|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 22, No 17 (September 1), 2004: pp. 3531-3539 © 2004 American Society of Clinical Oncology. DOI: 10.1200/JCO.2004.08.127 Papillary and Follicular Thyroid Carcinomas Show Distinctly Different Microarray Expression Profiles and Can Be Distinguished by a Minimum of Five GenesFrom the Human Cancer Genetics Program; Clinical Cancer Genetics Program; Comprehensive Cancer Center; Division of Human Cancer Genetics, Department of Molecular Virology, Immunology, and Medical Genetics; Division of Human Genetics, Department of Internal Medicine; and Department of Physiology and Cell Biology, The Ohio State University, Columbus, OH; Department of Surgery, Martin-Luther-University, Halle-Wittenberg, Halle/Saale, Germany; Division of Medical Genetics, University of Leicester, Leicester; and Cancer Research United Kingdom Human Cancer Genetics Research Group, University of Cambridge, Cambridge, United Kingdom Address reprint requests to Charis Eng, MD, PhD, Human Cancer Genetics Program, The Ohio State University, 420 W 12th Ave, Suite 690 TMRF, Columbus, OH 43210; e-mail: eng-1{at}medctr.osu.edu
PURPOSE: We have previously conducted independent microarray expression analyses of the two most common types of nonmedullary thyroid carcinoma, namely papillary thyroid carcinoma (PTC) and follicular thyroid carcinoma (FTC). In this study, we sought to combine our data sets to shed light on the similarities and differences between these tumor types. MATERIALS AND METHODS: Microarray data from six PTCs, nine FTCs, and 13 normal thyroid samples were normalized to remove interlaboratory variability and then analyzed by unsupervised clustering, t test, and by comparison of absolute and change calls. Expression changes in four genes not previously implicated in thyroid carcinogenesis were verified by reverse transcriptase polymerase chain reaction on these same samples, together with eight additional FTC tumors. RESULTS: PTCs showed two distinct groups of genes that were either over- or underexpressed compared with normal thyroid, whereas the predominant changes in FTCs were of decreased expression. Five genes could collectively distinguish the two tumor types. PTCs showed overexpression of CITED1, claudin-10 (CLDN10), and insulin-like growth factor binding protein 6 (IGFBP6) but showed no change in expression of caveolin-1 (CAV1) or -2 (CAV2); conversely, FTCs did not express CLDN10 and had decreased expression of IGFBP6 and/or CAV1 and CAV2. CONCLUSION: PTC and FTC show distinctive microarray expression profiles, suggesting that either they have different molecular origins or they diverge distinctly from a common origin. Furthermore, if verified in a larger series of tumors, these genes could, in combination with known tumor-specific chromosome translocations, form the basis of a valuable diagnostic tool.
Thyroid carcinoma represents 1% of all cancers and is the most common type of endocrine cancer. Nonmedullary thyroid carcinoma, which arises within the follicle cells of the thyroid, can be subdivided into the following three types: papillary thyroid carcinoma (PTC), follicular thyroid carcinoma (FTC), and anaplastic or undifferentiated thyroid carcinoma. PTC is by far the most common of these carcinomas and generally indolent in nature, whereas anaplastic carcinomas are rare and aggressive.1
The etiology of thyroid carcinoma is not well understood, although radiation exposure is a risk factor in childhood PTC, as evidenced by a dramatically increased incidence in the aftermath of the Chernobyl nuclear accident.2 Chromosome translocations that give rise to oncogenic fusion proteins have been characterized in both PTC (multiple different RET-PTC translocations, reviewed by Gimm1) and FTC (PAX8-peroxisome proliferator-activated receptor gamma [PPAR
Normal and tumor specimens used in this study and methods for microarray hybridization have been described in detail elsewhere.7,9 Briefly, RNA was extracted from tumors and normal thyroid samples (ie, pathologically uninvolved tissue from patients with thyroid cancer or benign thyroid disease) using standard methods. Biotinylated cRNA was prepared and hybridized to Affymetrix U95A_v2 GeneChips (Affymetrix, Santa Clara, CA) according to the manufacturer's protocols. The two studies were conducted independently, and whereas the tumor and normal thyroid samples used in the PTC study were obtained as paired tissues from the same patient, tissues used in the FTC study were predominantly unpaired. Overall, the 28 samples were derived from 20 different individuals. Raw data files, gene expression estimates, cluster diagrams, and other supplementary information are available at http://bioinformatics.med.ohio-state.edu/wn/ptc_ftc/index.html.
