Molecular signatures of eight thyroid tumor types

A76

Although fine needle aspiration (FNA) biopsy has proven to be an excellent diagnostic tool in the differential diagnosis of a thyroid nodule, it often cannot differentiate benign from malignant lesions. Indeed, 15-30% of FNAs fall into the 'indeterminate' or 'suspicious' category. As a consequence, many patients with thyroid nodules undergo surgery because of a 'suspicious', but not 'definite' diagnosis of malignancy. Therefore, other adjuncts, such as molecular-based diagnostic approaches are needed in the preoperative distinction of these lesions. To address this issue we chose to study by microarray analysis the 8 different thyroid tumor types that can present as suspicious lesions on FNA, including papillary thyroid cancer, follicular variant of papillary cancer, follicular cancer, Hürthle cell cancer, adenomatoid nodule, follicular adenoma, Hürthle cell adenoma, and lymphocytic thyroiditis. Based on sample size calculation we chose to examine 10 each of the 8 different tumor types. Cy-dye labeled c-RNA from the 8 tumor types and pooled normal thyroid control were competitively hybridized to oligonucleotide microarrays [Hs-OperonV3.0 National Cancer Institute/NIH (http://arraytracker.nci.nih.gov/index.shtml)] representing over 35,000 transcripts. Data was analyzed by using Partek Genomics Suite (Partek Inc, St. Louis, MO 63141) and BRB Array Tools, a statistical package developed by Biometric Research Branch, NCI/NIH (http://linus.nci.nih.gov/BRB-ArrayTools.html).ANOVA test with Bonferroni correction was used to identify genes that were statistically different among the eight tumor groups. Filtered data from the 80 thyroid tumor samples was used to build an expression ratio-based model capable of predicting benign vs malignant. Using K-Nearest Neighbor-based classification and Leave-one-out nested approach we cross-validated the model and determined the misclassification error. 87 genes were identified as 'classifiers' that could differentiate benign and malignant tumor types. An independent analysis of the same dataset using BRB Array Tools identified 99 classifier genes. Of these, 26 genes were also common to the Partek analysis. Principal Component Analysis (PCA) using these 26 genes revealed a clear separation of benign and malignant tumor types. A subset of 10 genes will be validated using a real time RT-PCR.