Molecular differential diagnosis of follicular thyroid carcinoma and adenoma based on gene expression profiling by using formalin-fixed paraffin-embedded tissues
- Aleksandra Pfeifer†1, 5,
- Bartosz Wojtas†1,
- Malgorzata Oczko-Wojciechowska1,
- Aleksandra Kukulska1,
- Agnieszka Czarniecka2,
- Markus Eszlinger6,
- Thomas Musholt7,
- Tomasz Stokowy1, 5,
- Michal Swierniak1, 8,
- Ewa Stobiecka4,
- Dagmara Rusinek1,
- Tomasz Tyszkiewicz1,
- Monika Kowal1,
- Michal Jarzab3,
- Steffen Hauptmann9,
- Dariusz Lange4,
- Ralf Paschke6 and
- Barbara Jarzab1Email author
© Pfeifer et al.; licensee BioMed Central Ltd. 2013
Received: 3 April 2013
Accepted: 30 September 2013
Published: 7 October 2013
Differential diagnosis between malignant follicular thyroid cancer (FTC) and benign follicular thyroid adenoma (FTA) is a great challenge for even an experienced pathologist and requires special effort. Molecular markers may potentially support a differential diagnosis between FTC and FTA in postoperative specimens. The purpose of this study was to derive molecular support for differential post-operative diagnosis, in the form of a simple multigene mRNA-based classifier that would differentiate between FTC and FTA tissue samples.
A molecular classifier was created based on a combined analysis of two microarray datasets (using 66 thyroid samples). The performance of the classifier was assessed using an independent dataset comprising 71 formalin-fixed paraffin-embedded (FFPE) samples (31 FTC and 40 FTA), which were analysed by quantitative real-time PCR (qPCR). In addition, three other microarray datasets (62 samples) were used to confirm the utility of the classifier.
Five of 8 genes selected from training datasets (ELMO1, EMCN, ITIH5, KCNAB1, SLCO2A1) were amplified by qPCR in FFPE material from an independent sample set. Three other genes did not amplify in FFPE material, probably due to low abundance. All 5 analysed genes were downregulated in FTC compared to FTA. The sensitivity and specificity of the 5-gene classifier tested on the FFPE dataset were 71% and 72%, respectively.
The proposed approach could support histopathological examination: 5-gene classifier may aid in molecular discrimination between FTC and FTA in FFPE material.
KeywordsFollicular thyroid adenoma Follicular thyroid cancer Gene expression Microarray Formalin-fixed paraffin-embedded blocks
Discrimination between malignant follicular thyroid cancer (FTC) and benign follicular thyroid adenoma (FTA) is the most difficult aspect of thyroid pathology. Postsurgical (post-thyroid unilateral lobectomy) FTC and FTA management algorithms are different: only cancer patients require completion total thyroidectomy, adjuvant radioiodine treatment, and long-term follow-up. The histological diagnosis of FTC remains a challenge for pathologists, as the diagnostic criteria of FTC, namely capsular invasion or angioinvasion, are prone to serious inter-observer variability . Thus, the discrimination of FTC from FTA is an important clinical problem, particularly for minimally invasive cases, and depends on the number of serial sections and tumour regions examined .
Several mutations play an important role in the biology of follicular tumours, such as paired box gene 8 (PAX8)/peroxisome proliferator-activated receptor gamma (PPARG) translocation and RAS point mutations. The first one occurs in 35–47% of FTC and up to 13% of FTA [3–5], the second occurs in approximately 20–50% and 19% of FTC and FTA, respectively [6–9].
Numerous single immunohistochemical markers have been proposed, and the most widely accepted single protein that improves diagnostic accuracy is galectin 3, even in the case of minimally invasive follicular carcinoma . Obviously, immunohistochemical panels can be extended by including other proteins e.g., cytokeratin 19 or p27 . Many other potential markers have been considered, such as HBME-1, extracellular matrix metalloproteinase inducer (EMMPRIN), growth arrest and DNA damage-inducible gene 153 (GADD153), thyroid transcription factor 1 (TFF-1), Ki-67, p63, and p53 [12–15], but the conclusions are limited by the relatively small size of tested populations.
Alternatively, the FTC and FTA differentiation problem was investigated in several miRNA-profiling studies. Some miRNAs are described as sensitive biomarkers of malignant and benign follicular thyroid tumours [16–18], but the overlap of differentiating miRNAs pointed out in these studies is limited.
A number of attempts have been made to improve the molecular diagnosis of FTC using specific mRNA signatures of malignant follicular tumours [19–26]. The most promising of these, by Borup and co-workers  was based on a 40-sample microarray dataset and led to the delineation of a 76-gene signature, which was highly sensitive and specific both for their own dataset and when tested with 2 previously published microarray datasets [21, 24]. Simultaneously, a recent publication by Chudova et al. proposed a 167-gene classifier that was able to diagnose thyroid nodules with indeterminate cytology . Although the classifier was trained on various types of thyroid benign and malignant nodules, the prospective multicentre validation study showed good specificity for follicular thyroid nodules . The proposed predictors might be an important step to increase the effectiveness of thyroid nodule diagnosis. The vast majority of recent studies have concentrated on preoperative differential diagnosis [26, 28], with molecular tests applied to fresh-frozen material with an aim to translate it into fine-needle aspiration biopsy specimen testing. This is evidently very important in the context of preoperative diagnosis, but it may not be the most efficient method for direct application to the analysis of formalin-fixed paraffin-embedded (FFPE) samples at the mRNA level, which in our opinion may aid postoperative differential diagnosis in controversial cases. In this study, we aimed to combine the gathered knowledge about the transcriptomes of FTC and FTA to derive a novel classifier of thyroid follicular malignancy applicable to FFPE material.
