The miR-146b-3p/PAX8/NIS Regulatory Circuit Modulates the Differentiation Phenotype and Function of Thyroid Cells during Carcinogenesis

Thyroid cancer has in general terms a good prognosis, as the majority of the patients with this disease can be cured with surgery and radioactive iodide treatment. The ability of differentiated thyroid cells to accumulate iodide is clinically highly relevant because it makes it possible for patients with thyroid cancer to be treated with ablative doses of radioactive iodide after an adequate stimulation by thyroid-stimulating hormone (TSH; ref. 1). However, some patients develop metastatic disease refractory to radioactive iodide treatment and their life expectancy significantly decreases. Thus, the maintenance of the thyroid differentiated phenotype during tumor transformation has a critical impact in thyroid cancer patient's survival (2).

The maintenance of the thyroid differentiated phenotype requires the presence of PAX8, a member of the paired box (PAX) family of transcription factors (TF; ref. 3). Together with the other thyroid TFs NKX2-1 and FOXE1, PAX8 is involved in thyroid follicular cell development and expression of thyroid-specific genes such as the sodium/iodide symporter (SLC5A5, also named as NIS), thyroglobulin (Tg), and thyroperoxidase (TPO; refs. 4, 5). These genes are essential for thyroid differentiation as they mediate the metabolism of iodide, leading to the synthesis of active thyroid hormone. One of the most important and well-established transcriptional targets of PAX8 is NIS. This symporter is a key plasma membrane protein that mediates active iodide transport in the thyroid and other tissues (6). In the healthy thyroid, NIS-mediated iodide uptake is the first step in thyroid hormone biosynthesis. NIS also plays a central role in thyroid cancer treatment. As long as a few NIS molecules are functionally expressed in thyroid cancer cells, thyroid cancer is successfully treated by thyroidectomy followed by radioactive iodide (¹³¹I) administration. Radioactive iodide selectively targets and destroys any remnant or metastatic NIS-expressing thyroid cancer cells. Radioactive iodide treatment is the most effectively targeted internal radiation anticancer therapy ever devised and it has been in use for over 60 years. Therefore, loss of the thyroid differentiated phenotype, particularly loss of NIS function, is one of the most important hallmarks of thyroid cancer progression, leading to iodide-refractory metastatic disease and a worse outcome of the patients (7, 8).

MicroRNAs (miRNA) inhibit translation or induce mRNA degradation in general by binding to the 3′-untranslated region (3′-UTR) of target mRNAs. Since initial observation, about 3,000 human miRNAs have been registered in miRBase (v.20.0; ref. 9) and are thought to regulate 50% of the transcriptome. Although a previous classical approach of cloning has partially revealed miRNA expression profile in thyroid tumors (10, 11), low throughput with low sensitivity and poor resolution makes these approaches limited to define miRNome. Genome-wide sequencing of miRNAs can identify miRNome in-depth, which reveals miRNA expression differences as well as the individual miRNA abundance. As a minimum threshold amount must be reached for miRNAs to repress target mRNAs (12–15), the abundance of miRNAs and their ratios in the entire miRNome of specific cell or tissue may be very important for their functions.

When preparing this article, results from The Cancer Genome Atlas (TCGA) project based on a comprehensive multiplatform analysis of 496 papillary thyroid carcinomas (PTC), the most common type of thyroid cancer, were published (16). In such work, several miRNAs were associated with less differentiated tumors. Here, we took an integrated approach and based on the abundance of the miRNAs and their ratios in the entire miRNome, we analyzed miRNAs and the transcriptome from paired normal and tumor thyroid tissue from 8 patients with PTC. Our aim was to uncover the underlying miRNA regulatory network that determines differentiation in PTCs and to provide functional studies to characterize those miRNAs involved in the modulation of genes essential for thyroid differentiation and iodide uptake, namely PAX8 and NIS.

For the sequencing experiment, fresh-frozen samples from PTC tumors (n = 8) and contralateral normal thyroid tissue from the same patient (n = 8) were collected at the Biobank of the Hospital Universitario La Paz (Madrid, Spain). The clinical characteristics of patients are summarized in Supplementary Table S1 and other details in Supplementary Materials and Methods. An independent cohort of 16 samples from PTC tumors was used for validation.

