|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
Journal of Clinical Oncology, Vol 24, No 31 (November 1), 2006: pp. 5043-5051 © 2006 American Society of Clinical Oncology. DOI: 10.1200/JCO.2006.06.7330 Meta-Analysis and Meta-Review of Thyroid Cancer Gene Expression Profiling Studies Identifies Important Diagnostic Biomarkers
From the Michael Smith Genome Sciences Centre, British Columbia Cancer Agency; Departments of Medical Genetics and Surgery, University of British Columbia; Genetic Pathology Evaluation Center, Prostate Research Center of Vancouver General Hospital & British Columbia Cancer Agency; Department of Surgery, St Paul's Hospital, Vancouver, Canada Address reprint requests to Sam M. Wiseman, MD, FRCSC, FACS, St Paul's Hospital, Room C302 Burrard Bldg, 1081 Burrard Street, Vancouver, BC, V6Z1Y6, Canada; smwiseman{at}providencehealth.bc.ca
Purpose: An estimated 4% to 7% of the population will develop a clinically significant thyroid nodule during their lifetime. In many cases, preoperative diagnoses by needle biopsy are inconclusive. Thus, there is a clear need for improved diagnostic tests to distinguish malignant from benign thyroid tumors. The recent development of high-throughput molecular analytic techniques should allow the rapid evaluation of new diagnostic markers. However, researchers are faced with an overwhelming number of potential markers from numerous thyroid cancer expression profiling studies. Materials and Methods: To address this challenge, we have carried out a comprehensive meta-review of thyroid cancer biomarkers from 21 published studies. A gene ranking system that considers the number of comparisons in agreement, total number of samples, average fold-change and direction of change was devised. Results: We have observed that genes are consistently reported by multiple studies at a highly significant rate (P < .05). Comparison with a meta-analysis of studies reprocessed from raw data showed strong concordance with our method. Conclusion: Our approach represents a useful method for identifying consistent gene expression markers when raw data are unavailable. A review of the top 12 candidates revealed well known thyroid cancer markers such as MET, TFF3, SERPINA1, TIMP1, FN1, and TPO as well as relatively novel or uncharacterized genes such as TGFA, QPCT, CRABP1, FCGBP, EPS8 and PROS1. These candidates should help to develop a panel of markers with sufficient sensitivity and specificity for the diagnosis of thyroid tumors in a clinical setting.
Thyroid nodules are extremely common, being palpable in 4% to 7% of the North American adult population, with new nodules detected at a yearly rate of 0.1%.1,2 Currently, fine-needle aspiration biopsy (FNAB) represents the most important initial test for diagnosing malignancy. The result of the FNAB cytology can be classified as benign (70% of cases), malignant (5% to 10%), indeterminate or suspicious (10% to 20%), or nondiagnostic (10% to 15%).3-5 Although nondiagnostic FNABs can be repeated, the indeterminate or suspicious group presents a dilemma for the clinician. In a recent report from our center on 80 patients who underwent thyroid resection for an indeterminate FNAB diagnosis of follicular neoplasm (FN), only 20% were confirmed as malignant.6 Thus, many patients undergo thyroid surgery for nodular disease that is eventually diagnosed as benign disease. Given the diagnostic limitations of FNAB when applied to thyroid tumors, multiple investigators have carried out expression profiling studies with hopes of identifying new diagnostic tools. Such analyses attempt to identify differentially expressed genes with an important role in disease development or progression using large-scale transcript-level expression profiling technologies such as cDNA microarrays,7 oligonucleotide arrays8 and Serial Analysis of Gene Expression (SAGE).9 Typically, dozens or hundreds of genes are identified, many of which are expected to be false positives, and only a small fraction useful as diagnostic/prognostic markers or therapeutic targets. A logical approach to distinguishing important genes from spurious genes, given a large number of candidate gene lists, is to search for the intersection of genes identified in multiple independent studies.10 It is expected that biologically relevant genes will be over-represented and system-specific spurious genes under-represented. As large numbers of cancer profiling studies have become available, the identification of such intersections has become increasingly popular10-12 but none have investigated thyroid cancer specifically. Such studies, although conceptually simple, face a number of technical challenges such as inconsistent gene identifiers, inaccessible data, and uncertain significance of results. Here, we attempt to overcome these challenges. Our approach involves a vote-counting strategy based on the number of studies reporting a gene as differentially expressed and further ranking based on total sample size and average fold-change. Similar strategies have been used to show that gene pairs consistently coexpressed in multiple platforms13 or data sets14 are more likely to share a common biologic process. Our objective was to use validation from multiple expression profiling data sets to identify high-confidence, differentially expressed genes as potential biomarkers for thyroid cancer. We present a novel meta-review method for ranking genes on the basis of published evidence, successfully validate our method against a more traditional meta-analysis approach, and provide a large number of highly significant multistudy genes. Such markers should prove a useful resource for further study by high-throughput molecular analytic techniques.
Data Collection and Curation Published lists of differentially expressed genes were processed to obtain the following information (wherever possible): unique identifier (probe/tag/accession); gene name/description; gene symbol; comparison conditions; sample numbers for each condition; fold-change; direction of change; and PubMed ID. All abbreviations used for sample descriptions are defined in Table 1.
