Comprehensive Characterization of Alternative Polyadenylation in Human Cancer

Abstract

Background: Alternative polyadenylation (APA) is emerging as a major post-transcriptional mechanism for gene regulation, and dysregulation of APA contributes to several human diseases. However, the functional consequences of APA in human cancer are not fully understood. Particularly, there is no large-scale analysis in cancer cell lines.

Methods: We characterized the global APA profiles of 6398 patient samples across 17 cancer types from The Cancer Genome Atlas and 739 cancer cell lines from the Cancer Cell Line Encyclopedia. We built a linear regression model to explore the correlation between APA factors and APA events across different cancer types. We used Spearman correlation to assess the effects of APA events on drug sensitivity and the Wilcoxon rank-sum test or Cox proportional hazards model to identify clinically relevant APA events.

Results: We revealed a striking global 3'UTR shortening in cancer cell lines compared with tumor samples. Our analysis further suggested PABPN1 as the master regulator in regulating APA profile across different cancer types. Furthermore, we showed that APA events could affect drug sensitivity, especially of drugs targeting chromatin modifiers. Finally, we identified 1971 clinically relevant APA events, as well as alterations of APA in clinically actionable genes, suggesting that analysis of the complexity of APA profiles could have clinical utility.

Conclusions: Our study highlights important roles for APA in human cancer, including reshaping cellular pathways and regulating specific gene expression, exemplifying the complex interplay between APA and other biological processes and yielding new insights into the action mechanism of cancer drugs.

Figure 1.

Global landscape of alternative polyadenylation (APA) events across patient samples and cancer cell lines. A) Global landscape of 7029 APA events across different cancer types in tumor samples (blue), normal samples (red), and cancer cell lines (purple). In the heatmap, red denotes transcripts with Percentage of Distal polyA site Usage Index (PDUI) greater than 0.5, while blue denotes transcripts with PDUI of less than 0.5. The columns are arranged by cancer type, and rows are arranged by mean PDUI of each APA event across all samples. B) Number of shortening (blue) and lengthening (red) APA events in cancer cell lines compared with tumor samples across 12 cancer types. C) Comparison of percentage of transcripts with PDUI greater than 0.5 divided by total number of APA events between tumor samples (red) and cancer cell lines (purple). The P value was calculated using a two-sided Wilcoxon test. The boxes show the median±1 quartile, with whiskers extending to the most extreme data point within 1.5 interquartile range from the box boundaries. APA = alternative polyadenylation; CCL = cancer cell lines; PDUI = Percentage of Distal polyA site Usage Index.

Figure 2.

Key factors in regulation of alternative polyadenylation (APA) events. A) Percentage of APA events positively correlated with expression level of APA factors in tumor samples across cancer types. B) Global shortening of APA events upon PABPN1 knockdown in U2OS and C2C12 cell lines. C) A schematic model for the important role of PABPN1 in suppressing the usage of proximal APA site. D) Chromatin-related gene ontology (GO) terms enrichment analysis of top 10% of 20 532 genes correlated to APA events in tumor samples. Gene lists were provided in Supplementary Table 4 (available online). The circle color represents the enrichment score, while the circle size represents the statistically significant P value. Enriched GO terms were listed in Supplementary Table 5 (available online). APA = alternative polyadenylation; CPSF = cleavage and polyadenylation specificity factor; CSTF = cleavage stimulation factor.

Figure 3.

Therapeutic liability of alternative polyadenylation (APA) events in cancer cell lines. A) The statistically significant correlation of APA events with IC50 value of Cancer Cell Lines Encyclopedia drugs across cancer cell lines for three or more drugs. The circle color represents Spearman coefficient value, while the circle size represents the false discovery rate value. B) Top 40 drugs ranked based on the number of correlated APA events using Cancer Therapeutics Response Portal data: pan-histone deacetylase inhibitors, red. All statistical tests were two-sided. APA = alternative polyadenylation; HADc = pan-histone deacetylase.

Figure 4.

Clinical relevance of alternative polyadenylation (APA) events. A) Statistically significant differentially regulated and/or clinically relevant APA events identified in at least four cancer types. For each cancer type, the red and blue box indicates the statistically significant differential APA events between paired tumor and normal samples (red: Diff ≥ 0.2, false discovery rate [FDR] < 0.05; blue: Diff ≤ –0.2, FDR < 0.05), the lemon yellow box indicates the statistically significant differential APA events among tumor subtypes (|Diff| ≥ 0.2, FDR < 0.05), the purple box indicates the statistically significant differential APA events among tumor stages (|Diff| ≥ 0.2, FDR < 0.05), and the pink and light blue box indicates the lengthening APA events associated with worse (pink) or better (light blue) overall survival (|Diff| ≥ 0.2, FDR < 0.05). Complete clinically relevant APA events were listed in Supplementary Table 7 (available online). B) Schematic representation of gene last exon structure and predicted alternative polyadenylation site, miRNA binding site, and AU-rich elements (top). APA profile of CSNK1D in paired tumor and normal samples (left) and overall patient survival time for lengthening and shortening group (right). C) APA profile of NDE1 in paired tumor and normal samples (left), among BRCA tumor subtypes (middle), and STAD stages (right). The boxes show the median±1 quartile, with whiskers extending to the most extreme data point within 1.5 interquartile range from the box boundaries. APA = alternative polyadenylation; PDUI = Percentage of Distal polyA site Usage Index.

Figure 5.

Alterations of alternative polyadenylation (APA) events in clinically actionable genes in tumor samples. A) Distribution of interquartile range (IQR) of Percentage of Distal polyA site Usage Index (PDUI) of clinically actionable genes across different cancer types. B) Recurrent clinically actionable genes with IQRs of PDUI of 0.2 or greater in tumor samples. C) Number of APA events of clinically actionable genes correlated with expression level of APA core factors. D) Selected correlations (two-sided Spearman correlation test) between APA events of clinically actionable genes and PABPN1 expression level (left: CTNNB1; middle: PIK3R1; right: FGFR2). Complete correlations were listed in Supplementary Table 8 (available online). APA = alternative polyadenylation; IQR = interquartile range; PDUI = Percentage of Distal polyA site Usage Index.