Introduction

MicroRNAs (miRs) are short, non-coding RNA molecules that posttranscriptionally regulate gene expression via RNA interference [1]. The mature miR molecules are approximately 16–29 nucleotides long and bind to the 3′UTR of mRNAs as antisense regulators to prevent translation or to cause mRNA degradation [2–4]. Until April 2010, 14,197 entries have been submitted to miRBase Release 15.0 representing hairpin precursor miRs giving rise to 15,632 mature miRs in 133 species. In Homo sapiens 940 miR sequences have been recorded. Their roles have been considered in development [5], cell proliferation [6], differentiation [7], apoptosis [8] and tumorigenesis (reviewed in Ref. 9) including leukaemia [10], lung [11], breast cancer [12], and various endocrine tumors e.g. thyroid carcinoma [13], endocrine pancreatic tumor [14], ovarian cancer [15] and pituitary tumor [16–18].

Pituitary adenomas account for at least 10% of intracranial tumors with a prevalence of 1 in 1,000 people in the general population [19]. Although these tumors are mostly benign, they may cause significant morbidity due to hormonal dysfunction and mass effect. Genetic background of sporadic pituitary adenomas, especial of NFPA is poorly characterized.

The activin/transforming growth factor-β (TGF-β) family consists of evolutionary conserved polypeptides which play a prominent role in regulating pituitary tumor growth and prolactin secretion from pituitary lactotrope cells. Members of this family, activin, TGFβ and bone morphogenetic proteins (BMPs) are dimeric molecules composed of two subunits linked by a disulfide bridge. The signal transduction begins with ligand binding to a single transmembrane serine/threonine kinase type II receptor, followed by transphosphorylation and activation of the type I receptor. The activated type I receptor phosphorylates intracellular mediators known as receptor-regulated Smads (R-Smads: Smad2 and Smad3). Two phosphorylated R-Smad subunits form a heterotrimer with Smad4 subunit that translocates to the nucleus, where it associates with various binding partners including menin, and binds to DNA [reviewed in Ref. 20, 21]. The tumor suppressor effect of the TGFβ pathway includes interference with cell cycle by inducing expression of p15Ink4b and p21Waf1, inhibition of cyclin D-CDK4/6 and cyclin E/A-CDK2 complexes [22, 23], and induction of apoptosis. Also, the TGFβ pathway represses expression of c-myc, a transcription factor that promotes cell proliferation, and Id protein nuclear factors which inhibit cell differentiation [24, 25].

Somatic mutations of genes encoding TGFβ receptors are rare in human pituitary tumors [26] and no difference in TGFβ RII expression has been found between NFPA and normal pituitary samples [27]. However, microarray studies have indicated that FSH, LH and TSH β-subunit, which are under TGFβ regulation, are underexpressed in NFPA samples suggesting that TGFβ signaling is repressed in NFPA [28, 29]. In addition, TGFβ administration decreased proliferation and increased apoptosis of HP75 cells, a cell line derived from a clinically non-functioning pituitary tumor [29, 30].

The aim of this study was to examine the miR expression profile of NFPA and normal pituitary samples and to identify alterations in biological pathways of NFPA. In addition, we studied the mRNA expression of Smads (Smad1-9) in the same samples in order to detect those miRs which may inhibit the TGFβ pathway through downregulation of Smad3 and may, therefore play a role in NFPA tumorigenesis.

Materials and methods

Patients

Ten NFPA and 10 normal pituitary tissue specimens were examined. Adenoma tissues were removed by transsphenoideal surgery at the Hungarian National Institute of Neurosurgery. All adenoma samples were gathered with the permission of the Local Committee on Human Research, after obtaining informed consent from each patient. The diagnosis of NFPA was based on clinical findings, hormone levels of patients and immunohistochemistry. Routine immunohistochemical examination included immunostaining for anterior pituitary hormones (GH, PRL, ACTH, LH, FSH and TSH) and staining for the MIB1 proliferation marker. All routine immunohistochemical studies were performed at the 1st Department of Pathology and Experimental Cancer Research, Semmelweis University, Budapest, Hungary. Normal pituitary samples were obtained by autopsy within 6 h of death from patients with no evidence of any endocrine disease (University Clinical Centre, Belgrade, Serbia). Demographic data of the patients are summarized in Table 1.

