Development of Novel mRNA Classifiers to Stratify Preoperative Thyroid Tumor Risk

Introduction: While molecular data can suggest the presence or absence of cancer for cytologically indeterminate thyroid nodules, current molecular markers to guide risk stratification in thyroid cancer are suboptimal. The Cancer Genome Atlas (TCGA) reported 5 mRNA gene expression clusters correlating with thyroid cancer progression free survival. This study’s aim was to refine these gene expression profiles using histopathologic data to create novel predictive gene signatures.

Methods: From 482 TCGA samples, the 5 mRNA clusters and 3 ATA pathology risk groups were merged to derive 3 new clusters: TCGA-low (L), intermediate (I) and high (H) risk. Differential gene expression analysis yielded an mRNA signature consisting of 145 genes that recapitulated the TCGA clusters. This signature was then applied to 2,205 molecularly tested Bethesda V/VI fine needle aspirates (FNAs) from the Afirma database and consensus clustering grouped the samples into 3 FNA mRNA clusters (FNA-L/I/H). These were then compared to the three TCGA clusters.

Results: TCGA-L (n=57) was enriched with BRAFV600E but only consisted of stage I/II disease, TCGA-I (n=272) was enriched in follicular cancers, and TCGA-H (n=120) had high rates of TERT and BRAFV600E, high MACIS scores, and stage III/IV disease. Analysis of 2,205 Afirma samples using the novel 145 gene signature was molecularly analogous to the TCGA data. FNA-L nodules were enriched with BRAFV600E (74%) and ALK/NTRK/RET fusions (10%). FNA-I was enriched with RAS mutations (11.7%) and lacked BRAFV600E alterations. FNA-H was enriched with TERT (15.4%) and BRAFV600E (87%) alterations.

Conclusions: Integration of mRNA clusters with ATA risk stratification produced novel prognostic gene signatures with concordance to the TCGA data set. These mRNA clusters may facilitate preoperative risk stratification with clinical management implications. Future studies with correlation of these new signatures with histopathologic and recurrence outcomes will be necessary to validate their clinical utility.