MiR-204 down-regulation elicited perturbation of a gene target signature common to human cholangiocarcinoma and gastric cancer

Background & Aims There is high need of novel diagnostic and prognostic tools for tumors of the digestive system, such as gastric cancer and cholangiocarcinoma. We recently found that miR-204 was deeply downregulated in gastric cancer tissues. Here we investigated whether this was common to other tumors of the digestive system and whether this elicited a miR-204-dependent gene target signature, diagnostically and therapeutically relevant. Finally, we assessed the contribution of the identified target genes to the cell cycle progression and clonogenicity of gastric cancer and cholangiocarcinoma cell lines. Methods We employed quantitative PCR and Affymetrix profiling for gene expression studies. In silico analysis aided us to identifying a miR-204 target signature in publicly available databases (TGCA). We employed transient transfection experiments, clonogenic assays and cell cycle profiling to evaluate the biological consequences of miR-204 perturbation. Results We identified a novel miR-204 gene target signature perturbed in gastric cancer and in cholangiocarcinoma specimens. We validated its prognostic relevance and mechanistically addressed its biological relevance in GC and CC cell lines. Conclusions We suggest that restoring the physiological levels of miR-204 in some gastrointestinal cancers might be exploited therapeutically.


INTRODUCTION
MicroRNAs (miRNAs) are small non-coding RNAs that control gene expression through base pairing with specific sequences in the 3'-UTR of target mRNAs.
For their ability to modulate simultaneously the levels of multiple gene products, microRNAs are invaluable tools at integrating intracellular signaling with changes in the extracellular and intracellular milieu, thus conferring spatial and temporal modulation to cell signaling [1].

Priority Research Paper
Oncotarget 29541 www.impactjournals.com/oncotarget MicroRNAs may play a role in cancer progression functioning as oncogenes or tumor suppressors [2][3][4][5]. In line with this, it is not surprising that specific miRNA signatures have been associated with specific stages of tumor progression, including response to chemotherapy. Consequently, microRNA signatures may be endowed with prognostic and diagnostic potential.
Gastric cancer (GC) is the fifth most frequently diagnosed cancer and the third leading cause of death related to cancer worldwide [6]. The two main histo-types of gastric cancer are the intestinal and the diffuse one [7,8].
The hepato-biliary tumors are a heterogeneous group of poor prognosis cancers localized to the liver, gall bladder and biliary tree [9]. Establishing the true incidence of bile duct cancer is very difficult because, to date, there are no specific and sensitive diagnostic tools. Based on the location of the tumor cholangocarcinoma can be classified in intrahepatic or extrahepatic cholangiocarcinoma [9,10].
MicroRNAs, as being exquisite transducers of micro-environmental signaling and sensitive sensors of the tissue homeostasis, can be ideal tools to finely pinpoint the molecular status of a tissue. In facts microRNAs are emerging as effective biomarkers for the differential diagnosis of the tumors, as being proven to discriminate tumoral and normal tissues of various cancer type such as, breast cancers, mesothelioma, head and neck squamous carcinoma and osteosarcomas [11][12][13][14].
In a previous study we profiled microRNA differentially expressed in gastric cancer vs matched normal tissue specimens and miR-204 was among those differentially modulated in a cohort of 123 matched tissues [14]. MiR-204 was also reported as down regulated in several cancers, including biliary-tree tumors [14][15][16][17]. However, little is known about the miR-204 gene targets in various tumors and, more, whether perturbation of miR-204-dependent gene signatures may contribute to tumor progression.
Taking the start from our previous observations, we here evaluated the expression of the miR-204 in a series of tumors of the digestive system and we mechanistically dissected its contribution to tumor progression. Our study hypothesis was that the deep downregulation of the miR-204 in tumor tissues may underlie a very relevant event in the progression of the disease, possibly by triggering gene expression changes, impinging on key protumorigenic properties of the transformed cells. As such, miR-204 levels may be diagnostically and prognostically relevant.
We first aimed at deeply characterize the expression of miR-204 in a sizable cohort of gastric cancer specimens (as compared to normal tissues) to identify, by combining transcription profiling and extensive in silico analysis, the potential targets of this microRNA. This allowed us to identify 37 gene targets with prognostic potential for gastric cancer in Kaplan Meier analysis. Next, we observed that miR-204 down regulation was a rather general feature of other tumors of the digestive system, also including colon, esophageal cancer and hepatocellular carcinoma, which all exhibited a similar degree of miR-204 perturbation. We focused on a rare type of digestive system cancer, the cholangiocarcinoma, because of the availability of a unique large repertoire of specimens in our network. We found that the miR-204 levels, deeply downregulated in cholangiocarcinomas as opposed to normal tissues, correlated with a gene signature almost identical to that observed in gastric cancer specimens. Among the identified gene targets (common to both gastric cancer and cholangiocarcinomas) we identified a 7-gene signature, which exhibited, collectively, prognostic value in Kaplan Meier analysis. We showed that manipulating the levels of miR-204 (by means of agonist molecules) in representative GC and CC cell lines, led to a reversal of the identified gene target signature. This affected the clonogenicity and cell cycle progression of the transfected cell lines, thus establishing a mechanistic link between the modulation of miR-204 targets and the protumorigenic properties of the tumor cells. Finally, we evaluated by siRNA-mediated silencing, the biological activity of each one of the targets toward cell cycle progression and clonogenicity of gastric tumor and cholangiocarcinoma cells.

