Abstract
The molecular determinants of thyroid follicular nodules are incompletely understood and assessment of malignancy is a diagnostic challenge. Since microRNA (miRNA) analyses could provide new leads to malignant progression, we characterised the global miRNA expression in follicular adenoma (FA) and follicular carcinoma (FC). Comparison of carcinoma and adenoma with normal thyroid revealed 150 and 107 differentially expressed miRNAs respectively. Most miRNAs were down-regulated and especially miR-199b-5p and miR-144 which were essentially lost in the carcinomas. Integration of the changed miRNAs with differentially expressed mRNAs demonstrated an enrichment of seed sites among up-regulated transcripts encoding proteins implicated in thyroid tumourigenesis. This was substantiated by the demonstration that pre-miR-199b reduced proliferation when added to cultured follicular thyroid carcinoma cells. The down-regulated miRNAs in FC exhibited a substantial similarity with down-regulated miRNAs in anaplastic carcinoma (AC) and by gene set enrichment analysis, we observed a significant identity between target mRNAs in FC and transcripts up-regulated in AC. To examine the diagnostic potential of miRNA expression pattern in distinguishing malignant from benign nodules we employed a supervised learning algorithm and leave-one-out-cross-validation. By this procedure, FA and FC were identified with a negative predicted value of 83% (data generated by microarray platform) and of 92% (data generated by qRT-PCR platform). We conclude that follicular neoplasia is associated with major changes in miRNA expression that may promote malignant transformation by increasing the expression of transcripts encoding tumourigenic factors. Moreover, miRNA profiling may facilitate the diagnosis of carcinoma vs adenoma.
Introduction
Thyroid nodules are found in up to 7% of the adult population (Hegedus et al. 2003). Although the majority of the nodules are benign, carcinoma of the thyroid gland is the most common malignancy of the endocrine system (Curado & Edwards 2007). Follicular adenomas (FA) are benign, encapsulated tumours and they are five times more frequent than follicular carcinomas (FC; Faquin 2008). FC represent 10–15% of all thyroid malignancies and mainly occur in middle-aged euthyroid women as a painless thyroid nodule (Faquin 2008). FA and FC are morphologically similar and carcinomas are distinguished by vascular and/or capsular invasion (Schmid & Farid 2006). As the latter features may be overlooked it is generally accepted, that application of biomarkers could improve diagnosis (Franc et al. 2003).
MicroRNAs (miRNAs) are non-coding single-stranded RNAs of about 22 nucleotides. miRNAs regulate translation and stability of particular target mRNAs by imperfect base pairing (Bartel 2004). It is estimated that the number of miRNAs may exceed 1000 (http://microrna.sanger.ac.uk/) and miRNAs regulate about one-third of the mammalian protein-coding mRNAs (Bartel 2009, Friedman et al. 2009). The expression of miRNAs is temporally and spatially regulated. Many are important for the differentiation processes during particular developmental stages, but miRNAs also exhibit important functions in mature cells (Schmittgen 2008). Moreover, miRNAs are aberrantly expressed or lost in a variety of cancers (Rosenfeld et al. 2008). Many target mRNAs encode oncogenes and tumour suppressors and in this way dysregulated miRNAs may play a causal role in malignant progression. Not surprisingly, miRNAs are therefore considered attractive candidates for classification of tumours. The role of miRNAs in thyroid cancer is incompletely understood. A number of miRNAs have previously been identified in various thyroid tumours (He et al. 2005, Pallante et al. 2006, Weber et al. 2006, Chen et al. 2008, Nikiforova et al. 2008). miR-197 and miR-346 are overexpressed in FC in comparison to adenoma and in vitro studies suggested that both miRNAs could have a significant impact on tumour cell proliferation (Weber et al. 2006).
In this study, we employed global miRNA analysis to identify differentially expressed miRNAs in follicular thyroid carcinoma and adenoma. miRNAs were integrated with differentially expressed mRNAs to identify and validate putative target mRNAs. The gene ontology and gene set enrichment analysis (GSEA) of these showed a significant enrichment of seed sites among transcripts encoding proteins involved in thyroid tumourigenesis and an overlap to transcripts up-regulated in anaplastic carcinoma (AC) respectively. As the data indicated that miRNAs could promote malignant progression, we further employed classification algorithms to distinguish carcinoma from adenoma.
Materials and methods
Thyroid tissue, FA and FC
Thyroid samples originated from a consecutive series of patients with a solitary or prominent scintigraphically cold thyroid nodule operated on, and with a histological diagnosis of an FA or FC. As a uniform histopathological evaluation was essential, the diagnosis was made by a particular pathologist specialising in thyroid pathology. All tumours were diagnosed and classified according to the WHO definition of histological criteria. Clinical data are listed in Table 1. Surgically removed thyroid samples were snap frozen at the Department of Pathology and stored at −120 °C over a 5-year period. The project was approved by the ethics committee and informed consent was obtained from all patients. Twelve FA, twelve FC and ten normal thyroid (NT) specimens were included. NT specimens were obtained by a thyroid pathologist to ensure that the tissue derived from macro- and micro-scopically normal tissue adjacent to the encapsulated tumours. The number of tumours was balanced to provide optimal power estimates and a similar number of samples in each diagnostic category.
Clinical data of patients with thyroid follicular neoplasia. Twelve patients with histopathologically verified follicular carcinomas (FC), minimal and widely invasiveness, and twelve patients with follicular adenomas (FA). The table depicts diagnosis, age, sex, tumour size, invasiveness of the examined tumours and status of known oncogenes. All tumours were negative for KRAS point mutation and examination of BRAF showed only one positive carcinoma sample positive for BRAF point mutation; K601E (c.1801A>G p.Lys601Glu). Examination for PAX8/PPARγ translocation exhibited no positive samples
| Diagnosis | Age (years) | Sex (M/F) | Nodule size (cm; max diameter) | Invasiveness (minimal vs widely) | BRAF gene mutation |
|---|---|---|---|---|---|
| FC_01 | 32 | F | 6 | Minimal | No mutation |
| FC_02 | 60 | F | 2.5 | Minimal* | No mutation |
| FC_03 | 32 | F | 1.5 | Minimal | No mutation |
| FC_04 | 48 | F | 5 | Widely | No mutation |
| FC_05 | 41 | M | 6 | Minimal | No mutation |
| FC_06 | 75 | F | 4 | Minimal* | No mutation |
| FC_07 | 35 | M | 7 | Minimal | No mutation |
| FC_08 | 59 | F | 2 | Widely | No mutation |
| FC_09 | 28 | M | 2.5 | Minimal | No mutation |
| FC_10 | 46 | F | 10.5 | Widely | No mutation |
| FC_11 | 32 | F | 4 | Minimal | Mutation |
| FC_12 | 63 | F | 3.5 | Minimal | No mutation |
| FA_01 | 55 | F | 3.5 | No mutation | |
| FA_02 | 41 | M | 4 | No mutation | |
| FA_03 | 65 | F | 4 | No mutation | |
| FA_04 | 47 | F | 2.5 | No mutation | |
| FA_05 | 63 | F | 4.5 | No mutation | |
| FA_06 | 33 | F | 2 | No mutation | |
| FA_07 | 37 | F | 3.5 | No mutation | |
| FA_08 | 46 | M | 3 | No mutation | |
| FA_09 | 37 | F | 2.5 | No mutation | |
| FA_10 | 51 | F | 4 | No mutation | |
| FA_11 | 65 | F | 4 | No mutation | |
| FA_12 | 37 | F | 4 | No mutation |
*indicates the two FC (FC_02 and FC_06) of the oncocytic type.
