Estimation of the Fraction of Cancer Cells in a Tumor DNA Sample Using DNA Methylation

Contamination of normal cells is almost always present in tumor samples and affects their molecular analyses. DNA methylation, a stable epigenetic modification, is cell type-dependent, and different between cancer and normal cells. Here, we aimed to demonstrate that DNA methylation can be used to estimate the fraction of cancer cells in a tumor DNA sample, using esophageal squamous cell carcinoma (ESCC) as an example. First, by an Infinium HumanMethylation450 BeadChip array, we isolated three genomic regions (TFAP2B, ARHGEF4, and RAPGEFL1) i) highly methylated in four ESCC cell lines, ii) hardly methylated in a pooled sample of non-cancerous mucosae, a pooled sample of normal esophageal mucosae, and peripheral leukocytes, and iii) frequently methylated in 28 ESCCs (TFAP2B, 24/28; ARHGEF4, 20/28; and RAPGEFL1, 19/28). Second, using eight pairs of cancer and non-cancer cell samples prepared by laser capture microdissection, we confirmed that at least one of the three regions was almost completely methylated in ESCC cells, and all the three regions were almost completely unmethylated in non-cancer cells. We also confirmed that DNA copy number alterations of the three regions in 15 ESCC samples were rare, and did not affect the estimation of the fraction of cancer cells. Then, the fraction of cancer cells in a tumor DNA sample was defined as the highest methylation level of the three regions, and we confirmed a high correlation between the fraction assessed by the DNA methylation fraction marker and the fraction assessed by a pathologist (r=0.85; p