Classification of Follicular Thyroid Tumors by Molecular Signature: Results of Gene Profiling1

1792 Vol. 9, 1792–1800, May 2003 Clinical Cancer Research

Catherine B. Barden, Katherine W. Shister, Conclusions: Gene profiling is a useful tool to predict Baixin Zhu, Gerardo Guiter, David Y. Greenblatt, the molecular diagnosis of follicular thyroid tumors. Genes Martha A. Zeiger, and Thomas J. Fahey III2 were identified that reliably differentiate follicular thyroid carcinoma from adenoma. This study provides insight into Departments of Surgery [C. B. B., K. W. S., B. Z., D. Y. G., T. J. F.] genes that may be important in the molecular pathogenesis and Pathology [G. G.], New York Presbyterian Hospital and Weill Medical College of Cornell University, New York, New York 10021; of follicular thyroid tumors, as well candidates for preoper- and Johns Hopkins University School of Medicine, Baltimore, ative diagnosis of follicular thyroid carcinoma. Maryland [M. A. Z.]; and Strang Cancer Prevention Center, New York, New York [T. J. F.] INTRODUCTION It is estimated that 5–10% of the population will develop a ABSTRACT clinically significant thyroid nodule during their lifetime (1). Purpose: Thyroid nodules are common, with a lifetime Although most thyroid nodules are benign, the most common risk of developing a clinically significant thyroid nodule of indication for surgical intervention is to exclude the diagnosis of 3 10% or higher. Preoperative diagnosis was greatly enhanced carcinoma. FNA was introduced as a preoperative test for by the introduction of fine needle aspiration in the 1970s, but thyroid nodules in the 1970’s and validated as a reliable test in there has been little advancement since that time. Discrim- numerous studies since that time (2, 3). With widespread adop- ination between benign and malignant follicular neoplasms tion of thyroid FNA, the likelihood of requiring surgery for a is currently not possible by fine needle aspiration and can thyroid nodule decreased ϳ50%, from 65–70 to 35–40%, with even be difficult after full pathologic review. The purpose of a concomitant decrease in the number of thyroidectomies per- these studies is to identify genes expressed in follicular ad- formed in the United States and elsewhere (4). enomas and carcinomas of the thyroid that will permit Although FNA is currently the best initial diagnostic test molecular differentiation of these neoplasms. for evaluation of a thyroid nodule, FNA cannot discriminate Experimental Design: Gene expression patterns of 17 between benign and malignant follicular thyroid tumors. Current thyroid follicular tumors were analyzed by oligonucleotide estimates indicate that carcinomas of the thyroid are ultimately array analysis. Gene profiles for follicular adenomas and found in 10–20% of lesions read as follicular tumors by FNA (5, carcinomas were identified, and the two groups were com- 6). Patients seen with follicular thyroid lesions are advised to pared for differences in expression levels. The differentially undergo surgery, generally a hemithyroidectomy, to provide an expressed genes were used to perform a hierarchical clus- accurate diagnosis and direct additional treatment. Those who tering analysis training set. Five follicular tumors with di- ultimately prove to have a carcinoma are frequently advised to agnosis undisclosed to the investigators and 2 minimally undergo a second operation, completion thyroidectomy. In es- invasive carcinomas were entered into the cluster analysis as sence, patients have to decide preoperatively whether to undergo a test set to determine whether diagnosis by gene profile thyroid lobectomy or total thyroidectomy based on inadequate correlated with that obtained by pathologic evaluation. clinical information. Results: Thyroid follicular adenomas and carcinomas Thus, the current algorithms for managing a patient with a showed strikingly distinct gene expression patterns. The thyroid nodule deemed a follicular lesion by FNA are confusing expression patterns of 105 genes were found to be signifi- and seemingly arbitrary. There is a clear need to develop more cantly different between follicular adenoma and carcinoma. accurate initial diagnostic tests for thyroid nodules, particularly Many uncharacterized genes contributed to the distinction for those nodules classified as follicular lesions on initial cyto- between tumor types. For five follicular tumors for which pathological review, and potentially to direct subsequent treat- the final diagnosis was undisclosed, the clustering algorithm ment as well. Herein, we show that molecular analysis of gave the correct diagnosis in all 5 cases. follicular thyroid nodules by microarray analysis allows differentiation of benign and malignant tumors and may prove useful in directing both the initial treatment, as well as follow-up.

Received 8/12/02; revised 12/30/02; accepted 12/30/02. MATERIALS AND METHODS The costs of publication of this article were defrayed in part by the Tissue Samples. Primary tumor tissues were obtained at payment of page charges. This article must therefore be hereby marked time of surgery from patients with a preliminary diagnosis of advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 Funded by the G. Tom Shires Faculty Scholar Award. 2 To whom requests for reprints should be addressed, at New York Presbyterian Hospital-Cornell University, Room F-2024, 525 East 68 3 The abbreviations used are: FNA, fine needle aspiration; RT-PCR, Street, New York, NY 10021. Phone: (212) 746-5130; Fax: (212) 746- reverse transcription-PCR; EMMPRIN, extracellular matrix metallopro- 8771; E-mail: [email protected]. teinase inducer.