Total RNA and DNA isolation
Total RNA was isolated from frozen samples using Trizol (Invitrogen) according to the manufacturer's protocol. RNA was precipitated using 100% isopropanol. Purified RNA was subsequently quantified on a NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA) and examined on a Bioanalyzer Nano RNA Chip (Agilent Technologies, Santa Clara, CA, USA) before labelling. Each sample (1 μg) was pooled and used as common reference. DNA from tissues was isolated by lysing the tissue in 20 μl proteinase K and 200 μl Tris/NaCl/EDTA/SDS, followed by overnight incubation at 55 °C. Finally, 5 M NaCl is added to the lysed tissues and DNA is precipitated by adding 200 μl ice-cold 96% ethanol. Total RNA for mRNA microarray was extracted from pre-miRNA transfected cells using NucleoSpin RNA/Protein from Macherey-Nagel according to their recommendations.
Detection of point mutations
Mutation analyses, including PCR-amplification and sequencing, were performed on all 24 tumour samples for KRAS and BRAF (V600E and K601E) and regular PCR-amplification, followed by gel-separation was used to analyse the fusion point of PAX8/PPARγ translocation. For detection of possible amplicons, a positive control was examined along with the 24 tumour samples. Primer sequences for KRAS, forward; 5′-TGTAAAACGACGGCCAGTCGATACACGTCTGCAGTCAA-3′ and reverse; 5′-CAGGAAACAGCTATGACCCTCATGAAAATGGTCAGAGA-3′, BRAF, forward; 5′-TGCTTGCTCTGATAGGAAAATG-3′ and reverse; 5′-GGATGGTAAGAATTGAGGCT-3′ and PAX8/PPARγ, forward; 5′-GCATTGACTCACAGAGCAGCA-3′ and reverse; 5′-AGACCACTCCCACTCCTTTG-3′ (Kroll et al. 2000).
miRNA and mRNA expression analyses
miRNA expression levels were determined by microarray analysis. Total RNA (1 μg) was labelled with fluorescent Hy3 (sample)/Hy5 (reference-sample) dye from the miRCURY LNA microRNA Array Power Labelling Kit (Exiqon, Vedbaek, Denmark) in accordance with the manufacturer's instructions. Using a TECAN hybridisation station, labelled samples were hybridised overnight to pre-printed miRCURY LNA microRNA Array, v.11.0 (Exiqon), containing probes for 841 human miRNAs, catalogued in the miRBase Sequence Database (Release 11.0; http://microrna.sanger.ac.uk/) and 428 proprietary human miRPlus sequences not yet annotated in miRBase.
Furthermore, miRNA expression levels from 30 thyroid specimens (12 FC, 12 FA and 6 NT), were generated using the miRCURY LNATM Universal RT miRNA PCR panels I and II, V2 (Exiqon). Total RNA (40 ng) was reverse transcribed using the Universal cDNA synthesis kit, mixed with SYBR Green master mix kit, and subsequently added to the pre-aliquoted miRNA PCR primer sets in two 384-well PCR plates enabling profiling of 742 human miRNAs. All reagents were from Exiqon and their recommendations were followed. Each plate contained an additional six primer sets for reference miRNAs and a set of negative controls. The amplification curves were analysed using the Roche LC software, both for determination of crossover point (Cp) and for melting curve analysis. One hundred and thirty-five miRNA assays were successfully assessed with sufficient signal (Cp <37, or 5 Cp less than negative controls) in all samples.
Analyses of global mRNA levels in RNA extracted from pre-miRNA transfected cells were performed by the Affymetrix platform. Total RNA (250 ng) was amplified and labelled using the Ambion WT Expression Kit (Applied Biosystems) according to the manufacturer's instructions. The labelled samples were hybridised to the Human Gene 1.0 ST GeneChip array (Affymetrix, Santa Clara, CA, USA). Arrays were washed and stained with phycoerythrin-conjugated streptavidin (SAPE) using the Affymetrix Fluidics Station 450, and subsequently scanned in the Affymetrix GeneArray 2500 scanner to generate fluorescent images, according to the Affymetrix GeneChip protocol. Cell intensity files (CEL files) were generated in the GeneChip Command Console Software (AGCC; Affymetrix). CEL files were modelled using the robust multichip average approach, followed by mean Probe Set Summarisation resulting in a single expression value for each gene. Modelled CEL files were generated with the software package Partek Genomics Suite 6.5.
Image analysis and normalisation of miRNA expression
Arrays were scanned in an Agilent DNA Microarray Scanner (Agilent Technologies) and resulting images were analysed with Genepix Pro 6.0 software (Molecular Devices). Background intensities were subtracted from foreground intensities and within array LOESS – normalised, followed by Aquantile normalisation between arrays as implemented in the Limma package in Bioconducter library (Gentleman et al. 2004) in R version 2.10 (R Development Core Team 2007). Mean values of the quadruplicate probes for each of the miRNAs were obtained and log2 ratios between sample and reference were used for further analyses.
Normalisation of the results derived from the RT miRNA PCR panels was performed based on the average of the assays detected in all samples; Normalised Cp=average samples Cp (n=135)−assay Cp. The normalised miRNA expression values were used for generating a diagnostic classifier between FC and FA as described in the section ‘Construction of classifier’.
Class comparison analyses
Class comparison was performed by the Limma package (http://bioconductor.org/packages/release/bioc/html/limma.html). For each miRNA standard errors and fold changes were estimated by fitting a linear model. Differentially expressed miRNAs were determined by applying empirical Bayes moderated t-statistics test (Wettenhall & Smyth 2004). miRNAs were defined to be differentially expressed if they had a Benjamini–Hochberg corrected P value below 0.05 and an absolute fold-change above 1.5.
mRNA expression data and miRNA target predictions followed by pathway analysis
The global mRNA expression data originating from FA, FC and AC tissue samples was based on previous work by our group (Borup et al. 2010). In addition, mRNA profiles of papillary and NT tissue samples were down loaded from the Array Express, ID: E-GEOD-6004 and E-GEOD-7307 respectively. To predict mRNA targets, based on the observed miRNA profiles, we implemented the method developed by Gallagher et al. (2010). Briefly, mRNA target predictions were based on the TargetScan miRNA target prediction database in combination with the observed changes in miRNAs. Only miRNAs that exhibited an absolute change >1.5-fold and mRNA with an average expression intensity >40 in FCs were included in the analysis. This method provides a weighted miRNA inhibitor score (sum of effects), predicting the transcripts, most likely to be regulated by miRNAs. As the majority of the differentially expressed miRNAs were down-regulated, only mRNAs with a positive composite score were considered in the analysis. A cutoff at +2 and a variance filtering of 0.176 were used to remove unhybridised probes in our weighted inhibitor score list (composite score). This resulted in the retrieval of 1381 probe sets, representing 528 genes. To examine whether the putative mRNA targets were associated with particular biological functions or pathways, the predicted mRNAs were examined in the Ingenuity Pathway Analysis software package (Ingenuity Systems, www.ingeuity.com). The most significant pathways were identified by the Fisher's exact test. GSEA of the target list was submitted to and performed using the open source software available at http://www.broadinstitute.org/gsea/msigdb/index.jsp. The gene list was loaded into the molecular signature database (MSigDB) and overlaps were detected in the C2 currated gene analysis. To determine the association to the gene set, the hypergeometric distribution of overlapping genes over all genes in a gene set was calculated. For this analysis a cutoff was set at composite score >+5 and only considering the differentially expressed genes, a cutoff was set at P<0.01, q=0.0036, resulting in 287 probes. Heatmaps were generated as supervised (two-way) cluster analysis.
