Discovery and validation of tissue-specific DNA methylation as noninvasive diagnostic markers for colorectal cancer

BACKGROUND: Noninvasive diagnostic markers that are capable of distinguishing patients with colorectal cancer (CRC) from healthy individuals or patients with other cancer types are lacking. We report the discovery and validation of a panel of methylation-based markers that specifically detect CRC. METHODS: This was a large-scale discovery study based on publicly available datasets coupled with a validation study where multiple types of specimens from six cohorts with CRC, other cancer types, and healthy individuals were used to identify and validate the tissue-specific methylation patterns of CRC and assess their diagnostic performance. RESULTS: In the discovery and validation cohort (N = 9307), ten hypermethylated CpG sites located in three genes, C20orf194, LIFR, and ZNF304, were identified as CRC-specific markers. Different analyses have suggested that these CpG sites are CRC-specific hypermethylated and play a role in transcriptional silencing of corresponding genes. A random forest model based on ten markers achieved high accuracy rates between 85.7 and 94.3% and AUCs between 0.941 and 0.970 in predicting CRC in three independent datasets and a low misclassification rate in ten other cancer types. In the in-house validation cohort (N = 354), these markers achieved consistent discriminative capabilities. In the cfDNA pilot cohort (N = 14), hypermethylation of these markers was observed in cfDNA samples from CRC patients. In the cfDNA validation cohort (N = 155), the two-gene panel yielded a sensitivity of 69.5%, specificity of 91.7%, and AUC of 0.806. CONCLUSIONS: Hypermethylation of the ten CpG sites is a CRC-specific alteration in tissue and has the potential use as a noninvasive cfDNA marker to diagnose CRC. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s13148-022-01312-9.