The sequencing procedure was carried out using the Genome Analyzer IIx Platform (Illumina) at the Genomics Core Unit of CNIO (the Spanish National Cancer Research Centre, Madrid, Spain) using the standard protocols recommended for small RNA-seq (Illumina TruSeq Small RNA) and mRNA-seq (Illumina TruSeq Stranded mRNA). Using Cutadapt software (v. 1.2.1), 3′ adapters from small RNA reads were removed, filtering reads with lengths of 17 to 34 nt. The remaining short reads were aligned to the human genome (UCSC, hg19 assembly) using Bowtie algorithm (v0.12.7; ref. 17) with the criteria of perfect match. Reads mapping on each mature miRNA genomic location were counted using HTSeq-count (v. 0.5.4), toward the miRNA annotation from miRBase version 20 (hg19 assembly, June 2013). Clean sequenced reads (excluding reads containing ambiguous base and adaptor contaminants) were used for further analysis and reads classified as small RNAs from other species [rRNA, tRNA, scRNA, snRNA, snoRNA, repeat-associated small RNAs and mRNAs (exons/introns)] were excluded (Supplementary Fig. S1). The mRNA reads were aligned to the human genome (UCSC, hg19 assembly) using TopHat v.2.0.4 (18), permitting two mismatches and a maximum of five multi-hits. The gene-level expression was calculated as the sum of all read counts over their exons by using Htseq-count and the gene annotation from the reference genome (hg19, UCSC; ref. 19). All sequencing data can be downloaded from Gene Expression Omnibus (GEO) under accession number GSE63511.

TargetScan algorithm (20) was used to predict hypothetical associations between miRNA and mRNA pairs. A second algorithm based on DIANAmicroT was used to validate the predictions (http://www.microrna.gr/microT-CDS). Inverse correlations between expression levels of miRNAs and their putative target genes were calculated using the Pearson correlation test, adjusting for multiple testing (Benjamini and Hochberg method). All significant predictions were integrated in addition of experimentally validated interactions collected from TarBase (v.6), miRecords (v.3), and miRTarbase (v.3) databases. Networks comprising genes and miRNAs were drawn by Cytoscape (21).

Total RNA was isolated from cells with TRIzol (Invitrogen) according to the manufacturer's instructions. RT-PCR and qRT-PCR were performed as described in Supplementary Materials and Methods.

PCCl3 cells were grown in Coon's modified Hams F12 medium supplemented with a six-hormone mixture (1 nmol/L TSH, 10 μg/mL insulin, 10 ng/mL somatostatin, 5 μg/mL transferrin, 10 nmol/L hydrocortisone, and 20 ng/mL glycyl-l-histidyl-l-lysine acetate) all from Sigma-Aldrich, and 5% donor calf serum (Life Technologies). Near-confluent cells were starved for TSH and insulin in the presence of 0.2% serum (starvation medium) for 4 days. Then, TSH (0.5 mU/mL) alone or together with IGF1 (100 ng/mL) or TGFβ (10 ng/mL; the last two from PeproTech) were added to the culture medium at 24 hours and RNA was then extracted.

Pax8 silencing was performed either transfecting the cells with Pax8 siRNA or Scrambled conditions (10 ng siRNA/mL; Dharmacon) as described previously (22) or transducing the cells with lentivirus containing short hairpin RNA against Pax8, using a shScramble as control (pGIPZ_Pax8 (clon#V3LHS_408573) and pGIPZ-Scramble, respectively (Open Biosystems). VSV-G pseudotyped lentivirus production was performed as previously described (23). MDCKhNIS cells have been described elsewhere (24) and the human thyroid follicular cell line Nthy-ori 3-1 was obtained from the European Collection of Cell Cultures (ECACC). HeLa cell were grown as described previously (23). Cells stably expressing miRNA are described in Supplementary Materials and Methods. Cell lines were routinely authenticated every 6 months by short tandem repeat (STR) profiles using the Applied Biosystems Identifier kit in the Genomic Facility at the Institute of Biomedical Research (IIBm; Madrid, Spain).