Table 1 Patient data, diagnosis Age (yr), Tumor-Node- Sex Tumor size AMESa,b Metastasis Cap inv. Vasc inv. Pathologic diagnosis A1 52 F 2.5 ϫ 1.5 ϫ 1.2 n/a n/a ϪϪFollicular adenoma A2 37 M 2.4 ϫ 1.7 ϫ 2.5 n/a n/a ϪϪFollicular adenoma A3 46 F 4.5 ϫ 4.0 ϫ 4.0 n/a n/a ϪϪFollicular adenoma A4 68 F 2.5 ϫ 1.1 ϫ 0.4 n/a n/a ϪϪFollicular adenoma A5 59 F 2.0 ϫ 2.0 ϫ 2.0 n/a n/a ϪϪFollicular adenoma A6 55 F 1.5 ϫ 1.0 ϫ 0.8 n/a n/a ϪϪFollicular adenoma A7 65 F 1.8 ϫ 1.5 ϫ 1.4 n/a n/a ϪϪFollicular adenoma A8 27 F 4.2 ϫ 3.7 ϫ 2.7 n/a n/a ϪϪFollicular adenoma A9 50 F 2.9 ϫ 2.8 ϫ 1.9 n/a n/a ϪϪFollicular adenoma A10 33 F 1.6 ϫ 0.6 ϫ 0.6 n/a n/a ϪϪFollicular adenoma ϫ ϫ Ϫϩ C1 77 M 10.0 6.5 5.5 High T3 Nx M0 Follicular carcinoma ϫ ϫ ϩϩ C2 61 F 1.5 1.0 1.0 Low T2 Nx Mx Follicular carcinoma ϫ ϫ ϩϩ C3 24 M 2.0 2.0 2.0 High T2 Nx Mx Follicular carcinoma ϫ ϫ ϩϩ C4 65 F 11.0 6.5 4.5 High T3 Nx M0 Follicular carcinoma ϫ ϫ ϩ Ϫ C5 39 F 3.0 2.5 2.0 Low T2 Nx Mx focal Minimally invasive; carcinoma follicular ϫ ϫ ϩϩ C6 74 M 2.9 2.0 2.0 High T4 Nx M0 Follicular carcinoma; poorly differentiated ϫ ϫ ϩϩ C7 59 F 8.5 5.3 5.3 High T4 Nx Mx Follicular carcinoma ϫ ϫ ϩϩ C8 71 M 4.0 4.0 3.0 High T4 N1 M0 Hurthle cell carcinoma ϫ ϫ ϩ Ϫ C9 38 F 2.4 2.3 1.5 Low T2 Nx Mx focal Minimally invasive; follicular carcinoma a AMES, age, metastasis, extracapsular invasion, size; Cap. inv., capsular invasion; Vasc inv., vascular invasion. n/a, not applicable. b AMES is a staging method for thyroid cancer based upon the four patient and tumor characteristics in footnote a.

Table 2 PCR conditions and primers Sense Antisense Size Reference c-Met GATTTTAGTCATCCCAATGTCC ATCCAGCATACAGTTTCTTGC 226 8 EMMPRIN CATGCTGGTCTGCAAGTCA CCTCTCAATGTGTAGCTCTG 141 9 Adrenomedullin AAGAAGTGGAATAAGTGGGCT TGGCTTAGAAGACACCAGAGT 410 10 Autotaxin CTCGTTCCAGTCGTGTCAGA CCATCCACGGAGAAGATGAT 504 Transforming growth factor ␤ II receptor ACTGTGCCATC TCCTGG GCAGGTTAGGTCGTTCTTCACG 535 11 Glyceraldehyde-3-phosphate GAAGGTGAAGGTCGGAGTC GAAGATGGTGATGGGATTTC 225 dehydrogenase

