Computational identification of preneoplastic cells displaying high stemness and risk of cancer progression

Evidence points towards the differentiate state of cells being a marker of cancer risk and progression, in line with the cancer-stem-cell hypothesis. Measuring the differentiation state of single cells in a preneoplastic population could thus enable novel strategies for early detection and risk prediction. Here we present a novel computational method called CancerStemID that estimates a stemness index of cells from single-cell RNA-Seq data. We validate CancerStemID in two human esophageal squamous cell carcinoma (ESCC) cohorts, demonstrating how it can identify undifferentiated preneoplastic cells whose transcriptomic state is overrepresented in invasive cancer. We demonstrate decreased differentiation activity of tissue-specific TFs in cancer cells compared to the basal cell-of-origin layer, and that the differentiation state correlates with differential DNA methylation at the promoters of such TFs independently of underlying NOTCH1 and TP53 mutations. In summary, these data support an epigenetic stem-cell model of oncogenesis and highlight a novel computational strategy in which to identify stem-like preneoplastic cells that undergo positive selection. We obtained 47 samples derived from 14 treatment-naïve esophageal squamous cell carcinoma (ESCC) patients, including 18 non-tumor tissues (≥5 cm from tumor), 15 dysplasia (≤2 cm to tumor margin), and 14 tumor. ****Raw sequencing data has been deposited to China Genomic Sequence Archive (GSA) under accession number HRA000776 **** NOTE FROM SUBMITTER: The raw data in fastq format has been deposited to China Genomic Sequence Archive (GSA) under accession number HRA000776. Due to patient privacy concerns, we can only publish the expression matrix of epithelial cells from de-identified patients.