A total of 2.5 × 10⁵, MDCKh NIS or Nthy-ori3-1, cells seeded in 12-well plates were transfected with 1 μg/well miRNA precursor expression vector or the null control vector using calcium phosphate or the lipid-based transfection agent Fugene6 according to the manufacturer's instructions (Promega). Transfected cells were selected with puromicin and the expression of miRNA detected by GFP and /or qRT-PCR.

miR-146b and unrelated miRNAs (miR-373, miR-15a) precursors (pre-mir) were cloned into pEGP-miR expression vector (Cell Biolabs), specifically into BamHI and NheI binding sites. The generated constructs were transfected to MDCKhNIS and Nthy-ori 3-1 cells as described in Supplementary Materials and Methods.

Cells were lysed in lysis buffer containing 1% Nonidet P-40, 1 mmol/L EDTA, 50 mmol/L Tris–HCl (pH 7.5), and 150 mmol/L NaCl, supplemented with complete protease inhibitors mixture (Roche Diagnostics). All samples were diluted in loading buffer and heated at 90°C for 5 minutes except for NIS, which was heated at 37°C for 30 minutes. Samples were then separated by SDS-PAGE, transferred onto nitrocellulose membrane (Schleicher & Schuell), and immunoreactive proteins visualized by enhanced chemiluminescence (GE Healthcare). Antibodies against hNIS and rNIS were kindly provided by Dr. Carrasco (Yale School of Medicine, New Haven, CT). The antibodies for actin and PAX8 were from Santa Cruz Biotechnology Inc. and BioPat, respectively.

PCCl3 and MDCKhNIS cells were assayed for iodide transport as previously described (25) and detailed in Supplementary Materials and Methods.

A differential expression analysis based on the read counts from the sequencing data was performed using the edgeR package (v.3.2.4; ref. 26). A Trimmed means of M values normalization method of the gene-level expression was applied (27). To dampen the multiple testing issues, low-expressed tags were filtered out based on a minimum cutoff level [miR-level of 0.6 “counts per million” (CPM) for small RNA-seq dataset, gene-level of 1 CPM for mRNA-seq dataset] in at least 50% of samples within any biologic group. The negative binomial Generalized Linear Model for a design of paired samples was used to detect genes differentially expressed between PTC and normal tissue. After correcting P values for Benjamini and Hochberg method, genes and miRNAs with false discovery rates (FDR) <0.05 were considered as differentially expressed.

The GOseq package (v. 1.12.0; ref. 28) was used to identify overrepresented gene sets defined in KEGG (Kyoto Encyclopedia of Genes and Genomes) among differentially expressed genes. The Wallnius noncentral hypergeometric distribution method was used to approximate the true distribution of members of each gene set among the list of deregulated genes. Gene sets with FDR <0.05 were considered as overrepresented. The results in qRT-PCR, luciferase activity, and iodide uptake are expressed as the mean ± SEM of at least three different experiments performed in triplicate. Statistical significance was determined by the t test analysis (two-tailed), and differences were considered significant at a P value of <0.05.

To carry out an in-depth analysis of the functional miRNome, we performed next-generation sequencing of small RNA in paired normal and tumor thyroid tissue from 8 patients with PTC, the most common type of thyroid cancer. The abundance value of each known miRNA was normalized using CPM in each small RNA library. Significance analysis of differential miRNA expression and hierarchical clustering analysis clearly separated tumors from normal tissue samples (Supplementary Fig. S1). The miRNAs with >1,000 CPMs were considered to be abundantly expressed and, thus, most likely to be functional (15). We found that approximately 15% of the detected miRNAs were abundantly expressed (81 of 577 in normal thyroid tissues and 77 of 574 in tumor tissues) accounting for 96% of all miRNAs reads (Supplementary Fig. S1 and Supplementary Dataset S1 and S2). The great majority of the most abundant miRNAs (87%) were present in both normal and tumor tissue with no significant variations identifying a thyroid-specific miRNome (Supplementary Fig. S1). Altogether, our sequence analysis accurately classifies normal and tumor tissue and shows that PTC and normal thyroid express a large, highly similar set of abundantly expressed miRNAs that corresponds to the functional miRNome of the thyroid.