We therefore used the aforementioned available dataset of Borup et al. for gene pre-selection, further selected genes and trained the classifier on very carefully selected FTC and FTA samples, and verified the classifier by real-time quantitative PCR (qPCR) analysis in formalin-fixed paraffin-embedded (FFPE) samples.
Tumour samples for microarray analysis (fresh-frozen [FF] material) were derived from 27 FTC (median age, 68 years) and 25 FTA (median age, 47 years) patients treated by thyroidectomy in Polish and German centers. The study was approved by the local ethics committees of the MSC Memorial Cancer Center and Institute of Oncology (Gliwice, Poland), University of Leipzig (Leipzig, Germany), University of Halle (Halle, Germany), and Mainz University Hospital (Mainz, Germany) and informed consent was obtained from all the patients.
Tumour samples for further validation (FFPE blocks) were derived from 31 FTC patients (median age of patients, 61 years) and 40 FTA patients (median age, 49 years). All patients were treated by thyroid surgery at MSC Memorial Cancer Center and Institute of Oncology, Gliwice Branch, and samples were subjected to routine histopathological examination at the Department of Pathology between 2006 and 2010.
Description of datasets analyzed in the study (see Additional file 1 for detailed information)
Borup et al. microarray dataset A
Training microarray dataset B
Validation qPCR dataset C
Validation microarray dataset D
Weber et al. validation microarray dataset E1
Hinsch et al. validation microarray dataset E2
RNA for microarray analysis was isolated using the RNeasy Mini kit (Qiagen, Hilden, Germany) after tumour content verification of the specimen by a pathologist. The standard Affymetrix microarray protocol was carried out, and samples were hybridized with the HG-U133 Plus 2.0 microarray. Microarray data analysis was performed in an R/Bioconductor environment. Datasets were pre-processed using the GC Robust Multiarray Average (GCRMA) method . The classifier was developed using the CMA package . Detailed methods are described in supplemental information (Additional file 1).
Validation experiment in FFPE samples
Blocks with sufficient tumour tissue in the specimen (approximately 80%) were selected. RNA was isolated using the FFPE RNeasy Mini Kit (Qiagen) from 5 slices of paraffin blocks selected by a histopathologist. Details are provided in Additional file 1.
Real-time quantitative PCR (qPCR) was carried out for 8 genes selected from the training dataset: carbonic anhydrase IV (CA4), engulfment and cell motility 1 (ELMO1), endomucin (EMCN), inter-alpha-trypsin inhibitor heavy chain family, member 5 (ITIH5), potassium voltage-gated channel, shaker-related subfamily, beta member 1 (KCNAB1), low density lipoprotein receptor-related protein 1B (LRP1B), pleckstrin homology domain containing, family G (with RhoGef domain) member 4B (PLEKHG4B), and solute carrier organic anion transporter family, member 2A1 (SLCO2A1). PCR amplification was performed with Universal Probe Library fluorescent probes (Roche, Basel, Switzerland) and a 5′-nuclease assay, starting with 200 ng of total RNA. Normalization was carried out in the GeNorm application . Details of the methods are provided in Additional file 1.
External microarray data and analysis pipeline
Three microarray datasets were downloaded from gene expression repositories: the datasets of Borup et al. (dataset A, 22 FTA and 18 FTC), Weber et al. (testing dataset E1, 12 FTA and 12 FTC), and Hinsch et al. (testing dataset E2, 4 FTA and 8 FTC). Details are provided in Table 1 and in Additional file 1.
Classifier construction and validation
Genes that exhibited significant differences between FTC and FTA were pre-selected from dataset A. The selection criteria were as follows: Student’s t-test (with equal variances assumed), non-corrected p-value <0.0005 (<0.09 when the p-value was corrected for multiple comparisons using the false discovery rate method), mean gene expression >5 in either the FTC or FTA group, and an absolute log ratio between the groups >1.5. Transcripts not fully annotated were filtered out. When more than one probe set per transcript was found, the probe set with the lowest p-value was selected.
Based on the subset of pre-selected genes, further gene selection using the Student's t-test (equal variances assumed), and classifier training were carried out on dataset B. Diagonal linear discriminant analysis (DLDA), a simple and reliable method based on linear combinations of genes, was chosen as the classification engine. However, other more sophisticated methods of classification were also used for comparison. Details of the analysis are provided in Additional file 1. The accuracy of the classification was assessed on dataset B using 10-fold cross-validation, repeated 10 times.
