Claudin-low bladder tumors are immune infiltrated and actively immune suppressed

Abstract

We report the discovery of a claudin-low molecular subtype of high-grade bladder cancer that shares characteristics with the homonymous subtype of breast cancer. Claudin-low bladder tumors were enriched for multiple genetic features including increased rates of RB1, EP300, and NCOR1 mutations; increased frequency of EGFR amplification; decreased rates of FGFR3, ELF3, and KDM6A mutations; and decreased frequency of PPARG amplification. While claudin-low tumors showed the highest expression of immune gene signatures, they also demonstrated gene expression patterns consistent with those observed in active immunosuppression. This did not appear to be due to differences in predicted neoantigen burden, but rather was associated with broad upregulation of cytokine and chemokine levels from low PPARG activity, allowing unopposed NFKB activity. Taken together, these results define a molecular subtype of bladder cancer with distinct molecular features and an immunologic profile that would, in theory, be primed for immunotherapeutic response.

Figure 1. Identification of a claudin-low subtype in bladder cancer.

(A) Unsupervised clustering of TCGA muscle-invasive UC samples. Samples were clustered on the basis of expression of tight-junction claudins, a bidirectional EMT signature, and a TIC signature. The tumors identified as claudin-low are highlighted in green on the dendogram. n = 408. (B) Waterfall plot showing correlation with the basal and luminal centroids as defined by BASE47 classification; claudin-low tumors are highlighted in green. Claudin-low tumors were significantly enriched in the BASE47 basal subtype (Fisher’s exact test P = 1.18 × 10^–16) and were highly correlated with the basal centroid (Pearson’s correlation P = 9.33 × 10^–15). n = 408. (C) Kaplan-Meier plot showing overall survival of bladder cancer by molecular subtype. Significance was determined by log-rank testing with a Bonferroni correction. n = 408. (D and E) Bar graphs showing the classification of TCGA UC tumors by TCGA mRNA cluster subtype (x axis) and our subtype classifications (y axis) by count and percentage. n = 129. EMT, epithelial-to-mesenchymal transition; TCGA, The Cancer Genome Atlas; TIC, tumor-initiating cell; UC, urothelial carcinoma.

Figure 2. Genomic characterization of bladder cancer subtypes.

(A) Oncoprint of genomic copy number alterations and mutations by bladder cancer subtype for genes previously identified as significantly mutated or copy number altered in more than 5% of bladder tumors. n = 408. (B) Bar plots of genes that were identified to have a significant (P < 0.05) difference in either gene mutation or copy number alteration (CNA) between the claudin-low and basal and/or luminal subtypes. *P < 0.05, **P < 0.01, and ***P < 0.001, by Fisher’s exact test.

Figure 3. Immune characterization of bladder cancer subtypes.

(A) Volcano plot of log₂ fold change of median gene expression and –log₁₀ P value of gene expression across bladder tumor subtypes. Dashed line across the plots corresponds to a significance threshold of P = 0.05. n = 408. Significance was calculated using Student’s t test with a Bonferroni correction. (B) Heatmaps of supervised clustering of bladder tumor subtypes across previously identified immune signatures. n = 408. (C) Heatmap of supervised clustering of bladder tumor subtypes across an immune suppression gene signature. n = 408. (D) Box plot of immune suppression gene signature z score across bladder tumor subtypes. n = 408. (E) Box plot of PD-L1 gene expression across the Pan-Cancer tumor types. n = 3,602. (F) Box plot of immune suppression gene signature z scores across the Pan-Cancer tumor types. n = 3,602. The box plots denote the interquartile range (IQR), with the box representing Q1 to Q3, the line denoting Q2, and the whiskers extending an additional 1.5 times the IQR beyond Q1 and Q3. The dots represent data points. BLCA, bladder urothelial carcinoma; BRCA, breast cancer; COAD, colon adenocarcinoma; GBM, glioblastoma multiforme; HNSC, head and neck squamous cell carcinoma; KIRC, kidney renal clear cell carcinoma; LAML, acute myeloid leukemia; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma; OV, ovarian serous cystadenocarcinoma; READ, rectum adenocarcinoma; UCEC, uterine corpus endometrial carcinoma; LUM, luminal; TCGA, The Cancer Genome Atlas; PanCan, Pan-Cancer.

Figure 4. Immune gene signatures have prognostic value across bladder cancer subtypes.

(A) Forest plot of Cox PH ratios of the immune gene signatures across all tumors, with a 95% CI indicated around the values. n = 408. (B) Forest plot of Cox PH ratios of the immune gene signatures within defined tumor subtypes, with a 95% CI indicated around the values. n = 408. *P < 0.05, prognostically significant signatures by Cox PH modeling. Cox PH, Cox proportional hazard.