follicular thyroid tumor. All tumor samples were obtained with subjected to in vitro transcription for5hat37°C. Fragmented permission of and in accordance with the guidelines of the cRNA was made in 5ϫ fragmentation buffer at 94°C for 35 min respective Institutional Review Boards. A pathologist dissected and was biotin labeled. Labeled RNA was hybridized to Af- the tumor tissue, and a 2-mm cube was obtained, snap frozen in fymetrix oligonucleotide arrays at 45°C overnight. An aliquot of liquid nitrogen, and stored at Ϫ80°C. Histological classification each sample was first hybridized to an Affymetrix Test Chip was confirmed, and diagnosis of adenoma or carcinoma was array to determine sample quality as outlined by the manufac- obtained from the final pathology report. The clinical and his- turer. All samples were of good quality and hybridized to tological features are summarized in Table 1. Affymetrix Hu95 GeneChips. Arrays were washed and stained RNA Purification, Labeling, and Hybridization. Fro- with streptavidin-phycoerythrin three times. Chips were scanned zen tissues were homogenized by sonication in Trizol reagent in a Hewlett Packard ChipScanner (Affymetrix) to detect hybrid- (Invitrogen, Carlsbad, CA). Total RNA was prepared according ization signals. to the manufacturer’s protocol. Integrity of the RNA was as- Data Analysis. The data were analyzed using GeneChip sessed by spectrophotometry. Twenty-four follicular thyroid Expression Analysis software (Affymetrix). The intensity of tumors were analyzed in all by oligonucleotide microarray each probe set of the array was captured, and expression values (GeneChip Hu95 array; Affymetrix, Santa Clara, CA). The were calculated. To determine the quantitative expression level, initial training set included 12 follicular adenomas and 7 follic- the average differences representing the perfectly matched ver- ular carcinomas. Subsequently, 5 follicular tumors of undis- sus the mismatched for each gene-specific probe were calcu- closed diagnosis and 2 minimally invasive carcinomas were lated. The data were normalized to account for variability in analyzed as the test set. The Hu95 array contains oligonucleo- hybridization for probe pairs and other hybridization artifacts. tides representing Ͼ12,000 known genes. All procedures were The analysis designates whether a transcript is reliably detected performed according to the instructions from Affymetrix. (present) or not detected (absent). Briefly, first-strand cDNA was synthesized from 5 ␮g of total Data analysis was performed to identify genes that were RNA with T7 (dT)24 primer. Second-strand synthesis was car- differentially expressed between the adenoma and carcinoma ried out using Escherichia coli polymerase DNA ligase, and groups. Data from the 12 adenomas and 7 carcinomas that RnaseH was added to the reaction. The cDNA was purified and comprised the training set were used. The minimally invasive

Table 3 Gene list Gene name Symbol Accession no. Up-regulated in carcinoma Activating transcription factor 5 ATF5 AB021663 Adrenomedullin ADM D14874 Asparagine synthetase ASNS M27396 Basigin (OK blood group) BSG X64364 Bcl-2-associated X protein BAX U19599 BTB domain containing 2 BTBD2 W68046 Carbonic anhydrase XII CA12 AF037335 CCAAT/enhancer binding protein (C/EBP), ␥ CEBPG U20240 Cdc42 effector protein 3 CEP3 AF094521 CED-6 protein CED-6 AL080142 Chromosome 18 open-reading frame 1 C18orf1 AF009426 Cystathionine-␤-synthetase CBS L00972 DNA damage-inducible transcript 3 DDIT3 S62138 Docking protein 1 (downstream of tyrosine kinase 1) DOK1 AF035299 Ectonucleotide pyrophosphatase/phosphodiesterase 2 (autotaxin) ENPP2 L35594 Ectonucleotide pyrophosphatase/phosphodiesterase 2 (autotaxin) ENPP2 L35594 Ectonucleotide pyrophosphatase/phosphodiesterase 1 ENPP1 D12485 Formyl peptide receptor-like 1 FPRL1 M84562 Fusion gene S79325 Glypican 1 CGPC1 X54232 HBV-associated factor XAP4 AA160708 Hexabrachion (tenascin-C) HXB X78565 Human clone 137308 mRNA partial cds AW006742 Hypothetical protein DJ971N18.2 AL021396 Hypothetical protein MCG3077 McG3077 AI620381 Insulin-like growth factor binding protein 3 IGFBP3 M35878 Insulin-like growth factor binding protein 3 IGFBP3 M35878 IFN, ␥ inducible protein 30 IFI30 J03909 Interferon, ␥-inducible protein 30 IFI30 J03909 KIAA0726 gene product KIAA0726 AB018269 KIAA0731 protein KIAA0731 AB018274 KIAA0843 protein KIAA0843 AB020650 Lysyl oxidase LOX L16895 Mal, T-cell differentiation protein MAL X76220 Met proto-oncogene (hepatocyte growth factor receptor) MET J02958 Methylene tetrahydrofolate dehydrogenase (NADϩ dependent), MTHFD2 X16396 methenyltetrahydrofolate cyclohydrolase Myelin-associated glycoprotein MAG M29273 Nuclear receptor subfamily 2, group F, member 6 NR2F6 X12794 Paternally expressed 10 PEG10 AB028974 PCTAIRE protein kinase 1 PCTK1 X66363 PCTAIRE protein kinase 1 PTCK1 X66363 Phospholipase C, ␤ 4 PLCB4 L41349 Polymerase (RNA) mitochondrial (DNA directed) POLRMT U75370 Pre-B-cell colony-enhancing factor PBEF U02020 Protein S (␣) PROS1 M15036 Proteolipid protein 2 PLP2 U93305 Pyruvate kinase, muscle PKM2 M26252 Retinol-binding protein 1, cellular RBP1 M11433 Serine protease inhibitor, Kunitz type 1 SPINT1 AB000095 Signal transducer and activator of transcription 6, interleukin-4 induced STAT6 U16031 Similar to vaccinia virus HindIII K4L ORF HU_K4 U60644 Thyroid hormone receptor-associated protein 240kDa subunit TRAP240 AB011165 TLS/CHOP fusion gene product HG-2724-HT2820 Transforming growth factor, ␣ TGFA X70340 Trinucleotide repeat containing 5 TNRC5 U80744 TU3A protein TU3A AF035283 Tubulin, ␤ 4 TUBB4 U47364 Ubiquitin carrier protein E2-EPF M91670 Zinc finger protein 91 (HPF7, HTF10) ZNF91 L11672 Up-regulated in adenoma Sorbitol dehydrogenase SORD L29254 5Ј iodothyronine deiodinase S48220 cAMP-responsive element modulator CREM S68271 Carbonic anhydrase IV CA4 M83670 Collagen, type XV, ␣ 1 COL15A1 L25286 DKFZP434C212 protein DKFZP434C212 AL080196 Early growth response 2 EGR2 J04076