The 5-gene classifier was validated by qPCR on dataset C. First, the normalized expression data were log2-transformed. Then, the classifier performance was assessed using leave-one-out cross-validation (LOOCV; in each iteration, the DLDA classifier was trained on n-1 samples and tested on the remaining one). All 5 genes (validated by qPCR, details in results) from dataset C were used in each iteration and no further gene selection was used in the cross-validation loops.
Dataset C was also used to create a receiver operating characteristic (ROC) curve and to assess the diagnostic efficacy of the classifier. In the leave-one-out loop, for each sample, we calculated the probability that the sample belongs to the FTC class. Varying the threshold for the probability, the ROC curve was plotted.
An additional validation of the 5-gene classifier was performed using microarray data (datasets D, E1, E2). Both training dataset B and testing dataset D contained microarrays of the same type. Therefore, the final classifier, which had been trained on dataset B, was directly tested on dataset D. This approach was not possible for testing datasets E1 and E2 as they contained different microarray types. Therefore, LOOCV was used to assess the accuracy of the classifier on those 2 datasets. In all validation steps, only 5 genes amplified by qPCR were analyzed in the microarray test datasets and no further gene selection was used in the cross-validation loops.
Comparison of the classifiers
To compare the 5-gene classifier developed by us to other published classifiers (the 76-gene classifier developed by Borup et al. , 3-gene classifier developed by Weber et al. , and 5-gene classifier developed by Foukakis et al. ), we calculated the accuracies of those classifiers in 4 different datasets: dataset B (our own), dataset D (our own), dataset E1 , and dataset E2 .
Borup’s classifier was originally created on dataset A and was based on 76 differentially expressed genes and the SVM method with a radial kernel. To calculate its accuracy on dataset B, we performed a 10-fold cross-validation of such a classifier with a fixed list of 76 genes. To calculate its accuracy on dataset D, we trained this classifier on dataset B and tested it on dataset D. The accuracy of this classifier on datasets E1 and E2 had already been reported by Borup et al. and we used those values for comparison.
To calculate the accuracies of the classifiers developed by Weber and Foukakis, we applied a DLDA classification method. Analogous to the analysis above, to calculate the accuracies of these classifiers on dataset B, we performed a 10-fold cross-validation. To calculate their accuracies on dataset D, we trained those classifiers on dataset B and tested on dataset D. To calculate their accuracies on dataset E1 and E2, we performed LOOCV analysis.
The accuracy of the 5-gene DLDA classifier on dataset B was calculated by 10-fold cross-validation (genes were selected by t-test). The calculation of the accuracy of our classifier on datasets D, E1, and E2 is described above.
Developing a robust classifier for FTC and FTA differentiation
Because we aimed to obtain the most robust classifier possible, and had access to 2 reliable datasets (our own dataset with 26 samples [training dataset B] and the dataset of Borup et al. with 40 samples [dataset A]), we decided to use both of them sequentially for classifier construction. First, from dataset A (large enough to represent the variability of FTC) we selected 99 genes with high significance and large magnitude of difference (see Methods; the genes are listed in Additional file 2, and the raw expression values of these genes in dataset B are attached in Additional file 3). Second, based on the pre-filtered genes, we used dataset B (with histological diagnosis verification for each sample, based on the consensus diagnosis of 2 independent histopathologists) to build the classifier.
Performance measures of classifiers in different datasets
Dataset (origin, method of analysis)
Prevalence of FTC in dataset (%)
B (own, microarray)
DLDA classification based on the 8 best genes chosen from 99 preselected ones.*
DLDA classification based on 45 (optimal number) best genes chosen from 99 preselected ones.*
C (own, FFPE qPCR)
5-gene DLDA classification (cut-off 0.5)**
72(95% CI: 60–82)
67(95% CI: 48–82)
76(95% CI: 60–89)
71(95% CI: 52–86)
72(95% CI: 56–85)
5-gene DLDA classification (cut-off 0.12)**
D (own, microarray)
5-gene DLDA classifier trained on dataset B, tested on D
E1 (Weber et al. microarray)
5-gene DLDA classifier.**
E2 (Hinsch et al. microarray)
5-gene DLDA classifier.**
For the classifier, we selected 8 transcripts that were most significant in the analysis of 99 preselected genes on dataset B: CA4, ELMO1, EMCN, ITIH5, KCNAB1, LRP1B, PLEKHG4B, and SLCO2A1. The classifier built on these genes was referred to as the 8-gene classifier. Although we did not use information regarding the direction of the gene expression change, all of these most significant genes were downregulated in FTC. Detailed information about the significance of these genes in all the datasets used in the paper is included in Additional file 4.
Classifier testing on FFPE material and qPCR
We assessed the expression of all the 8 selected genes in serial dilutions of calibrator RNA (mixture of excellent quality RNA from FTA and FTC). Because of the poor amplification of LRP1B, the gene was excluded from further analysis (insufficient qPCR efficiency). Next, the genes of the 8-gene classifier were analyzed in test dataset C (40 FTA and 31 FTC samples, RNA from FFPE blocks). CA4 and PLEKHG4B did not amplify sufficiently in FFPE samples (CA4 was amplified in 4 FTA and none of the FTC samples, and PLEKHG4B in 3 FTAs and 1 FTC samples; data not shown). This was probably due to the low constitutive expression of these genes or their high susceptibility to degradation (Additional file 1: Figure C). To verify this, we evaluated the expression of these genes by transcriptome sequencing of 2 FTCs and noted that LRP1B, CA4, and PLEKHG4B exhibited low abundance in FTC, compared to other genes (Additional file 1: Figure D). Considering the poor amplification of the 3 genes, we decided to proceed with the 5-gene classifier.