Construction of classifier
Normalised data from 24 samples consisting of the expression level of 545 different probes (377 annotated miRNAs and 168 hsa miRPlus, all with an A value >7, A=½(log2Red+log2Green)) were included in the analysis. The classifier was constructed to classify FC and FA. The validation of the classifiers' performance on the training set was done by leave-one-out-cross-validation (LOOCV), that provides an unbiased estimate of the prediction error by going over all training examples in turn using the complement training set (of size 24−1=23) to perform training (including probe ranking, selection and model fitting) and using the last left out sample for prediction. In this way, we obtained as many predictions as there were samples in the training set. These predictions formed unbiased estimates of the prediction error quantified in terms of the confusion matrix. The training of the classifiers inside the LOO loop consisted of a univariate selection step followed by applying support vector machine learning (SVM; Vapnik 1998). Probes were ranked according to their differential expression by the Student's t-test with a threshold of P<0.001 (Dudoit et al. 2003). A permutation test was performed to determine if the cross-validated misclassification rate is lower than expected by chance (Tusher et al. 2001, Simon et al. 2007). In 1000 random permutations of the class label, the entire cross-validation was repeated for classifying the random classes of samples. The proportion of the 1000 random permutations, that gives a smaller or similar cross-validation misclassification rate as obtained with the real data, determined the permutation P value. The statistical significance of the error rate was determined for the SVM classifier.
Quantitative reverse transcription PCR of miRNA and mRNA
cDNA was prepared from 25 ng total RNA from 34 tumour samples using TagMan microRNA reverse transcription kit and TagMan microRNA assays containing predesigned primers for miR-221, miR-182, miR-96, miR-199a3p, miR-144*, miR-199b5p and miR-1826 was added. MiR-191 was used for endogenous control. Quantitative reverse transcription PCR (qRT-PCR) was performed by TagMan universal PCR master mix No AmpEras UNG, according to the manufacturer's instructions, all from Applied Biosystems. Each amplification reaction was performed in triplicate, and median value of the three cycle threshold was used for further analyses. For calculations of fold changes we used the 
For validation of miRNA predicted transcripts 2 μg total RNA from 24 tumour samples (12 FC and 10 NT) was reverse transcribed to cDNA using High-Capacity cDNA Reverse Transcription kit (Applied Biosystems). Synthesised cDNA (200 ng/μl) diluted 1/100-fold and 2 μl aliquot used in 10 μl total PCR containing fast SYBR green PCR Master Mix (Applied Biosystems) and 10 pmol/μl of each primer for target genes. Primer sequences for VPS26A, forward; 5′-GCTTGCCACCTATCCTGATG-3′ and reverse; 5′-CTGGGTCCAATTCCTGTGAT-3′ and ABCC1, forward; 5′-TAAAAGAGGATGCCCAGGTG-3′ and reverse; 5′-ACCCTGTGATCCACCAGAAG-3′. β-Actin was measured for endogenous control. The PCR comprised 40 cycles of a four-step programme: step 1, 50 °C for 2 min, step 2, 95 °C for 2 min, step 3 (40 cycles), 95 °C for 10 s and 60 °C for 45 s and step 4, 95 °C for 15 s, 60 °C for 15 s and 95 °C for 15 s. qPCRs were performed in triplicate.
miRNA transfections and proliferation experiments
Human thyroid follicular R082 W-1 cancer cells were grown at 37 °C, 5% CO2 in RPMI 1640 with GlutaMax (Invitrogen) supplemented with 10% FBS (Invitrogen), 1% penicillin and streptomycin. Cells were transfected with 50 nM pre-miR-199b and pre-miR-144, both from Applied Biosystems and for control miRNA, we used miRIDIAN microRNA mimic negative control#2 (Dharmacon/Thermo Scientific, hafayette, CO, USA). Cells were transfected with Turbofect transfection reagent (Dharmacon/Thermo Scientific) according to the manufacturer's instructions. Cell proliferation was studied real-time using the XCELLigence system as described in the manufacturer's instructions (Roche Applied Science and ACEA Bioscience). This allowed continuous measurements of the electronic impedance by detecting the physiological changes of the cells on the electrodes attached to the E-96. Five hours post-transfection with pre-miRNAs, transfected cells were plated in quadruplicates and monitored for 5 days. The experiment was repeated three times. For mRNA microarray analysis, total RNA was extracted 24 h after pre-miRNA transfections and labelled as described in the section ‘miRNA and mRNA expression analyses’.
Results
Tumour samples and oncogenic mutations
The thyroid samples originated from a consecutive series of patients and included tumours from 19 women and 5 men and all tumours were classified according to the WHO definition of histological criteria. The median age was 44 years in carcinoma patients and 47 years in adenoma patients. The size of the tumours ranged from 1.5 to 10.5 cm and the median diameter was 4 cm in the carcinoma patients and 3.75 cm in the adenoma patients (Table 1). All tumours were examined for KRAS and BRAF mutations and this showed that only one carcinoma sample was positive for BRAF and examination for PAX8/PPARγ translocation with PCR followed by gel-separation showed no positive amplicons (Table 1). Taken together, the thyroid specimens in the two diagnostic groups were comparable both with respect to the clinical features and the presence of oncogenic mutations.
miRNA expression in FC and FA
The following class comparison analysis and miRNA target analysis are based on the derived microarray expression data since this platform counts the largest number of miRNAs. Class comparison analysis revealed 150 annotated and differentially expressed human miRNAs – 37 up-regulated and 113 down-regulated miRNAs – in FCs compared with NT. The fold change ranged from 3.1- to −39-fold. Owing to the substantial 39-fold down-regulation of miR-199b-5p (also named miR-199b); this miRNA is essentially lost in FC. MiR-144*, miR-199b-3p, miR-199a-5p and miR-144 were also strongly reduced almost to the background. Of the most up-regulated miRNAs, miR-221, miR-96 and miR-182 exhibited fold changes of 3.1, 2.9 and 2.6 respectively. The comparison of FA to NT tissue revealed 107 differentially expressed miRNAs. Forty-two were up-regulated and 65 were down-regulated. Finally, the comparison of carcinoma to adenoma showed that 56 miRNAs were differentially expressed between these groups. Twelve miRNAs were up-regulated and 44 were down-regulated in the carcinoma. The five most up- and down-regulated miRNAs in each comparison are listed in Table 2 and the complete list of miRNAs that changed more than twofold is listed in Supplementary Table 1 (see section on supplementary data given at the end of this article). Fifty-eight miRNAs were differentially expressed in both the carcinoma and adenoma compared with NT (Fig. 1A). The overlap was marked among the down-regulated miRNAs, where 49 of the 65 down-regulated miRNAs in the adenoma were identical, compared with 9 out of 42 for the up-regulated miRNAs. With the exception of miR-199b-5p, that were progressively down-regulated in carcinoma, the shared miRNAs exhibited basically the same levels in adenoma and carcinoma compared with NT (Fig. 1B). Taken together, the results imply that changes in miRNA expression predominantly occur during transition from NT epithelial cell to adenoma and not during malignant transformation.