Table 3 Continued Gene name Symbol Accession no. Engulfment and cell motility 1 ELMO1 D87457 Far upstream element (FUSE) binding protein 3 FUBP3 U69127 Fc fragment of IgG binding protein FCGBP D84239 Fc fragment of IgG, receptor, transporter ␣ FCGRT U12255 Fibrillin 1 FBN1 X63556 Fibulin 5 FBLN5 AF093118 Friedriech ataxia region gene X123 X123 L27479 Gap junction protein, ␣ 1, 43 kDa (connexin 43) GJA1 X52947 Gelsolin GSN X04412 Glucosyltransferase AD-017 AD017 L13435 Histone cell cycle regulation-defective homolog A HIRA X89887 HLA class II region-expressed gene KE4 HKE4 AL031228 Hypothetical protein MGC14797 MCG14797 AL035306 IMAGE clone 3140802 AL050002 Immunoglobulin heavy constant ␥ 3 marker IGHG3 Y14737 Inhibitor of DNA binding 4 ID4 AL022726 Inositol 1, 4, 5-triphosphate receptor, type 1 ITPR1 D26070 Inositol 1, 4, 5-triphosphate receptor, type 1 ITPR1 D26070 Insulin-like growth factor binding protein 7 IGFBP7 L19182 Interleukin 11 receptor, ␣ IL11RA U32324 KIAA0084 protein KIAA0084 D42043 KIAA0758 protein KIAA0758 AB018301 Laminin, ␤ 1 LAMB 1 M69196 LSM 3 protein LSM3 N98670 Metallothionein 1 F (functional) MT1F M10943 MKP-1 like protein kinase phosphatase MKP-L AF038844 Nidogen 2 NID2 D86425 Tumor necrosis factor receptor superfamily, member 11b TNFRSF11B AB008822 (osteoprotegerin) Putative emu1 LOC129080 AL031186 Ribonuclease Rnase A family 1 (pancreatic) RNASE1 D26129 Selenoprotein P, plasma 1 SEPP1 Z11793 Serine palmitoyltransferase, long chain base subunit 1 SPTLC1 Y08685 Serum response factor (c-fos serum response element-binding SRF J03161 transcription factor) T-complex-associated-testis-expressed 1-like 1 TCTEL1 D50663 Transforming growth factor, ␤ receptor II TGFBR2 D50683 Trefoil factor 3 (intestinal) TFF3 AI985964 Trefoil factor 3 (intestinal) TFF3 L08044 v-Jun avian sarcoma virus 17 oncogene homolog JUN J04111

tumors were not included in the original training set because cluster analysis. These tumors were also collected by an Insti- they behave differently than most follicular carcinomas. First, tutional Review Board-approved protocol. Gene profiles for the the data were screened to identify signals counted as present by 105 differentially expressed genes of the 5 tumors were used to the Affymetrix software. Second, signals with at least 2-fold include these tumors to the test set. Additionally, the two min- differences in expression were identified because it has been imally invasive follicular carcinomas were entered into the test established that a 2-fold or greater change is significant and set cluster. accurate (7). These results were screened with a nonparametric Semiquantitative RT-PCR. To verify the results ob- t test, with the P set at Ͻ0.01. One hundred five genes were tained from the Affymetrix chip hybridization, 5 genes, which identified as differentially expressed and were used as a gene list showed robust, Ͼ3-fold overexpression and for which primers for cluster analysis. Data were exported to GeneSpring (Silicon were readily available, were chosen. One ␮g of total RNA was Genetics, Redwood City, CA), which was used for unsupervised reverse transcribed with oligo(dT) primer. A 2-␮l aliquot of the hierarchical clustering and statistical analysis. Cluster analysis cDNA was used for PCR. The primers and PCR conditions are was used to group the tumors based upon their similarities listed in Table 2. Products were electrophoresed on an agarose measured across expression of 105 genes. gel and visualized by ethidium bromide under UV light. Evaluation of Unknown Samples. Once the hierarchical Protein Purification and Western Blotting. Frozen tis- cluster analysis was established using gene expression profiles sue was thawed in ice-cold homogenization buffer containing of differentially expressed genes in 17 tumors, the same analysis 150 mM NaCl, 50 mM Tris-buffered saline (pH 7.4), 10% was performed on 5 follicular tumors with investigators blinded glycerol, 1% Triton-X, 2 mM EGTA, 2 mM MgCl2,1mM to the final diagnosis. The unknown specimens were obtained diethyldithiocarbamate, 1 mM phenylmethylsulfonyl fluoride, from a different institution than the tumors used to create the 10 ␮g/ml leupeptin, 10 ␮g/ml aprotinin, 5 ␮g/ml pepstatin, 3