Results for the 8 genes included in the classifier and chosen for qPCR validation on FFPE samples
Signal log ratio (FTC vs. FTA)
Training dataset B
Validation dataset C (FFPE)
Validation dataset C (FFPE)
Validation dataset C (FFPE)
We tested the final 5-gene classifier on the qPCR/FFPE-derived test dataset C using a cross-validation approach and obtained a classification accuracy of 72%, sensitivity of 72%, specificity of 71%, positive predictive value (PPV) of 72%, and negative predictive value (NPV) of 67% (Table 2). Given that both PPV and NPV depend on the composition of the dataset, we also calculated the positive and negative likelihood ratios, which were 2.58 and 0.4, respectively.
Classifier testing on microarray datasets
The 5-gene classifier was further tested on 3 microarray datasets: D, E1, and E2. The classifier was trained on dataset B and tested on dataset D, resulting in an accuracy of 73%. Cross-validation of the classifier on datasets E1 and E2 provided accuracies of 92% and 83%, respectively (Table 2). These results further confirmed the reliability of the classifier.
Comparison of the classifiers
Accuracy comparison for various classifiers
Accuracy on dataset B
Accuracy on dataset D
Accuracy on dataset E1
Accuracy on dataset E2
76-gene Borup's classifier
5-gene own classifier
3-gene Weber's classifier
5-gene Foukakis classifier
Although follicular tumours are routinely diagnosed post-surgically by classic histological criteria (tumour capsule infiltration and/or angioinvasion), this field of pathology is facing numerous challenges. These problems were recognized early but are still not fully resolved. A study from 1978 in Scandinavia has shown observer disagreement of nearly 30% when examining follicular carcinoma . A report of agreement between 5 pathologists diagnosing follicular thyroid tumours reported their final consensus diagnosis in the range of 0.11-0.69 . The inter-observer agreement for FTC diagnosis in that paper was estimated to 0.23, with large intra-observer variability (0.68) . Similar data were obtained by our group . This stresses the need for extensive training of pathologists involved in the diagnosis of follicular tumours, and the importance of reference centers, but also points to the necessity for tools that can improve pathologist accuracy. When mining for novel molecular markers, it is important that a consensus diagnosis by experienced pathologists be used, as was done in our work.
A number of gene expression profiling studies have attempted to identify transcripts that are differentially expressed between FTA and FTC [19–25]. However, none of these studies resulted in a simple, efficient gene signature that was applicable in clinical practice.
Recently, a promising gene expression classifier based on 167 genes was created  and validated  in a large multicentre trial. It was used for the discrimination between benign and malignant thyroid nodules that cannot be determined by cytology. It achieved a sensitivity of 92% and specificity of 52%, which is similar to our results (sensitivity 90% and specificity 55%). While these studies as well as other ones [28, 35] lead towards a classifier applicable to fine-needle aspiration biopsy material, our approach is, in fact, different from these other published reports. We aim to support the diagnosis of post-thyroidectomy FFPE material carried out by routine histopathology with additional mRNA-based markers. Such a tool could be used as a pre-selection test to identify tumours that need meticulous histopathological evaluation. The genes selected in our study are differentiating in degraded mRNA specimens from FFPE blocks, and thus might potentially serve as a molecular indicator of malignancy. Both approaches (preoperative small sample cytology molecular testing  and post-operative whole section FFPE-based analysis, as proposed here) are in fact complementary and might be applied sequentially to provide the optimal final diagnosis.
In our opinion, it is necessary that the classifier be limited in the terms of genes tested, as cost-effectiveness issues may hamper the clinical application of multigene signatures; thus, selecting the optimal panel of markers for further testing is of utmost importance, especially from the perspective of health systems not reimbursing the cost of complex genomic diagnostic methods.
One of the important limitations of many previous studies concerning FTC/FTA differences is their relatively small sample size. To our knowledge, in all microarray studies published before 2010, the number of follicular tumours ranged from 7 to 28 [20, 22]. The Borup et al. study, published in 2010, was based on a dataset of 40 follicular tumours, the largest non-custom microarray dataset of follicular tumours to date. Thus, the gene pre-selection in our study was based on a re-analysis of the Borup dataset.