Two independent methods of bioinformatic analysis were used for the FTC-PTC combined study, as outlined in Figure 1. First, model-based gene expression estimates were obtained using dChip software (http://www.dChip.org) according to the Li-Wong perfect-match-only model,10 as follows. Gene expression estimates for 16 FTC data set samples and 12 PTC data set samples were obtained separately after invariant set normalization11 at the probe level; a normal sample with median intensity was selected as the baseline array in each data set for normalizing other arrays. It should be noted that the PTC data set originally contained 16 samples, but two PTC-normal pairs were found to have different probe density distributions compared with the rest of the samples, with a high proportion of high-intensity values. These four samples were removed from the combined study before obtaining the estimates to eliminate any potential bias as a result of these samples. The two data sets were then combined using a transformation based on the normal samples in both PTC and FTC data sets, under the hypothesis that a given gene in PTC-normal samples and FTC-normal samples should have same median gene expression. All of the gene expression estimates were multiplied by the scaling factors
In parallel, the data were also analyzed with the Microarray Suite v5.0 and Data Mining Tool software packages (Affymetrix) in a manner that circumvented the need to normalize the two data sets (Fig 1). Absolute data calls were analyzed to determine which probe sets were called as absent across all normal samples and FTCs but present in PTCs (ie, are overexpressed in PTC, relative to FTC and normal samples) and, conversely, which probe sets were called as present across all normal samples and PTCs but absent in FTCs (ie, are underexpressed in FTC). In addition, a two-stage comparative analysis was made. Stage 1 compared tumors only with the normal samples hybridized within the same study. Comparative.chp files were generated with Microarray Suite (Affymetrix) for normal-tumor pairs in the PTC study (six comparisons) and by multiple pairwise comparisons of all normal (n = 6) and tumor samples (n = 9) in the FTC study (54 comparisons). Previous analysis of the FTC data by this method had demonstrated that, of the genes showing consistent changes across all FTCs, most showed decreased expression compared with normal thyroid and few were consistently overexpressed.9 Thus, for stage 2, we took this subset of FTC-underexpressed genes and examined whether they were under- or overexpressed or showed no change among the PTC-normal comparisons. Lists of genes generated by the two analyses were compared for results that were in common between them. Confirmatory reverse transcriptase polymerase chain reaction (RT-PCR) studies were then conducted on a subset of these genes that had a known or putative role in tumorigenesis but were not previously implicated in the pathogenesis of thyroid cancer. The RNA panel used for RT-PCR analysis comprised the samples used for microarray analysis, with the exception of one PTC in which insufficient RNA remained, together with up to 10 additional FTC tumors. Aliquots (1 µg) of RNA were first treated with DNase-I (Invitrogen, Carlsbad, CA) and then reverse transcribed using random hexamer priming with the SuperScript-II system (Invitrogen). The following four genes were selected for study on the basis that the array data indicated distinct behaviors between PTC and FTC: caveolin-1 (CAV1), caveolin-2 (CAV2), claudin-10 (CLDN10), and insulin-like growth factor binding protein (IGFBP6). CLDN10 and IGFBP6 were amplified for 29 cycles in a duplex reaction with glyceraldehyde-3-phosphate dehydrogenase and visualized by agarose gel electrophoresis. PCR was conducted using HotStar mastermix (Qiagen, Valencia, CA) with 1.0 µmol/L of each primer, except glyceraldehyde-3-phosphate dehydrogenase (0.25 µmol/L), in a final volume of 20 µL. Primer sequences are available upon request. CAV1 and CAV2 were analyzed using a previously established real-time quantitative PCR assay.9
Before normalization, tumor and normal samples from each of the two data sets clustered together according to their lab of origin (see online supplementary data at http://bioinformatics.med.ohio-state.edu/wn/ptc_ftc/index.html), an effect that likely reflects nonbiologic differences in their processing, as previously described.14 This artifact could be removed by normalization of the data, as described earlier. FTC and PTC then clustered into two distinct groups, whereas normal thyroid samples fell into a third grouping that, importantly, showed admixture of samples derived from the two laboratories, indicating the successful removal of lab-specific effects. Figure 2 illustrates three clusters of genes that were overexpressed in PTC (Fig 2A), underexpressed in FTC (Fig 2B), or underexpressed in both FTC and PTC (Fig 2C). Analyses conducted using the Affymetrix software package were designed to circumvent interlaboratory differences without requiring normalization. Although this approach was restrictive and thus disadvantageous for discovery of entirely new patterns of gene expression, it proved helpful in focusing on previously characterized differences in expression pattern between the two tumor types.