Fig. 1 Gene cluster 17. Dendrogram of cluster analysis of 12 follicular adenomas and 7 follicular carcinomas based upon the pattern of expression of 105 differentially expressed genes. Red indicates relative high expression; blue indicates low levels of expression.

mM hydrogen peroxide, 50 mM sodium fluoride, 1 mM sodium the levels of gene expression for the two groups were compared. orthovandate, and 10 mM sodium molybdate. Tissues were ho- One hundred five genes were differentially expressed in ade- mogenized using a glass-on-glass tissue homogenizer. Homo- noma and carcinoma. Fifty-nine genes were up-regulated in genates were centrifuged at 11,750 ϫ g for 10 min at 4°Cto carcinoma, whereas 49 genes were up-regulated in adenoma. remove the particulate material. The protein concentration of the The list of genes differentially expressed is outlined in Table 3. supernatant was measured using the Lowry protein assay kit. Two-Way Clustering. The dendrogram displays the Immunoblot analysis for EMMPRIN was performed. Fifty length and subdivision of the branches, which correlated to the ␮g/lane of protein from tissue were loaded on a 10% SDS- relatedness of the thyroid tumors. Two distinct groups of tumors PAGE gel; after transfer, the membrane was blocked in 3% BSA are apparent in this two-dimensional analysis, suggesting that and incubated overnight with primary antibody. ␤-Actin was the tumors can be divided into two types on the basis of the used as an internal control. expression level of these 105 genes (Fig. 1). Notably, all of the samples in the top cluster had a pathologic diagnosis of follic- RESULTS ular adenoma. Differentially Expressed Genes. To determine genes Diagnosis of Unknown Samples. Five additional follic- differentially expressed in follicular carcinoma and adenoma, ular tumor samples were obtained, all of which had the diagno-

Fig. 2 Gene cluster 19 ϩ unknowns. Dendrogram of cluster analysis of 17 follicular tumors, plus 5 unknown samples (U1–5) and 2 minimally invasive carcinomas (C5*, C9*) based upon the pattern of expression of 105 differentially expressed genes. Red indicates relative high expression; blue indicates low levels of expression.

sis undisclosed to the investigators. Gene expression analysis sion of EMMPRIN is greater in the follicular carcinomas than in and subsequent cluster analysis identified 4 adenomas and 1 the follicular adenomas. carcinoma (Fig. 2). Release of the pathologic diagnosis to the investigators revealed that the diagnosis predicted by clustering was correct in all 5 cases. Interestingly, the two minimally DISCUSSION invasive carcinomas clustered with the follicular adenomas. Accurate preoperative diagnosis of thyroid nodules has Corroboration of Gene Expression. We validated the been a goal of thyroid researchers and clinicians for decades. differential expression of 5 genes identified in this study by The widespread introduction of FNA Ͼ 20 years ago changed semiquantitative PCR on 19 of 24 tumors analyzed. One of these the management of thyroid nodules substantially and resulted in genes has been previously associated with thyroid carcinomas a significant decrease in rates of thyroidectomy for benign (met), and the remainder are not known to be involved in the thyroid nodules (4). Recent advances in molecular biology have molecular pathogenesis of follicular thyroid tumors. For each yet to translate into more refined preoperative diagnosis and gene tested, the level of expression by RT-PCR correlated with management of thyroid nodules and, in particular, follicular the data obtained by microarray analysis (Fig. 3). neoplasms of the thyroid. The data presented here suggest that Additionally, differential expression of the protein follicular adenomas can be reliably distinguished from follicular EMMPRIN was confirmed by immunoblot (Fig. 4). The expres- carcinomas by molecular analysis. Furthermore, we have iden-

Fig. 3 Semiquantitative PCR. Verification of carcinoma and adenoma-specific gene expression of ADM (adrenomedullin), ENPP2 (autotaxin), BSG (EMMPRIN), MET (c-met), and transforming growth factor ␤II receptor by RT- PCR. Primers specific for the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase were used as controls. Samples A1–A10, adenomas, samples C1–C9 carcinomas, and C5* and C9* are minimally invasive carcinomas.

