In the past decade, molecular testing has emerged as a promising method to increase the accuracy of the preoperative diagnosis of malignant thyroid tumors. Several molecular diagnostic tests, including RNA based gene expression and multi-panel mutation genotyping analysis are commercially available for clinical use (17, 20). Available tests remain, however, limited by relatively low specificity and PPV (21, 22). Furthermore, the recent reclassification of a subgroup of malignant FVPTC to clinically “benign” NIFTP has further complicated the situation, since prior publications assessing the performance of commercially available molecular tests were based upon NIFTP being categorized as malignant. Indeed, most NIFTPs were reported as suspicious/malignant by expression based Afirma or mutation and gene fusion based ThyroSeq in several studies (14, 21, 23, 24).
Our group previously carried out a series of studies to identify genetic markers for distinguishing cancer from benign thyroid tumors using genome-wide gene expression arrays (15, 16). In this study, we have developed a molecular test for evaluating preoperative thyroid FNAs taking NIFTP lesions into consideration, by further characterizing isoforms of our previously profiled gene candidates.
Taking advantage of the improved annotation of the human genome over the last decade, we first explored expression of a broad range of splice variants of our 12 gene set in frozen thyroid tumor samples to select candidates which were differentially expressed in benign and malignant neoplasms. In this study, the 5 isoforms of CDH3, FNDC4, HMGA2, KLK7, and PLAG1 we identified showed potential in differentiating different thyroid tumor subtypes, and importantly, were chosen because they were not expressed in PBMCs. In FNA samples, quantitative molecular tests must address the contribution from PBMCs. Positive or negative selection using antibody-coated magnetic beads can minimize their contribution, but their use decreases overall assay sensitivity (data not shown). In the absence of selection, the load control reference gene needs to reflect the number of thyrocytes in FNA samples rather than a standard total RNA content measure. In our study, we tested two isoforms of TPO and two of thyroglobulin. Only TPO1 was found to have constant levels of expression across differentiated thyroid tumor subtypes (Fig. 1a) and importantly, was undetectable by PCR after 40 cycles in any of the 31 PBMC samples obtained perioperatively (Figure S3). We did observe, however, a decrease in TPO1 expression in one large HC metastasis, possibly a consequence of loss of differentiation in this recurrent advanced tumor.
The 5-isoform panel was tested using our intraoperative FNA samples. In this study, only a single pass of needle aspirate from each tumor was used for our analysis. To ensure the reliability of the assay, only samples reaching a TPO1 threshold of detectability of 30 cycles or less were included in this study. Only 7 (4.4%) of FNA samples showed no detectable TPO1, and 137 (86.2%) of 159 FNA samples met our threshold and produced reliable gene expression profiles. This compares favorably with reported 2–20% of cases yielding cytopathologically non-diagnostic results with two to five FNA passes (25).
In our study, quantitative PCR data generated from 137 qualified FNA samples demonstrated a differential expression among benign and malignant thyroid tumors. The ROC analysis shows our 5-isoform panel has an 86% ability to distinguish benign thyroid tumors (AN, FA, and HA) and NIFTP from malignant tumors (FVPTC and PTC), with a specificity of 91% at a sensitivity of 75%, resulting in an NPV of 91% and a PPV of 74%.
The thyroid follicular-patterned lesions, FA, NIFTP, and FVPTC, contribute the most to the indeterminate cytologies. Molecularly, they all are frequently associated with RAS mutations (26). Currently, histologic evaluation of capsule and vascular invasion is necessary for diagnosis of NIFTP. Thus, an accurate diagnosis of NIFTP is impossible by preoperative cytology or mutation based molecular tests. Nevertheless, NIFTP is an indolent lesion with < 1% risk of recurrence (27), and should not be treated as thyroid cancer, although it may warrant resection as potential premalignant lesion.
The NIFTP classification is new. Currently, the reclassified indolent NIFTP is still considered as a surgical disease by most endocrinologists. Its separation from malignant FVPTC may significantly impact clinical treatment decisions, leading to lesser surgical and other ablative procedures, and potentially simple observation as therapeutic options, which are currently under active investigation. More studies, especially prospective long-term follow-up studies, are needed to evaluate its behavior, progression, and optimal management. Finding tools for accurate preoperative identification of NIFTP will promote the study and management for this lesion. The data presented here show our newly developed 5-isoform panel may reduce cytologically and molecularly indeterminate diagnoses.
Our study also has a number of limitations, foremost the number of available intraoperative FNAs, which also do not exactly replicate the standard preoperative diagnostic FNAs typically obtained percutaneously in a clinic setting. We limited ourselves to epithelial thyroid tumors, and were unable to obtain FNA samples from the infrequent FC and HC cases encountered in the timespan samples were collected for this study. Therefore, although the available FC and HC tissue samples had high scores, suggesting the selected isoform panel may do well diagnostically, the performance of the panel in these cases remains unknown for FNAs.