Table 1 Clinical and pathological findings in patients with clinically non-functioning pituitary adenomas
Full size table

RNA isolation

Tissue specimens were immediately frozen in liquid nitrogen after the adenoma removal, and stored at −80°C until use. Total RNA was extracted with miRNeasy Mini Kit (Qiagen Inc., Chatsworth, CA). RNA integrity and concentration was measured using Agilent Bioanalyzer 2100 System (Agilent Tech Inc., Santa Clara, USA).

miR profiling in pituitary tissues using TLDA cards

miR expression profile was analyzed in 8 NFPA and 4 normal pituitary samples using TaqMan Low Density Array (TLDA) Human MicroRNA Panel v.2 (Applied Biosystems, Foster City, CA). All procedures were performed following the manufacturer’s instructions. Briefly, 300 ng total RNA per sample was reverse transcribed using Megaplex RT primer Pool A and B up to 380 stem-looped primers per pool and TaqMan MicroRNA Reverse Transcription Kit (P/N: 4366596). Preamplification was carried out by Megaplex Preamp Primers and TaqMan PreAmp Master Mix. Real-time PCR amplifications were run in TaqMan Human MicroRNA Array A and B (PN: 4398965, 4398966) using Megaplex Primer Pools, Human Pools Set v3.0 (PN: 4444750) on 7900HT Real Time PCR System (Applied Biosystems). Real-time RT–PCR data analysis was performed using Real-Time StatMiner™ software (Integromics, Granada, Spain). Expression level was calculated by the ddCt method, and fold changes were obtained using the formula 2−ddCt.

Real-time quantification of individual miRs revealed by TLDA miR array

Expression level of hsa-miR-135a (PN: 4427975, Assay ID: 000460), hsa-miR-140-5p (PN: 4427975, Assay ID: 001187), hsa-miR-582-3p (PN: 4427975, Assay ID: 002399), hsa-miR-582-5p (PN: 4427975, Assay ID: 001983), hsa-miR-938 (PN: 4427975, Assay ID: 002181), RNU44 (PN: 4427975, Assay ID: 001094), RNU48 (PN: 4427975, Assay ID: 001006), U6 snRNA (MammU6, PN: 4427975, Assay ID: 001973) were determined in pituitary samples using individual TaqMan MicroRNA Assays (Applied Biosystems) according to the protocol provided by the supplier. Briefly, reverse transcriptions were performed with miRNA specific stem-loop RT primer using TaqMan MicroRNA Reverse Transcription Kit (P/N: 4366596) followed by qRT–PCR using TaqMan assays and TaqMan Universal PCR Master Mix. Reactions were run in triplicate in 384-well plates on 7900 HT RealTime PCR System (Applied Biosystems). Expression level was calculated by ddCt method, and fold changes were obtained using the formula 2−ddCt.

Pathway analysis

Significantly differentially expressed miRs were uploaded into the DIANA-mirPath tool. In silico target prediction by TargetScan v5. followed by enrichment analysis of multiple miR target genes by Pearson’s Chi-squared test comparing each set of miR targets to all known KEGG pathways was performed. KEGG [31] is a database resource that provides knowledge on several genomes as well as their relationships to biological systems. It has been utilized as a systematic knowledge base for molecular and network biology [32]. For further investigation we excluded miRs with −ln(P-value) < 3 (P < 0.05).