Fig. 4 Immunoblot. The differential expression of EMMPRIN was confirmed by Western blot. Fifty ␮g of protein were loaded/lane. The first four samples (left) are from follicular adenomas, the next four (right) are from follicular carcinomas.

tified new genes that might be useful in the distinction of benign surface glycoprotein of human tumor cells and stimulates neigh- and malignant follicular tumors. boring fibroblasts to produce expression of several matrix met- Follicular adenomas and carcinomas are two closely re- taloproteinase (19). EMMPRIN has been associated with en- lated neoplasms, with potentially markedly different clinical hanced expression in invasive human tumors. Zucker et al. (20) outcomes. Our current understanding of the molecular steps demonstrated that overexpression of EMMPRIN by transfection leading to a malignant phenotype in follicular tumors is limited. of the MDA-MB-436 breast cancer cell line with greatly in- Previous studies have suggested a number of possibly important creased its tumorigenicity and invasiveness in vivo. genetic alterations, including expression of galectin-3, met/HGF The two remaining genes that we found to be overex- receptor, reactivation of telomerase, and PAX-8/peroxisome pressed in follicular carcinomas, autotaxin and adrenomedullin, proliferator-activated receptor␥ fusion translocation (12–15). have not been previously analyzed in thyroid carcinomas, al- However, none of these molecular changes have translated into though they have been recognized as potential markers of car- clinically useful markers to differentiate benign and malignant cinogenesis in other solid tumors. Autotaxin was originally tumors to date. We turned to gene expression profiling in an isolated as a tumor motility-stimulating protein and is a member attempt to identify new genes responsible for the differences in of the nucleotide pyrophosphatase and phosphodoiesterase en- biological behavior between follicular adenomas and carci- zyme family (21). It is overexpressed in several human cancers nomas. such as melanoma, hepatocellular carcinoma, and neuroblas- Of the 105 genes identified with significantly different toma (22). Recent work has revealed that it also has angiogenic expression levels between adenomas and carcinomas, most have properties and can stimulate the proliferation of several cancer not been previously described as important in the molecular cell lines (22, 23). pathogenesis of thyroid carcinoma. We selected 6 genes for Adrenomedullin was isolated from pheochromocytoma and additional analysis to validate the microarray expression data. has recently been shown to play a role in cancer cell growth and These genes were chosen because their expression profiles were survival (24). It is abundantly expressed in tumor cell lines among the most significantly different between adenomas and of epithelial origin (lung, breast, colon ovary, and prostate; carcinomas. Of the 6 genes analyzed, we confirmed differential Ref. 10). Furthermore, a clinical study performed on patients gene expression in 4 genes expressed preferentially in carcino- with ovarian carcinoma revealed an association with adreno- mas, and 2 genes expressed preferentially in adenomas com- medullin expression and poor prognosis (25). pared with carcinomas. One of the 4 genes, the met gene, has Additional insight into the biological behavior of mini- been previously reported to be overexpressed in papillary and mally invasive carcinomas may be provided by examination of follicular thyroid carcinoma (16). Recently, Huang et al. (17) the differences in expression of the 105 genes in these tumors. demonstrated overexpression of met in papillary thyroid carci- We elected not to include the minimally invasive tumors in the noma by microarray analysis. original training set because of their unusually benign behavior A recent investigation into genes differentially regulated in for a cancer. Thus, it is noteworthy that these tumors were metastatic versus nonmetastatic follicular carcinomas identified clustered with the adenomas because the clinical outcome for EMMPRIN as a marker for metastasis (18). EMMPRIN is a cell patients with minimally invasive carcinomas is much more