RESULTS
We recently identified microRNAs differentially modulated in a limited set of matched samples of tumor and normal tissues from gastric carcinoma patients [14]. The microRNA-204 (miR-204) was among the differentially modulated microRNAs and a component of a prognostically relevant signature [14]. To deepen those previous findings, we specifically evaluated the expression of microRNA-204 in a novel discovery set of 20 frozen gastric specimens (15 tumor and 5 normal gastric mucosa) (Table 1), by quantitative PCR ( Figure 1A). This showed that the miR-204 levels were much lower in tumor tissues as compared to normal tissues, in line with our previous observations ( Figure 1A) [14]. To statistically strengthen this finding, we employed a larger cohort of 276 Tumor and 38 Normal samples from publicly available gene expression databases (TCGA), as a validation set (Table 1) ( Figure 1B). Again, we observed similar modulation of the miR-204 in the validation set ( Figure 1B). Of note, we also found that a deeper downregulation of the miR-204 levels was a feature of the "intestinal" histotype as opposed to the "diffuse" histotype of gastric cancer (Supplementary Figure 1A). The Intestinal-type gastric cancer is by far the most common variant of gastric cancer (accounting for 50-70% of the cases) (WHO classification of tumours of the digestive system. Lyon: International Agency for Research on Cancer; 2010). Next, we set out to better understand the contribution of miR-204 downregulation to the progression of GC. As microRNAs work mainly by targeting the expression of multiple mRNAs through base paired interactions within their 3'-UTR region, we set out to determine the targets of miR-204 in the tumor tissues analyzed. To do this, we aimed at identifying mRNAs whose expression anti-correlated with that of the microRNA-204 and which contained a microRNA-204 seed sequence in their 3'-UTR region by means of an Affymetrix-based profiling in our discovery set as assessed in Figure 1A (Table 1) ( Figure 1C). Extensive in silico analysis (Supplementary Figure 1B) allowed us to identify 37 gene targets, which obeyed to the mentioned criteria (Table 3). Indeed, such gene targets, belonging to distinct biological processes (Table 4A-4B), exhibited increased expression in tumoral vs normal samples in both our discovery set ( Figure 1C, heat map) and in the validation set from TGCA database examined ( Figure 1D, heat map). Principal component analysis (PCA) performed on the gene expression levels of the 37 genes in both our discovery set and in the TGCA samples revealed that the expression levels of the microRNA-204 targets could effectively discriminate between tumor and normal samples ( Figure 1E, left panel) and between tumor and normal tissue ( Figure 1F, right panel), respectively. This was true for both the "intestinal" and "diffuse" histotype of gastric cancer (data not shown). We further refined our analysis and we focused on seven out of 37 identified targets (NOTCH1, FOXM1, RAD51, CENP-A, CENP-E, KIF15, SHCBP1), which showed a significantly high   Table 1) in both the discovery set C. and in the TGCA dataset. Normalized intensity values reported. D. E-F. Principal Component Analysis (PCA) of the distribution of the 37 genes (as from 1C-1D) in the discovery set E. and in the TGCA dataset F. The first component (PC1) represents inter-group variability (x-axis) while the second component (PC2) represents the intra-group variability (y-axis).
Oncotarget 29544 www.impactjournals.com/oncotarget Pearson anti-Correlation value in the cohorts analyzed and belonged to similar biological processes (Table 3 and  Table 4).
We choose to further study the expression of CENP-A, SHCBP-1, FOXM1, KIF15, in a cohort of 40 matched frozen tissue specimens (Table 1) ( Figure  2A), previously characterized for the miR-204 levels (Supplementary Figure 2A). We observed that the expression of two out these four genes targets (FOXM1 and KIF15) significantly and positively correlated with the node (N) status of the patient (Supplementary Figure  2B). Next, we evaluated whether the higher levels of the representative genes examined would correlate with patient survival, by Kaplan-Meier analysis (KMplotter. com). This showed that the identified genes contributed prognostic potential, with higher expression of the gene correlating with decreased patient survival. This was true when analyzing both the 37 and the 7 genes signature ( Figure 2B left and right panel, respectively).
We also identified a subset of 20 of the 37 genes (including 15 genes common to TGCA and KM plotter datasets) significantly associated to the overall survival of the patients when the latter were stratified according to the tumor histotype, by univariate analysis ( Figure 2C left and right panel, Supplementary Table 1). This correlated with a deeper downregulation of miR-204 in the "intestinal" histotype (Supplementary Figure 1A), and with higher levels of its representative candidate targets (Table 5 and Supplementary Figure 5, respectively).