Gene expression

mRNA expression of genes involved in identified pathways was evaluated in 10 NFPA and 10 normal pituitary tissues including each of the 12 samples used for the miR expression experiment. Reverse transcription of 1 μg of total RNA was performed using the Superscript III First Strand Synthesis Kit (Invitrogen, USA). The mRNA expression was determined with qRT–PCR using the 7500 Fast Real-Time PCR System (Applied Biosystems). Quantitative RT-PCR was carried out in triplicate in 20 μl reaction volume containing 10 μl TaqMan Fast Universal PCR Master Mix, 250 pM TM probe, and 10-fold diluted cDNA in RNase-free water. The used inventoried probes were synthesised by Applied Biosystems (P/N: 4331182, SMAD1: Hs00195432_m1, SMAD2: Hs00183425_m1, SMAD3: Hs00232219_m1, SMAD4: Hs00929647_m1, SMAD5: Hs00195437_m1, SMAD6: Hs00178579_m1, SMAD7: Hs00178696_m1, SMAD9: Hs00195441_m1, DLK1: Hs00171584_m1, MEG3: Hs01087966_m1, and endogenous controls: 18S: Hs99999901_s1, ACTB: Hs99999903_m1, GAPDH: Hs99999905_m1, B2M: Hs99999907_m1, HPRT1: Hs99999909_m1). The most appropriate housekeeping genes were determined by Normfinder software. Based on these results, the geometric mean of GAPDH and ACTB was used as a computed endogenous control. Expression level was calculated by the ddCt method, and fold changes were obtained using the formula 2−ddCt.

In silico target prediction for Smad3

In addition to the DIANA-mirPath tool for target prediction, we used three computational target prediction algorithms: microrna.org (http://www.microrna.org/microrna/home.do), Targetscan 4.0 (http://www.targetscan.org) and MicroCosm (http://www.ebi.ac.uk/enright-srv/microcosm/htdocs/targets/v5/).

Statistical analysis

MiR expression data obtained by TLDA were normalized using three housekeeping genes, RNU44, RNU48 and MammalianU6 selected by Normfinder integrated in StatMiner. Normfinder calculates the overall stability for all candidate housekeeping genes tested on a sample set indifferent from its composition. It generates an overall stability rank from distinct intragroup and intergroup measures of variability. By grouping the samples into tumoral and normal tissues, the best combination of MammU6, RNU44 and RNU48 has been identified, and an endogenous control has been computed by their geometric mean. The same computed endogenous control was used in single miR expression experiments. After data normalisation, miR expression levels in NFPA were compared to those found in normal pituitary tissues using T-test or Mann–Whitney U test depending on the results of Shapiro-Wilks normality test. Statistical analysis was performed using Integromics RealTime StatMiner and Statistica 7.0 software (StatSoft Inc., Tulsa, USA). Hierarchical cluster analysis was performed by UPGMA clustering method using Euclidean distance for similarity measuring. Results were visualized as heatmaps. For correlation analysis between miR expression, Smad3 expression and tumor size Pearson’s or Spearman rank correlation tests were used depending on the data distribution evaluated by normality test. A value of P < 0.05 was considered to be significant.

Results

MiR expression profile of non-functioning pituitary adenomas

miRs with threshold cycle (CT) ≥35 were removed from the analysis and were considered as unexpressed. Of the 670 miRs studied, 205 were unexpressed in any pituitary tissues while 3 miRs (miR-124*, miR-515-5p and miR-872) were expressed only in NFPA tissues and 5 miRs (miR-198, miR-299-5p, miR-497*, miR-548c-3p and miR-622) only in normal pituitary tissues. Of the 457 miRs expressed in both normal and tumor tissues, 92 were significantly underexpressed and 70 significantly overexpressed in NFPA compared to normal pituitary tissues (Suppl. Table 1). Unsupervised hierarchical cluster analysis using significantly altered miRs correctly classified each of the 12 pituitary samples and it separated 3 groups among the NFPA samples (Fig. 1).

Fig. 1
figure 1

Hierarchical clustering of the 12 different pituitary samples based on their miR expression patterns. 162 miRs with significant differences in expression were subjected to hierarchical clustering by UPGMA clustering method using Euclidean distance for similarity measuring. Data were normalized using 3 housekeeping genes: RNU44, RNU48 and MammalianU6. a cluster analysis of dct values, b cluster analysis of -ddct values. Overexpression is indicated with red/pink, whereas underexpression is coded with green/blue. NFPA clinically non-functioning pituitary adenoma sample, NP normal pituitary sample (Color figure online)

Full size image

Validation of the TLDA miR array

Five overexpressed miRs predicted to target Smad3 (miR-135a, miR-140-5p, miR-582-3p, miR-582-5p and miR-938) were analysed by individual miR assays. All these miRs were significantly overexpressed in NFPA samples in agreement with findings obtained from the TLDA miR array (Fig. 2).

