Comprehensive Molecular Characterization of Muscle-Invasive Bladder Cancer

Abstract

We report a comprehensive analysis of 412 muscle-invasive bladder cancers characterized by multiple TCGA analytical platforms. Fifty-eight genes were significantly mutated, and the overall mutational load was associated with APOBEC-signature mutagenesis. Clustering by mutation signature identified a high-mutation subset with 75% 5-year survival. mRNA expression clustering refined prior clustering analyses and identified a poor-survival "neuronal" subtype in which the majority of tumors lacked small cell or neuroendocrine histology. Clustering by mRNA, long non-coding RNA (lncRNA), and miRNA expression converged to identify subsets with differential epithelial-mesenchymal transition status, carcinoma in situ scores, histologic features, and survival. Our analyses identified 5 expression subtypes that may stratify response to different treatments.

Keywords: APOBEC mutation; DNA methylation; basal mRNA subtype; lncRNA transcriptome; luminal mRNA subtype; microRNA; muscle-invasive bladder cancer; neoantigen; neuronal subtype; regulon.

Conflict of interest statement

Potential conflicts of interest

S.P.L. has received investigator initiated research funding from Endo Pharmaceuticals; support for a clinical trial from FKD and Viventia; is a consultant for UroGen, Vaxiion, Nucleix and BioCancell. A.D.C., G.G. and M.M have received research funding from Bayer AG. E.A.G. is now employed by GenomeDx Biosciences. M.M. has no current conflicts; was previously an equity holder in and consultant for Foundation Medicine. G.B.M. is a member of the scientific advisory board and receives research support from AstraZeneca. D.J.M. has stock options in ApoCell, Inc. J.B. has a paid consultancy with Pfizer; has received advisory board and lecture fees from Merck; has received advisory board fees from Genentech; has given uncompensated presentations at Genentech. C.J.W. is a cofounder and advisory board member of Neon Therapeutics. No other conflicts of interest declared.

Copyright © 2017 Elsevier Inc. All rights reserved.

Figures

Figure 1. Landscape of mutational signatures, mutations and copy number alterations

(A) Alteration landscape for 412 primary tumours. Top to bottom: Synonymous and non-synonymous somatic mutation rates, with one ultra-mutated sample with a POLE signature. Mutational signature (MSig) cluster, APOBEC mutation load, and neoantigen load by quartile. Normalized activity of 4 mutational signatures. Combined tumor stage (T1,2 vs. T3,4) and node status, papillary histology, and gender. Somatic mutations for significantly mutated genes (SMGs) with frequency ≥ 7%. Copy number alterations for selected genes, and FGFR3 and PPARG gene fusions. (B) Kaplan-Meier plots for overall survival (L to R): Overall mutation burden (SNVs); Mutation signature clusters (MSig1–4); APOBEC-mediated mutation load; Neoantigen load;

Figure 2. mRNA expression subtypes

Top, L to R: 5 mRNA expression subtypes: luminal-papillary, luminal-infiltrated, luminal, basal-squamous and neuronal. Covariates: 4 previously reported TCGA subtypes; selected clinical covariates and key genetic alterations; normalized expression for miRNAs and proteins; log2 (fold change against the median expression across samples) for selected genes, for labeled gene sets. Samples within the three luminal subtypes, the basal-squamous subtype, and the neuronal subtype are ordered by luminal, basal, and neuroendocrine signature scores, respectively. Genes that are down-regulated* vs. up-regulated** in CIS.

Figure 3. Somatic alterations in signaling pathways across mRNA subtypes

Somatic alterations include mutations and copy-number changes (i.e. deep deletions and high-level amplifications, from GISTIC results). Missense mutations are counted only if they have known oncogenic function based on OncoKB (http://oncokb.org) annotations, or have previously been reported in COSMIC, or occur at known mutational hotspots. The table shows the fraction of samples with alterations in selected signaling pathways. In the pathway diagrams, edges show pairwise molecular interactions; boxes outlined in red denote alterations leading to pathway activation, while boxes outlined in blue denote predicted pathway inactivation. The oncoprint illustrates type and frequency of alteration, as well as patterns of co-occurrence, for selected genes from the pathways highlighted in the table.

Figure 4. LncRNA expression subtypes

(A) Heatmap and covariates for four unsupervised lncRNA consensus clusters. Top to bottom: normalized abundance heatmap for 171 lncRNAs; profile of silhouette width calculated from the consensus membership heatmap, Wcm; covariates for clinical parameters, molecular subtypes, purity, mutations in TP53 and FGFR3, FGFR3 and PPARG gene fusions; row-scaled mRNA levels for 3 genes; collapsed CIS gene sets (Dyrskjot et al., 2004) (Methods; CIS up = genes up-regulated in CIS; CIS down = genes down-regulated in CIS); row-scaled regulon activity profiles (showing activated, undefined, or repressed status) for 23 regulators; RNA-seq-based EMT scores (Mak et al., 2016). The following p values are Bonferroni-corrected: for mutated genes (for 58 SMGs), gene fusions (for 23 fusions), regulon activity (for 23 regulators), and mRNA-seq (for 12 genes). (B) A Kaplan-Meier plot for overall 5-year survival according to lncRNA subtype.

Figure 5. MicroRNA expression subtypes

(A) Heatmap and covariates for a 4-cluster unsupervised consensus clustering solution. Top to bottom: Normalized heatmap showing a subset of 142 miRNAs that had a mean RPM ≥ 50 and an absolute value of tumour-vs-normal fold change ≥ 1.5. Profile of silhouette width calculated from the consensus membership heatmap, Wcm, with lower values indicating samples that are atypical cluster members. Covariate tracks for clinical parameters, genomic platform subtypes, purity, mutations in TP53 and FGFR3, and FGFR3 and PPARG gene fusions. Row-scaled regulon activity profiles for 23 regulators that have been associated with bladder cancer. Row-scaled mRNA levels for 12 genes, then for collapsed CIS gene sets (Dyrskjot et al., 2004) (Methods; CIS up = genes up-regulated in CIS; CIS down = genes down-regulated in CIS); and RNA-seq-based EMT scores (Mak et al., 2016). The following p values are Bonferroni-corrected: for mutated genes (for 58 SMGs), gene fusions (for 23 fusions), regulon activity (for 23 regulators), and mRNA-seq (12 genes). (B) A Kaplan-Meier plot for overall survival data that has been censored at 5 years.

Figure 6. Integrated analysis

(A) Cluster of cluster assignments analysis (COCA). Unsupervised clustering of subtype calls. Subtype calls for mRNA (red), lncRNA (black), and miRNA (blue) are colored by separate data type. Annotations at the right of and below the heatmap use colors for mRNA subtypes. (B,C) Multivariate Cox analysis for overall survival. (B) Coefficients (β) from the LASSO-penalized multivariate Cox regression on 15 covariates that were significant (corrected p

Figure 7

Proposed schema of expression-based, subtype-stratified therapeutic approach as a framework for prospective hypothesis testing in clinical trials. * For luminal-papillary cases, the low predicted likelihood of response is based on preliminary data from (Seiler et al., 2017). See Discussion.