Dysregulated miRNAs. The five most up- and down-regulated miRNAs in the three comparisons, follicular carcinomas (FC) vs normal thyroid (NT), follicular adenomas (FA) vs NT and FC vs FA are listed. miRNAs were defined to be differentially expressed if they had a Benjamini–Hochberg corrected P value below 0.05 and an absolute fold-change above 1.5
| miRNA | Fold change | P value | Adjusted P value |
|---|---|---|---|
| FC vs NT | |||
| miR-199b-5p | −39 | 3.15×10−31 | 5.98×10−28 |
| miR-144* | −15.1 | 7.10×10−9 | 2.50×10−7 |
| miR-199b-3p | −12.1 | 1.33×10−28 | 1.26×10−25 |
| miR-199a-5p | −9 | 1.66×10−14 | 1.97×10−12 |
| miR-144 | −6.9 | 3.34×10−21 | 1.06×10−18 |
| miR-221 | 3.1 | 1.12×10−5 | 0.0001 |
| miR-96 | 2.9 | 5.92×10−6 | 8.72×10−5 |
| miR-182 | 2.6 | 2.11×10−6 | 3.46×10−5 |
| miR-597 | 2.4 | 0.002 | 0.0114 |
| miR-222 | 2.3 | 0.0007 | 0.0049 |
| FA vs NT | |||
| miR-199b-5p | −26 | 4.69×10−27 | 8.92×10−24 |
| miR-144* | −8.1 | 6.76×10−6 | 0.0001 |
| miR-663 | −6.2 | 5.48×10−12 | 4.51×10−10 |
| miR-199b-3p | −6.1 | 1.26×10−18 | 4.80×10−16 |
| miR-142-3p | −4.6 | 2.34×10−15 | 4.46×10−13 |
| ebv-miR-BART8 | 5.3 | 0.0083 | 0.0489 |
| miR-517a | 4.5 | 0.0043 | 0.0285 |
| miR-512-3p | 3.6 | 1.70×10−3 | 1.37×10−2 |
| miR-301b | 3.2 | 5.08×10−6 | 8.55×10−5 |
| miR-518a-3p | 2.60 | 6.36×10−3 | 3.94×10−2 |
| FC vs FA | |||
| miR-512-3p | −3.2 | 0.0032 | 0.0395 |
| miR-886-5p | −3 | 2.58×10−6 | 0.0003 |
| miR-450a | −3 | 6.92×10−7 | 0.0001 |
| miR-301b | −2.6 | 5.78×10−5 | 0.0032 |
| miR-429 | −2.4 | 0.0001 | 0.0046 |
| miR-637 | 2.1 | 7.09×10−5 | 0.0036 |
| miR-631 | 2 | 0.0028 | 0.0364 |
| miR-219-2-3p | 1.8 | 0.0041 | 0.0467 |
| miR-662 | 1.8 | 0.0017 | 0.0272 |
| miR-744 | 1.8 | 2.21×10−7 | 7.00×10−5 |
miRNA expression in follicular carcinoma (FC) and adenoma (FA). (A) Venn diagram showing differential and common miRNAs among FC and FA in relation to normal thyroid (NT) tissue. The total number of differentially expressed miRNAs is shown in black, and the number of up- and down-regulated miRNAs is shown in green and red respectively. (B) The graphs show the fold change of 9 common up-regulated (green) and 49 common down-regulated (red) miRNAs, respectively, in relation to NT.
Citation: Journal of Molecular Endocrinology 48, 1; 10.1530/JME-11-0039
Computational identification of putative target mRNAs and pathway analysis
Results from the global expression profiling of NT and FA and FC samples were used for integration of mRNA and miRNA array data. In a pilot analysis, we counted predicted seed sites corresponding to the changed miRNAs from the differentially expressed transcripts using the Targetscan miRNA–mRNA integration software in Partek Genomic Suite (http://www.partek.com/partekgs). Based on the fact that the vast majority of mRNAs were down-regulated by associated miRNAs (Guo et al. 2010), only mRNAs and miRNAs that exhibited an inverse expression pattern were considered. We detected a significant enrichment (Fisher's exact test P<0.01, assuming that the total number of 904 miRNA could target 22.000 transcripts at a global level) of seed sites corresponding to down-regulated miRNAs. The down-regulated miRNAs in the FC group exhibited putative seed sites in almost 85% of the up-regulated transcripts, which distinguished carcinoma from NT and adenoma respectively. All limitations of the computational approach taken into consideration, the results suggest that the changed miRNAs could have an impact on the observed changes in the transcriptome during progression to carcinoma.
To identify the biological pathways and further substantiate the possibility that miRNA changes had an impact on the gene expression pattern in the carcinoma, we employed the previously reported ranking system for mRNA target identification (Gallagher et al. 2010). Following miRNA target identification, 528 of the highest ranked transcripts with composite score >+2, was loaded into the geneontology and pathway analysis feature of the Ingenuity Software. Almost 200 of the putative targets mRNAs encoded factors categorised in ‘neoplasia’, ‘cancer’ and ‘tumourigenesis’ (P values of 1.95×10−5, 2.16×10−5 and 2.25×10−5 respectively). When we explored the ‘biological functions’, a group of 17 transcripts (APC, ATM, CCDC6, CCND1, COPS2, ETS1, FN1, GABBR2, LIFR, NRAS, PPARGC1A, PTPN4, RAP2A, S1PR1, SMAD2, TGFBR1 and VEGFA; P=4.42×10−5) encoded factors previously described in thyroid cancer. Corresponding to the down-regulation of miRNAs, 9 out of 11 target transcripts within the thyroid cancer specific mRNAs were increased (82%; P<0.05).
