Proteomic analysis of non-muscle invasive and muscle invasive bladder cancer highlights distinct subgroups with metabolic, matrisomal, and immune hallmarks and emphasizes importance of the stromal compartment

Abstract

We present the proteomic profiling of 79 bladder cancers, including treatment-naïve non-muscle-invasive bladder cancer (NMIBC, n = 17), muscle-invasive bladder cancer (MIBC, n = 51), and neoadjuvant-treated MIBC (n = 11). Proteins were extracted from formalin-fixed, paraffin-embedded samples and analyzed using data-independent acquisition, yielding >8,000 quantified proteins. MIBC, compared to NMIBC, shows an extracellular matrix (ECM) and immune response signature as well as alteration of the metabolic proteome together with concomitant depletion of proteins involved in cell-cell adhesion and lipid metabolism. Neoadjuvant treatment did not consistently impact the proteome of the residual tumor mass. NMIBC presents two proteomic subgroups that correlate with histological grade and feature signatures of cell adhesion or lipid/DNA metabolism. Treatment-naïve MIBC presents three proteomic subgroups with resemblance to the basal-squamous, stroma-rich, or luminal subtypes and signatures of metabolism, immune functionality, or ECM. The metabolic subgroup presents an immune-depleted microenvironment, whereas the ECM and immune subgroups are enriched for markers of M2-like tumor-associated macrophages and dendritic cells. Markers for natural killer cells are exclusive for the ECM subgroup, and markers for cytotoxic T cells are a hallmark of the immune subgroup. Endogenous proteolysis is increased in MIBC alongside upregulation of matrix metalloproteases, including MMP-14. Genomic panel sequencing yielded the prototypical profile of prevalent FGRF3 alterations in NMIBC and TP53 alterations in MIBC. Tumor-stroma interactions of MIBC were investigated by proteomic analysis of patient-derived xenografts, highlighting specific tumor and stroma contributions to the matrisome and tumor-induced stromal proteome phenotypes. © 2024 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Keywords: bladder cancer; proteomics; tumor microenvironment.

Figure 1

Overview of study. (A) Overview of BC cohort. (B) Overview of experimental workflow (created with Biorender.com). (C) Quantified proteins per sample. (D) Principal component analysis of treatment‐naïve BC samples. (E) Differential abundance of proteins in treatment‐naïve MIBC versus NMIBC at sparsity reduction with a 90% cutoff, resulting in 7,356 proteins. We used linear models of microarray analysis (LIMMA), applying a significance cutoff using a Benjamini‐Hochberg adjusted p value (p‐adj) < 0.05 and a minimal fold‐change of 1.5. (F) Comparison of downregulated (top) or upregulated (bottom) biological processes in treatment‐naïve MIBC versus NMIBC. (G) Immune and stromal cell markers across NMIBC and MIBC. The selected proteins shown are either significantly enriched/depleted (Fisher's exact test, p < 0.05) or differentially regulated (quantitative difference >1.5‐fold‐change and p‐adj < 0.05) in at least one proteomic group. The Z score was calculated based on sample intensity subtracted by the mean intensity and divided by the standard deviation across all samples.

Figure 2

Proteomic characterization of NMIBC. (A) Unsupervised hierarchical clustering analysis with Euclidean distance and Ward's minimum variance method of NMIBC identified two distinct clusters. The analysis was performed after sparsity reduction to max. 25%, leaving 4,800 proteins in the dataset, and solely on NMIBC samples. (B) Heatmap of selected differentially regulated proteins in proteomic NMIBC subgroups. Only two proteomes were classifiable into one of the UROMOL subtypes [8]. CUR, complete ureter resection; PUR, partial ureter resection. (C) Principal component analysis of NMIBC subgroups and MIBC. The Z score was calculated based on sample intensity subtracted by the mean intensity and divided by the standard deviation across all samples. (D) Comparison of our NMIBC subgroups with proteomic subtypes of Stroggilos et al [19]. The Cell Adhesion subgroup is enriched with 10 signature proteins of the NPS3 group.

Figure 3

Proteomic characterization of MIBC. (A) Unsupervised hierarchical clustering of MIBC proteomes (proteins: quantitative difference >1.5‐ fold‐change and p‐adj < 0.05) performed after sparsity reduction to maximum 25%, leaving 4,800 proteins in the dataset, and solely on MIBC samples. (B) Heatmap including previously established candidates for molecular characterization of MIBC [9, 15]. Proteins were enriched/depleted (Fisher's exact test, p < 0.05) or differentially regulated (p‐adj < 0.05) in one or more proteomic MIBC subgroup. Neuroendocrine markers were not enriched in any subtype. The Z score was calculated based on sample intensity subtracted by the mean intensity and divided by the standard deviation across all samples. (C) Distribution of samples across proteomic MIBC subgroups, based on T stage. (D) Distribution of recurrence‐free survival or follow‐up time, respectively. Five cases presented with recurrent tumor or metastasis during surgery and were removed from this analysis. (E) Cox proportional hazard analysis. (F) Heatmap of possible therapeutic targets. FGFR3 intensities were taken from an additional analysis using a library‐predicted approach and are specifically enriched in the metabolic subgroup of MIBC (Fisher's exact test, p = 8 × 10⁻⁴). (G) Enrichment of matrisome proteins across proteomic MIBC subgroups.