Gene Mapping The National Center for Biotechnology Information's Entrez gene identifier was chosen as the common target identifier for the overlap analysis. SAGE tags were mapped to transcripts by the first position (3'-most NlaIII anchoring enzyme recognition site), sense-strand tag predicted from Refseq15 or MGC16 sequences and then mapped to Entrez using the DiscoverySpace software package (Varhol et al, unpublished data; http://www.bcgsc.ca/discoveryspace/). Affymetrix probes were mapped using Affymetrix annotation files (Santa Clara, CA). Clone accession ids were mapped using the DAVID Resource (http://david.abcc.ncifcrf.gov/).17 If no tag, probe, or accession ID was available, the entry was mapped using gene symbol or gene synonyms.
Ranking Genes were ranked according to several criteria in the following order of importance: (1) number of comparisons in agreement (ie, listing the same gene as differentially expressed and with a consistent direction of change); (2) total number of samples for comparisons in agreement; and (3) average fold-change reported for comparisons in agreement. Total sample size was considered more important than average fold-change because many studies do not report a fold-change. Therefore, average fold-change was based solely on the subset of studies for which a fold-change value was available.
Assessment of Significance
Meta-Analysis of Affymetrix Data Additional information on methods appears in the Appendix (online only).
A total of 34 comparisons were available from 21 studies, utilizing 10 different expression platforms (Table 2). Of the 1,785 genes reported as differentially expressed in these studies (827 up- and 958 downregulated), 1,562 could be mapped to an Entrez gene identifier (723 up- and 839 downregulated). In all overlap analysis groups considered except for one, we identified genes that were reported in multiple studies with a level of overlap found to be significant by Monte Carlo simulation (P < .05; Table 3). The cancer-versus-noncancer group is provided as an example. In this case, a total of 755 genes were reported from 21 comparisons, and of these, 107 genes were reported more than once with consistent fold-change direction (Fig 1). In some cases (MET, TFF3, and SERPINA1), genes were independently reported as many as six times with a consistent fold-change direction. Only 18 genes were found to be reported in multiple studies with inconsistent fold-change. This in itself is an encouraging result. Given that approximately equal numbers of genes were reported as up- versus downregulated (723 up, 839 down) we might expect that multistudy genes with inconsistent fold-change direction would be as common as (or more common than) genes with consistent direction (under random expectation). Instead, we see that in most cases (85.6%), studies that report the same gene agree on the direction, even for large numbers of studies.
The total amount of overlap observed was assessed by Monte Carlo simulation and found to be highly significant (P < .0001; 10,000 permutations). In the simulation, an average of 18.2 (95% CI, 18.12 to 18.28) genes were observed with an overlap of two (same gene identified in two comparisons) compared with 68 in the actual data. For overlap of three, only 0.3 (95% CI, 0.29 to 0.31) genes were observed on average compared with 27 for real data. In 10,000 permutations, the simulated data never produced an overlap greater than three, whereas real data identified 12 genes with overlap of four, five, or six. The probability of observing one or more genes with an overlap of two or more was P = .99. For overlap of three or more P = .037, and for four or more P < .0001. The total number of genes with overlap of two was still highly significant, but we expect at least some false positives to occur by chance. Therefore, we have provided only those genes (top 39) with overlap of three or more and consider those with four or more to be the most reliable (Table 4).
If the cancer-versus-noncancer group is broken into two categories, cancer versus normal and cancer versus benign, we find that most of the top genes were found in both types of comparisons. A small number of genes were found in only one of the two categories. A comparison of genes with multistudy evidence based on published lists versus the smaller subset reanalyzed from raw Affymetrix microarray data showed a highly significant level of agreement (P < .0001). The 107 cancer-versus-noncancer multistudy genes showed a concordance of 0.177 (95% CI, 0.129 to 0.225) with the 179 multistudy genes identified from the reanalyzed Affymetrix subset (Fig 2). In total, there were 43 genes identified by both methods. Given that the two lists of genes were produced by very different subsets of data, in addition to the potential differences in processing, this was an encouraging result. However, it does appear that reprocessing the microarray data in a consistent manner would certainly alter the results and would likely increase the total number of multistudy genes.
Additional information on results appears in the Appendix (online only).