Fig. 2
figure 2

Expression of miRs by individual Taqman miR assays. Left column NP, right column NFPA samples. Columns are means ± SE. *P < 0.05

Full size image

Chromosomal localization of the differentially expressed miRs

Chromosomal localization of differentially expressed miRs showed that the 162 under- or overexpressed miRs map to 35 chromosomal regions. In the underexpressed group, 4 hotspot regions with 43 miRs, while in the overexpressed group one hotspot region with 8 miRs were identified (Fig. 3, Suppl. Table 1). mRNA expression of two genes (MEG3 and DLK1) mapped to chromosome region 14q32 was measured by qRT-PCR. Both genes were significantly underexpressed in NFPA samples (Fig. 4).

Fig. 3
figure 3

Chromosomal localization of significantly differentially expressed miRs in clinically non-functioning pituitary tumors (each spot represents one miR; red spot overexpressed miR, blue spot underexpressed miR) (Color figure online)

Full size image
Fig. 4
figure 4

Chromosomal region of 14q32.2–32.31, harboring MEG, DLK1 and miRs (a). 51 miRs and 53 small nucleolar RNA (snoRNA) encoded here. We measured 46 of the 51 miRs, of those 26 miRs showed underexpression in pituitary tumors compared to normal tissue. Underexpression of MEG and DLK1 in pituitary tumors (right columns) compared to normal tissues (left columns). Columns are means ± SE. *P < 0.05 (b)

Full size image

Correlation between miR expression and clinicopathological findings

There was a significant negative correlation between tumor size and the expression level of 18 miRs (Suppl. Table 2). Of these, 6 miRs (miR-450b-5p, miR-424, miR-503, miR-542-3p, miR-629 and miR-214) were significantly underexpressed (Fig. 5), one miR (miR-592) was significantly overexpressed in NFPA compared to normal pituitary tissues and 11 miRs (miR-224, miR-581, miR-221*, miR-504, miR-215, miR-143, miR-103, miR-510, miR-181a, miR186, miR-708) did not differ significantly in tumor samples compared to normal. Four underexpressed miRs (miR-450b-5p, miR-424, miR-503, miR-542-3p) clustered together in chromosomal region Xq26.3.

Fig. 5
figure 5

Significant correlations between expression of underexpressed miRs and tumor size

Full size image

In silico pathway analysis of differentially expressed miRs

To highlight the potential pathogenetic role of differentially expressed miRs in NFPA and to identify potentially altered biological processes via miRs, an in silico pathway analysis was performed. Using the DIANA-mirPath tool several possible pathways modified by differentially expressed miRs were identified (Suppl. Tables 3, 4). Most importantly, comparison of overexpressed miRs indicated the possible involvement of the TGFβ signaling pathway with 205 miR-mRNA interactions involving 51 genes and 39 miRs in NFPA tissues.

mRNA expression of Smads and in silico analysis of miRs potentially targeting Smad3 mRNA

To further explore the possible involvement of the TGFβ signaling pathway in the pathogenesis of NFPA, mRNA expression of Smads (Smad1-9) were analyzed in pituitary tissues. As shown in Fig. 6., mRNAs of Smad3, Smad6 and Smad9 (=Smad8) were significantly (P < 0.001) underexpressed in NFPA compared to normal pituitary tissues. Expression of Smad3 but not Smad6 and Smad9 showed a tendency to correlate with tumor size (P = 0.08, P = 0.89 and P = 0.33).

Fig. 6
figure 6

Expression of Smads in clinically non-functioning pituitary tumors. Y axis represents each of Smads expression (log2RQ). Columns are means ± SE. *P < 0.05

Full size image

In order to reveal a possible link between Smad3 mRNA expression and differentially expressed miRs in NFPA, we performed correlation analysis. This analysis revealed that the expression of 80 miRs negatively correlated with Smad3 expression. Using 4 different algorithms for in silico target prediction, we were able to narrow this list to 19 overexpressed miRs potentially targeting Smad3 in NFPA (Table 2).