We next wanted to know which miRNAs are the major players in PTC pathogenesis. As only a few miRNAs are abundantly expressed in the miRNome and they seem to be the most important in thyroid biology, only miRNAs with CPM >1,000 and more than 1.5-fold change were considered most likely to be important in PTC pathogenesis in this study. We identified a set of 12 upregulated and eight downregulated miRNAs after pair-wise comparison between the eight malignant tumors and their corresponding normal samples (Table 1). In the list of upregulated miRNAs, the most abundant and upregulated by far was miR-146b-5p. Interestingly, miR-146b-3p, formerly thought to be a passenger strand, was also among the most abundant upregulated miRNAs. Because miR-155 and miR-34a are close to the threshold of 1,000 CPMs and have been previously shown to be deregulated in PTC, we decided to consider them as more likely to be functional. On the other hand, as miR-21-3p and miR-181a-2-3p are not included in TargetScan database, they were not considered in this work. In the list of downregulated miRNAs, the three most abundant were miR-451a, miR-486, and miR-100b. Low abundant miRNAs being significantly deregulated in PTC are shown in Supplementary Dataset S2 and, although its biologic relevance remains to be proved, we cannot rule out that they play a functional role. Subsequent validation through qRT-PCR of several deregulated miRNAs was performed in an independent cohort of 16 patients with PTC. (Supplementary Fig. S2A). We also performed next-generation sequencing in eight extra tumors with PTC obtaining similar results (Supplementary Fig. S2B). Together, the miRNomes of human normal thyroid and PTC led us to identify the consistently deregulated miRNA in thyroid cancer.

To identify target genes of the most abundant deregulated miRNAs, we first performed next-generation sequencing of mRNA in the paired normal and tumor thyroid tissue from the same 8 patients with PTC. Overall, 910 genes were significantly deregulated (536 were overexpressed and 374 were downregulated; Supplementary Dataset S3). We further characterized the biologic signatures associated with our sequencing analysis. Overrepresentation analysis (ORA) based on the KEGG pathways database revealed enrichment of cell membrane–mediated pathways (Supplementary Table S2), including ECM–receptor interaction, cell adhesion molecules (CAM) and focal adhesion. This finding highlights the role of integrins among other membrane-bound mediators in thyroid malignancy, similar to what has been previously described (29). ORA also revealed enrichment of small cell lung cancer and TGFβ signaling pathways, highlighting the role of these canonical pathways in thyroid cancer.

A single miRNA has been shown to target multiple mRNAs to regulate gene expression. Conversely, the expression of a single gene can be regulated by several miRNAs. Thus, the generation of miRNAs-gene regulatory networks is believed to affect essential cellular functions. To identify target genes of the upregulated miRNAs in PTC, we used the computational target prediction program TargetScan and searched into the 374 downregulated genes in our sequencing analysis. This revealed a shorter list of 217 potential target genes for the upregulated miRNAs in PTC. In addition to having binding sites in the 3′-UTRs of their predicted targets, these 12 miRNAs were also inversely correlated with the expression levels of their predicted targets (FDR < 0.05, based on linear regression model, see Materials and Methods). As a result, we identified a master miRNA-gene regulatory network for PTC consisting of 12 miRNAs that can be visualized in Fig. 1. Such network reveals new candidate genes important for thyroid carcinogenesis. It also shows that some genes essential for thyroid differentiation are under the influence of this network (Fig. 1A, circles in red). The three miRNAs predicted to regulate the largest number of targets are miR-182, miR-146b-3p, and miR-34a (regulating 71, 66, and 65 targets, respectively; Fig. 1B).

Network analysis have shown that genes that are subject to extensive miRNAs regulation, referred as to target hubs, are more likely to be biologically relevant (30, 31). In addition, it has been shown that many TFs are among these target hubs and there is increasing evidence of reciprocal regulation between miRNAs and TFs as well as the existence of extensive coordination in the regulation of shared targets genes at the genome-scale level (32, 33). Considering only bioinformatics prediction, we found that 48 out of 217 (22%) of the target genes in PTC were predicted to be regulated by five or more miRNAs and were thus considered target hubs (Fig. 1A, circles with a thick edge in the center of the network and Fig. 1C). Similar results were obtained using a second algorithm, DIANAmicroT (Supplementary Fig. S3). Among these genes, several important tumor suppressors and TFs involved in human cancer were found (Supplementary Table S3). Of note, PAX8 is predicted to be modulated by five miRNAs revealing this nuclear protein as a target hub.