A common criticism of expression profiling studies is a lack of agreement between studies. However, by applying our meta-review method to a large number of published studies, we observe that many genes are consistently reported at a highly significant rate. These genes may represent real biologic effects that, through repeated efforts, have overcome the issues of noise and error typically associated with such experiments. A comparison of our meta-review method (using published gene lists) to a meta-analysis of a smaller subset of studies (for which raw data were available) showed a strong level of concordance. Thus, we believe our approach represents a useful alternative for identifying consistent gene expression markers when raw data are unavailable (as is generally the case). However, a limitation of our method resulting from unavailability of raw data is that we are unable to assign a measure of confidence at the gene level. We can identify consistently reported genes and rank them according to simple criteria such as total sample size and average fold-change, but we can not calculate a true combined fold-change or P value. In order for more powerful meta-analysis methods to be applied researchers must provide access to their raw data. Also, we remind the reader that although we have focused on the cancer-versus-noncancer comparisons, a large number of other comparison groups were analyzed (Table 3). As a means of further assessing our results, we review the top 12 cancer-versus-noncancer candidates to identify which markers have been previously confirmed as differentially expressed or having diagnostic/prognostic utility in thyroid cancer (Table 5). In total, 10 of 12 markers have been confirmed at the RNA level and six of these have gone on to be validated at the protein level. For discussion purposes we have broken the genes into two categories, well-characterized and novel or uncharacterized. We also compare our results to a previous review of promising thyroid biomarkers.
We defined well-characterized genes as those that have been validated in more than one follow-up study and at both the RNA and protein level, such as MET, TFF3, SERPINA1, TIMP1, FN1, and TPO. Several studies have implicated MET protein expression in thyroid cancer as both a diagnostic tool24-28 and prognostic tool.24,26-28 Increased MET expression has been associated with higher risk for metastasis26 and recurrence in PTC26,27 and negative prognosis in FTC.28 However, in another study, decreased MET was shown to be an effective predictor of distant metastases among PTC cases.24 Although no reports have evaluated TFF3 at the protein level, numerous studies have suggested TFF3 as a useful biomarker at the RNA level.23,29-33 A two-gene panel of SFTPB and TFF3 was shown to correctly diagnose PTC with a sensitivity of 88.9%, specificity of 96.7%, and accuracy of 94.9%30 and TFF3/LGALS3 mRNA ratio was shown to distinguish FA from FTC with sensitivity and specificity of 80.0% and 91.5% respectively.33 An antibody study of SERPINA1 reported immunoreactivity in nine of 10 PTCs with no staining in the adjacent normal thyroid tissues.34 TIMP1 upregulation was confirmed by immunohistochemistry (IHC) with positive immunostaining in 68% of PTC cases and none of the normal cases.35 Another IHC study of TIMP1 for 86 PTC specimens showed increased immunoreactivity in the tumor regions versus nontumor regions in 92% cases and significant correlations with unfavorable prognostic variables.36 FN1 has been proposed as a useful reverse transcriptase (RT-) polymerase chain reaction (PCR) marker of differentiated thyroid cancer (DTC)37 and an important modulator of thyroid cell adhesiveness and neoplastic cell growth.38 An IHC study of 85 FTCs and 21 FAs reported that coexpression of FN1 and GAL3 or FN1 and HBME1 was restricted to cancer, although their concurrent absence was highly specific for benign lesions (96%).39 A large number of studies have investigated TPO as a marker for thyroid carcinoma. Lazar et al40 found that higher thyroid cancer stage was associated with lower TPO mRNA expression. Segev et al41 reviewed five IHC studies involving nearly 400 follicular lesions and found that 93% of FAs and 97% of FTCs were accurately diagnosed by TPO antibody staining. Studies using FNAB samples, however, have proved less promising with false-positive rates as high as 32%.41 For the most part, the six genes reviewed above appear promising as thyroid cancer candidates and suggest our meta-analysis method is producing reasonable results. For four genes (TGFA, QPCT, CRABP1, and FCGBP) we could find only a single follow-up study or validation experiment confirming their potential importance in thyroid cancer. Bergstrom et al42 suggest that increased expression of TGFA may be responsible for aberrant activation of epidermal growth factor receptor and ultimately an overexpression and activation of MET. Jarzab et al43 built a classifier capable of discriminating between PTC and nonmalignant samples in 90% of cases. This classifier included QPCT (along with 18 other genes). QPCT was considered a novel gene and was validated by quantitative PCR in that study, but has been studied little further since. CRABP1 downregulation was confirmed by RT-PCR (in one of the original microarray studies),31 and another study reported that hypermethylation of promoter CpG islands for CRABP1 in PTC may explain the reduced expression.44 Differential expression of FCGBP was confirmed in a separate study by restriction-mediated differential display and real-time RT-PCR.45 For two genes (EPS8 and PROS1) we could find no confirmation beyond the initial microarray experiment. In our meta-analysis, five studies identified EPS823,43,46-48 and four identified PROS123,46,48,49 as upregulated in comparisons of cancer with noncancer. And yet, to our knowledge, no follow-up study has confirmed either of these genes (even at the RNA level). It is unclear whether genes such as EPS8 and PROS1 have not been further validated because they are false positives or simply because they have not yet been chosen for further study. These genes and the other less characterized candidates may represent novel diagnostic markers for thyroid cancer and warrant further investigation. Comparison to a previous meta-review by Segev et al41 of mainly single-gene, protein-level thyroid cancer studies found that four of their 12 markers identified as promising preoperative diagnostic markers were identified as high-ranking candidates (top 30) in our meta-analysis (TPO, CD44, KRT19, and LGALS3). Two of their candidates were either not represented (HBME-1) or can not be reliably assayed by the microarray platforms (RET/PTC rearrangements). However, six other promising markers (CDKN1B, TERT, CP/LTF, DLGAP4, HMGA1, and PAX8) do have representation on at least some of the expression platforms, and yet were not identified as differentially expressed in even a single study in our meta-analysis. These genes may have displayed some differential expression but not reached the required thresholds for inclusion in the published lists. Or, they may represent cases in which changes in RNA levels do not correlate well with changes in protein levels. Segev et al41 concluded that large-scale thyroid tumor expression profiling of multiple markers on tumors from large and diverse patient cohorts are still required to identify a panel of markers with sufficient sensitivity and specificity to accurately diagnose indeterminate thyroid lesions. We agree and believe that our meta-review of thyroid cancer gene expression profiling studies provides a high-quality list of candidates from which to identify such a panel.