MiR-204 is downregulated in GI tumors
In order to assess whether the miR-204 downregulation and its consequent alteration of gene transcription would be a gastric cancer specific lesion, we analyzed the levels of expression of miR-204 in matched tissues from other tumors of the digestive system, by

Hepatic cholangiocarcinomas exhibited deep miR-204 downregulation and high levels of its target gene signature
We evaluated the levels of miR-204 in 26 frozen cholangiocarcinoma tissues and in 19 FFPE samples as opposed to matched normal biliary duct tissues ( Table  2). This revealed that the miR-204 was lower in the tumor as opposed to matched normal tissues ( Figure  3A). This happened irrespectively whether the RNA was derived from FFPE material or from frozen tissues ( Figure 3A-3B). Analysis of the miR-204 expression in 45 cholangiocarcinoma specimens from the TCGA database (Table 2) confirmed our observation ( Figure  3C). Next, we assessed the levels of the seven miR-204 gene targets (those already characterized in the gastric cancer specimens). Again, consistently with the prior observation, quantitative PCR of 26 frozen matched cholangiocarcinoma specimens revealed that the 7 genes selected exhibited significantly higher levels in the tumor tissues as opposed to the normal ones ( Figure 3D). The analysis of samples in our cohort also showed that the degree of downregulation of the miR-204 and of some of its targets (SHCBP1, CENPA, RAD51) was higher in the intrahepatic cholangiocarcinoma samples as opposed to the extrahepatic ones (Supplementary Figure 5A). We extended the analysis of the levels of the whole set of the previously identified miR-204 targets on the 45 CC samples from TGCA database. This revealed that the selected 36 out 37 targets (expression data for MMP3 was not available) were differentially modulated among tumor and normal tissues derived from the cholangiocarcinoma patients ( Figure 4A), again consistently to what observed in the gastric cancer specimens (Table 3). PCA analysis showed that the identified 36-target gene signature could effectively discriminate between tumor and normal tissue ( Figure 4B). This was clearly illustrated, for gastric cancer and cholangiocarcinoma, by the Volcano plots in Figure 4C-4D ( Figure 4C-4D, respectively) and was true also for esophageal cancer in Supplementary Figure 6 (Supplementary Figure 6). This revealed a high degree of similarity about the modulation of miR-204 and the perturbation of its target genes among three unrelated tumors of the digestive system.

