Integrated analysis of recurrent properties of cancer genes to identify novel drivers

Abstract

The heterogeneity of cancer genomes in terms of acquired mutations complicates the identification of genes whose modification may exert a driver role in tumorigenesis. In this study, we present a novel method that integrates expression profiles, mutation effects, and systemic properties of mutated genes to identify novel cancer drivers. We applied our method to ovarian cancer samples and were able to identify putative drivers in the majority of carcinomas without mutations in known cancer genes, thus suggesting that it can be used as a complementary approach to find rare driver mutations that cannot be detected using frequency-based approaches.

Figure 1

Mutation occurrence and correlation with gene length of known, candidate and rest of mutated genes. Occurrence of mutated genes in 20 cancer types (A) and 3,052 samples (B). None of the 10,681 genes is mutated in all 20 cancer types or samples; TP53 is the only gene to be mutated in 19 cancer types, while >40% of genes are mutated only in one cancer type. (C) Dependence of the recurrence of mutations on the gene length. Plotted is the length distribution of the coding portion for all genes that were found mutated in one to 20 cancer types. The interpolation line and R² were calculated using the LM function in R.

Figure 2

Expression of known, candidate and rest of mutated genes in cancer and normal tissues. (A) Breadth of expression of mutated genes in healthy tissues. Since the data were not normally distributed (P value 10^-42, Shapiro-Wilk test, Additional file 1, Figure S1), distributions were compared using the Wilcoxon test. (B) Fraction of housekeeping and tissue-specific mutated genes. Housekeeping genes were defined as genes expressed in 107/109 tissues (98%). Tissue specific genes were defined as genes expressed in 27/109 tissues (<25%). Fisher's exact test with one degree of freedom was used to determine statistical significance. (C) Volcano plot showing the log2ratios between the fractions of expressed genes in each group of mutated genes and in non-mutated genes. For each log2ratio, the corresponding P value from the chi-squared test with one degree of freedom is also shown. None of the three studies used for this analysis [36,43,44] identified candidate cancer genes, thus only the expression of known cancer genes and other mutated genes could be checked. (D) Volcano plot showing the log2ratios between the fractions of mutated genes (known cancer genes, candidates and other mutated genes) and non-mutated genes that are expressed in the normal counterparts of the 20 tumour types. The P value from the chi-squared test, one degree of freedom for each log2ratio is also shown. For assignment of normal tissues to tumour types see Additional file 2, Table S3. (E) Volcano plot showing the log2ratios of the faction of highly expressed mutated genes compared with the rest of highly expressed human genes. Highly expressed genes were identified as those genes with expression higher than the median expression for that tissue (see Methods).

Figure 3

Identification of novel driver genes in ovarian carcinoma. (A) Pipeline to identify putative driver genes on the basis of patient and gene properties. Starting from all tumour samples with mutation and expression data, the first filters removes samples with mutations in known cancer genes and with mutated genes that are not expressed. Then, only short genes with damaging mutations are retained. Finally, genes with properties that resemble those of known cancer genes are identified as putative drivers. (B) Volcano plot for the expression of mutated genes in ovarian carcinomas. Of the 7,048 total mutated genes, only 4,723 had expression data. Of those, 223 were known cancer genes of the Cancer Gene Census [58]; 36 were previously defined as candidate cancer genes in ovarian cancer [11,62]; all remaining 4,464 mutated genes had no putative involvement in cancer. (C) Identification of novel drivers in ovarian carcinomas. Following our pipeline, we identified 56 genes that may favour cancer development in 23 ovarian cancer patients.

Figure 4

Properties of putative drivers in ovarian cancer. (A) Gene silencing effects of 395 known cancer genes with available shRNA data in 102 cancer cell lines. The distributions of log2ratios of the shRNA concentrations in the final cell population and the initial DNA pool (log2ratio_shRNA, see Methods) were compared between known cancer genes, oncogenes, tumour suppressors and the non-mutated genes using Wilcoxon test. Complete data are reported in Additional file 2, Table S8. (B) Gene silencing effects of the 40 putative drivers identified with our pipeline, seven tumour suppressors and eight oncogenes with available shRNA data in 25 ovarian cancer cell lines. The list of known tumour suppressors and oncogenes associated with ovarian cancer was derived from the Cancer Gene Census [58]. Complete data are reported in Additional file 2, Table S9. (C) Confirming evidence of the effect of RNAi on three putative drivers. The block of RB1CC1 and KDM5B via RNAi leads to RB1 repression, with a consequent loss of the ability of RB1 to promote cell differentiation [92] and senescence [93], respectively. Interestingly, the Rb pathway is a known key player in ovarian cancer [62]. Similarly, anti-PRKCQ siRNAs inactivate CASP8. As a consequence, the CASP8/BCL10/MALT1 complex cannot be formed, thus preventing the cells to enter apoptosis [94]. (D) Effect of putative drivers on cell proliferation and survival. Reported are the links with pathways involved in gene proliferation of 19 out of 56 putative drivers mutated in 13 out of 23 tumour samples. The sample ID where the gene is mutated is provided together with the number of ovarian cancer cell lines over the total that displayed increased proliferation upon gene silencing, when available.