Materials and Methods Data collection and curation. Lists of differentially expressed genes were collected and curated from publications, supplementary Web pages, or files provided by authors. The following information was recorded wherever possible: Unique identifier (probe/tag/accession); gene name/description; gene symbol; comparison conditions; sample numbers for each condition; fold-change; direction of change; and PubMed ID. For consistency, all fold-change values were converted to whole-number fold-changes with ± sign indicating direction of change. For example a 0.5 fold-change for cond1/cond2 was converted to –2.0. If signal log ratio (SLR) was provided, this was also converted (2SLR = fold-change). Individual data tables were stored in separate files and then combined into a single master file. In one study, Giordano et al (Oncogene 24:6646-6656, 2005) were interested in identifying expression profiles that correlated with specific mutational status (eg, BRAF V600E point mutation, RET/PTC rearrangement, and so on). The comparisons conducted were unique to this study and not applicable for our overlap analysis. Therefore we reanalyzed their data using the log-normalized data file to calculate differentially expressed genes between cancer (PTC) and noncancer (normal) samples. A total of 90 upregulated and 151 downregulated genes were chosen with a fold-change of greater than 2.0 and Bonferroni corrected P value (t test, two-sided) of less than .001. Gene mapping. To determine the amount of overlap between the various published studies, it was first critical to obtain common, consistent gene identifiers. Entrez gene identifier from NCBI was chosen as the target identifier. SAGE tags were mapped to genes by the first position (3'-most NlaIII anchoring enzyme recognition site), sense-strand tag predicted from Refseq (Pruitt KD, Katz KS, Sicotte H, et al. Trends Genet 16:44-47, 2000) or MGC (Mammalian Gene Collection Program Team, Strausberg RL, Feingold EA, et al. Proc Natl Acad Sci U S A 99:16899-16903, 2002) sequences and then mapped to Entrez using the DiscoverySpace software package (Varhol et al, unpublished data, http://www.bcgsc.ca/discoveryspace/). Only unambiguous mappings were allowed. Affymetrix probes were mapped to Entrez gene ids using the Affymetrix annotation files (Santa Clara, CA). Clone accession IDs were mapped to Entrez gene ids using the DAVID Resource (http://david.abcc.ncifcrf.gov/; Dennis G Jr, Sherman BT, Hosack DA, et al. Genome Biol 4:P3, 2003). If no tag, probe, or accession ID was available, the entry was mapped from gene symbol directly to Entrez or indirectly using gene synonyms. In 43 cases, Entrez gene IDs were manually determined on the basis of gene description. Any genes still not mapped were assigned an Entrez ID of "NA." Gene history was checked to identify Entrez IDs that have been retired or replaced. If retired, Entrez ID was changed to NA. If replaced, the new replacement Entrez ID was used. Ranking. Each published study consists of one or more comparisons between pairs of conditions (eg, PTC v normal). A method of ranking potential molecular markers was devised for each comparison subset of interest. A comparison subset refers to a list of comparisons for which the two conditions address a question of interest. For example, to identify markers that have consistently been shown to distinguish cancer samples from noncancer samples (normal or benign) we would first define two condition categories. The cancer category would include samples labeled as PTC, FTC, ATC, and so on. The noncancer category would include samples labeled as normal, GT, FA, and so on. From these categories, we identify all studies/comparisons that contrasted samples from the cancer category against the noncancer. Each study/comparison yields a list of genes found to have been differentially expressed between the conditions. It should be noted that two studies by Finley et al (Finley DJ, Arora N, Zhu B, et al. J Clin Endocrinol Metab 89:3214-3223, 2004; and Finley DJ, Zhu B, Barden CB, et al. Ann Surg 240:425-437, 2004) appear to have a high amount of redundancy between the actual samples analyzed for gene expression. Therefore, to prevent artificially high overlap between these data sets, only one or the other was included in each comparison group (depending on the nature of the comparison). Genes are ranked according to several criteria in the following order of importance: (1) number of comparisons in agreement (ie, listing the same gene as differentially expressed and with a consistent direction of change); (2) total number of samples for comparisons in agreement; and (3) average fold-change reported for comparisons in agreement. Our primary ranking criteria is thus simply the number of independent studies that have identified the same gene as a potential marker of some condition. Sample sizes and fold-changes are used to further rank genes with evidence from equal numbers of studies/comparisons. Total sample size was considered more important than average fold-change because many studies do not report a fold-change. The average fold-change was based solely on those studies for which a fold-change