To further examine the putative biological significance of miRNA changes, we performed a GSEA (http://www.broadinstitute.org/gsea/msigdb/index.jsp). The target list was filtered by a composite score >+5 and only differentially expressed genes were included (P<0.01, q=0.0036) in the analysis, as shown in Fig. 2B. This resulted in 287 probe sets of which 260 probe sets were up-regulated. The 260 probes corresponded to 135 different genes, which were uploaded into GSEA and underwent a C2 currated genes analysis. The ‘Rodrigues_thyroid_carcinoma_anaplastic_up-regulated_cluster’ that contains 17 (82 probes) of our 135 genes was significantly connected to our target list (P=1.55×10−6). The 17 overlapping transcripts are GSK3B, FXR1, WAPAL, ITCH, NUPL1, SLC7A11, DDX3X, KPNA4, STK4, MAP3K7IP3, SSX2IP, TBL1XR1, SIX4, EIF5B, PRPF40A, NARG1 and USP31. The relative expression levels of the overlapping transcripts and of the entire Rodrigues cluster in FC and NT are depicted in heatmap C and D, Fig. 2. As shown in heatmap D, Fig. 2, more than 90% of the transcripts included in the Rodriques cluster were also up-regulated in FC. The significance of the overlap between FC and AC was further reinforced by a comparison of the recent characterisation of miRNAs in AC (Braun et al. 2010), since miR-199b, miR-100, miR-101, miR-31, miR-17, miR-99a, miR-125b, miR-26a, miR-708, miR-486, let-7i and let-7g, are down-regulated in both FC and AC. The overlap between AC and FC indicated that these cancers exhibited common features in their miRNA expression patterns, and we therefore examined the expression pattern of the 135 miRNA target transcripts in FC, papillary carcinoma (PC) and AC, see heatmap E, Fig. 2. In accordance with the GSEA, the 135 transcripts showed a unique expression pattern for FC and AC as opposed to the PC that were clearly distinguished by this gene set.
Heatmaps of predicted miRNA target transcripts in thyroid tumours. (A) depicts an unfiltered principal component analysis (PCA) of the gene expression in follicular carcinoma (FC) and normal thyroid tissue (NT). The heatmap in (B) shows the relative expression of the 260 up-regulated probe sets, representing 135 gene symbols, derived from the miRNA predicted targets with a composite score >+5 and filtered, so only differentially expressed genes were included (P<0.01, q=0.0036) in the analysis. The heatmap of (C) illustrates the relative expression level of the overlapping 17 transcripts (GSK3B, FXR1, WAPAL, ITCH, NUPL1, SLC7A11, DDX3X, KPNA4, STK4, MAP3K7IP3, SSX2IP, TBL1XR1, SIX4, EIF5B, PRPF40A, NARG1 and USP31) represented by 82 probes from the ‘Rodrigues_thyroid_carcinoma_anaplastic_up-regulated_cluster’ based on the gene set enrichment analysis. The relative expression of the entire Rodrigues cluster in FC and NT is depicted in the heatmap of (D). The expression pattern of the 135 miRNA target transcripts in FC, papillary (PC) and anaplastic carcinoma (AC) is shown in (E).
Citation: Journal of Molecular Endocrinology 48, 1; 10.1530/JME-11-0039
Proliferative effects of selected miRNAs and functional target validation
To examine the functional significance of the miR-199b-5p and mir-144 that were strongly down-regulated in FC, we determined their effect on proliferation of thyroid carcinoma R082 W-1 cells. Cells were transfected with pre-miR-199b and pre-miR-144 and a miRIDIAN microRNA Mimic was included as negative control. Whereas, the doubling time based on the normalised cell index showed a significant ∼23% (P<0.004) decrease upon pre-miR-199b-5p transfections, the effect of pre-miR-144 was only 8% (P=0.056; Fig. 3). The results showed that down-regulation of miR-199b-5p, may be causally involved in the growth of FC. To validate the computational predicted targets, we subsequently transfected R082 W-1 cells with miR-199b-5p and examined the genome-wide transcriptome to identify transcripts that were regulated by the miRNA. In total 581 transcripts were down-regulated and 733 were up-regulated. Out of the 190 computationally predicted miR-199b-5p targets 56 were represented among the down-regulated mRNAs. Among the up-regulated only seven transcripts were predicted targets. Thus, down-regulated target mRNAs were enriched in the list of computational predicted target transcripts (Fisher's exact test, P<0.0005), whereas there was no enrichment (P=0.34) among up-regulated transcripts. Our data show that about 30% of the computational predicted targets are likely to represent in vivo targets. We subsequently explored the expression levels of the 56 transcripts in the FC. Data were filtered by a P value of 0.05 resulting in 39 differentially expressed transcripts where 87% were significantly up-regulated, corresponding to the loss of 199b-5p in FC (Fig. 4 and Table 3). Finally, we validated two of the computational predicted target genes of miR-199b-5p, VPS26A and ABCC1, in the tumour set, that was used for miRNA analysis and demonstrated a significant increased expression in FC vs NT, P=0.007 and P=0.0008 respectively (Fig. 5).
Effect of proliferation in follicular thyroid cancer cell transfected with pre-miR-199b and -144. A representative screen-plot from the proliferation-experiment, illustrating a marked reduction in growth of the follicular thyroid cell line (R082 W-1) after transfecting with pre-miR-199b. The doubling time of the transfected cells are 16 h for pre-miR-199b, 14 h, for pre-miR-144 and 13 h for control miRNA. The increase in doubling time is significant for the pre-miR-199b transfected cells (P=0.004) and non-significant for the pre-miR-144 (P=0.056) in comparison with the doubling time of the control miRNA transfected cells. Transfected cells were plated in quadruplicates for each type of miRNA transfection. The experiment was repeated in biological triplicates. The X-axis represents time in hours and the Y-axis represents the normalised cellular index.
Citation: Journal of Molecular Endocrinology 48, 1; 10.1530/JME-11-0039
Expression patterns of miR-199b-5p target transcripts in histopathological samples. Two-way cluster illustrating the expression of 39 transcripts in follicular thyroid (FC) cancer specimens and normal thyroid (NT) tissue. These transcripts are both experimentally verified and computational predicted targets of miR-199b-5p (Fisher's exact test, P<0.0005). The expression data are filtered by a P value of 0.05. Blue square represents NT and FC is labelled in green squares. Each transcript is labelled by its unique identifier symbol and a list of the equivalent gene symbols is shown in Table 3.
Citation: Journal of Molecular Endocrinology 48, 1; 10.1530/JME-11-0039
miR-199b-5p target transcripts in follicular cancer specimens. A complete list of 39 target transcripts of miR-199b-5p, matching the unique identifier probes as illustrated in the two-way cluster, Fig. 4. The listed targets are a subset of the initially 190 computational predicted targets of miR-199b-5p that were correlated to the down-regulated transcripts in follicular thyroid cancer cells upon transfecting with pre-miR-199b. This comparison resulted in 56 identical transcripts, i.e. 29.5% of the computational predicted targets were actual targets. The expression pattern of the 56 transcripts was examined in the data from the histopathological follicular thyroid cancer specimens and upon filtering the data (P<0.05) the final list resulted in 39 transcripts. Nearly all of the 39 transcripts (87%) were significantly up-regulated as expected in the follicular thyroid cancer samples
| miR-199b-5p target transcripts in follicular thyroid cancer specimens |
|---|
| AP1G1, ARHGAP12, ARHGAP21, CCDC43, CELSR1, C1GALT1, C9orf5, ECE1, ERLIN1, ETS1, EXTL3, FZD6, GIT1, GPD2, IPO8, LARP4, LIN7C, MAP3K11, MGAT4B, MPP5, NLK, NPAS2, PARP12, PCYOX1, PLXND1, POMGNT1, PPARGC1A, PPFIBP1, PPP1R2, RBBP4, R3HDM2, SLC24A3, SOS2, STK4, TAF9B, TSPAN6, VPS26A, WDTC1, ZNF468 |
Relative expression level of two predicted miR-199b-5p targets by QRT-PCR. Relative expression level of two of the computational predicted targets of miR-199b-5p, VPS26A and ABCC1 examined in the tumour set (12 follicular carcinoma (FC) and 10 normal thyroid (NT) samples) also used for miRNA analysis. VPS26A and ABCC1 showed a relative increase of 3- (P=0.007) and 3.5 (P=0.0008)-fold in FC vs NT respectively. Blue colour bars represent NT and green ones represent FC.