Cluster analysis and t tests identified 68 transcripts, corresponding to 61 genes that showed more than two-fold overexpression in PTC compared with both FTC and normal samples. A subset of these genes is illustrated in Figure 2A. Using Affymetrix analysis, seven probe sets were called present in at least two thirds of PTCs but called absent in all normal thyroid samples and FTCs (Table 1). Three of these seven genes were in common with those from the t test results (Table 1), and four genes (LAMB3, COMP, IL1RAP, and CLDN10) clustered closely together among genes overexpressed in PTC (Fig 2A). CLDN10 was further analyzed by RT-PCR, which confirmed that the transcript showed increased expression in four of five PTCs tested but was not detectable in 13 of 13 FTCs (Fig 3A). Thus, overexpression of CLDN10 may be a highly discriminatory marker for PTC. In contrast, three genes (ADORA1, sciellin, and CITED1), which were previously found to be overexpressed in PTCs compared with normal thyroid,7 were also called present in two or three of the nine FTCs (Table 1 and Fig 2A). Similarly, galectin-3 was overexpressed in all PTCs and a proportion of FTCs (Fig 2A), which was consistent with previous reports.15,16
Using the same analytic approaches, 142 genes showed significant underexpression in FTCs using the t test, and 52 genes were called absent in two thirds or more of FTCs but present in normal samples and PTCs by Affymetrix analysis. A subset is illustrated in Figure 2B. These lists showed significant overlap with genes previously identified as underexpressed in FTC in multiple pairwise tumor-normal comparisons.9 We have previously shown that two of these genes, CAV1 and CAV2, were coordinately underexpressed in 15 of 19 FTCs by quantitative PCR, Western blot analysis, and immunohistochemistry.9 In this study, isoform-specific, real-time quantitative PCR analysis was extended to the five PTC-normal pairs. No significant difference in expression of CAV1 and CAV2 alpha- and beta-isoforms was found for PTCs, thus supporting the results of our comparative microarray analysis. Furthermore, immunohistochemical studies have shown that caveolin-1 protein is detectable in 29 (76%) of 38 PTCs but only in 22 (45%) of 49 FTCs.9 Most interestingly, the two-stage comparative analysis with Affymetrix software identified the following four genes that were underexpressed in FTCs but overexpressed in PTCs: IGFBP6 (1736_at), cystatin E/M (CST6; 33128_s_at), dihydropyrimidinase related protein-3 (DPYSL3; 36149_at), and a putative lymphocyte G0/G1 switch gene (G0S2; 38326_at). Three of these genes (IGFBP6, G0S2, and CST6) also showed significant results by t test. CST6 and G0S2 were both overexpressed relative to FTCs or normal samples with similar fold changes, whereas for IGFBP6, the fold change for PTC versus FTC was much greater in magnitude than PTC versus normal samples (7.91 for PTC v FTC; 3.80 for PTC v normal), which was consistent with relative underexpression in FTCs. IGFBP6 and CST6 also clustered with other PTC-overexpressed genes (Fig 2A), whereas DPYSL3 and G0S2 localized to related parts of the tree (data not shown). Semiquantitative RT-PCR with IGFBP6 confirmed overexpression in four of five PTCs and decreased expression in 15 of 18 FTCs compared with normal samples (Fig 3B). Overexpression of CST6 in PTCs was reported previously,7 although confirmatory RT-PCR was not performed. Table 2 lists the microarray and RT-PCR data for CAV1, CAV2, IGFBP6, and CLDN10, together with Cbp/p300-interacting transactivator (CITED1), another gene previously identified as a potentially good discriminator between FTC and PTC.7 Immunohistochemical analysis of protein levels has previously been described for CITED17 predominantly in PTCs and for CAV19 in a wide range of nonmedullary thyroid tumors but was not possible for the remaining proteins because of lack of commercial antibody (CLDN10) or nonspecificity of available antibodies (IGFBP6 and CAV2). To summarize, by RT-PCR, PTCs showed increased expression of CITED1, CLDN10, and IGFBP6 but showed no change in expression of CAV1 and CAV2. In contrast, FTCs did not express CLDN10 and, with the exception of one tumor that was a lung metastasis, all primary FTCs showed decreased expression of IGFBP6 and/or CAV1 and CAV2. Thus collectively, these five genes can discriminate between the two tumor types.