As miR-146b-5p and -3p are two of the most abundant upregulated miRNA in PTC, we focused our analysis on both strands generated by the same pre-miR. We found that miR-146b-5p and miR-146b-3p target 54 and 66 genes, respectively, being 22 of these genes targeted by both strands (Fig. 2A and B). The majority of these target genes are related to ECM–receptor interaction, CAMs and focal adhesion (Fig. 2C and Supplementary Table S4). Of note, miR-146b-3p targets two genes essential for thyroid differentiation, namely PAX8 and SLC5A5 (NIS; Fig. 2B, circles in red). In addition, miR-146b-5p targets DIO2, the iodothyronine deiodinase that converts T4 into T3; and both miR-146b-3p and -5p target IYD (DEHAL1), the iodotyrosine deiodinase that controls the recycling of iodide for thyroid hormone synthesis (Fig. 2B, circles in red). Overall, these results uncover the target genes of miR-146b-5p and -3p in PTC, revealing that this miRNA precursor is a key regulator of iodide-metabolizing genes. Interestingly, SLC5A5 (NIS), IYD (DEHAL1), and DIO2 are downstream targets of PAX8 (22, 34). Therefore, miR-146b regulates both the TF (PAX8) and its downstream target genes (NIS, DEHAL1, and DIO2), suggesting that miR-146b forms a feed-forward loop that controls gene expression through a direct and indirect mechanism (Fig. 2D).

We wanted to extend our analysis to other miRNAs involved in thyroid differentiation. According to TargetScan, we found that a group of 11 upregulated miRNAs were predicted to target genes essential for thyroid differentiation (Supplementary Table S5). As mentioned before, PAX8 is predicted to be modulated by five miRNAs (miR-146b-3p, miR-182, miR-221, miR-222, and let-7e). However, downregulated miRNAs are also targeting PAX8, which might be counteracting the inhibitory effects mediated by the upregulated miRNAs. NIS is predicted to be modulated by miR-146b-3p and by miR-21-5p. FOXE1 is predicted to be modulated by miR-155 and TSHR by miR-34a, though other downregulated miRNAs are involved. TG seems not to be modulated by any deregulated miRNAs in PTC.

Both TFs and miRNAs can exert a widespread impact on gene expression. Network studies have revealed that miRNAs and TFs act coordinately to regulate sets of common target genes at the genome-scale level through the existence of reciprocal regulation between each other. More locally, such miRNAs–TF pairs may form small regulatory circuits that constitute successful mechanisms of gene expression in one particular organism or tissue. As mentioned before, we observed that miR-146b-3p was predicted to regulate both the TF (PAX8) and its downstream target gene (NIS). We therefore wanted to explore whether miR-146b and PAX8 shared more target genes in the sequencing data derived from the tumor samples. For this purpose, we used our previously characterization of the downstream targets genes of PAX8 in the rat thyroid cell line PCCl3 (22). We first confirmed that PAX8 was significantly downexpressed in the tumors (FC, 0.5; P < 0.05; not shown). We next searched in our sequence analysis and found that 51 downregulated genes in the tumors were downstream target genes of PAX8. Interestingly, PAX8 and miR-146b-3p shared 15 target genes including SLC5A5 (NIS) and IYD (DEHAL; Fig. 3A and B).

We next wanted to study whether PAX8 and miR-146b were regulating each other. To explore whether PAX8 regulates miR-146b expression, we first searched for putative binding sites of PAX8 in the miR-146b regulatory region. We found several PAX8-binding elements upstream the transcription start site of both rat and human miR-146b (Fig. 4A and Supplementary Dataset S4). We next studied the effect of Pax8 silencing on miR-146b expression after siRNA transfection in PCCl3 cells (Fig. 4B). We observed that the levels of miR-146b significantly decreased after silencing Pax8, suggesting that this TF positively regulates the transcription of miR-146b. The same results were obtained when Pax8 was silenced by shRNA (not shown).