Citation: Journal of Molecular Endocrinology 48, 1; 10.1530/JME-11-0039
miRNA-based classification of follicular nodules
To provide an overview of the miRNA expressions across follicular neoplasia (FA and FC) and NT, we generated a principal component analysis (PCA) using all expressed miRNAs (Fig. 6A). At this stage, the two populations could be discerned reflecting the relatively large and consistent differences in miRNA expression between the groups. Filtering the expression values by a t-test, we reached a subset of 179 miRNAs (P<0.01), where follicular neoplasia could easily be distinguished from NT tissue. We subsequently attempted to separate FC from adenoma by the Student's t-test for feature selection and applied the supervised learning algorithm SVM and LOOCV to generate a classifier. The optimal signature for classification of FC and FA consist of two miRNAs, miR-1826 and miR-Eplus-1078, and based on expression values of the two classification miRNAs, a PCA plot was generated (Fig. 6B). Both the negative predictive value (NPV) and the positive predictive value (PPV) for carcinoma is 83%. The LOOCV was used to compute misclassification rate. Based on 1000 random permutations, the SVM classifier comprise a P value of 0.001. The SVM can be turned into a probabilistic classifier giving an estimate of the probability of the predicted class label, i.e. assess the prediction uncertainty (Platt 1999). The predictive probabilities for all samples are listed in Table 4. FA sample 11, although correctly classified exhibited a probability of 0.5 and the misclassified FA sample 12 had a probability for FC of 0.9 indicating that FA_11 is highly uncertain, whereas FA_12 is most likely a misdiagnosed carcinoma. It was not feasible to generate a classifier that could distinguish FA and minimally invasive carcinoma. Even so, widely invasive carcinoma can be distinguished from minimally invasive carcinoma by the expression of miRPlus-E1001 (P<0.01; average threefold up-regulation) and lower expression of miRNA-410 (average fivefold) compared with the minimally invasive carcinoma. Lastly, expression values of miR-1826, miR-Eplus-1078, miR-221, miR-182, miR-96, miR-199b-3p, miR-144* and miR-199b-5p were also examined by qRT-PCR and this confirmed the microarray results (Supplementary Figure 1, see section on supplementary data given at the end of this article). To validate whether miRNAs can classify thyroid follicular malignancies by a diverse method, we examined miRNA expression levels by panels of qRT-PCR – assays in 30 (12 FC, 12 FA and 6 NT) of the included thyroid samples. An overview based on the total number of expressed miRNAs across all samples is illustrated in a PCA – plot (Fig. 6C). We focused on building a diagnostic classifier to differentiate between FC and FA and found a signature comprising of 14 miRNAs to be most favourable. As a consequence of a different miRNA analysing tool, the optimal signature for classification of FC and FA consists of 14 miRNAs, miR-19a, -501-3p, -17, -335, -106b, -15a, -16, -374a, -542-5p, -503, -320a, -326, -330-5p and let-7i. A PCA plot based on expression values of the 14 miRNAs is illustrated in Fig. 6D. Applying this signature resulted in only one misclassified carcinoma and derived a NPV of 92% and a PPV of 100%, both for malignancies.
Principal component analysis. (A) shows the projection of follicular carcinomas (FC) and follicular adenomas (FA) and normal thyroid (NT) employing all miRNAs derived from the microarray analysis. The projection of FC and FA employing the expression values of only miR-1826 and miR-Eplus-1078 is seen in (B). (C) shows the projection of FC and FA and NT employing all miRNAs derived from the qRT-PCR panels. In (D) the projection of FC and FA employing the expression values of the 14 miRNAs that was found to be the optimal signature for classification of FC is seen.
Citation: Journal of Molecular Endocrinology 48, 1; 10.1530/JME-11-0039
Predictive probabilities of miRNA classifier. Predictive probabilities for follicular adenoma (FA) and follicular carcinoma (FC) samples using SVM classifier, Cost =10, gamma =0.01.
| Sample | Predictive probability | Sample | Predictive probability |
|---|---|---|---|
| FC_01 | 0.04 | FA_01 | 0.73 |
| FC_02 | 0.2 | FA_02 | 0.66 |
| FC_03 | 0.07 | FA_03 | 0.84 |
| FC_04 | 0.41 | FA_04* | 0.45 |
| FC_05 | 0.18 | FA_05 | 0.9 |
| FC_06* | 0.67 | FA_06 | 0.94 |
| FC_07 | 0.13 | FA_07 | 0.93 |
| FC_08* | 0.62 | FA_08 | 0.87 |
| FC_09 | 0.3 | FA_09 | 0.96 |
| FC_10 | 0.46 | FA_10 | 0.8 |
| FC_11 | 0.22 | FA_11 | 0.5 |
| FC_12 | 0.08 | FA_12* | 0.1 |
*Misclassified samples (FA: n=2, FC: n=2)
Discussion
In contrast to papillary and medullary thyroid cancers, where many tumours exhibit defined mutations in oncogenes, the causal mutations leading to follicular neoplasia are incompletely understood. Consequently, efforts have been devoted towards defining biomarkers such as miRNAs that would allow the clinicians to distinguish carcinoma from adenoma as well as other thyroid cancers and obtain information of the molecular pathways that drive thyroid tumourigenesis.
Compared with NT, FA and FC exhibited widespread changes in their miRNA expression. We confirmed previously reported up-regulations of miR-197, -346, -187, -221, -222, -224 and -155 in carcinoma and up-regulation of miR-339, -210, -328 and -342 in adenoma (Weber et al. 2006, Nikiforova et al. 2008) and identified a number of previously undescribed thyroid miRNAs including miR-199b-5p, miR-144*, miR-199b-3p, miR-199a-5p, miR-144, miR-96, miR-182 and miR-597, to mention the most striking. Of particular significance, among the novel thyroid miRNAs, miR-199b-5p was found to be lost in the carcinoma. Loss of miR-199b-5p (also known as miR-199b) has previously been described in chorioncarcinoma (Chao et al. 2009) and in meduloblastoma, where the loss increases metastasis (Garzia et al. 2009). MiR-96 was among the few up-regulated miRNAs in the carcinoma and this is in line with urothelial carcinomas, where miR-96 is a promising tumour marker in urine (Yamada et al. 2010). MiR-182 was also noteworthy in that increased expression of miR-182 has also been observed in malignant melanomas and gliomas (Segura et al. 2009, Jiang et al. 2010), where it is associated with metastasis and poor prognosis.