value was provided in the original publication. Also, it should be noted that the average fold-change is based on all reported fold-changes irrespective of the method used to calculate them (normalization, logging, etc). In most cases fold-changes are calculated as a ratio of mean expression in one condition versus another. But, in other cases where matched normal and tumor patient samples were available it could represent the median fold-change observed for all individual tumor/normal ratios (Jarzab B, Wiench M, Fujarewicz K, et al. Cancer Res 65:1587-1597, 2005). Despite this caveat, we believe that the average and range of reported fold-changes gives a good idea of the relative magnitude of differential expression for each gene of interest. Assessment of significance. Significance of the observed level of overlap between studies for a comparison subset was assessed by Monte Carlo simulation. Where possible, the actual gene lists produced by mapping each expression technology to Entrez gene ID were utilized. For studies with custom arrays (Arnaldi LA, Borra RC, Maciel RM, et al. Thyroid 15:210-221, 2005; Chevillard S, Ugolin N, Vielh P, et al. Clin Cancer Res 10:6586-6597, 2004; Onda M, Emi M, Yoshida A, et al. Endocr Relat Cancer 11:843-854, 2004; and Yano Y, Uematsu N, Yashiro T, et al. Clin Cancer Res 10:2035-2043, 2004), a comparable number of genes was chosen from the combined gene list of all other platforms. Because the custom arrays may in reality have features/genes unique to the array; this simplifying step may actually increase the chance of observing overlap in the random simulations. Thus, the final P value is likely an overestimate. For most platforms, mapping of features to genes was not perfect. On average, 91.4% of features were successfully mapped to a gene identifier. Thus, for the custom arrays, we randomly chose this same fraction of genes. For SAGE, three thyroid libraries from the Cancer Genome Anatomy Project (normal: SAGE_Thyroid_normal_B_001, benign: SAGE_Thyroid_follicular_adenoma_B_TT005, carcinoma: SAGE_Thyroid_follicular_carcinoma_B_TT004) were used to create a realistic total tag set and then mapped to Entrez as above. Once total gene lists were created for each platform type, we randomly chose gene sets of the same size observed in our review of the literature. For example, if a study/comparison reported 17 upregulated and 40 downregulated genes we would randomly choose 17 and 40 genes and label them "up" and "down," respectively. The random selection was repeated for all comparisons in the comparison subset. Finally, the amount of overlap between comparisons was tallied and the result compared to that actually observed. This random permutation process was repeated 10,000 times, and an estimate of significance determined. Meta-analysis of Affymetrix data. The method presented in the preceding paragraphs makes use of reported lists of differentially expressed genes from published literature. An obvious disadvantage of this approach is that each publication may make use of different methods to ascertain differential expression (eg, scaling, filtering, normalization, significance thresholds, P value estimation, multiple testing corrections, etc). Collecting and reanalyzing 21 sets of raw data from 10 different platforms in a consistent manner would be an immense task and most likely impossible, because many raw data sets are unavailable. The majority of data have not been placed in public databases. However, to assess our method, we did reanalyze a subset of data from raw image files using a standard methodology and compared with our results. The Affymetrix platforms were chosen for their ease of analysis and because they were most represented in the published studies (nine studies). Two of the data sets were freely available on the Internet (Giordano TJ, Kuick R, Thomas DG, et al. Oncogene 24:6646-6656, 2005; and Huang Y, Prasad M, Lemon WJ, et al. Proc Natl Acad Sci U S A 98:15044-15049, 2001), and the other seven were requested by e-mail. Ultimately, we were able to obtain only four data sets (two requests were successful) representing five comparisons with a total of 117 samples (65 PTC, 15 FTC, 25 normal, and 12 FA; Giordano et al; Huang et al; Aldred MA, Huang Y, Liyanarachchi S, et al. J Clin Oncol 22:3531-3359, 2004; and Weber F, Shen L, Aldred MA, et al. J Clin Endocrinol Metab 2005).
Cel files were loaded and analyzed with the DChip software. Arrays were normalized and modeled using default settings. Probes were filtered based on variation (0.50 < standard deviation/mean < 1000.00) and P call across samples ( Gene Ontology analysis. A Gene Ontology (Ashburner M, Ball CA, Blake JA, et al. Nat Genet 25:25-29, 2000) analysis was performed for the genes with multistudy confirmation in the cancer-versus-noncancer overlap analysis group using the BiNGO (Maere S, Heymans K, Kuiper M. Bioinformatics 21:3448-3349, 2005) plug-in for the Cytoscape (Shannon P, Markiel A, Ozier O, et al. Genome Res 13:2498-2504, 2003) software package. Significance was calculated using the hypergeometric test, corrected with a Benjamini & Hochberg FDR correction, and a cutoff of 0.05 applied to the result.
