Abstract 6194: DNA methylation-based immune response signature for thyroid nodule diagnostics

Background: Thyroid cancer (TC) account for 90% of endocrine malignant tumor. The association between thyroid inflammation and carcinogenesis has long been recognized. However, the differential immunological mechanisms underlying pathogenesis of papillary thyroid cancers (PTCs) and benign thyroid nodules (BTNs) have not been well characterized. In this study, we aimed to explore the immunological differences between PTCs and BTNs. Further, we attempted to develop a novel classifier based on DNA methylation of immune response genes for thyroid cancer precise diagnosis. Methods: A publicly available tissue Reduced Representative Bisulfite Sequencing dataset composed of 80 PTCs and 65 BTNs was collected (GSE107738), including a training cohort (39 PTCs, 28 BTNS) and a testing cohort (41 PTCs, 37 BTNs). We identified differentially methylated regions (DMRs) between PTCs and BTNs. Functional enrichment analyses of DMR-associated genes were conducted by hypergeometric test. DNA methylation markers of immune response genes were extracted. A diagnostic classifier based on these markers was developed using a training cohort and was validated in an independent testing cohort. These markers were further confirmed by two independent datasets (GSE53051: 20 PTCs, 32 BTNs; GSE97466: 60 PTCs, 17 BTNs) generated by the Infinium Methylation 450K array. A new classifier based on the corresponding 15 array-based methylation markers was built using these two datasets, and was validated in the 450K array data of 499 thyroid malignant tissue samples in the TCGA database. Results: We identified 220 DMRs between PTCs and BTNs. DMR-associated genes were significantly enriched in immune response biological processes. The classifier comprised of 15 DNA methylation markers from immune response genes achieved 100% sensitivity, 76% specificity, 82% positive predictive value (PPV), 100% negative predictive value (NPV) and 88% accuracy in the testing cohort. This classifier also demonstrated high accuracy for cytologically indeterminate thyroid nodules (25 PTCs, 29 BTNs). Leave-one-out cross validation using the corresponding array methylation markers in two independent datasets achieved prediction accuracy of 92% and 95%, respectively. An array-based classifier trained on all samples of the methylation array data achieved a sensitivity of 89% in classifying 499 thyroid malignant tissue samples in the TCGA database. Conclusion: Our study demonstrated that DNA methylation signature of immu