DNA methylation profiling to determine the primary sites of metastatic cancers using formalin-fixed paraffin-embedded tissues [TrainingSet]

Accurate identification of the primary site of metastatic cancer is critical to guide the subsequent treatments. There is a significant portion of patients whose primary sites are initially classified as uncertain, and 3-9% of cancer patients are diagnosed with cancer of unknown primary (CUP) even after comprehensive diagnostic workups. Yet, a widely accepted molecular test is still not available. Here, we presented the combination of a novel DNA methylation sequencing-based method and an algorithm to predict the tissues of origin for metastatic cancers. The assay applied degraded DNA from formalin-fixed, paraffin-embedded (FFPE) tissues to generate reduced represent bisulfite sequencing libraries (FFPE-RRBS). Comparable DNA methylation metrics were obtained for the paired fresh frozen (FF) RRBS and FFPE-RRBS libraries and the FFPE-RRBS libraries of matched primary and metastatic cancer tissues. We generated and systemically evaluated 28 molecular classifiers built on four methylation evaluation methods and seven machine-learning approaches from a training data set of 498 primary cancer patients. Of those classifiers, the beta values-based (mean methylation) linear support vector (BELIVE) performed the best, achieving overall accuracies of 81-95% for identifying the primary sites of 215 metastatic cancer patients by utilizing the top-k predictions (k=1, 2, 3). The prediction accuracies ranged from 92% to 98% for 4702 patients with primary tumors in a cross-validation cohort. Lastly, BELIVE successfully identified the tissues of origin for approximately 81-93% of cases in a cohort of 68 patients initially diagnosed with CUP. We sequenced RRBS libraries from 503 samples of primary tumor, 218 samples of metastatic tumor, and 68 samples of CUP patients.The assay was conducted using genomic DNA purified either from 10-20 mg of FF tissues or five to eight 5-10 µm FFPE tissue sections. Board-certified pathologists checked one H&E stained slide to ensure that tumor cells accounted for 10% or more cell population and that the necrosis area was less than 50%. NOTE FROM SUBMITTER: Raw data (FASTQ files) are controlled to protect patient privacy.