Results Gene Ontology analysis. A gene ontology analysis of multistudy genes from the cancer-versus-noncancer overlap analysis group identified 12 significant terms (Table A1). Genes tended to be localized in the extracellular region and, more specifically, the extracellular matrix. Significant biologic processes involved hormone metabolism and generation. Significant molecular functions involved binding (cadmium, selenium, and copper), mitogen-activated protein kinase phosphatase activity, and retinoic acid receptor activity.
The authors indicated no potential conflicts of interest.
We would like to acknowledge the generosity of the researchers who shared their data with us through personal communications including, Frank Weber, Carrie Ris-Stalpers, Charis Eng, Sandya Liyanarachchi, and Twyla Pohar.
Supported by the BC Cancer Foundation. Canadian Institutes of Health Research (O.L.G), the Natural Sciences and Engineering Council of Canada, and the Michael Smith Foundation for Health Research (S.J.M.J. and S.M.W.). Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
1. Gharib H, Goellner JR: Fine-needle aspiration biopsy of the thyroid: An appraisal. Ann Intern Med 118:282-289, 1993 2. Greenlee RT, Hill-Harmon MB, Murray T, et al: Cancer statistics, 2001. CA Cancer J Clin 51:15-36, 2001 3. Goellner JR, Gharib H, Grant CS, et al: Fine needle aspiration cytology of the thyroid, 1980 to 1986. Acta Cytol 31:587-590, 1987[Medline] 4. Caraway NP, Sneige N, Samaan NA: Diagnostic pitfalls in thyroid fine-needle aspiration: A review of 394 cases. Diagn Cytopathol 9:345-350, 1993[Medline] 5. Ravetto C, Colombo L, Dottorini ME: Usefulness of fine-needle aspiration in the diagnosis of thyroid carcinoma: A retrospective study in 37,895 patients. Cancer 90:357-363, 2000[CrossRef][Medline] 6. Wiseman SM, Baliski C, Irvine R, et al: Hemithyroidectomy: The optimal initial surgical approach for individuals undergoing surgery for a cytological diagnosis of follicular neoplasm. Ann Surg Oncol 13:425-432, 2006 7. Schena M, Shalon D, Davis RW, et al: Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270:467-470, 1995 8. Lockhart DJ, Dong H, Byrne MC, et al: Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat Biotechnol 14:1675-1680, 1996[CrossRef][Medline] 9. Velculescu VE, Zhang L, Vogelstein B, et al: Serial analysis of gene expression. Science 270:484-487, 1995 10. Rhodes DR, Yu J, Shanker K, et al: Large-scale meta-analysis of cancer microarray data identifies common transcriptional profiles of neoplastic transformation and progression. Proc Natl Acad Sci U S A 101:9309-9314, 2004 11. Cahan P, Ahmad AM, Burke H, et al: List of lists-annotated (LOLA): A database for annotation and comparison of published microarray gene lists. Gene 360:78-82, 2005[CrossRef][Medline] 12. Shih W, Chetty R, Tsao MS: Expression profiling by microarrays in colorectal cancer. Oncol Rep 13:517-524, 2005[Medline] 13. Griffith OL, Pleasance ED, Fulton DL, et al: Assessment and integration of publicly available SAGE, cDNA microarray, and oligonucleotide microarray expression data for global coexpression analyses. Genomics 86:476-488, 2005[CrossRef][Medline] 14. Lee HK, Hsu AK, Sajdak J, et al: Coexpression analysis of human genes across many microarray data sets. Genome Res 14:1085-1094, 2004 15. Pruitt KD, Katz KS, Sicotte H, et al: Introducing RefSeq and LocusLink: Curated human genome resources at the NCBI. Trends Genet 16:44-47, 2000[CrossRef][Medline] 16. Mammalian Gene Collection Program Team, Strausberg RL, Feingold EA, et al: Generation and initial analysis of more than 15,000 full-length human and mouse cDNA sequences. Proc Natl Acad Sci U S A 99:16899-16903, 2002 17. Dennis G Jr, Sherman BT, Hosack DA, et al: DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol 4:P3, 2003[CrossRef][Medline] 18. Arnaldi LA, Borra RC, Maciel RM, et al: Gene expression profiles reveal that DCN, DIO1, and DIO2 are underexpressed in benign and malignant thyroid tumors. Thyroid 15:210-221, 2005[CrossRef][Medline] 19. Chevillard S, Ugolin N, Vielh P, et al: Gene expression profiling of differentiated thyroid neoplasms: Diagnostic and clinical implications. Clin Cancer Res 10:6586-6597, 2004 20. Onda M, Emi M, Yoshida A, et al: Comprehensive gene expression profiling of anaplastic thyroid cancers with cDNA microarray of 25 344 genes. Endocr Relat Cancer 11:843-854, 2004 21. Yano Y, Uematsu N, Yashiro T, et al: Gene expression profiling identifies platelet-derived growth factor as a diagnostic molecular marker for papillary thyroid carcinoma. Clin Cancer Res 10:2035-2043, 2004 22. Strausberg RL, Buetow KH, Greenhut SF, et al: The cancer genome anatomy project: Online resources to reveal the molecular signatures of cancer. Cancer Invest 20:1038-1050, 2002[CrossRef][Medline] 23. 