Although our final 5-gene classifier is much smaller, it exhibits accuracy comparable to that obtained using the large, 76-gene classifier of Borup et al. On 2 of 4 datasets used for comparison, it gave the same accuracy as the large classifier, and in the other 2, the loss of accuracy was lower than 10%. When compared to other classifiers with a similar number of genes (3-gene classifier of Weber and 5-gene classifier of Foukakis), our classifier gives better accuracies on validation datasets. Our test set (dataset C), derived from FFPE samples and analyzed using qPCR, reproduces cases obtained in a routine setting. The classifier accuracy obtained on this dataset is 72%, with a sensitivity of 71% and specificity of 72%. For the detection of cancer, the sensitivity is of utmost importance; shifting the test threshold gives a sensitivity of 90%, with an acceptable specificity of 55%. This approach seems justified, as the gain in FTC diagnosis sensitivity would outweigh the drop in specificity via careful histopathological assessment.
From the initial 8-gene classifier, 3 genes: CA4, LRP1B, and PLEKHG4B, could not be amplified with qPCR, probably due to their low expression in FFPE specimens. The remaining 5 genes, positively validated in our analysis, are downregulated in FTC compared to adenoma. ELMO1, EMCN, KCNAB1, and SLCO2A1 are plasma membrane components and ITIH5 is an extracellular matrix (ECM) component. Some of these proteins exhibit functional similarities: SLCO2A1 and KCNAB1 exhibit transporter activity [36, 37], and ELMO1, EMCN, and ITIH5 are involved in cell movement and ECM stability [38–40]. Some of the genes from our 5-gene classifier have already been mentioned in other studies of high-throughput gene expression analysis of follicular tumours. ITIH5, KCNAB1, and SLCO2A1 are mentioned in the Borup et al. study. Besides, ELMO1 is differentially expressed in Barden et al.  and KCNAB1 is differentially expressed in 2 other papers by Takano  and Weber , respectively.
There are some potential limitations to our study. The most important issue, applying both to our analysis as well to all other studies carried out to date, is that we do not understand the reason for sample misclassification. Are the misclassified samples the same that would be misclassified by histopathologists? The excellent prognosis of follicular cancer treated with adequate surgical and adjuvant therapy makes this assessment difficult, given that patients exhibiting disease recurrence or dissemination initially or during follow up are relatively rare. This issue will be addressed in future studies, preferably using gene expression profiling of FTCs with metastases or local recurrence. We also cannot exclude the possibility that the reason for incorrect classification of some samples is that the signal from a small amount of neoplastic tissue might be dominated by the surrounding normal thyroid tissue.
Because routine diagnostic applications of the test is our goal, we did not apply any methods, such as paraffin slide micro dissection, to enrich the tumour content of our samples, as these methods are difficult to apply in the clinical setting. We believe that our signature is useful, as it was constructed using 2 large datasets and validated on an independent dataset using a different method. Although it is not efficient enough to be used as a clinical diagnostic test by itself, it would improve the process of diagnosis . It is also a step towards developing a highly powerful classifier, as each microarray dataset generated and analyzed improves our understanding of gene expression differences between different types of follicular tumours. Additional studies also should be undertaken to elucidate the molecular and clinical aspects of the discovered markers.
We must stress that the problem of FTC and FTA differentiation has also been investigated by several miRNA-profiling studies. miRNAs are known to be a good disease markers, and can be easily detected in frozen tissues, paraffin blocks (FFPE), fine-needle aspiration biopsies (FNAB), or even serum . Some miRNAs are described as sensitive biomarkers of malignant and benign follicular thyroid tumours [16–18]. In future, a possible FNA or FFPE-block test might combine degradation-resistant mRNA markers or protein markers and a panel of miRNAs to provide optimal analytical parameters.
In summary, we developed a simple 5-gene classifier that can distinguish FTC from FTA with good accuracy. It is based on the 2 largest FTC-FTA microarray datasets and was validated on an independent FFPE sample set, with a potential sensitivity of up to 90%. In future, such a molecular test may serve as an important tool for assisting pathologists in cases of thyroid follicular neoplasms where a clear clinical decision cannot be made based on histopathology.
Follicular thyroid cancer
Follicular thyroid adenoma
Quantitative real-time PCR
Fine needle aspiration biopsy
GC robust multiarray average
Diagonal linear discriminant analysis
Carbonic anhydrase IV
Engulfment and cell motility 1
Inter-alpha-trypsin inhibitor heavy chain family, member 5
Potassium voltage-gated channel, shaker-related subfamily, beta member 1
Low density lipoprotein receptor-related protein 1B
pleckstrin homology domain containing, family G (with RhoGef domain) member 4B
Solute carrier organic anion transporter family, member 2A1
Positive predictive value
Negative predictive value
Receiver operating characteristic
Peroxisome proliferator-activated receptor gamma
Poorly differentiated thyroid carcinoma
Paired box gene 8
Extracellular matrix metalloproteinase inducer
Growth arrest- and DNA damage-inducible gene 153
Thyroid transcription factor 1
We thank Dr. Krzysztof Fujarewicz for critically analysing the manuscript and for his helpful suggestions. This work was supported by the Ministry of Science and Higher Education grants nr N N401 072637, N N403 194340 and Foundation for Polish Science MPD Program “Molecular Genomics, Transcriptomics and Bioinformatics in Cancer” and Postgraduate School of Molecular Medicine (BW and TS). This research was also supported by a DFG grant to Markus Eszlinger (ES162/4-1). Author of this publication (AP) is a scholar in SWIFT project POKL.08.02.01-24-005/10, which is co-financed by European Union within European Social Fund.