Differential diagnosis of thyroid nodules rests predominantly on their histology, often initially performed on a limited tissue sample obtained by fine-needle aspiration. Although the histologies of PTC and FTC are usually quite distinct, molecular markers are still of potential clinical value. Microarray expression profiling offers a powerful means of identifying such markers and may also shed light on the etiologic differences between these tumors. Here, we have demonstrated that PTC and FTC show distinct expression profiles. In keeping with a previous comparison of loss of heterozygosity rates in different types of thyroid carcinoma,6,17 we found a much larger number of genes to be underexpressed in FTCs than in PTCs. Conversely, several genes were strongly overexpressed in PTCs but not in FTCs. The most intriguing result was the identification of four genes that are potentially underexpressed in FTC but overexpressed in PTC (CST6, DPYSL3, G0S2, and IGFBP6), the latter confirmed by RT-PCR. Although it seems counterintuitive that either over- or underexpression of the same gene can both be associated with cancer, similar results have been documented for CAV1, which is underexpressed in sarcomas, ovarian carcinoma, and prostate cancer18-21 but overexpressed in esophageal squamous cell carcinomas, pancreatic ductal adenocarcinoma, and lung adenocarcinoma.22-25
In terms of the etiology of thyroid carcinomas, it remains controversial as to whether thyroid tumors of different histologies share a common origin and whether they follow an adenoma-carcinoma progression. We have previously shown that PPAR It is also important to note that, although we were successful in combining two independent microarray data sets, normalization was essential before combining them to remove nonbiologic differences. The inclusion of normal thyroid samples from the two independent studies was valuable in monitoring this; without normalization, both tumor and normal samples clustered tightly by processing laboratory, whereas after normalization, normal samples from the two studies combined into a single admixed cluster, as would be expected biologically (Fig 2). Our experience suggests that the most powerful studies remain those in which samples are processed at the same time under highly standardized laboratory conditions. Nevertheless, this combined analysis corroborated the results of our independent studies for numerous genes, whereas RT-PCR analysis on a limited subset of additional, novel genes confirmed the predicted gene expression patterns, thus validating our approaches. In conclusion, we have made a comparative analysis of microarray expression data for PTC and FTC, the two most common forms of nonmedullary thyroid carcinoma. Although there was some commonality between the two tumor types among underexpressed genes, distinct expression patterns could be observed. Five genes (CITED1, CAV1, CAV2, IGFBP6, and CLDN10) were identified that collectively distin-guish PTCs from FTCs. If validated in larger independent panels, these genes, together with known tumor-specific chromosomal translocations, could prove to be a powerful molecular adjunct to differentiating histologic subtypes of thyroid carcinoma, even from fine-needle aspiration.
The authors indicated no potential conflicts of interest.
Supported by an Advanced Training Fellowship from the Wellcome Trust (M.A.A.; ref: 064271/Z/01/Z). C.E. is a recipient of the Doris Duke Distinguished Clinical Scientist Award. This work was partially funded by National Cancer Institute grant No. P30CA16058 to The Ohio State University Comprehensive Cancer Center, and by a generous gift from the Brown family in memory of Welton D. Brown (C.E.). Authors' disclosures of potential conflicts of interest are found at the end of this article.
