AUTHOR=Wang Xuezhu , Dong Yucheng , Wu Zilong , Wang Guanqun , Shi Yue , Zheng Yongchang TITLE=Machine Learning-Based Comparative Analysis of Pan-Cancer and Pan-Normal Tissues Identifies Pan-Cancer Tissue-Enriched circRNAs Related to Cancer Mutations as Potential Exosomal Biomarkers JOURNAL=Frontiers in Oncology VOLUME=11 YEAR=2021 URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2021.703461 DOI=10.3389/fonc.2021.703461 ISSN=2234-943X ABSTRACT=

A growing body of evidence has shown that circular RNA (circRNA) is a promising exosomal cancer biomarker candidate. However, global circRNA alterations in cancer and the underlying mechanism, essential for identification of ideal circRNA cancer biomarkers, remain under investigation. We comparatively analyzed the circRNA landscape in pan-cancer and pan-normal tissues. Using co-expression and LASSO regularization analyses, as well as a support vector machine, we analyzed 265 pan-cancer and 319 pan-normal tissues in order to identify the circRNAs with the highest ability to distinguish between pan-cancer and pan-normal tissues. We further studied their expression in plasma exosomes from patients with cancer and their relation with cancer mutations and tumor microenvironment landscape. We discovered that circRNA expression was globally reduced in pan-cancer tissues and plasma exosomes from cancer patients than in pan-normal tissues and plasma exosomes from healthy controls. We identified dynein axonemal heavy chain 14 (DNAH14), the top back-spliced gene exclusive to pan-cancer tissues, as the host gene of three pan-cancer tissue-enriched circRNAs. Among these three circRNAs, chr1_224952669_224968874_+ was significantly elevated in plasma exosomes from hepatocellular carcinoma and colorectal cancer patients. It was also related to the cancer mutation chr1:224952669: G>A, a splice acceptor variant, and was increasingly transcription-driven in cancer tissues. Moreover, pan-cancer tissue-enriched and pan-normal tissue-enriched circRNAs were associated with distinct tumor microenvironment patterns. Our machine learning-based analysis provides insights into the aberrant landscape and biogenesis of circRNAs in cancer and highlights cancer mutation-related and DNAH14-derived circRNA, chr1_224952669_224968874_+, as a potential cancer biomarker.