However, in order to substantiate the biological significance of the changed miRNAs, we performed a weighted target identification (Gallagher et al. 2010), followed by a molecular pathway analysis and an analysis of the expression of the target mRNAs, since ∼80% of miRNA regulations are reflected by changes in mRNA expression levels (Guo et al. 2010). Moreover, the function of miR-199b-5p and miR-144 was examined by introducing the two miRNAs in thyroid carcinoma cells.
The pathway analyses depicted an enrichment of transcripts encoding proteins directly involved in thyroid carcinogenesis and tumourigenesis, and, in the GSEA, we discovered a significant overlap with genes up-regulated in AC. The resemblance to AC may not be surprising, because there is a substantial overlap between the miRNAs that change in FC and those altered in AC (Braun et al. 2010). Moreover, in a PCA of all genes, FCs tend to distribute as a continuum ranging from relatively well-differentiated FCs to FCs that are located in close proximity to ACs in contrast to papillary cancers that form their own homogenous group (Borup et al. 2010). The common changes in miRNAs may also be interpreted as an indication that FC and ACs share a common tumourigenic pathway that differs from the papillary cancers. Moreover, loss of miRNAs is likely to represent an early event in follicular neoplasia since the majority of the miRNAs are also down-regulated in adenoma. Taken together, the results imply that the observed changes in miRNA signatures do have functional consequences in the tumours and may participate in the tumourigenesis. This was further reinforced by the demonstration that miR-199b-5p reduced cell-growth and that almost 30% of the computational predicted targets were in fact down-regulated by pre-miR199b in the cultured thyroid cells and correspondingly up-regulated in the thyroid carcinoma lacking the miRNA. MiR-144 had no significant effect on proliferation and we cannot exclude that loss of this miRNA reflects alterations in the vascularisation of the carcinoma (Rasmussen et al. 2010).
The tumours collected from consecutively referred patients whose sex and age were in accordance with that of larger epidemiological studies. Furthermore, there was no preponderance of oncogenic mutations in the tumour sets. To exploit if the miRNAs could be useful to depict FC from adenoma, we generated a diagnostic signature from two different technical platforms, the qRT-PCR assays was included as an independent validation (Git et al. 2010). Although the results need to be confirmed by independent studies, the performance of either platform was acceptable, as the classifiers exhibited an NPV of 83 and 92% for malignancies with the microarray and qRT-PCR-based platform respectively. The qRT-PCR platform provided a better separation of FA and FC than the microarray platform, which is reflected by higher accuracy. From a clinical point of view, the predicted probabilities derived from each individual sample are essential since they provide a quantitative value (on the reliability) to the extent in the accuracy of the diagnosis based on the classification. According to this, we found that most samples are classified with high accuracy.
Since we showed that miRNA-based classification of histopathological follicular thyroid specimens is possible, the next obvious step is to examine whether it is feasible to implement miRNA-based classification as an additional preoperative diagnostic tool. Taking the limited sensitivity and reproducibility of the histopathological diagnosis into account, the consistency between miRNA-based classification and the pathological diagnosis is surprisingly high. This could reflect the fact that all samples were examined by the same dedicated endocrine pathologist. Studies of inter-observer variations amongst pathologists in assessment of follicular lesions have demonstrated an observer variation for FC of 27%, where the carcinomas tended to be misdiagnosed as adenomas (Hirokawa et al. 2002, Kakudo et al. 2002). In a similar study an overall agreement between American and Japanese pathologists of 33 and 52% respectively, was found (Hirokawa et al. 2002).
All results considered, we conclude that thyroid follicular neoplasia is accompanied by major changes in miRNA expression that may be directly implicated in malignant transformation and may facilitate diagnosis of follicular thyroid cancer.
Supplementary data
This is linked to the online version of the paper at http://dx.doi.org/10.1530/JME-11-0039.
Declaration of interest
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.
Funding
This study was supported by the Agnes and Knut Mørk Foundation, the Research Foundation of Herlev University Hospital, the Toyota Foundation and the Novo Nordisk Foundation.
References
Bartel DP 2004 MicroRNAs: genomics, biogenesis, mechanism, and function. Cell 116 281–297. doi:10.1016/S0092-8674(04)00045-5.
Bartel DP 2009 MicroRNAs: target recognition and regulatory functions. Cell 136 215–233. doi:10.1016/j.cell.2009.01.002.
Borup R, Rossing M, Henao R, Yamamoto Y, Krogdahl A, Godballe C, Winther O, Kiss K, Christensen L & Hogdall E et al. 2010 Molecular signatures of thyroid follicular neoplasia. Endocrine-Related Cancer 17 691–708. doi:10.1677/ERC-09-0288.
Braun J, Hoang-Vu C, Dralle H & Huttelmaier S 2010 Downregulation of microRNAs directs the EMT and invasive potential of anaplastic thyroid carcinomas. Oncogene 29 4237–4244. doi:10.1038/onc.2010.169.
Chao A, Tsai CL, Wei PC, Hsueh S, Chao AS, Wang CJ, Tsai CN, Lee YS, Wang TH & Lai CH 2009 Decreased expression of microRNA-199b increases protein levels of SET (protein phosphatase 2A inhibitor) in human choriocarcinoma. Cancer Letters 291 99–107. doi:10.1016/j.canlet.2009.10.005.
Chen YT, Kitabayashi N, Zhou XK, Fahey TJ III & Scognamiglio T 2008 MicroRNA analysis as a potential diagnostic tool for papillary thyroid carcinoma. Modern Pathology 21 1139–1146. doi:10.1038/modpathol.2008.105.
Curado MP & Edwards B 2007 Cancer Incidence in Five Continents Vol. IX. MP Curada, B Edwards, HR Shin, H Storm, J Ferlay, M Heanue & P Boyle (eds). IARC Scientific Publication. No. 160, Lyon: IARC
Dudoit S, Shaffer JP & Boldrick JC 2003 Multiple hypothesis testing in microarray experiments. Statistical Science 18 71–103. doi:10.1214/ss/1056397487.
Faquin WC 2008 The thyroid gland: recurring problems in histologic and cytologic evaluation. Archives of Pathology and Laboratory Medicine 132 622–632.
Franc B, de la Salmoniere P, Lange F, Hoang C, Louvel A, de Roquancourt A, Vilde F, Hejblum G, Chevret S & Chastang C 2003 Interobserver and intraobserver reproducibility in the histopathology of follicular thyroid carcinoma. Human Pathology 34 1092–1100. doi:10.1016/S0046-8177(03)00403-9.
Friedman RC, Farh KK, Burge CB & Bartel DP 2009 Most mammalian mRNAs are conserved targets of microRNAs. Genome Research 19 92–105. doi:10.1101/gr.082701.108.