- Franc B, De-la SP, Lange F, Hoang C, Louvel A, De RA, Vilde F, Hejblum G, Chevret S, Chastang C: Interobserver and intraobserver reproducibility in the histopathology of follicular thyroid carcinoma. Hum Pathol. 2003, 34: 1092-1100. 10.1016/S0046-8177(03)00403-9.View ArticlePubMedGoogle Scholar
- Lang W, Georgii A, Stauch G, Kienzle E: The differentiation of atypical adenomas and encapsulated follicular carcinomas in the thyroid gland. Virchows Arch A Pathol Anat Histol. 1980, 385: 125-141. 10.1007/BF00427399.View ArticlePubMedGoogle Scholar
- Cheung L, Messina M, Gill A, Clarkson A, Learoyd D, Delbridge L, Wentworth J, Philips J, Clifton-Bligh R, Robinson BG: Detection of the PAX8-PPAR gamma fusion oncogene in both follicular thyroid carcinomas and adenomas. J Clin Endocrinol Metab. 2003, 88: 354-357. 10.1210/jc.2002-021020.View ArticlePubMedGoogle Scholar
- Fagin JA, Mitsiades N: Molecular pathology of thyroid cancer: diagnostic and clinical implications. Best Pract Res Clin Endocrinol Metab. 2008, 22: 955-969. 10.1016/j.beem.2008.09.017.View ArticlePubMedPubMed CentralGoogle Scholar
- Sahin M, Allard BL, Yates M, Powell JG, Wang XL, Hay ID, Zhao Y, Goellner JR, Sebo TJ, Grebe SK, et al: PPARgamma staining as a surrogate for PAX8/PPARgamma fusion oncogene expression in follicular neoplasms: clinicopathological correlation and histopathological diagnostic value. J Clin Endocrinol Metab. 2005, 90: 463-468.View ArticlePubMedGoogle Scholar
- Esapa CT, Johnson SJ, Kendall-Taylor P, Lennard TW, Harris PE: Prevalence of Ras mutations in thyroid neoplasia. Clin Endocrinol (Oxf). 1999, 50: 529-535. 10.1046/j.1365-2265.1999.00704.x.View ArticleGoogle Scholar
- Shi YF, Zou MJ, Schmidt H, Juhasz F, Stensky V, Robb D, Farid NR: High rates of ras codon 61 mutation in thyroid tumors in an iodide-deficient area. Cancer Res. 1991, 51: 2690-2693.PubMedGoogle Scholar
- Suarez HG, du Villard JA, Severino M, Caillou B, Schlumberger M, Tubiana M, Parmentier C, Monier R: Presence of mutations in all three ras genes in human thyroid tumors. Oncogene. 1990, 5: 565-570.PubMedGoogle Scholar
- Vasko V, Ferrand M, Di CJ, Carayon P, Henry JF, De MC: Specific pattern of RAS oncogene mutations in follicular thyroid tumors. J Clin Endocrinol Metab. 2003, 88: 2745-2752. 10.1210/jc.2002-021186.View ArticlePubMedGoogle Scholar
- Saggiorato E, De PR, Volante M, Cappia S, Arecco F, Dei Tos AP, Orlandi F, Papotti M: Characterization of thyroid 'follicular neoplasms' in fine-needle aspiration cytological specimens using a panel of immunohistochemical markers: a proposal for clinical application. Endocr Relat Cancer. 2005, 12: 305-317. 10.1677/erc.1.00944.View ArticlePubMedGoogle Scholar
- Abulkheir IL, Mohammad DB: Value of immunohistochemical expression of p27 and galectin-3 in differentiation between follicular adenoma and follicular carcinoma. Appl Immunohistochem Mol Morphol. 2012, 20: 131-140. 10.1097/PAI.0b013e318228de00.View ArticlePubMedGoogle Scholar
- Bryson PC, Shores CG, Hart C, Thorne L, Patel MR, Richey L, Farag A, Zanation AM: Immunohistochemical distinction of follicular thyroid adenomas and follicular carcinomas. Arch Otolaryngol Head Neck Surg. 2008, 134: 581-586. 10.1001/archotol.134.6.581.View ArticlePubMedGoogle Scholar
- Cochand-Priollet B, Dahan H, Laloi-Michelin M, Polivka M, Saada M, Herman P, Guillausseau PJ, Hamzi L, Pote N, Sarfati E, et al: Immunocytochemistry with cytokeratin 19 and anti-human mesothelial cell antibody (HBME1) increases the diagnostic accuracy of thyroid fine-needle aspirations: preliminary report of 150 liquid-based fine-needle aspirations with histological control. Thyroid. 2011, 21: 1067-1073. 10.1089/thy.2011.0014.View ArticlePubMedGoogle Scholar
- Tan A, Etit D, Bayol U, Altinel D, Tan S: Comparison of proliferating cell nuclear antigen, thyroid transcription factor-1, Ki-67, p63, p53 and high-molecular weight cytokeratin expressions in papillary thyroid carcinoma, follicular carcinoma, and follicular adenoma. Ann Diagn Pathol. 2011, 15: 108-116. 10.1016/j.anndiagpath.2010.11.005.View ArticlePubMedGoogle Scholar
