Integrated genomic and gene expression profiling identifies two major genomic circuits in urothelial carcinoma

Abstract

Similar to other malignancies, urothelial carcinoma (UC) is characterized by specific recurrent chromosomal aberrations and gene mutations. However, the interconnection between specific genomic alterations, and how patterns of chromosomal alterations adhere to different molecular subgroups of UC, is less clear. We applied tiling resolution array CGH to 146 cases of UC and identified a number of regions harboring recurrent focal genomic amplifications and deletions. Several potential oncogenes were included in the amplified regions, including known oncogenes like E2F3, CCND1, and CCNE1, as well as new candidate genes, such as SETDB1 (1q21), and BCL2L1 (20q11). We next combined genome profiling with global gene expression, gene mutation, and protein expression data and identified two major genomic circuits operating in urothelial carcinoma. The first circuit was characterized by FGFR3 alterations, overexpression of CCND1, and 9q and CDKN2A deletions. The second circuit was defined by E3F3 amplifications and RB1 deletions, as well as gains of 5p, deletions at PTEN and 2q36, 16q, 20q, and elevated CDKN2A levels. TP53/MDM2 alterations were common for advanced tumors within the two circuits. Our data also suggest a possible RAS/RAF circuit. The tumors with worst prognosis showed a gene expression profile that indicated a keratinized phenotype. Taken together, our integrative approach revealed at least two separate networks of genomic alterations linked to the molecular diversity seen in UC, and that these circuits may reflect distinct pathways of tumor development.

Figure 1. DNA copy number alterations in 146 bladder tumors.

A) Whole genome heatmap representing relative copy number profiles of the samples. Segments of gains or deletions are color-coded according their relative log2 copy number ratios. B) DNA copy number frequency plot of gains (red) and losses (blue). Above: Recurrent high-level focal amplifications (FGA; red) are indicated by red bars and labeled according to their cytogenetic localization. Below: Recurrent homozygous deleted regions (HD; blue) and recurrent minimal regions of deletions (MRD; green) labeled according to their respective cytogenetic localization.

Figure 2. Associations between chromosomal aberrations visualized by MDS.

Recurrent FGAs, HDs, and MRDs, as well as recurrent large chromosome arm deletions were included in the analysis. FGAs and HDs present in <5% of samples were excluded. Aberrations with significant positive associations, as determined by hypergeometric tests, are indicated in red and connected with green lines. Aberrations located to the same chromosomes are circled in gray for visualization purposes.

Figure 3. Genomic complexity is associated with UC gene expression subtypes.

A) Hierarchical cluster analysis (HCA) on gene expression data segregated the tumors into five clusters, HC1 to HC5. Samples within each HCA group are ordered according to their relative FGFR3 expression (high expression left, low expression right). For each individual tumor, molecular signature (MS) type, pathological grade, stage, nFGA, fBAC, and TP53/MDM2 status is indicated. The relative expression levels for genes within the CIN signature are indicated by a heatmap below (green, low expression; red high expression). B) Boxplot illustrating the number of FGAs (nFGA) and frequency of genomic imbalances (fBAC) for samples within each HCA group. C) Boxplot of nFGA (left) and fBAC (right) for tumor samples when grouped on TP53/MDM2 status and MS type. P-values obtained by Wilcoxon statistics. n.s., not significant. D) Boxplot illustrating increased CIN score for samples with increased nFGAs (left) and fBAC (right). P-values obtained by ANOVA. E and F) Disease specific survival (DSS) analysis with tumors grouped according to nFGAs (0 vs ≥1 FGA) and CIN pathway score (above or below median), respectively.

Figure 4. Integrated analysis of genomic alterations, gene mutations, and gene expression data.

A) Recurrent genomic alterations with significant association to gene expression subtypes. Activating mutation of FGFR3, PIK3CA, and RAS, and inactivating mutations of CDKN2A, as well as amplifications of FGFR3 and RAF1, and TP53/MDM2 status is also displayed. Within each HCA group, samples are ordered according to their relative FGFR3 expression. Dashed vertical lines, which define subsets within each HCA group, are drawn with respect to the E2F3, RB1, 5p, RAF1, and RAS alterations pattern and FGFR3 expression. Amplifications and homozygous deletions are indicated in black. Gains and deletions are indicated in gray. Activating/inactivating mutations are indicated in black. B) Heatmap representing relative expression levels of selected genes and the Bild E2F3 signature . Green, low expression; red high expression. C) Frequencies of selected genomic alterations in gene expression subgroups.

Figure 5. Validation of gene expression data by IHC on tissue microarray.

A) Barplots summarizing tumor cell protein scores of selected proteins in tumors stratified according to the gene expression subtypes. Error bars represent ±SEM. B) IHC stainings of two representative HC1 (top) and HC5 samples (bottom).

Figure 6. Genomic networks and survival analysis of genomic subtypes.

A) MDS plot based on the subset of genomic alterations (Figure 4A) and categorized gene expression data (Figure S3B) that showed at least one instance of significant positive or negative association in a pair-wise hypergeometric tests. Green lines, significant positive association; red lines, significant negative association. B) Kaplan-Meier analysis of tumors grouped according to a combination of gene expression and genomic alteration patterns using disease specific survival (DSS) as endpoint.