From multi-omics integration towards novel genomic interaction networks to identify key cancer cell line characteristics

Abstract

Cancer is a complex disease where cancer cells express epigenetic and transcriptomic mechanisms to promote tumor initiation, progression, and survival. To extract relevant features from the 2019 Cancer Cell Line Encyclopedia (CCLE), a multi-layer nonnegative matrix factorization approach is used. We used relevant feature genes and DNA promoter regions to construct genomic interaction network to study gene-gene and gene-DNA promoter methylation relationships. Here, we identified a set of gene transcripts and methylated DNA promoter regions for different clusters, including one homogeneous lymphoid neoplasms cluster. In this cluster, we found different methylated transcription factors that affect transcriptional activation of EGFR and downstream interactions. Furthermore, the hippo-signaling pathway might not function properly because of DNA hypermethylation and low gene expression of both LATS2 and YAP1. Finally, we could identify a potential dysregulation of the CD28-CD86-CTLA4 axis. Characterizing the interaction of the epigenome and the transcriptome is vital for our understanding of cancer cell line behavior, not only for deepening insights into cancer-related processes but also for future disease treatment and drug development. Here we have identified potential candidates that characterize cancer cell lines, which give insight into the development and progression of cancers.

Figure 1

(A) Silhouette score for cluster sizes 6 to 11. (B) Number of samples in each cluster. (C) Consensus plot for cluster size 8.

Figure 2

(A) The histology of each cluster member as defined by the 2019 CCLE metadata. Here, it becomes apparent that there are a number of clusters with mixed cancer types, but more importantly, there are clusters that show strong homogeneity. (B) Histogram for the number of feature genes and DNA promoter regions with Kim score and a more stringent feature scoring.

Figure 3

Distribution of the cancer tissues in cluster 5 and cluster 8. All cancer tissues in cluster 5 are also present in cluster 8 but cluster 8 also contains some cancer tissues that are not a member of cluster 5.

Figure 4

(A) Median Log2(TPM + 1) values of genes in cluster 5 and cluster 8. This shows the main drivers behind the stratification of cluster 5 and cluster 8. (B) DNA promoter region of cluster 5 and cluster 8. Although cluster 5 and cluster 8 both contain the same cancer tissues, a different methylation pattern is observed for certain regions.

Figure 5

Genomic interaction network modules for cluster 7 (lymphoid neoplasm). (A) Subnetwork for the genes connected to EGFR. (B) Subnetwork for the region of genes connected to YAP1, TEAD4, JAG1, and SMAD1. (C) Subnetwork for the set of genes connected with CD28, CD86, and CTLA4. CTLA4 is a seeding node which is highlighted by the light color and the dotied interactions. Gene–gene interactions are shown in orange, DNA promoter region–gene interactions in cyan, protein–protein interactions in purple, and transcription interactions in green (activation) or red (inhibition).