1. Gimm O: Thyroid cancer. Cancer Lett 163:143-156, 2001[CrossRef][Medline] 2. Nikiforov YE, Gnepp DR: Pediatric thyroid cancer after the Chernobyl disaster. Pathomorphologic study of 84 cases (1991-1992) from the Republic of Belarus. Cancer 74:748-766, 1994[CrossRef][Medline]
3. Kroll TG, Sarraf P, Pecciarini L, et al: PAX8-PPAR
4. Marques AR, Espadinha C, Catarino AL, et al: Expression of PAX8-PPAR 5. Nikiforova MN, Biddinger PW, Caudill CM, et al: PAX8-PPARgamma rearrangement in thyroid tumors: RT-PCR and immunohistochemical analyses. Am J Surg Pathol 26:1016-1023, 2002[CrossRef][Medline]
6. Ward LS, Brenta G, Medvedovic M, et al: Studies of allelic loss in thyroid tumors reveal major differences in chromosomal instability between papillary and follicular carcinomas. J Clin Endocrinol Metab 83:525-530, 1998
7. Huang Y, Prasad M, Lemon WJ, et al: Gene expression in papillary thyroid carcinoma reveals highly consistent profiles. Proc Natl Acad Sci U S A 98:15044-15049, 2001 8. Aldred MA, Morrison CD, Gimm O, et al: Peroxisome proliferator-activated receptor gamma is frequently downregulated in a diversity of sporadic nonmedullary thyroid carcinomas. Oncogene 22:3412-3416, 2003[CrossRef][Medline]
9. Aldred MA, Ginn-Pease ME, Morrison CD, et al: Caveolin-1 and caveolin-2, together with three bone morphogenetic protein-related genes, may encode novel tumor suppressors downregulated in sporadic follicular thyroid carcinogenesis. Cancer Res 63:2864-2871, 2003
10. Li C, Wong W: Model-based analysis of oligonucleotide arrays: Expression index computation and outlier detection. Proc Natl Acad Sci U S A 98:31-36, 2001 11. Li C, Hung Wong W: Model-based analysis of oligonucleotide arrays: Model validation, design issues and standard error application. Genome Biol 2:research0032.1-0032.11, 2001
12. Eisen MB, Spellman PT, Brown PO, et al: Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95:14863-14868, 1998
13. Tusher VG, Tibshirani R, Chu G: Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 98:5116-5121, 2001 14. Nielsen TO, West RB, Linn SC, et al: Molecular characterisation of soft tissue tumours: A gene expression study. Lancet 359:1301-1307, 2002[CrossRef][Medline]
15. Orlandi F, Saggiorato E, Pivano G, et al: Galectin-3 is a presurgical marker of human thyroid cancer. Cancer Res 58:3015-3020, 1998
16. Nikiforova MN, Lynch RA, Biddinger PW, et al: RAS point mutations and PAX8-PPAR gamma rearrangement in thyroid tumors: Evidence for distinct molecular pathways in thyroid follicular carcinoma. J Clin Endocrinol Metab 88:2318-2326, 2003 17. Huang Y, de la Chapelle A, Pellegata N: Hypermethylation, but not LOH, is associated with the low expression of MT1G and CRABP1 in papillary thyroid carcinoma. Int J Cancer 104:735-744, 2003[CrossRef][Medline]
18. Wiechen K, Sers C, Agoulnik A, et al: Down-regulation of caveolin-1, a candidate tumor suppressor gene, in sarcomas. Am J Pathol 158:833-839, 2001
19. Wiechen K, Diatchenko L, Agoulnik A, et al: Caveolin-1 is down-regulated in human ovarian carcinoma and acts as a candidate tumor suppressor gene. Am J Pathol 159:1635-1643, 2001 20. Cui J, Rohr LR, Swanson G, et al: Hypermethylation of the caveolin-1 gene promoter in prostate cancer. Prostate 46:249-256, 2001[CrossRef][Medline] 21. Fiucci G, Ravid D, Reich R, et al: Caveolin-1 inhibits anchorage-independent growth, anoikis and invasiveness in MCF-7 human breast cancer cells. Oncogene 21:2365-2375, 2002[CrossRef][Medline] 22. Kato K, Hida Y, Miyamoto M, et al: Overexpression of caveolin-1 in esophageal squamous cell carcinoma correlates with lymph node metastasis and pathologic stage. Cancer 94:929-933, 2002[CrossRef][Medline]
23. Yang G, Truong LD, Wheeler TM, et al: Caveolin-1 expression in clinically confined human prostate cancer: A novel prognostic marker. Cancer Res 59:5719-5723, 1999 24. Suzuoki M, Miyamoto M, Kato K, et al: Impact of caveolin-1 expression on prognosis of pancreatic ductal adenocarcinoma. Br J Cancer 87:1140-1144, 2002[CrossRef][Medline]
25. Ho CC, Huang PH, Huang HY, et al: Up-regulated caveolin-1 accentuates the metastasis capability of lung adenocarcinoma by inducing filopodia formation. Am J Pathol 161:1647-1656, 2002 Submitted August 19, 2003; accepted March 22, 2004. This article has been cited by other articles:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
INTRODUCTION
]),
in PTC and FTC studies, respectively, where mp is the median of the gene expression estimate among the normal samples in PTC study, mf is the corresponding value in the FTC study, and mc is the median of mp and mf. The U95A GeneChip (Affymetrix) contains approximately 12,000 probes. A subset of 1,358 transcripts were selected for cluster analysis, filtered by standard deviation and coefficient of variation (standard deviation, > 50 and coefficient of variation, > 0.3). Two-dimensional hierarchical clustering based on average linkage and centered correlation similarity matrix was performed using CLUSTER and TREEVIEW software.