similar to adenomas than invasive carcinomas. It is also inter- 9. Klein, C. A., Seidl, S., Petat-Dutter, K., Offner, S., Geigl, J. B., esting to note the expression of autotaxin and adrenomedullin by Schmidt-Kittler, O., Wendler, N., Passlick, B., Huber, R. M., Schlimok, PCR in these tumors. The pattern of expression is unlike the G., Baeuerle, P. A., and Riethmuller, G. Combined transcriptome and genome analysis of single micrometastatic cells. Nat. Biotechnol., 20: angioinvasive carcinomas, rather, they resemble more closely 387–392, 2002. the adenomas. 10. Miller, M. J., Martinez, A., Unsworth, E. J., Thiele, C. J., Moody, Molecular classification of tumors is an emerging technol- T. W., Elsanner, T., and Cuttita, F. Adrenomedullin expression in ogy that will undoubtedly change the way patients are managed human tumor cell lines. J. Biol. Chem, 271: 23345–23351, 1996. (26, 27). Follicular lesions of the thyroid are one group of 11. Bartlett, J. M., Langdon, S. P., Scott, W. N., Love, S. B., Miller, neoplasms that may benefit greatly from molecular analysis as a E. P., Katsaros, D., Smyth, J. F., and Miller, W. R. Transforming growth part of the work-up. It is possible to determine the level of gene factor ␤ isoform expression in human ovarian tumours. Eur. J. Cancer, expression by PCR on RNA isolation from as few as 10 cells 33: 2397–4032, 1997. isolated as part of the FNA biopsy (28). A recent study has 12. Saggiorato, E., Cappia, S., De Giuli, P., Mussa, A., Pancani, G., Caraci, P., Angeli, A., and Orlandi, F. Galectin-3 as a presurgical examined this technique for preoperative diagnosis of medullary immunocytodiagnostic marker of minimally invasive follicular thyroid thyroid carcinoma (29). carcinoma. J. Clin. Endocrinol. Metab., 86: 5152–5158, 2001. A clearer understanding of the nature of a follicular neo- 13. Ippolito, A., Vella, V., La Rosa, G. L., Pellegriti, G., Vigneri, R., plasm may allow physicians and surgeons to more accurately and Belfiore, A. Immunostaining for Met/HGF receptor may be useful advise patients as to the necessity for surgery, as well as the to identify malignancies in thyroid lesions classified suspicious at fine- needle aspiration biopsy. Thyroid, 11: 783–787, 2001. extent of surgery. This could result in a decrease in thyroid 14. Saji, M., Xydas, S., Westra, W. H., Liang, C-K., Clark, D. P., surgery for nodules predicted to be benign on molecular anal- Udelsman, R., Umbricht, C. B., Sukumar, S., and Zeiger, M. A. Human ysis, as well as appropriate comprehensive surgery for nodules telomerase reverse transcriptase (hTERT) gene expression in thyroid predicted to be aggressive. Furthermore, decisions regarding neoplasms. Clin. Cancer Res., 5: 1483–1489, 1999. postoperative adjuvant radioactive iodine treatment might be 15. Kroll, T. G., Sarraf, P., Pecciarini, L., Chen, C. J., Mueller, E., made with greater confidence. We are optimistic that future Spiegelman, B. M., and Fletcher, J. A. PAX8-PPAR␥1 fusion oncogene evaluation of thyroid nodules will include molecular analysis, in in human thyroid carcinoma. Science (Wash. DC), 25: 289, 1357–1360, 2000. addition to the cytological diagnosis, obtained by thyroid FNA. 16. Trovato, M., Villari, D., Bartolone, L., Spinella, S., Simone, A., The data presented here provide a host of new genes that should Violi, M. A., Trimarchi, F., Batolo, D., and Benvenga, S. Expression of permit molecular analysis to become clinical reality in the future the hepatocyte growth factor and c-met in normal thyroid, non-neoplas- management of follicular thyroid tumors. tic, and neoplastic nodules. Thyroid, 8: 125–131, 1998. 17. Huang, Y., Prasad, M., Lemon, W. J., Hampel, H., Wright, F. A., Kornacker, K., LiVolsi, V., Frankel, W., Kloos, R. T., Eng, C., Pelle- ACKNOWLEDGMENTS gata, N. S., and de la Chapelle, A. Gene expression in papillary thyroid We thank Dr. Jenny Z. Xiang and Xianchun Huang of the Microar- carcinoma reveals highly consistent profiles. Proc. Natl. Acad. Sci. ray Core Facility at Weill Medical College of Cornell University for USA, 98: 15044–15049, 2001. assistance with performing the microarrays and helpful discussions. 18. Chen, K-T., Lin, J-D., Chao, T-C., Hsueh, C., Chang, C. A., Weng, H-F., and Chan, E-C. Identifying differentially expressed genes associated with metastasis of follicular thyroid cancer by cDNA expression REFERENCES array. Thyroid, 11: 41–46, 2001. 1. Mazzaferri, E. L. Management of a solitary thyroid nodule. N. Engl. 19. Kanekura, T., Chen, X., and Kanzaki, T. Basigin (CD147) is ex- J. Med., 328: 553–559, 1993. pressed on melanoma cells and indices tumor cell invasion by stimu- 2. Goldstein, R. E., Netterville, J. L., Burkey, B., and Johnson, J. E. lating production of matrix metalloproteinases by fibroblasts. Int. J. Implications of follicular neoplasms, atypia, and lesions suspicious for Cancer, 99: 520–528, 2002. malignancy diagnosed by fine-needle aspiration of thyroid nodules. 20. Zucker, S., Hymowitz, M., Rollo, E. E., Mann, R., Conner, C. E., Ann. Surg., 235: 656–664, 2002. Cao, J., Foda, H. D., Tompkins, D. C., and Toole, B. P. Tumorigenic 3. Gharib, H., and Goellner, J. R. Fine-needle aspiration biopsy of the potential of extracellular matrix metalloproteinase inducer. Am. J. thyroid: an appraisal. Ann. Intern. Med., 118: 282–289, 1993. Pathol, 158: 1921–1928, 2001. 4. Hamberger, B., Gharib, H., Melton, L. J., III, Goellner, J. R., and 21. Clair, T., Lee, H. Y., Liotta, L. A., and Stracke, M. L. Autotaxin is Zinsmeister, A. R. Fine-needle aspiration biopsy of thyroid nodules: an exoenzyme possessing 5Ј-nucleotide phosphodiesterase/ATP pyro- impact on thyroid practice and cost of care. Am. J. Med., 73: 381–384, phosphatase and ATPase activities. J. Biol. Chem., 272: 996–1001, 1982. 1997. 5. Baloch, Z. W., Fleisher, S., LiVolsi, V. A., and Gupta, P. K. Diag- 22. Nam, S. W., Clair, T., Kin, Y-S., McMarlin, A., Schiffman, E., nosis of “follicular neoplasm”: a gray zone in thyroid fine-needle Liotta, L. A., and Stracke, M. Autotaxin (NPP-2), a metastasis, enhanc- aspiration cytology. Diagn. Cytopathol., 26: 41–44, 2002. ing motogen, is an angiogenic factor. Cancer Res., 61: 6938–6944, 6. Greaves, T. S., Olvera, M., Florentine, B. D., Raza, A. S., Cobb, 2001. C. J., Tsao-Wei, D. D., Groshen, S., Singer, P., Lopresti, J., and Martin, 23. Umezu-Goto, M., Kishi, Y., Taira, A., Hama, K., Dohmae, N., S. E. Follicular lesions of the thyroid: a 5-year fine-needle aspiration Takio k, Yamori, T., Mills, G. B., Inoue, K., Aoki, J., and Arai, H. experience. Cancer (Phila.), 90: 335–341, 2000. Autotaxin has lysophosphatase D activity to tumor cell growth and 7. Wodicka, L., Dong, H., Mittman, M., Ho, M. H., and Lockhart, D. J. motility by lysophosphatidic acid production. J. Cell Biol., 158: 227– Genome-wide expression monitoring in Saccharomyces cerevisiae. Nat. 233, 2002. Biotechnol., 15: 1359–1367, 1997. 24. Cuttitta, F., Pio, R., Garayoa, M., Zudaire, E., Julian, M., Elsasser, 8. Shin, J. H., Chung, J., Kim, H. O., Kim, Y. H., Hur, Y. M., Rhim, T. H., Montuenga, L. M., and Martinez, A. Adrenomedullin functions as J. H., Chung, H. K., Park, S. C., Park, J. G., and Yang, H. K. Detection an important tumor survival factor in human carcinogenesis. Microsc. of cancer cells in peripheral blood of stomach cancer patients using Res. Tech., 57: 110–119, 2002. RT-PCR amplification of tumour-specific mRNAs. Aliment. Pharma- 25. Hata, K., Takebayashi, Y., Akiba, S., Fujiwaki, F., Iida, K., Na- col. Ther., 16: 137–144, 2002. kayama, K., Nakayama, S., Fukomoto, M., and Miyazaki, K. Expression