Conversely, to explore whether miR-146b regulates PAX8 expression, Nthy-ori cells were transfected with an expression vector encoding for the miR-146b precursor, which expresses the two strands miR-146b-5p and miR-146-3p. Overexpression of miR-146b decreased the protein levels of endogenous PAX8 (Fig. 4C), yet no significant effect was observed in the mRNA levels (Fig. 4E). TargetScan-based bioinformatics predictions showed that there were three putative binding sites in the 3′-UTR of PAX8 for miR-146b-3p (Fig. 4E). Moreover, we addressed whether miR-146b-3p was specifically targeting the 3′-UTR of PAX8. We thus generated a luciferase reporter vector encoding for the entire 3′-UTR and another one for the 3′-UTR reverse, used as a negative control. Using HeLa cells, we observed that miR-146b overexpression induced a 30% decrease of the luciferase activity on average of the entire 3′-UTR of PAX8 (Fig. 4F). These results show that miR-146b-3p directly represses PAX8 expression by targeting the 3′-UTR. And overall, we provide evidence whereby in thyroid cells operate a negative feedback loop in which the TF (PAX8) limits its own activity by inducing a repressor (miR-146b; Fig. 4G).

To explore whether miR-146b-3p was able to downregulate NIS expression, we first used MDCKhNIS. This cell system expresses NIS in the cell membrane and mediates active transport of iodide into the cytoplasm (24). We transfected these cells an expression vector encoding for the miR-146b precursor. We observed that miR-146b decreases protein levels of NIS (Fig. 5A), yet no significant effect was observed in the mRNA levels (Fig. 5B). In addition, an 80% reduction in iodide uptake was seen (Fig. 5C). These results show that miR-146b impairs NIS mRNA translation and subsequently decreases protein levels and transport activity of NIS.

We next wanted to address whether miR-146b-3p was specifically targeting the 3′-UTR of hNIS. The bioinformatics predictions informed us that one putative target site was present at positions 3-9 in the 3′-UTR (Fig. 5E). We generated a luciferase reporter vector encoding for the entire 3′-UTR and another one encoding for a short region of 7nt in tandem harboring the target site at positions 3–9 (called 3–9 nt segm). The reverse constructs of the entire 3′-UTR and the 3–9 nt segm were used as negative controls. After transfecting HeLa cells, we observed a 25% decrease of the luciferase activity on average both with the entire 3′-UTR and the 3–9 nt segm when we transfected cells with miR-146b (Fig. 5E). These results provide evidence that the direct repression of NIS expression by miR-146b-3p requires a binding site at positions 3–9 of the 3′-UTR.

We finally wanted to investigate whether miR-146b was hormonally regulated in differentiated thyroid cells. TSH is essential for the thyroid cell to express the genes involved in the uptake and metabolism of iodide and hence, allows the thyroid cell to be fully differentiated. Thus, we assumed that those miRNAs more likely to be involved in targeting thyroid differentiating genes are necessarily downregulated by TSH in normal thyroid cells. To explore whether the expression of miR-146b was dependent on TSH regulation, we used rat PCCl3 cells. miR-146b was strongly downregulated after stimulating the cells with TSH (Fig. 5F). In addition, this downregulation was recovered by IGFI and TGFβ, known repressors of TSH-induced NIS expression (Fig. 5G; refs. 35, 36). Thus, endogenous levels of miR-146b are tightly controlled by TSH in differentiated thyroid cells.

In this work, we identified a master miRNA regulatory network in PTC that is involved in essential biologic processes such as differentiation. Functional studies characterized miR-146b-3p as a repressor of PAX8 and NIS and showed that miR-146b and PAX8 regulate each other, unveiling a regulatory circuit that determines the differentiated phenotype in PTC.

Identifying the molecular events leading to the loss of thyroid differentiation in thyroid cancer attracts much attention, particularly loss of NIS expression. NIS mediates active transport of iodide into the thyroid cell, the basis for the diagnosis and therapeutic management of thyroid cancer patients with radioiodide. We demonstrate for the first time that NIS is posttranscriptionally modulated by miRNAs in thyroid cancer. Very recently, it has been shown that miR-339 modulates NIS expression in rat normal thyroid cells (37), yet this miRNA has not been shown to be deregulated in thyroid cancer in previous studies including ours (16, 38, 39). We consistently show that miR-146b-3p binds to the 3′-UTR of NIS (site 3–9 nt), leading to an impaired translation of the protein and subsequently decreasing the iodide uptake of the cells. Several mechanisms have been proposed to explain NIS repression in thyroid cancer in order to establish new therapeutic approaches to try to reinduce iodide uptake (8, 40). All of them are based on the transcriptional repression induced by aberrant signaling pathways and/or epigenetic events. Here, we show an additional mechanism based on the posttranscriptional regulation of NIS mediated by miRNAs.

