Integrated Genomic Analysis of the Ubiquitin Pathway across Cancer Types

Abstract

Protein ubiquitination is a dynamic and reversible process of adding single ubiquitin molecules or various ubiquitin chains to target proteins. Here, using multidimensional omic data of 9,125 tumor samples across 33 cancer types from The Cancer Genome Atlas, we perform comprehensive molecular characterization of 929 ubiquitin-related genes and 95 deubiquitinase genes. Among them, we systematically identify top somatic driver candidates, including mutated FBXW7 with cancer-type-specific patterns and amplified MDM2 showing a mutually exclusive pattern with BRAF mutations. Ubiquitin pathway genes tend to be upregulated in cancer mediated by diverse mechanisms. By integrating pan-cancer multiomic data, we identify a group of tumor samples that exhibit worse prognosis. These samples are consistently associated with the upregulation of cell-cycle and DNA repair pathways, characterized by mutated TP53, MYC/TERT amplification, and APC/PTEN deletion. Our analysis highlights the importance of the ubiquitin pathway in cancer development and lays a foundation for developing relevant therapeutic strategies.

Keywords: FBXW7; The Cancer Genome Atlas; biomarker; cancer prognosis; pan-cancer analysis; therapeutic targets; tumor subtype; ubiquitin pathway.

Figure 1. Frequently Mutated UBQ and DUB Genes as Potential Cancer Drivers

(A) UBQ and DUB genes are plotted as fractions of hotspot versus LoF mutations among all non-silent mutations across cancer types. Genes enriched with hotspot mutations are shown in red, genes enriched with LoF mutations are in blue, and FBXW7 is shown in orange, because it is enriched with both hotspot and LoF mutations. The circles represent UBQs, and the squares represent DUBs. (B) Significantly mutated genes identified by MutSigCV in each cancer type are shown. The circles represent UBQs, and the squares represent DUBs; the circle or square size is proportional to the significance level. The fraction of patients harboring non-silent mutations in each gene is shown by color scale. (C) UBQ and DUB genes enriched with hotspot and LoF mutations are mapped to different gene categories in the ubiquitin pathway. See also Figure S2.

Figure 2. FBXW7 Is Enriched with Both Hotspot and Loss-of-Function Mutations

(A) Fractions of hotspot mutations versus LoF mutations among all non-silent mutations in FBXW7 are plotted for different cancer types. Cancer types enriched with hotspot mutations are shown in red, those enriched with LoF mutations are in blue, and those enriched with both hotspot and LoF mutations are in gray. (B) WD40 domain structure of FBXW7 protein in which three arginines (R465, R479, and R505) are mutation hotspots and located at the substrate binding surface. (C) Distributions of FBXW7 non-silent mutations in cancer types enriched with hotspot mutations (UCEC and UCS) and cancer types enriched with LoF mutations (ESCA, LUAD, LUSC, READ, SKCM, and STAD). (D) FBXW7 mutations show mutually exclusive patterns with PIK3CA mutations in BLCA, CESC, and LUSC. (E) Compared to tumors without mutations in FBXW7 or PI3K pathway genes, tumors with either FBXW7 or PI3K pathway mutations show elevated PI3K-Akt pathway activity, with *, p < 0.05. The bottom and top of the box are the first and third quartiles, and the whiskers extend to 1.5 IQR of the lower quartile and the upper quartile, respectively.

Figure 3. Somatic Copy-Number Alterations of UBQ and DUB Genes

(A) Fractions of UBQ and DUB genes residing in the amplification or deletion peaks (identified by GISTIC2, q < 0.25) compared to non-UBQ/DUB genes in different cancer types. Significant deletion enrichments are detected with *p < 0.01. (B) Most frequently amplified or deleted UBQ and DUB genes in multiple cancer types. The circle size is proportional to the significance level of GISTIC2 results. (C) MDM2 amplification shows a mutually exclusive pattern with BRAF mutations in SKCM. TP53 mutations are shown for comparison. Each bar represents one patient; significance was assessed by Fisher’s exact test. (D) TP53 protein and mRNA expression of tumor samples with MDM2 amplification versus those with BRAF mutations or wild-type (WT) samples, with *p < 0.05. The bottom and top of the box are the first and third quartiles, and the whiskers extend to 1.5 IQR of the lower quartile and the upper quartile, respectively. (E) Graphical model showing the synergistic effect of MDM2 inhibitor and BRAF inhibitor. See also Figure S3.

Figure 4. Multiple Mechanisms Contribute to Upregulation of UBQ and DUB Genes in Cancer

(A) UBQ and DUB genes showed upregulation in tumor samples in seven cancer types (GSEA, q < 0.1). (B) Proportions of upregulated, neutral, and downregulated UBQ/DUB genes in the seven cancer types (Wilcoxon signed rank test, q < 0.1). (C) Top: proportions of copy-number amplification, neutral level, and deletions in upregulated and neutral UBQ/DUB gene groups in each cancer type. Middle: proportions of significantly decreased (paired t test, p < 0.05), decreased, and other expression of miRNA regulators in tumor samples relative to matched normal samples in upregulated and neutral UBQ/DUB gene groups. Bottom: proportions of significantly decreased (paired t test, p < 0.05), decreased, and otherwise DNA methylation level in tumor samples relative to matched normal samples in upregulated and neutral UBQ/DUB gene groups. The asterisks indicate the significant proportion difference between the two groups (chi-square test, *q < 0.01). Right: Venn diagram showing the proportions of upregulated UBQ/DUB genes affected by different regulatory mechanisms. See also Figures S4 and S5.

Figure 5. Integrative Genomic Clustering and Patient Survival Analysis

(A–C) Heatmaps of consensus clustering for three platforms: RNA sequencing (RNA-seq)-based mRNA expression (A), somatic copy-number alterations (B), and DNA methylation (C). (D) Consensus matrix of integrative clustering showing three robust clusters (COCA1, COCA2, and COCA3). (E) COCA clusters correlate with patient overall survival and disease-specific survival times in 10 and 9 cancer types, respectively. (F) Kaplan-Meier plots of nine cancer types showing overall survival curves for three clusters of patients with log-rank p values. See also Figure S6.

Figure 6. Biological Pathways and Somatic Drivers Associated with the Poor Prognostic Tumor Subtypes (COCA2)

(A) Association of COCA2 with GSEA hallmark gene sets. Significant positive associations are shown in red, significant negative associations are shown in blue, and non-significant ones are shown in gray. (B) Reverse-phase protein array (RPPA)-based pathway scores of cell-cycle and DNA repair between COCA2 samples (red box) and other samples (blue box). The bottom and top of the box are the first and third quartiles, and the whiskers extend to 1.5 IQR of the lower quartile and the upper quartile, respectively. (C) Significantly mutated genes identified by MutSigCV, in which the mutations are significantly enriched (red) or depleted (blue) in COCA2 compared to COCA1 and COCA3 in different cancer types (q < 0.1). (D) SCNA drivers identified by GISTIC2, in which amplifications (for oncogenes) and deletions (for tumor suppressors) are significantly enriched in COCA2 compared to COCA1 and COCA3 in different cancer types (q < 0.001).

Figure 7. Mechanistic Model Describing the Biological Process Underlying COCA2 Subtypes

Somatic drivers identified for COCA2 subtypes (top) cause the expression-level changes of key UBQ and DUB genes in SCF complex, APC/C complex, and DNA damage response that underlie the aberrant activities of cell-cycle and DNA damage pathways (middle), thereby leading to poor patient survival of COCA2 subtypes (bottom). See also Figure S7.