of the adrenomedullin gene in epithelial ovarian cancer. Mol. Hum. sion profile that distinguished a unique leukemia. Nat. Genet., 30: Reprod., 6: 867–872, 2000. 41–47, 2001. 26. Van’t Veer, L. J., Dai, H., van ge Vijver, M. J., He, Y. D., Hart, 28. Winzer, R., Schmutzler, C., Jakobs, T. C., Ebert, R., Rendl, J., A. M., Mao, M., Peterse, H. L., van der Kooy, K., Marton, M. J., Reiners, C., Jakob, F., and Kohrle, J. Reverse transcriptase-polymerase Witteveen, A. T., Schreiber, G. J., Kerkhoven, R. M., Roberst, C., chain reaction analysis of thyrocyte-relevant genes in fine-needle aspi- Linsley, P. S., Bernards, R., and Friend, S. H. Gene expression profiling ration biopsies of the human thyroid. Thyroid, 11: 981–987, 1998. predicts clinical outcome of breast cancer. Nature (Lond.), 415: 530– 29. Takano, T., Miyauchi, A., Matsuzuka, F., Liu, G., Higashiyama, T., 536, 2002. Yokozawa, T., Kuma, K., and Amino, N. Preoperative diagnosis of 27. Armstrong, S. A., Staunton, J. E., Silverman, L. B., Pieters, R., den medullary thyroid carcinoma by RT-PCR using RNA extracted from Boer, M. L., Minden, M. D., Sallan, S. E., Lander, E. S., Golub, T. R., leftover cells within a needle used for fine needle aspiration biopsy. and Korsmeyer, S. J. MLL translocations specify a distinct gene expres- J. Clin. Endocrinol. Metab., 84: 951–955, 1999.

Downloaded from clincancerres.aacrjournals.org on September 27, 2021. © 2003 American Association for Cancer Research. Classification of Follicular Thyroid Tumors by Molecular Signature: Results of Gene Profiling

Catherine B. Barden, Katherine W. Shister, Baixin Zhu, et al.

Clin Cancer Res 2003;9:1792-1800.

Updated version Access the most recent version of this article at: http://clincancerres.aacrjournals.org/content/9/5/1792

Cited articles This article cites 26 articles, 6 of which you can access for free at: http://clincancerres.aacrjournals.org/content/9/5/1792.full#ref-list-1

Citing articles This article has been cited by 17 HighWire-hosted articles. Access the articles at: http://clincancerres.aacrjournals.org/content/9/5/1792.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/9/5/1792. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.