- Paunovic I, Isic T, Havelka M, Tatic S, Cvejic D, Savin S: Combined immunohistochemistry for thyroid peroxidase, galectin-3, CK19 and HBME-1 in differential diagnosis of thyroid tumors. APMIS. 2012, 120: 368-379. 10.1111/j.1600-0463.2011.02842.x.View ArticlePubMedGoogle Scholar
- Nikiforova MN, Tseng GC, Steward D, Diorio D, Nikiforov YE: MicroRNA expression profiling of thyroid tumors: biological significance and diagnostic utility. J Clin Endocrinol Metab. 2008, 93: 1600-1608. 10.1210/jc.2007-2696.View ArticlePubMedPubMed CentralGoogle Scholar
- Weber F, Teresi RE, Broelsch CE, Frilling A, Eng C: A limited set of human MicroRNA is deregulated in follicular thyroid carcinoma. J Clin Endocrinol Metab. 2006, 91: 3584-3591. 10.1210/jc.2006-0693.View ArticlePubMedGoogle Scholar
- Rossing M, Borup R, Henao R, Winther O, Vikesaa J, Niazi O, Godballe C, Krogdahl A, Glud M, Hjort-Sorensen C, et al: Down-regulation of microRNAs controlling tumourigenic factors in follicular thyroid carcinoma. J Mol Endocrinol. 2012, 48: 11-23. 10.1530/JME-11-0039.View ArticlePubMedGoogle Scholar
- Borup R, Rossing M, Henao R, Yamamoto Y, Krogdahl A, Godballe C, Winther O, Kiss K, Christensen L, Hogdall E, et al: Molecular signatures of thyroid follicular neoplasia. Endocr Relat Cancer. 2010, 17: 691-708. 10.1677/ERC-09-0288.View ArticlePubMedGoogle Scholar
- Chevillard S, Ugolin N, Vielh P, Ory K, Levalois C, Elliott D, Clayman GL, El-Naggar AK: Gene expression profiling of differentiated thyroid neoplasms: diagnostic and clinical implications. Clin Cancer Res. 2004, 10: 6586-6597. 10.1158/1078-0432.CCR-04-0053.View ArticlePubMedGoogle Scholar
- Hinsch N, Frank M, Doring C, Vorlander C, Hansmann ML: QPRT: a potential marker for follicular thyroid carcinoma including minimal invasive variant; a gene expression, RNA and immunohistochemical study. BMC Cancer. 2009, 9: 93-10.1186/1471-2407-9-93.View ArticlePubMedPubMed CentralGoogle Scholar
- Lubitz CC, Gallagher LA, Finley DJ, Zhu B, Fahey TJ: Molecular analysis of minimally invasive follicular carcinomas by gene profiling. Surgery. 2005, 138: 1042-1048. 10.1016/j.surg.2005.09.009.View ArticlePubMedGoogle Scholar
- Takano T, Miyauchi A, Yoshida H, Kuma K, Amino N: 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. 2004, 90: 1600-1605. 10.1038/sj.bjc.6601702.View ArticlePubMedPubMed CentralGoogle Scholar
- Weber F, Shen L, Aldred MA, Morrison CD, Frilling A, Saji M, Schuppert F, Broelsch CE, Ringel MD, Eng C: Genetic classification of benign and malignant thyroid follicular neoplasia based on a three-gene combination. J Clin Endocrinol Metab. 2005, 90: 2512-2521. 10.1210/jc.2004-2028.View ArticlePubMedGoogle Scholar
- Fryknas M, Wickenberg-Bolin U, Goransson H, Gustafsson MG, Foukakis T, Lee JJ, Landegren U, Hoog A, Larsson C, Grimelius L, et al: Molecular markers for discrimination of benign and malignant follicular thyroid tumors. Tumour Biol. 2006, 27: 211-220. 10.1159/000093056.View ArticlePubMedGoogle Scholar
- Cerutti JM, Delcelo R, Amadei MJ, Nakabashi C, Maciel RM, Peterson B, Shoemaker J, Riggins GJ: A preoperative diagnostic test that distinguishes benign from malignant thyroid carcinoma based on gene expression. J Clin Invest. 2004, 113: 1234-1242.View ArticlePubMedPubMed CentralGoogle Scholar
- Chudova D, Wilde JI, Wang ET, Wang H, Rabbee N, Egidio CM, Reynolds J, Tom E, Pagan M, Rigl CT, et al: Molecular classification of thyroid nodules using high-dimensionality genomic data. J Clin Endocrinol Metab. 2010, 95: 5296-5304. 10.1210/jc.2010-1087.View ArticlePubMedGoogle Scholar
- Alexander EK, Kennedy GC, Baloch ZW, Cibas ES, Chudova D, Diggans J, Friedman L, Kloos RT, LiVolsi VA, Mandel SJ, et al: Preoperative diagnosis of benign thyroid nodules with indeterminate cytology. N Engl J Med. 2012, 367: 705-715. 10.1056/NEJMoa1203208.View ArticlePubMedGoogle Scholar