Table 2 List of the 19 overexpressed miRs targeting Smad3 in clinically non-functioning pituitary adenomas identified by multiple in silico target prediction algorithms, expression level of the identified miRs and their correlations with Smad3 expression
Full size table

Discussion

In this study, we have assessed the miR expression profiles of NFPA and normal pituitary tissues using TLDA. This method has been successfully applied in earlier studies including ours for the identification of significantly altered miRs in different tumor types such as adrenal and pituitary tumors [33–35]. Because appropriate normalisation of data is critical to ensure accurate and reliable results, miR expression data were normalised to a computed housekeeping gene. Using this approach, we have identified several miRs with significantly different expression between NFPA and normal pituitary. Of the significantly altered miRs, 4 overexpressed (miR-128a, miR-93, miR-378 and miR-429) and one underexpressed miR (miR-543) have been validated previously by qRT-PCR in 18 NFPA and 10 normal pituitary samples (data not shown) [34, 36, 37], ensuring that the expression profile determined by TLDA was accurate. In addition, we measured 5 of 19 miRs potentially targeting Smad3 (miR-135a, miR-140-5p, miR-582-3p, miR-582-5p and miR-938) by individual miR assays. All of them and the previously measured miR-429 showed significant overexpression in NFPA samples compared to normal pituitary.

We found that unsupervised hierarchical clustering of miR array results rendered all normal tissues into one group, while NFPA samples were separated into three groups. Three of the 8 NFPA tissues belonged to a group clustered closer to normal pituitary tissues than the other 5 NFPA tissues. However, we were unable to reveal differences in clinical, hormonal or immunohistochemical findings between these three groups of NFPA. This finding may support previous proteomic and mRNA expression data showing that NFPA tissues obtained from different patients are not homogeneous [38–40].

Chromosomal localization of differentially expressed miRs indicated several regions, including 14q32.2–14q32.31 for 32 underexpressed miRs (Suppl. Table 1, Fig. 3), suggesting that this chromosomal region may contain tumor suppressor gene(s). Indeed, MEG3, a tumor suppressor gene maps to this region and it has been found to be underexpressed by promoter hypermethylation in NFPA tumors [41, 42]. Beside MEG3, DLK1 is also mapped to 14q32.31 and in our experiment both genes were underexpressed in NFPA samples (Fig. 4), supporting that expression of those miRs is regulated along with their host genes or their environment [2, 43].

Another interesting locus, Xq26.3 was identified in our study with highly underexpressed miRs (miR-424, miR-450b-5p, miR-542-3p, miR-542-5p and miR-503). Three of these miRs, miR-424, miR-542-3p and miR-503 have been predicted to target cyclin D1 and cyclin D2 which are frequently overexpressed in NFPA [44–47]. In addition, miR-503 has been validated to target directly cyclin D1 and it has been suggested to be a putative tumor suppressor [48]. Also, a potentially important target of miR-503 revealed by in silico prediction is the cell cycle regulator CDC25. CDC25 and Wee1 kinase have opposite effects in cell cycle regulation. In our previous study, we found that Wee1 kinase was downregulated by a specific subset of overexpressed miRs in pituitary tumors including NFPA [37]. These findings support the view that cell cycle alteration affecting the regulatory loop between Wee1 and CDC25 is influenced by miRs and that loss of expression of these miRs may result in tumor growth [37].

We have found 19 miRs correlating with tumor size (Suppl. Table 2). Of these miRs, miR-629 and miR-214 potentially target Bcl2, an antiapoptotic molecule and therefore, they may contribute to tumorigenesis through improper regulation of apoptosis. MiR-592 has been reported to be underexpressed in colon tumors deficient in mismatch repair compared to those with proficient mismatch repair [49]. We observed the upregulation of this miR in NFPA compared to normal pituitary tissues, which supports the findings of Zhu et al. [50] showing that genomic instability frequently found in colon cancer is very unusual in pituitary tumors.

