cohort (HR: 0.318, 95% CI: 0.170–0.594, log-rank P < 0.001, Fig. 2a), as well as the ZSHS cohort (HR: 0.139, 95% CI: 0.018–1.105, log-rank P = 0.029, Fig. 2b).

We further explored the therapeutic implications for immune checkpoint blockade (ICB) within this paradigm. The Luminal subtype indicated a pronounced survival advantage derived from PD-L1 blockade therapy (log-rank P = 0.028), while the Basal subtype did not show similar benefits (log-rank P = 0.200, Fig. 2a and Supplementary Fig. 2B). Our exploration further delved into the intricate connection between the four molecular subtype/CD8⁺ T cells-stratified subgroups and their responsiveness to immunotherapy. Utilizing data from the IMvigor210 cohort, our findings highlighted a distinct pattern that solely patients featured by Luminal-CD8⁺T^high exhibited significantly enhanced OS (log-rank P = 0.009, Fig. 2b) and a heightened objective response rate (ORR, defined as the proportion of patients with complete response and partial response; P = 0.002, Fig. 2c) subsequent to treatment with atezolizumab.

We next investigated data from the NCT03179943 phase II clinical trial, encompassing 15 patients subjected to guadecitabine in combination with atezolizumab. Specifically, among these patients, those falling under the Basal-CD8⁺T^high subgroup (median OS: 20.1 months) exhibited a noteworthy extension in OS through immuno-chemotherapy in comparison with counterparts in the Basal-CD8⁺T^low (median OS: 7.1 months), Luminal-CD8⁺T^high (median OS: 2.9 months), and Luminal-CD8⁺T^low (median OS: 3.7 months, Fig. 3e) subgroups. Importantly, it is worth noting that all individuals who achieved stable disease (SD) belonged to the Basal-CD8⁺T^high subgroup (Fig. 3d).

In an effort to unravel the underlying rationale behind the predictive significance of integrating molecular subtype and CD8⁺ T cells infiltration, we next characterized the immune contexture within the context of the four subgroups. As illustrated, patients belonging to the Luminal-CD8⁺T^high subgroup exhibited significantly elevated tumor neoantigen burden across both the TCGA and IMvigor210 cohorts (Fig. 4a and Supplementary Fig. 3B). Furthermore, the subgroup characterized by the Luminal subtype and high CD8⁺ T cells infiltration demonstrated notably heightened infiltration of effector CD8⁺ T cells relative to exhausted CD8⁺ T cells (Fig. 4b). To corroborate these outcomes, we further investigated the presence of CD103⁺CD8⁺ T cells and CXCR5⁺CD8⁺ T cells, which demonstrated heightened levels among patients with Luminal-CD8⁺T^high subtype. In contrast, the Basal-CD8⁺T^high subgroup within the ZSHS cohort exhibited a notable enrichment of TIGIT⁺CD8⁺ T cells (as illustrated in Fig. 4c).

Additionally, patients featured by Luminal-CD8⁺T^high exhibited elevated IFN-γ-related and IFN-γ response signatures, while demonstrating reduced TGF-β response and pan fibroblast TGF-β signatures (Fig. 4b) in the TCGA cohort. These findings were replicated in the ZSHS cohort. Patients categorized as Luminal-CD8⁺T^high exhibited the highest infiltration of IFN-γ⁺ cells and GZMB⁺ cells. Conversely, TGF-β⁺ cells and IL-10⁺ cells were more abundant within the Basal-CD8⁺T^high subgroup (Fig. 4d). Immunohistochemistry analysis further validated these trends. Notably, the Luminal-CD8⁺T^high subgroup within the ZSHS cohort displayed an increased presence of various immune checkpoints, including PD-1⁺ cells and PD-L1⁺ cells. Conversely, patients in the Basal-CD8⁺T^high subgroup showcased elevated expression of exhaustion markers such as TIM-3⁺ cells and CTLA-4⁺ cells (Fig. 4e).

Through analyses of the TCGA cohort, we found that as the TMB increased, the Luminal-CD8⁺T^high subgroup exhibited an augmentation in both APOBEC mutational signatures and aneuploidy scores (Fig. 4f). Furthermore, beyond the prevalent chromosome 9p21.3 deletion and 11q13.3 amplification characteristic of bladder cancer, we observed unique alterations. Specifically, the Luminal-CD8⁺T^high subgroup displayed 3p25.2 PPARG amplification and 13q14.2 RB1 deletion, while the Luminal-CD8⁺T^low subgroup exhibited 1q23.3 NECTIN4 amplification, 4p16.3 FGFR3 amplification, and 7p11.2 EGFR amplification (Fig. 4f).

In pursuit of further therapeutic avenues for MIBC patients, we sought to establish a connection between emerging biomarkers and the four molecular subtype/CD8⁺T cells-stratified subgroups. Specifically, patients featured by Luminal-CD8⁺T^low exhibited a propensity to possess FGFR3 single nucleotide variants (SNVs), which were classified under level 1 of actionability according to the OncoKB database (Fig. 5a). Moreover, FGFR3 hotspot mutations and FGFR3-TACC fusion events were notably enriched in patients characterized by the Luminal subtype and low CD8⁺ T cells infiltration (Fig. 5b). This finding underscored the potential benefits that patients with Luminal-CD8⁺T^low subtype could derive from FGFR3 inhibitors.

In terms of the potential benefit of antibody–drug conjugate (ADC) drugs, we identified that patients featured by Luminal-CD8⁺T^low displayed elevated NECTIN4 mRNA expression and a greater proportion of NECTIN4 amplifications (Fig. 5c). These observations proposed a potential clinical advantage for patients with Luminal-CD8⁺T^low subtype through the utilization of NECTIN4-ADC drugs. Moreover, we ascertained that TACSTD2 mRNA expression was prominently enriched in the Luminal-CD8⁺T^low subgroup (Fig. 5d). This finding indicated that patients within this subgroup might benefit from TROP2-ADC drug treatments. Furthermore, the high expression of HER2 protein within the Luminal-CD8⁺T^low subgroup (Fig. 5e) indicated the sensitivity of these patients to HER2-targeted drug interventions as well.