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 24. Belfiore A, Gangemi P, Costantino A, et al: Negative/low expression of the Met/hepatocyte growth factor receptor identifies papillary thyroid carcinomas with high risk of distant metastases. J Clin Endocrinol Metab 82:2322-2328, 1997 25. Ippolito A, Vella V, La Rosa GL, et al: Immunostaining for Met/HGF receptor may be useful to identify malignancies in thyroid lesions classified suspicious at fine-needle aspiration biopsy. Thyroid 11:783-787, 2001[CrossRef][Medline] 26. Ramirez R, Hsu D, Patel A, et al: Over-expression of hepatocyte growth factor/scatter factor (HGF/SF) and the HGF/SF receptor (cMET) are associated with a high risk of metastasis and recurrence for children and young adults with papillary thyroid carcinoma. Clin Endocrinol (Oxf) 53:635-644, 2000[CrossRef][Medline] 27. Inaba M, Sato H, Abe Y, et al: Expression and significance of c-met protein in papillary thyroid carcinoma. Tokai J Exp Clin Med 27:43-49, 2002[Medline] 28. Di Renzo MF, Olivero M, Ferro S, et al: Overexpression of the c-MET/HGF receptor gene in human thyroid carcinomas. Oncogene 7:2549-2553, 1992[Medline] 29. Finley DJ, Arora N, Zhu B, et al: Molecular profiling distinguishes papillary carcinoma from benign thyroid nodules. J Clin Endocrinol Metab 89:3214-3223, 2004 30. Hamada A, Mankovskaya S, Saenko V, et al: Diagnostic usefulness of PCR profiling of the differentially expressed marker genes in thyroid papillary carcinomas. Cancer Lett 224:289-301, 2005[CrossRef][Medline] 31. Hawthorn L, Stein L, Varma R, et al: TIMP1 and SERPIN-A overexpression and TFF3 and CRABP1 underexpression as biomarkers for papillary thyroid carcinoma. Head Neck 26:1069-1083, 2004[CrossRef][Medline] 32. Takano T, Miyauchi A, Yoshida H, et al: High-throughput differential screening of mRNAs by serial analysis of gene expression: Decreased expression of trefoil factor 3 mRNA in thyroid follicular carcinomas. Br J Cancer 90:1600-1605, 2004[CrossRef][Medline] 33. Takano T, Miyauchi A, Yoshida H, et al: Decreased relative expression level of trefoil factor 3 mRNA to galectin-3 mRNA distinguishes thyroid follicular carcinoma from adenoma. Cancer Lett 219:91-96, 2005[CrossRef][Medline] 34. Poblete MT, Nualart F, del Pozo M, et al: Alpha 1-antitrypsin expression in human thyroid papillary carcinoma. Am J Surg Pathol 20:956-963, 1996[CrossRef][Medline] 35. Wasenius VM, Hemmer S, Kettunen E, et al: Hepatocyte growth factor receptor, matrix metalloproteinase-11, tissue inhibitor of metalloproteinase-1, and fibronectin are up-regulated in papillary thyroid carcinoma: A cDNA and tissue microarray study. Clin Cancer Res 9:68-75, 2003 36. Maeta H, Ohgi S, Terada T: Protein expression of matrix metalloproteinases 2 and 9 and tissue inhibitors of metalloproteinase 1 and 2 in papillary thyroid carcinomas. Virchows Arch 438:121-128, 2001[CrossRef][Medline] 37. Hesse E, Musholt PB, Potter E, et al: Oncofoetal fibronectin–a tumour-specific marker in detecting minimal residual disease in differentiated thyroid carcinoma. Br J Cancer 93:565-570, 2005[CrossRef][Medline] 38. Liu W, Asa SL, Ezzat S: 1alpha,25-Dihydroxyvitamin D3 targets PTEN-dependent fibronectin expression to restore thyroid cancer cell adhesiveness. Mol Endocrinol 19:2349-2357, 2005 39. Prasad ML, Pellegata NS, Huang Y, et al: Galectin-3, fibronectin-1, CITED-1, HBME1 and cytokeratin-19 immunohistochemistry is useful for the differential diagnosis of thyroid tumors. Mod Pathol 18:48-57, 2005[CrossRef][Medline] 40. Lazar V, Bidart JM, Caillou B, et al: Expression of the Na+/I- symporter gene in human thyroid tumors: A comparison study with other thyroid-specific genes. J Clin Endocrinol Metab 84:3228-3234, 1999 41. Segev DL, Clark DP, Zeiger MA, et al: Beyond the suspicious thyroid fine needle aspirate: A review. Acta Cytol 47:709-722, 2003[Medline] 42. Bergstrom JD, Westermark B, Heldin NE: Epidermal growth factor receptor signaling activates met in human anaplastic thyroid carcinoma cells. Exp Cell Res 259:293-299, 2000[CrossRef][Medline] 43. Jarzab B, Wiench M, Fujarewicz K, et al: Gene expression profile of papillary thyroid cancer: Sources of variability and diagnostic implications. Cancer Res 65:1587-1597, 2005 44. Huang Y, de la Chapelle A, Pellegata NS: 