Identification of mRNA isoform switching in breast cancer

Abstract

Background: Alternative splicing provides a major mechanism to generate protein diversity. Increasing evidence suggests a link of dysregulation of splicing associated with cancer. While previous genomic-based studies demonstrated the expression of a handful of tumor-specific isoforms, genome-wide alterations in the balance between isoforms and cancer subtypes is understudied.

Result: We systematically analyzed the isoform-level expression patterns and isoform switching events of 819 breast tumor and normal samples assayed by mRNA-seq from TCGA project. On average, 2.2 isoforms per gene were detected and 67.5 % of detected genes (i.e. expressed) showed 1-2 isoforms only. While the majority of isoforms for a given gene were positively correlated with each other and the overall gene level, 470 pairs of isoforms displayed an inverse correlation suggesting a switching event. Most of the isoform switching events were associated with molecular subtypes, including a Basal-like-associated switching in CTNND1. 88 genes showed switching independent of subtypes, among which the isoform pattern of PRICKLE1 was associated with a large genomic signature of biological significance.

Conclusion: Our results reveal that the majority of genes do not undergo complex mRNA splicing within breast cancers, and that there is a general concordance in isoform and gene expression levels in breast tumors. We identified hundreds of isoform switching events across breast tumors, most of which were associated with differences in tumor subtypes. As exemplified by the detailed analysis of CTNND1 and PRICKLE1, these isoform switching events potentially provide new insights into the post-transcriptional regulatory mechanisms of tumor subtypes and cancer biology.

Fig. 1

General isoform expression characteristics across 819 breast samples. a Proportion of detected isoforms using distinct cutoff to call isoform detection in 819 breast samples (blue lines) and on average (red line). b Distribution of isoforms expression across samples using different cutoff for isoform detection. c Distribution of the number of detected isoforms for each gene. d Pearson correlation of the abundance of detected isoform pairs of the same gene

Fig. 2

Hierarchical clustering analysis of the top 6000 most variably expressed transcripts. a Overview of the clustering diagram of 728 breast tumors and 91 normal samples. Gene signatures of (b) basal, (c) luminal and (d) Her2 subtype were identified, with multiple isoforms from the same biomarker genes co-clustered

Fig. 3

Isoform expression and gene structure of CTNND1. a Gene and isoform expression of CTNND1 in the TCGA BRCA dataset. The total gene expression of CTNND1 provides comparable level in luminal and basal tumors, whereas the abundance of two CTNND1 isoforms, uc001nlo.3 and uc001nlt.3, exhibit inversely correlated differential expression in luminal and basal tumors. b Isoform expression of CTNND1 in the TCGA PanCan12 data set. The expression pattern of uc001nlo.3 and uc001nlt.3 pairs is similarly observed in the PanCan12 Luminal BRCA and Basal BRCA tumors, as well LUAD and Squamous clusters. c CTNND1 gene structure. The two isoforms are transcribed from the same transcription start site and provide alternative splicing in exon 20 (red arrow)

Fig. 4

Isoform expression and gene structure of PRICKLE1. a Expression pattern of PRICKLE1 isoforms uc010skw.1 and uc001rnl.2. Two expression patterns (red and black) were identified by K-means clustering in subsets of breast samples. b PRICKLE1 gene structure.uc001skw.1 and uc001rnl.2 contain alternative 3’ splice site in exon 7 (red). c 5’ UTR structure of PRICKLE1 and putative RNA binding proteins. Alternative 3’ splice site (red) and hairpin structure at the flanking region (blue) are enriched with putative binding sites of RNA binding proteins

Fig. 5

Hierarchical clustering analysis of the gene signatures associated with the isoform expression pattern of PRICKLE1. 1059 genes are significantly correlated with the ratio of uc010skw.1 and uc001rnl.2, and are not associated with PAM50 subtype or RIN score. PRICKLE1-based clusters identified by K-means clustering were color coded consistent with Fig. 3a. RIN group displays samples with high (>8.7, median RIN) and low RIN score (≤8.7). Four gene expression signatures (a-d, shown by side color bars) are identified to be involved in cancer hallmarks