Figure 5. BCR and TCR segment expression is prognostic.

(A) Number of TCR gene segments by subtype in which increased expression was significantly associated with improved survival by Cox PH model fit. Null distributions (gray bars) with 95% CIs were generated for each by bootstrap resampling of non-TCR genes and calculation of the number of significant P values that were similarly associated with prolonged survival. n = 292. (B) Number of BCR gene segments by subtype in which increased expression was significantly associated with improved survival by Cox PH model fit. Null distributions (gray bars) with 95% CIs were generated for each by bootstrap resampling of non-TCR genes and calculation of the number of significant P values that were similarly associated with prolonged survival. n = 292. (C) Specific TCR gene segments in which increased expression was significantly associated with improved survival by Cox PH model fit for all tumors (gray boxes), basal tumors (red boxes), claudin-low tumors (green boxes), and luminal tumors (blue boxes). (D) Specific BCR gene segments in which increased expression was significantly associated with improved survival by Cox PH model fit for all tumors (gray boxes), basal tumors (red boxes), claudin-low tumors (green boxes), and luminal tumors (blue boxes). (E) Log base 10 number of reads supporting any BCR V(D)J rearrangement are shown by subtype. n = 181. Mann-Whitney U–Wilcoxon test with an FDR multiple testing correction was used to determine significance. (F) Repertoire diversity by subtype. The box plots in E and F denote the interquartile range (IQR), with the box representing Q1 to Q3, the line denoting Q2, and the whiskers extending an additional 1.5 times the IQR beyond Q1 and Q3. The dots represent data points. n = 150. Mann-Whitney U–Wilcoxon test with an FDR multiple testing correction was used to determine significance. BCR, B cell receptor; Cox PH, Cox proportional hazard; TCR, T cell receptor.

Figure 6. Predicted neoantigen burden by bladder cancer subtype.

(A) Stacked bar plot showing the number of predicted neoantigens in each bladder tumor with a predicted IC₅₀ of less than 50 nm (red bars) and less than 150 nm (yellow bars). Numbers of predicted neoantigens are shown in the left y axis. Blue line and right y axis show the number of missense mutations per tumor. n = 289. (B) Scatter plot of somatic missense mutations (log₂) versus predicted neoantigen burden (log₂) across TCGA data set. Significance and correlation were determined using Spearman’s rank test. n = 289. (C) Box plot showing the number of predicted neoantigens with an IC₅₀ of less than 50 nm by tumor molecular subtype. Subtypes were not significantly different (P > 0.05). Significance was determined by 1-way ANOVA. n = 289. The box plots denote the interquartile range (IQR), with the box representing Q1 to Q3, the line denoting Q2, and the whiskers extending an additional 1.5 times the IQR beyond Q1 and Q3. The dots represent data points. (D) Kaplan-Meier plot showing survival of bladder cancer patients with high (greater than median value, blue line) versus low (less than median value, red line) predicted numbers of neoantigens. Vertical hash marks indicate censored data. Significance was determined by log-rank test. n = 289. TCGA, The Cancer Genome Atlas.

Figure 7. Cytokine and chemokine regulation across bladder cancer subtypes.

(A and B) Volcano plots of log₂ fold change of median gene expression and –log₁₀ P value of gene expression for cytokines and chemokines across claudin-low/basal and claudin-low/luminal subtypes. Dashed lines across plots correspond to P = 0.05. Significance was calculated using Student’s t test with a Bonferroni correction. n = 408. (C) GSEA enrichment plots indicating that NF-κB signatures were decreased in rosiglitazone-treated UMUC7 and UMUC9 bladder cancer cell lines. Significance was determined using GSEA software. (D) Box plots showing that immunosuppression gene signature expression was significantly decreased across UMUC7 and UMUC9 cell lines after rosiglitazone treatment. Significance was determined using Student’s t test. n = 6. (E) Correlation plot of immunosuppression and EMT gene signature expression. n = 408. Significance and correlation were calculated using a Spearman’s rank test. (F) Box plots showing that EMT gene signature expression was decreased across UMUC7 and UMUC9 cell lines after rosiglitazone treatment. Significance was determined using Student’s t test. n = 6. The box plots in D and F denote the interquartile range (IQR), with the box representing Q1 to Q3, the line denoting Q2, and the whiskers extending an additional 1.5 times the IQR beyond Q1 and Q3. The dots represent data points. ES, enrichment score; EMT, epithelial-to-mesenchymal transition; GSEA, gene set enrichment analysis.

Figure 8. Model of immune infiltration across bladder cancer subtype.

Proposed model of immune response regulation through PPARγ and NF-κB signaling.