We also show that miR-146b-3p is strongly downregulated by TSH in normal thyroid cells (PCCl3) and its expression is recovered by IGF1 and TGFβ, two known repressors of NIS expression via PI3K and Smads, respectively (35, 36). This highlights the importance of a tight hormonal regulation of miRNA expression in the thyroid cell. In accordance, it has been previously demonstrated that TSH signaling through CREB1 represses several miRNAs by physically interacting with their regulatory region (41). As TSH signaling is impaired during thyroid carcinogenesis, the miRNAs are able to escape from this TSH-mediated repression. In addition to the impairment of TSH signaling, TGFβ and PI3K signaling are known to be hyperactivated in thyroid cancer (8, 42). Thus, we suggest that aberrant signaling of key growth factors and cytokines (i.e., TSH, TGFβ, and IGF1) are contributing to the strong upregulation of miR-146b seen in thyroid cancer (up to 35-fold).

Apart from regulating NIS, the most abundant miRNAs are involved in the modulation of other genes essential for thyroid differentiation. PAX8 is a TF that specifically governs the transcription of NIS and is necessary for the development of the thyroid gland and for maintaining the differentiated state in the thyroid of the adult. Moreover, this TF is a master regulator of other key cellular processes such as cell-cycle regulation, DNA repair, replication, metabolism, and cell polarity (22, 43). We show that PAX8 is predicted to be targeted by five of the most abundant upregulated miRNAs in thyroid cancer (Fig. 1C and Supplementary Table S3), suggesting that this TF is a target hub that is in accordance with its biologic importance in the thyroid gland. One of such miRNAs, miR-146b-3p, effectively represses PAX8 through binding to its 3′-UTR. Intriguingly, we further show that PAX8 and miR-146b regulate each other and share common targets, including iodide metabolizing genes. This provides evidence that such miR-TF pair forms small regulatory circuits that constitutes a successful mechanism of regulating gene expression in the thyroid.

Regulatory circuits involving miRNAs and TFs are prevalent mechanisms of gene expression at the genome scale-level (44, 45). One of such regulatory circuits is the feed-forward loop where the miRNA simultaneously represses the TF and its target genes. This circuit functions to suppress leaky transcription and hence target genes will only be expressed whether the levels of the miRNA decrease. Here, we describe a miR-146b–PAX8–NIS circuit as an example of a feed-forward loop (Figs. 2E and 6) in which the miRNA (146b) represses the TF (PAX8) and its downstream target (NIS; Fig. 2E). In addition and complementary to this feed-forward loop, we also describe another regulatory circuit in which miR-146b and PAX8 form a negative feedback loop where the TF (PAX8) limits its own activity by inducing a repressor (miR-146b; Fig. 4G). Such a loop can result in coexpression of both components, either in steady-state levels or in oscillation. (46). The expression of both components depends on additional input signals. As a consequence, network circuits allow the spreading of regulatory effects. An upstream signal activates the miRNA, which, in turn, represses all its direct targets including the TF. As a result, all downstream targets of the TF are also repressed. In thyroid cells, the spreading of regulatory effects concerning PAX8 and miR-146b affect key biologic functions such as thyroid differentiation. Particularly, common target genes, such as SLC5A5 (NIS), IYD (DEHAL), and DIO2, reveal the importance of controlling the expression of a set of genes regulating the metabolism of iodide.

Taken together, we propose an integrated model that may be operating to determine the thyroid differentiated phenotype and in the end, the uptake of iodide (Fig. 6). Thyroid cells are under the control of TSH and show lower levels of miR-146b and higher levels of PAX8. Upon an oncogenic signal (e.g., BRAF), TSH signaling is impaired and TGFβ and PI3K signaling hyperactivated, leading to higher levels of miR-146b and lower levels of PAX8. As a result, tumor cells adopt a less differentiated state that, for instance, makes them more insensitive to radioactive iodide. However, this state can change if an external signal transiently overcomes the oncogenic signal by decreasing the levels of the miR-146b. An example of such an external signal can be exogenous administration of recombinant human TSH, which is widely used in the clinical practice to enhance iodide uptake in thyroid cancer patients. Another example of an external signal that may promote a change in the differentiated state is kinase inhibitors (i.e., selumetinib).