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] 45. O'Donovan N, Fischer A, Abdo EM, et al: Differential expression of IgG Fc binding protein (FcgammaBP) in human normal thyroid tissue, thyroid adenomas and thyroid carcinomas. J Endocrinol 174:517-524, 2002[Abstract] 46. Finley DJ, Zhu B, Barden CB, et al: Discrimination of benign and malignant thyroid nodules by molecular profiling. Ann Surg 240:425-437, 2004[Medline] 47. Weber F, Shen L, Aldred MA, et al: Genetic classification of benign and malignant thyroid follicular neoplasia based on a 3-gene combination. J Clin Endocrinol Metab 90:2512-2521, 2005 48. Giordano TJ, Kuick R, Thomas DG, et al: Molecular classification of papillary thyroid carcinoma: Distinct BRAF, RAS, and RET/PTC mutation-specific gene expression profiles discovered by DNA microarray analysis. Oncogene 24:6646-6656, 2005[CrossRef][Medline] 49. Barden CB, Shister KW, Zhu B, et al: Classification of follicular thyroid tumors by molecular signature: Results of gene profiling. Clin Cancer Res 9:1792-1800, 2003 50. Chen KT, Lin JD, Chao TC, et al: Identifying differentially expressed genes associated with metastasis of follicular thyroid cancer by cDNA expression array. Thyroid 11:41-46, 2001[CrossRef][Medline] 51. Aldred MA, Huang Y, Liyanarachchi S, et al: Papillary and follicular thyroid carcinomas show distinctly different microarray expression profiles and can be distinguished by a minimum of five genes. J Clin Oncol 22:3531-3539, 2004 52. Cerutti JM, Delcelo R, Amadei MJ, et al: A preoperative diagnostic test that distinguishes benign from malignant thyroid carcinoma based on gene expression. J Clin Invest 113:1234-1242, 2004[CrossRef][Medline] 53. Eszlinger M, Krohn K, Paschke R: Complementary DNA expression array analysis suggests a lower expression of signal transduction proteins and receptors in cold and hot thyroid nodules. J Clin Endocrinol Metab 86:4834-4842, 2001 54. Zou M, Famulski KS, Parhar RS, et al: Microarray analysis of metastasis-associated gene expression profiling in a murine model of thyroid carcinoma pulmonary metastasis: Identification of S100A4 (Mts1) gene overexpression as a poor prognostic marker for thyroid carcinoma. J Clin Endocrinol Metab 89:6146-6154, 2004 55. Mazzanti C, Zeiger MA, Costouros NG, et al: Using gene expression profiling to differentiate benign versus malignant thyroid tumors. Cancer Res 64:2898-2903, 2004 56. Takano T, Hasegawa Y, Matsuzuka F, et al: Gene expression profiles in thyroid carcinomas. Br J Cancer 83:1495-1502, 2000[CrossRef][Medline] 57. Pauws E, Veenboer GJ, Smit JW, et al: Genes differentially expressed in thyroid carcinoma identified by comparison of SAGE expression profiles. Faseb J 18:560-561, 2004 58. Scarpino S, Cancellario d'Alena F, Di Napoli A, et al: Increased expression of Met protein is associated with up-regulation of hypoxia inducible factor-1 (HIF-1) in tumour cells in papillary carcinoma of the thyroid. J Pathol 202:352-358, 2004[CrossRef][Medline] 59. Nardone HC, Ziober AF, LiVolsi VA, et al: C-Met expression in tall cell variant papillary carcinoma of the thyroid. Cancer 98:1386-1393, 2003[CrossRef][Medline] 60. Mineo R, Costantino A, Frasca F, et al: Activation of the hepatocyte growth factor (HGF)-Metsystem in papillary thyroid cancer: Biological effects of HGF in thyroid cancer cells depend on Met expression levels. Endocrinology 145:4355-4365, 2004 61. Shi Y, Parhar RS, Zou M, et al: Tissue inhibitor of metalloproteinases-1 (TIMP-1) mRNA is elevated in advanced stages of thyroid carcinoma. Br J Cancer 79:1234-1239, 1999[CrossRef][Medline] 62. Umeki K, Tanaka T, Yamamoto I, et al: Differential expression of dipeptidyl peptidase IV (CD26) and thyroid peroxidase in neoplastic thyroid tissues. Endocr J 43:53-60, 1996[Medline] Submitted March 27, 2006; accepted August 30, 2006. This article has been cited by other articles:
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
TOP
ABSTRACT
INTRODUCTION
20% of arrays). Probes were determined to be differentially expressed if they demonstrated fold-change greater than 2 and P value less than .05 (after false-discovery rate [FDR] –based multiple testing correction). Finally, the five comparisons (three PTC v normal; one FTC versus normal; and one FTC v FA) were analyzed for overlapping genes as herein and the results compared to the cancer-versus-noncancer comparison analysis for concordance using the LOLA tool (Cahan P, Ahmad AM, Burke H, et al. Gene 360:78-82, 2005). 
, and 
were used to identify which part of the Venn diagram each gene in 