In order to predict the function of miRs in biological processes and to identify pathogenetic pathways potentially modified by posttranscriptional regulation, we performed an in silico pathway analysis for significantly differentially expressed miRs using DIANA-mirPath tool. Several pathways have been identified in NFPA including the TGFβ signaling pathway with 205 miR-mRNA interactions involving 51 genes and 39 overexpressed miRs suggesting the complexity of the regulatory network altered in pituitary tumors. For analyzing TGFβ pathway in our samples, we measured the expression profile of Smads. These molecules are central regulators of the TGFβ and BMP pathways in the pituitary; and their alterations are critical steps towards tumor formation and progression [51]. In previous studies no differences have been found at the level of TGFβ receptors between NFPA and normal pituitary tissues, but FSH, LH and TSHβ-subunit which are under TGFβ regulation [28] were underexpressed in NFPA [27]. Among Smads, we found that Smad3, Smad6 and Smad9 (=Smad8) were significantly underexpressed. Smad3 is a downstream molecule of the TGFβ signaling, as it forms a complex with Smad4 which translocates to the nucleus. Smad6 and Smad7 can inhibit this process as well as BMP signaling transduction. We found that Smad6 was significantly underexpressed in NFPA samples, however, Smad7 expression did not change. In addition, BMP signaling downstream molecules (Smad1, Smad5 and Smad8) were not altered in these NFPA samples.

Based on these results, we have focused on Smad3, the main TGFβ downstream molecule, and based on our in silico pathway analysis we performed a correlation analysis between Smad3 and miRs expression followed by complex bioinformatical target prediction. We identified 19 miRs overexpressed and potentially targeting Smad3, of which miR-140-5p has already been experimentally validated to target Smad3 directly [52]. Interestingly, it has been shown that not only miRs are able to influence Smad expression but Smads are also capable to enhance miRs maturation by binding p68, a component of Drosha microprocessor complex [53]. TGFβ/BMP regulated miRs (T/B miRs) contain in their stem region of primary transcripts a conserved sequence similar to Smad binding element (CAGAC or CAGGG). BMP specific Smad1 and Smad5, as well as TGFβ specific Smad3 interact with the p68 and enhance T/B miRs’ expression posttranscriptionally [54].

Previous studies showed that 44 T/B miRs exist in the human genome [54]. Our finding of 5 underexpressed miRs (miR-199a-5p, miR-214, miR-509-5p, miR-487b and miR-877) may indicate that downregulation of these miRs may be the result of a global inhibition of the TGFβ/BMP pathway and may suggest that a negative feed-back regulation loop, similar to that reported for miR-20a and E2F [55], may exist between some miRs and the TGFβ/BMP downstream molecule Smads. However, the T/B miR miR-140-3p (the antisense strand of miR-140-5p) showed overexpression in tumors, perhaps indicating that overexpression of this miR may involve other mechanism(s).

Posttranscriptional regulation of the TGFβ pathway by miRs has been implicated in both development [56, 57] and tumorigenesis [58, 59]. The TGFβ tumor suppressor pathways are affected by miR-106b-25 and the miR-17-92 cluster cooperatively. MiR-106b, miR-93, miR-17-5p and miR-20a suppress p21Waf1 expression, which is required for TGFβ-induced cell cycle arrest, and earlier studies showed that p21Waf1 expression can be detected only in less than 5% of cells of NFPA [60]. MiR-25 and miR-92 inhibit BIM expression, which is essential for TGFβ-dependent apoptosis [58, 59]. In our previous study, we found that miR-93, -20a and -17-5p were significantly overexpressed in NFPA compared to normal pituitary tissues.

Another important link between Smad3 and pituitary tumorigenesis arises from the direct interaction between Smad3 and the tumor suppressor menin. Inactivation of menin blocks TGFβ and activin signaling, antagonizing their growth-inhibitory properties in anterior pituitary cells [61]. It is well established that MEN1 gene mutations play a role in MEN1-related pituitary tumorigenesis, but MEN1 gene mutations seem to be extremely rare in sporadic pituitary adenomas [62, 63]. However, tumor transformation is associated with a significant reduction in menin protein levels in a high percentage of pituitary adenomas [64] and studies by several groups using RT-PCR [65–67] showed no differences in MEN1 mRNA levels between neoplastic and normal pituitary tissues. These data together may raise the possibility that posttranscriptional mechanisms, at least partly determined by MEN1 targeting miRs, may play an important role in the regulation of menin expression. Indeed, 4 miRs (miR-149, miR-570, miR-592, miR-769-5p) in our study potentially targeting the MEN1 transcript showed significant overexpression but the direct regulation of menin expression by these miRs needs further studies.

