Identification of DNA methylation signatures in follicular-patterned thyroid tumors

Follicular-patterned thyroid tumors (FPTTs) are frequently encountered in thyroid pathology, encompassing follicular adenoma (FA), follicular thyroid carcinoma (FTC), noninvasive follicular thyroid neoplasm with papillary-like nuclear features (NIFTP), and follicular variant of papillary thyroid carcinoma (fvPTC). Recently, a distinct entity termed differentiated high-grade thyroid carcinoma has been described by the 5th edition of the WHO classification of the thyroid tumors, categorized as either high-grade fvPTC, high-grade FTC or high-grade oncocytic carcinoma of the thyroid (OCA). Accurate differentiation among these lesions, particular between the benign (FA), borderline (NIFTP) and malignant neoplasms (FTC and fvPTC), remains a challenge in both histopathological and cytological diagnoses. This study aimed to develop a novel molecular diagnostic approach utilizing DNA methylation to distinguish between these thyroid tumors.

DNA methylation signatures and machine learning were employed to construct classification models for FPTTs. A total of 178 thyroid samples from the Gene Expression Omnibus were analyzed. The models were validated using two independent cohorts.

13 cytosine-guanine dinucleotides (CpGs) exhibited significant differences in methylation levels among FA, FTC, NIFTP and fvPTC. Notably, NIFTP showed hypomethylation compared to other subtypes. A Random Forest classifier, based on the methylation status of these 13 CpGs, effectively categorized the four tumor subtypes (AUC = 0.86, accuracy = 0.70 for internal data, and AUC approximately 0.80 for validation data). The selected CpGs were significantly associated with the tumor progression pathway.

This study established a robust method for categorizing FPTTs based on DNA methylation patterns. The identified DNA methylation approach holds clinical promise for efficiently diagnosing thyroid neoplasms.