Cancer driver topologically associated domains identify oncogenic and tumor-suppressive lncRNAs

Abstract

Cancer long noncoding RNAs (lncRNAs) have been identified by experimental and in silico methods. However, current approaches for identifying cancer lncRNAs are not sufficient and effective. To uncover them, we focus on the core cancer driver lncRNAs, which directly interact with cancer driver protein-coding genes (PCGs). We investigate various aspects of cancer lncRNAs, including their expression patterns, genomic locations, and direct interactions with cancer driver PCGs, and developed a pipeline to identify candidate cancer driver lncRNAs. Finally, we validate the reliability of potential cancer driver lncRNAs through functional analysis of bioinformatics data and CRISPR-Cas9 knockout experiments. We find that cancer lncRNAs are more concentrated in cancer driver topologically associated domains (CDTs), and CDT is an important feature in identifying cancer lncRNAs. Moreover, cancer lncRNAs show a high tendency to be coexpressed with and bind to cancer driver PCGs. Utilizing these distinctive characteristics, we develop a pipeline CAncer Driver Topologically Associated Domains (CADTAD) to identify candidate cancer driver lncRNAs in pan-cancer, including 256 oncogenic lncRNAs, 177 tumor-suppressive lncRNAs, and 75 dual-function lncRNAs, as well as in three individual cancer types, and validate their cancer-related functions. More importantly, the function of 10 putative cancer driver lncRNAs in prostate cancer is subsequently validated to influence cancer phenotype through cell studies. In light of these findings, our study offers a new perspective from the 3D genome to study the roles of lncRNAs in cancer. Furthermore, we provide a valuable set of potential lncRNAs that could deepen our understanding of the oncogenic mechanism of cancer driver lncRNAs.