The complexity of the regulatory network altered in pituitary tumors is underlined by our pathway analysis as well. In addition to the TGFβ pathway, we found the potential involvement of other known pathways, such as MAPK (over-activated in GH producing and NFPA as shown in B-Raf overexpression [68] and proteomics experiments [69]), Akt/mTOR [70], ERK1/2 [71] and Wnt and Notch pathways (Suppl. Tables 3, 4).

Related to pituitary, it has been found in earlier studies that Wnt inhibitors are down-regulated in NFPA [45]. The Wnt downstream effector β-catenin has been shown to exhibit an increased immunostaining in 57% of pituitary adenomas. The Wnt-β-catenin pathway stabilizes PITX2 mRNA, and PITX2 directly activates cyclin D. The changes in WIF1, SFRP1, PITX2, and cyclin D1 expression support the possibility that elevated Wnt-β-catenin signaling is important for pituitary tumorigenesis [45]. In our pathway analysis, Wnt pathway are concerned by miRs with 249 interactions between 72 genes and 44 miRs including overexpressed miRs potentially targeting Wnt inhibitors; WIF1 (miR-128, miR-137, miR-140-3p, miR-200b, miR-330-5p, miR-373, miR-429), SFRP2 (miR-128, miR-340, miR-373) and SFRP4 (miR135a, miR-153, miR-182, miR-183, miR-26a, miR-340, miR-342-3p, miR-500, miR-501-3p, miR-519c-3p, miR-532-5p) Both Wnt and Notch signaling has been described to be modulated by miRs. MiR-17-5p, an overexpressed miR in pituitary, has been found to result in Wnt signaling activation through targeting HBP1 [72]. Mir-148a has been shown to repress Wnt signaling, and we found it underexpressed in pituitary tissues, suggesting its role in the activation of this pathway [73]. Recently, miR-449 has been described as a modifier of the Wnt pathway in primary pigmented nodular adrenocortical disease (PPNAD) through targeting WISP2 [74] and, therefore, its underexpression in NFPA may promote activation of the pathway. We found Mir-199b to be underexpressed in pituitary tissues and in vitro experiments has confirmed its inhibitory effect on cell proliferation rate [75].

Recent studies using microarrays and proteomics have emphasized the importance of Notch signaling in pituitary tumorigenesis. Notch3 signaling has been already connected to lung cancer and aggressive T-cell neoplasias. Notch has been shown to be strongly overexpressed and DLK1, a potential ligand of Notch3, has been found to be one of the most markedly down-regulated genes in NFPA [39]. ASCL1, another component of the Notch pathway, has been also strongly down-regulated in NFPA. It has been suggested that high levels of Notch3 could repress ASCL1 leading to the loss of DLK1 expression in these tumors. In our experiments, miR-197 and -33b, which are predicted to target DLK1, were significantly overexpressed in NFPA tissues and, therefore, they could result in a loss of DLK1 expression.

In conclusion, our findings support the possibility that the TGFβ signaling is repressed and modulated by miRs in NFPA via Smad3 whose expression can be directly regulated by several overexpressed miRs including the experimentally validated miR-140. In addition, the complexity of the regulatory network identified by miR profile may suggest that the decrease of TGFβ signaling via Smad3 may result in a shift toward alternative, non-Smad pathways, including Ras-MAPK, p38, c-Jun, and PI3 K-Akt (Fig. 7). These alternative pathways have been already considered as contributing factors in pituitary tumorigenesis.

Fig. 7
figure 7

The TGFβ pathway influenced by miRs. Blue codes represent underexpressed, and red codes illustrate overexpressed component of the TGFβ pathway in clinically non-functioning pituitary adenomas (Color figure online)

Full size image