Gallagher IJ, Scheele C, Keller P, Nielsen AR, Remenyi J, Fischer CP, Roder K, Babraj J, Wahlestedt C & Hutvagner G et al. 2010 Integration of microRNA changes in vivo identifies novel molecular features of muscle insulin resistance in type 2 diabetes. Genome Medicine 2 9 doi:10.1186/gm130.
Garzia L, Andolfo I, Cusanelli E, Marino N, Petrosino G, De Martino D, Esposito V, Galeone A, Navas L & Esposito S et al. 2009 MicroRNA-199b-5p impairs cancer stem cells through negative regulation of HES1 in medulloblastoma. PLoS ONE 4 e4998 doi:10.1371/journal.pone.0004998.
Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis B, Gautier L, Ge Y & Gentry J et al. 2004 Bioconductor: open software development for computational biology and bioinformatics. Genome Biology 5 R80 doi:10.1186/gb-2004-5-10-r80.
Git A, Dvinge H, Salmon-Divon M, Osborne M, Kutter C, Hadfield J, Bertone P & Caldas C 2010 Systematic comparison of microarray profiling, real-time PCR, and next-generation sequencing technologies for measuring differential microRNA expression. RNA 16 991–1006. doi:10.1261/rna.1947110.
Guo H, Ingolia NT, Weissman JS & Bartel DP 2010 Mammalian microRNAs predominantly act to decrease target mRNA levels. Nature 466 835–840. doi:10.1038/nature09267.
He H, Jazdzewski K, Li W, Liyanarachchi S, Nagy R, Volinia S, Calin GA, Liu CG, Franssila K & Suster S et al. 2005 The role of microRNA genes in papillary thyroid carcinoma. PNAS 102 19075–19080. doi:10.1073/pnas.0509603102.
Hegedus L, Bonnema SJ & Bennedbaek FN 2003 Management of simple nodular goiter: current status and future perspectives. Endocrine Reviews 24 102–132. doi:10.1210/er.2002-0016.
Hirokawa M, Carney JA, Goellner JR, DeLellis RA, Heffess CS, Katoh R, Tsujimoto M & Kakudo K 2002 Observer variation of encapsulated follicular lesions of the thyroid gland. American Journal of Surgical Pathology 26 1508–1514. doi:10.1097/00000478-200211000-00014.
Jiang L, Mao P, Song L, Wu J, Huang J, Lin C, Yuan J, Qu L, Cheng SY & Li J 2010 miR-182 as a prognostic marker for glioma progression and patient survival. American Journal of Pathology 177 29–38. doi:10.2353/ajpath.2010.090812.
Kakudo K, Katoh R, Sakamoto A, Asa S, DeLellis RA, Carney JA, Naganuma H, Kameyama K & Takami H 2002 Thyroid gland: international case conference. Endocrine Pathology 13 131–134. doi:10.1385/EP:13:2:131.
Kroll TG, Sarraf P, Pecciarini L, Chen CJ, Mueller E, Spiegelman BM & Fletcher JA 2000 PAX8–PPARgamma1 fusion oncogene in human thyroid carcinoma [corrected]. Science 289 1357–1360. doi:10.1126/science.289.5483.1357.
Nikiforova MN, Tseng GC, Steward D, Diorio D & Nikiforov YE 2008 MicroRNA expression profiling of thyroid tumours: biological significance and diagnostic utility. Journal of Clinical Endocrinology and Metabolism 93 1600–1608. doi:10.1210/jc.2007-2696.
Pallante P, Visone R, Ferracin M, Ferraro A, Berlingieri MT, Troncone G, Chiappetta G, Liu CG, Santoro M & Negrini M et al. 2006 MicroRNA deregulation in human thyroid papillary carcinomas. Endocrine-Related Cancer 13 497–508. doi:10.1677/erc.1.01209.
Platt JC 1999 Probabilities for SV machines. In Advances in Large Classifiers, pp 72–85. Eds PJ Bartlett, B Schölkopf, D Schuurmans & AJ Simola. Cambridge, MA: MIT Press
Rasmussen KD, Simmini S, Abreu-Goodger C, Bartonicek N, Di GM, Bilbao-Cortes D, Horos R, Von LM, Enright AJ & O'Carroll D 2010 The miR-144/451 locus is required for erythroid homeostasis. Journal of Experimental Medicine 207 1351–1358. doi:10.1084/jem.20100458.
R Development Core Team. R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria, ISBN 3-900051-07-0. URL. Downloads. 2007
Rosenfeld N, Aharonov R, Meiri E, Rosenwald S, Spector Y, Zepeniuk M, Benjamin H, Shabes N, Tabak S & Levy A et al. 2008 MicroRNAs accurately identify cancer tissue origin. Nature Biotechnology 26 462–469. doi:10.1038/nbt1392.
Schmid KW & Farid NR 2006 How to define follicular thyroid carcinoma? Virchows Archiv 448 385–393. doi:10.1007/s00428-006-0162-0.
Schmittgen TD 2008 Regulation of microRNA processing in development, differentiation and cancer. Journal of Cellular and Molecular Medicine 12 1811–1819. doi:10.1111/j.1582-4934.2008.00483.x.
Schmittgen TD & Livak KJ 2008 Analyzing real-time PCR data by the comparative C(T) method. Nature Protocols 3 1101–1108. doi:10.1038/nprot.2008.73.
Segura MF, Hanniford D, Menendez S, Reavie L, Zou X, Alvarez-Diaz S, Zakrzewski J, Blochin E, Rose A & Bogunovic D et al. 2009 Aberrant miR-182 expression promotes melanoma metastasis by repressing FOXO3 and microphthalmia-associated transcription factor. PNAS 106 1814–1819. doi:10.1073/pnas.0808263106.
Simon R, Lam A, Li MC, Ngan M, Menenzes S & Zhao Y 2007 Analysis of gene expression data using BRB-ArrayTools. Cancer Informatics 3 11–17.
Tusher VG, Tibshirani R & Chu G 2001 Significance analysis of microarrays applied to the ionizing radiation response. PNAS 98 5116–5121. doi:10.1073/pnas.091062498.
Vapnik V 1998 Statistical Learning Theory. New York, NY, USA: Wiley-Interscience
Weber F, Teresi RE, Broelsch CE, Frilling A & Eng C 2006 A limited set of human MicroRNA is deregulated in follicular thyroid carcinoma. Journal of Clinical Endocrinology and Metabolism 91 3584–3591. doi:10.1210/jc.2006-0693.
Wettenhall JM & Smyth GK 2004 limmaGUI: a graphical user interface for linear modeling of microarray data. Bioinformatics 20 3705–3706. doi:10.1093/bioinformatics/bth449.
Yamada Y, Enokida H, Kojima S, Kawakami K, Chiyomaru T, Tatarano S, Yoshino H, Kawahara K, Nishiyama K & Seki N et al. 2010 MiR-96 and miR-183 detection in urine serve as potential tumour markers of urothelial carcinoma: correlation with stage and grade, and comparison with urinary cytology. Cancer Science 102 522–529. doi:10.1111/j.1349-7006.2010.01816.x.
*(F N Bennedbæk and F C Nielsen contributed equally to this work)