We also unveiled for the first time the main target genes of miR-146b-5p and 146b-3p in PTC, both of them strands generated by the miR-146b precursor. As previously shown by other studies (39, 47), both strands are significantly upregulated in PTC, indicating that the 3p strand, although suffering degradation, reaches the minimum threshold amount to be considered as functional. miR-146b has been demonstrated to be a prognostic factor for PTC as it is associated with aggressive clinicopathologic features and a poor clinical outcome (48). It has also been proven to be elevated in the serum of patients with PTC (49). Some of its target genes have been validated in thyroid cancer (50). Thus, although more functional studies are needed, particularly in animal models, we believe that miR-146b can be currently considered as an oncomir in thyroid cancer. Our results contribute to this notion as we demonstrate that one of the strands specifically targets NIS and PAX8, key players in the dedifferentiation process occurring in thyroid cancer.

In contrast to expression profiling based on microarrays, next-generation sequencing enables the study of miRNAs at unprecedented quantitative and qualitative accuracy. As mentioned before, quantitative analysis helps to discriminate those miRNAs more functionally prominent. In fact, over 60% of detected miRNAs have no real activity, indicating that the functional miRNome of a cell is considerably smaller than previously thought (15). In addition, it allows the characterization of both the 3p and 5p mature strands of a given miRNA. The first study performing next-generation sequencing of miRNAs in PTC mainly focused on isomer analysis (39). In another recent study, Huang and colleagues (38) analyzed the miRNAs expression datasets of PTC from TCGA Data Portal. The authors emphasized that immune responses are significantly enriched and under specific regulation in the direct miRNA–target network, but miRNAs regulating thyroid differentiation genes were not addressed. Very recently, while preparing this article, a study based on TCGA (16) showed that three selected miRNAs that were epigenetically regulated (miR-21, miR-146b, and miR-204) contributed to loss of thyroid differentiation as they correlated to gene expression data for nine thyroid genes including NIS (but not PAX8). However, functional studies proving the impact of miRNAs in thyroid differentiation are needed. In our work, we went one step further and provide functional evidence for the requirement of miR-146b-3p as a key player in loss of thyroid differentiation in PTC.

In summary, we present in this work the functional miRNomes of normal thyroid and PTC tumors, providing resources for investigating the regulatory roles of miRNAs in thyroid biology and thyroid carcinogenesis. Such miRNAs play important functions as they generate regulatory circuits that modulate genes essential for thyroid differentiation. We provide evidence to consider the regulatory circuit based on miR-146b-3p–PAX8–NIS as a regulator of the differentiated state in thyroid tumor cells. Therefore, inhibition of the miR-146b, one of the key miRNAs in the network, may represent an important strategy for the treatment of PTC, particularly to reinduce the uptake of radioactive iodide in patients.

No potential conflicts of interest were disclosed.

Conception and design: G. Riesco-Eizaguirre, L. Wert-Lamas, P. Santisteban

Development of methodology: G. Riesco-Eizaguirre, L. Wert-Lamas, L.P. Fernández

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Riesco-Eizaguirre, L. Wert-Lamas, A. Sastre-Perona, P. Santisteban

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Riesco-Eizaguirre, L. Wert-Lamas, J. Perales-Patón, P. Santisteban

Writing, review, and/or revision of the manuscript: G. Riesco-Eizaguirre, L. Wert-Lamas, A. Sastre-Perona, P. Santisteban

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G. Riesco-Eizaguirre, J. Perales-Patón, P. Santisteban

Study supervision: G. Riesco-Eizaguirre, P. Santisteban

The authors thank Antonio De la Vieja for critical reading of the article. They also thank Dr. Nancy Carrasco (Department of Cellular and Molecular Physiology, Yale School of Medicine) for providing the human and rat NIS antibody.

This work was supported by Grants BFU2010-16025 and SAF2013-44709-R from the Dirección General de Proyectos de Investigación; RD12/0036/0030 from FIS, Instituto de Salud Carlos III (ISCIII); and S2011/BMD-2328 TIRONET project from the Comunidad de Madrid (Spain) to P. Santisteban; and FIS-ISCIII; PI14/01980 to G. Riesco-Eizaguirre. L. Wert-Lamas and A. Sastre-Perona hold a predoctoral FPI and FPU fellowship, respectively, from the Spanish Government and L.P. Fernández was a JAE.doc postdoctoral from CSIC.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.