- Wu ZJ, Irizarry RA, Gentleman R, Martinez-Murillo F, Spencer F: A model-based background adjustment for oligonucleotide expression arrays. J Am Stat Assoc. 2004, 99: 909-917. 10.1198/016214504000000683.View ArticleGoogle Scholar
- Slawski M, Daumer M, Boulesteix AL: CMA: a comprehensive Bioconductor package for supervised classification with high dimensional data. BMC Bioinformatics. 2008, 9: 439-10.1186/1471-2105-9-439.View ArticlePubMedPubMed CentralGoogle Scholar
- Vandesompele J, De PK, Pattyn F, Poppe B, Van RN, De PA, Speleman F: Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol. 2002, 3: RESEARCH0034.View ArticlePubMedPubMed CentralGoogle Scholar
- Foukakis T, Gusnanto A, Au AY, Hoog A, Lui WO, Larsson C, Wallin G, Zedenius J: A PCR-based expression signature of malignancy in follicular thyroid tumors. Endocr Relat Cancer. 2007, 14: 381-391. 10.1677/ERC-06-0023.View ArticlePubMedGoogle Scholar
- Saxen E, Franssila K, Bjarnason O, Normann T, Ringertz N: Observer variation in histologic classification of thyroid cancer. Acta Pathol Microbiol Scand A. 1978, 86A: 483-486.PubMedGoogle Scholar
- Lange D, Sporny S, Sygut J, Kulig A, Jarzab M, Kula D, Jarzab B: [Histopathological diagnosis of thyroid cancer in a multicenter trial]. Endokrynol Pol. 2006, 57: 336-342.PubMedGoogle Scholar
- Karger S, Krause K, Gutknecht M, Schierle K, Graf D, Steinert F, Dralle H, Fuhrer D: ADM3, TFF3 and LGALS3 are discriminative molecular markers in fine-needle aspiration biopsies of benign and malignant thyroid tumours. Br J Cancer. 2012, 106: 562-568. 10.1038/bjc.2011.578.View ArticlePubMedPubMed CentralGoogle Scholar
- England SK, Uebele VN, Kodali J, Bennett PB, Tamkun MM: A novel K+ channel beta-subunit (hKv beta 1.3) is produced via alternative mRNA splicing. J Biol Chem. 1995, 270: 28531-28534. 10.1074/jbc.270.48.28531.View ArticlePubMedGoogle Scholar
- Lu R, Kanai N, Bao Y, Schuster VL: Cloning, in vitro expression, and tissue distribution of a human prostaglandin transporter cDNA(hPGT). J Clin Invest. 1996, 98: 1142-1149. 10.1172/JCI118897.View ArticlePubMedPubMed CentralGoogle Scholar
- Gumienny TL, Brugnera E, Tosello-Trampont AC, Kinchen JM, Haney LB, Nishiwaki K, Walk SF, Nemergut ME, Macara IG, Francis R, et al: CED-12/ELMO, a novel member of the CrkII/Dock180/Rac pathway, is required for phagocytosis and cell migration. Cell. 2001, 107: 27-41. 10.1016/S0092-8674(01)00520-7.View ArticlePubMedGoogle Scholar
- Hamm A, Veeck J, Bektas N, Wild PJ, Hartmann A, Heindrichs U, Kristiansen G, Werbowetski-Ogilvie T, Del MR, Knuechel R, et al: Frequent expression loss of Inter-alpha-trypsin inhibitor heavy chain (ITIH) genes in multiple human solid tumors: a systematic expression analysis. BMC Cancer. 2008, 8: 25-10.1186/1471-2407-8-25.View ArticlePubMedPubMed CentralGoogle Scholar
- Kinoshita M, Nakamura T, Ihara M, Haraguchi T, Hiraoka Y, Tashiro K, Noda M: Identification of human endomucin-1 and -2 as membrane-bound O-sialoglycoproteins with anti-adhesive activity. FEBS Lett. 2001, 499: 121-126. 10.1016/S0014-5793(01)02520-0.View ArticlePubMedGoogle Scholar
- Barden CB, Shister KW, Zhu B, Guiter G, Greenblatt DY, Zeiger MA, Fahey TJ: Classification of follicular thyroid tumors by molecular signature: results of gene profiling. Clin Cancer Res. 2003, 9: 1792-1800.PubMedGoogle Scholar
- Eszlinger M, Krohn K, Hauptmann S, Dralle H, Giordano TJ, Paschke R: Perspectives for improved and more accurate classification of thyroid epithelial tumors. J Clin Endocrinol Metab. 2008, 93: 3286-3294. 10.1210/jc.2008-0201.View ArticlePubMedGoogle Scholar
- Yu S, Liu Y, Wang J, Guo Z, Zhang Q, Yu F, Zhang Y, Huang K, Li Y, Song E, et al: Circulating microRNA profiles as potential biomarkers for diagnosis of papillary thyroid carcinoma. J Clin Endocrinol Metab. 2012, 97: 2084-2092. 10.1210/jc.2011-3059.View ArticlePubMedGoogle Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1755-8794/6/38/prepub
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.