Discovery of therapeutic targets in cancer using chromatin accessibility and transcriptomic data

Abstract

Most cancer types lack targeted therapeutic options, and when first-line targeted therapies are available, treatment resistance is a huge challenge. Recent technological advances enable the use of assay for transposase-accessible chromatin with sequencing (ATAC-seq) and RNA sequencing (RNA-seq) on patient tissue in a high-throughput manner. Here, we present a computational approach that leverages these datasets to identify drug targets based on tumor lineage. We constructed gene regulatory networks for 371 patients of 22 cancer types using machine learning approaches trained with three-dimensional genomic data for enhancer-to-promoter contacts. Next, we identified the key transcription factors (TFs) in these networks, which are used to find therapeutic vulnerabilities, by direct targeting of either TFs or the proteins that they interact with. We validated four candidates identified for neuroendocrine, liver, and renal cancers, which have a dismal prognosis with current therapeutic options.

Keywords: ATAC-seq; cancer; drug repurposing; functional genomics; networks; regulatory networks; therapy.

Figure 1.

Construction, clustering, and identification of key regulatory TFs from patient regulatory networks. A) Construction of edges between TFs and regulatory elements and via those regulatory elements, target genes. “O” is the Outdegree of a TF (the number of genes it regulates) “B” is the betweenness centrality of a TF (the frequency of a TF falling on the shortest paths between other TFs and their regulatory targets to identify bottlenecks) B) Schematic of TF-Score for TFs across patients. Units are within-sample ranks of each component (Methods) C) t-SNE of top 15 PCs of TF-scores in 371 patients with 22 identified clusters and their cancer types. Clusters in figure are labeled with the Top 3 TFs in that cluster by relative TF-Score.

Figure 2:

Features of cancer subtypes using TF-Score. A) Expression of neuroendocrine marker genes across 22 clusters identified using TF-Score. Expression of each gene in log2TPM converted to z-scores across all 371 samples. Mean expression of all marker genes plotted on y-axis for each sample. B) Expression of known marker genes for WNT activation and MSI-H cancer phenotypes in GI cancer patients belonging to Cluster 18 (COAD only) and Cluster 4 (mixed GI) samples. WNT activation is commonly seen in COAD cancer.

Figure 3:

A) Assignment of cell lines to representative clusters using a nearest template prediction based approach.) B) Detailed assignments of 335 cell lines present in both DepMap and CCLE datasets to TF-Score clusters by cell line cancer type. Heatmap is normalized by total number of cell lines in that cancer type. C) Concordance of NTP approach to assigned cell lines to patient clusters by cancer type. Detailed assignments of cancer types with less than 50% concordance showing likely phenotypic convergence of LUAD cell lines to squamous-like and neuroendocrine cancers. Uterine cancers also show noteworthy similarity to renal cancers likely arising from their shared lineage as mesoderm-derived cell types.

Figure 4:

CaRNetS workflow. A) Flowchart summarizing CaRNetS approach B) CaRNetS scoring schema showing the features used in druggable gene prioritization. C) t-SNE of top 15 principal components of TF-scores in 371 patients with 22 identified clusters; top candidates for each cluster indicated with known candidates highlighted. D) Known candidate genes in 9 clusters and associated CaRNetS scores.

Figure 5:

Effect of inhibitors of 4 candidate genes across 3 clusters. P-values for comparison of treatment with control obtained using a two-way ANOVA test. Error bars indicate standard deviation of viability across 3 biological replicates of 3 technical replicates each. A) Cell viability assay for Mercaptopurine in DMS53 cells. B) RG7112 in A498 cells. C) Teglarinad Chloride in DMS53 cells. D) TP-472 in HEPG2 cells.