Figure 4

YAP1 staining; genomic characterization of cohort. (A) IHC of YAP1 in treatment‐naïve BC (NMIBC, n = 17; MIBC, n = 50) comparing differences across MIBC subtypes and NMIBC. (B) Quantification of YAP1 in different tumor areas. YAP1 intensity score of 0 represents no signal, and a score of 3 indicates high signal intensity. Chi‐squared tests show a correlation between YAP1 and BC subclusters in both stroma (left) and tumor (middle). The right panel depicts the percentage of tumor nuclei with positive YAP1 signal. (C) Count and types of genomic alterations per gene. (D) Frequency of genes with likely pathogenic mutations (FATHMM [74], ClinVar [75]) and different prevalence in BC subgroups (Fisher's exact test, p < 0.05).

Figure 5

Tumor‐infiltrating immune cells in MIBC; association of BSCL2 with recurrence‐free survival. (A) Fraction of tumor‐infiltrating lymphocytes in MIBC subgroups (Kruskal–Wallis test, *p = 0.014), given in [%] as intratumoral stromal area occupied by mononuclear inflammatory cells over total intratumoral stromal area. (B) Differential regulation of immune checkpoint proteins. PD‐L1 intensities were taken from the library‐predicted approach (Fisher's exact test, p = 1.3 × 10⁻³). (C) Representative images of CD8 (pink) and PD1 (turquoise) in MIBC with either high or low PD‐L1 (green) (n = 10) and (D) quantitative data (*p = 0.045, unpaired Wilcoxon test). The bar plot shows the mean of cells per μm², error bars represent SEM. (E) Principal component analysis of proteomic BC subgroups. (F and G) Mass spectrometric intensities of E‐ and P‐cadherin in proteomic BC subgroups (E‐cadherin: p = 1.9 × 10⁻⁹; P‐Cadherin: p = 1.3 × 10⁻³, ANOVA). (H) Proteins associated with time to recurrence. (I) MS intensities of BSCL2 comparing recurrence within 1 year (n = 11) to cases with recurrence‐free survival for more than 3 years (n = 8) (*p = 0.007, one‐sided t‐test; results of two‐sided t‐test are given in supplementary material, Figure S13). (J) Representative IHC of BSCL2. (K) IHC quantification (h‐score [83]) comparing recurrence within 1 year (n = 11) to cases with recurrence‐free survival for more than 3 years (n = 8) (p = 0.045, one‐sided t‐test; results of two‐sided t‐test are given in supplementary material, Figure S13). (L) Kaplan–Meier plot of BSCL2 (p = 0.016, log‐rank test). Cutoff point was determined by maximally selected rank statistics [80]. Vertical lines indicate censored data.

Figure 6

Elevated proteolysis and MMP levels in MIBC; proteomic characterization of MIBC PDX models. (A) Number of semi‐tryptic peptides and fully tryptic peptides in treatment‐naïve cases. (B) Number of semi‐tryptic peptides per protein. (C) Proportional intensity of semi‐tryptic peptides in MIBC and NMIBC (p = 3.2 × 10⁻⁵, t‐test). (D) Mean abundance of MMPs and TIMPs [86] (error bars represent SD). (E) MMP‐14 levels (p = 6.0 × 10⁻⁶, t‐test). (F) Representative MMP‐14 IHC. The arrows highlight staining in stromal regions. (G) IHC quantification (h‐score [83]) of MMP‐14 in tumor and stroma of treatment‐naïve bladder cancer (NMIBC, n = 11; MIBC, n = 46; p = 7.9 × 10⁻⁶ unpaired Wilcoxon rank‐sum test). (H) Immunoblot of T24 cells with shRNA‐based knockdown of MMP‐14 using two different constructs (sh_A and sh_B) and scrambled control (shCtrl). (I) Transwell migration (mean number of cells, ±SD) of T24 cells bearing MMP‐14 shRNA‐knockdown versus control with 0% or 10% FCS chemoattractant [left panel; shCtrl versus sh_A (p = 2.8 × 10⁻⁶, t‐test) or sh_B (p = 2.7 × 10⁻⁵); right panel; shCtrl versus sh_A (p = 6.1 × 10⁻⁸) or sh_B (p = 5.4 × 10⁻⁸)]. Ten images were taken per condition. (J) Transwell matrigel invasion assay (mean number of cells, ±SD) of T24 cells bearing MMP‐14 shRNA‐knockdown versus control with 0% FCS chemoattractant, shCtrl versus sh_A (p = 1.1 × 10⁻⁵, t‐test), and shCtrl versus sh_B (p = 2.9 × 10⁻⁵, t‐test). The bar plot shows the mean of 10 images for each condition. (K) Mass spectrometry‐based proteomics of the MIBC PDX specimens showing numbers of identified human (tumor) and murine (stroma) proteins per sample. (L) Summed protein intensities of either human (tumor) or mouse (stroma) origin for core matrisome (top) or matrisome‐associated (bottom) proteins; *p < 0.05; **p < 0.01; ***p < 0.001; NS, not significant (t‐test). (M) Cross‐species correlation analysis of PDX proteomics data using a partial least‐squares (PLS) model and depicted as a correlation circle plot, highlighting four distinct sets of proteins and displaying the corresponding top 100 proteins of human or murine origin. Box‐and‐whisker plots show summed protein intensities of tumor and stroma proteins of three differentially colored sets, highlighting their direct or inverse correlation. Key biological features of tumor and stroma proteins are stated.