Transcriptional landscape of oncogene-induced senescence: a machine learning-based meta-analytic approach

Oncogene-induced senescence (OIS) is highly heterogeneous, varying by oncogenic signals and cellular context. While its dual role, in the initial inhibition potentially later leading to promotion of tumors through the senescence-associated secretory phenotype, is still a matter of debate, it is undeniable that OIS is critical to understanding tumorigenesis. A major obstacle to OIS research is the absence of a universally accepted marker. Here, we present a robust OIS-specific transcriptomic secretory phenotype, termed oncogene-induced senescence secretory phenotype (OIS-SP), which can identify OIS across multiple biological contexts from in vitro datasets to in vivo human samples. We apply a meta-analytic machine learning pipeline to harmonize a deliberately varied selection of Ras-Raf-MEK-induced senescence datasets of differing origins, oncogenic signals and cell types. Finally we make use of bypass data to identify key genes and eliminate genes associated with quiescence, so identifying 40 OIS-SP genes. Within this set, we determined a robust core of five OIS-SP genes (FBLN1, CXCL12, EREG, CST1 and MMP10). Importantly, these 5 OIS-SP genes showed clear, consistent regulation patterns across various human Ras-Raf-MEK-mutated tumor tissues, which suggests that OIS-SP may be a novel cancer driver phenotype with an unexpectedly critical role in tumorigenesis.