Restoring miR-204 level in GC and CC cell lines impinged on the expression of common gene targets
In order to investigate a causal relationship between the lower levels of the miR-204 and the higher levels of its anti-correlated targets, we manipulated the levels of  Oncotarget 29550 www.impactjournals.com/oncotarget the miR-204 in two representative GC and CC cell lines (GTL-16 and HUCCT1, respectively). We transfected both cell lines with a control (ctrl) or a mimic-204 molecule (mimic-204) and transfection of the latter led to a dose dependent increase of the intracellular levels of the miR-204 (Supplementary Figure 7A). Next, we assessed, by quantitative PCR, the levels of the seven target genes (CENP-A, SHCBP-1, FOXM1, KIF15, CENP-E, RAD51, NOTCH1) at 24hr, 48hr and 72hr from the transfection (10nM). Overexpression of the miR-204 agonist led to time-dependent, decreased expression of the levels of all 7 representative gene targets ( Figure 5A, left and right panels). Despite we observed a different kinetics in the GTL-16 as opposed to the HUCCT1 cells transfected with the mimic-204 molecule, in both cases we witnessed effective downregulation of the gene targets, compatible with a reversal of the miR-204 dependent gene expression ( Figure 6A, left and right panels, respectively) (p < 0.05 at 72hrs). Notably, the effect of transfecting the mimic-204 into both cell lines was conserved even when much lower doses of the agonist were employed, 4nM and 2,5nM respectively (supplementary Figure 7B-7C) as opposed to10nM-shown in Figure 5A. On the other hand, when we transfected GTL-16 and HUCCT1 cells, stably expressing miR-204 with a miR-204 antagomiR molecule (and a control molecule), we witnessed upregulation of 6 out of the 7 miR-204 targets (Supplementary Figure 8). Next, we evaluated the cell cycle profile of the CTRL-and mimic-204-transfected cells. FACS staining of both GTL-16 and HUCCT1 cells, stained with propidium iodide, revealed a time dependent effect of the agonist treatment on the cell cycle, consisting of an increased G1/S phase, which was maximal at 72hrs ( Figure 5B upper and lower panels). We also evaluated, by western blotting of whole cell lysates from GTL-16 and HUCCT1 cells treated with the mimic-204, the expression of the p21 protein, known to increase in cells in G1/S. Compatible with the flow cytometry data, we observed a sharp increase of the p21 protein in the mimic-204 treated cells, as compared to their control counterparts ( Figure 5C, left and right panel).

Altering the miR-204 levels affects clonogenicity
Next, we assessed whether manipulating the levels of the miR-204 gene targets in the GTL-16 and HUCCT1 cells, by means of transfecting mimic-204, could affect some of their protumorigenic properties. To this aim we performed clonogenic assays. This strategy revealed that the clonogenicity of the both cell lines was affected upon overexpression of the miR-204 mimic molecule as compared to the control-transfected cells ( Figure  5D). These results are compatible with the previously mentioned involvement of most of the miR-204 candidate targets in the cell cycle control.

Interference of single miR-204 target genes affects cell cycle and clonogenicity
It is known that the biological effects of the expression of a given microRNA, are mediated by the modulation of clusters of target genes rather than single targets. However, to detail the mechanism of action of the mimic-204, we aimed to investigate which of the miR-204 target gene(s) could mostly influence the miR-204 phenotype. Therefore we transfected the HUCCT1 and the GTL-16 cells with siRNAs targeting each gene of the 7-target signature (Supplementary Figure 9) and we evaluated the effect of the silencing by means of FACS analysis (cell cycle) and clonogenic assays ( Figure  6). Staining of the cells with propidium iodide at 72h following siRNA transfection revealed that downregulation of each miR-204 target affected to a varying degree the Oncotarget 29553 www.impactjournals.com/oncotarget cell cycle of the transfected cells, compatible with the different biological activities of each of the targets. In detail, silencing of SHCBP1 in HUCCT1 cells strongly increased the Sub-G1 cells ( Figure 6A). Similar effects were also consequent to silencing of CENP-E and FOXM1 and for NOTCH1, RAD51 and CENP-A ( Figure 6A-6B). Silencing of the mentioned genes affected very similarly both the GTL-16 and HUCCT1 cells, with the exception of FOXM1 in the GTL-16 cells ( Figure 6B). Clonogenic assays revealed that silencing of each of the target caused a reduction in colony formation, thus fully matching what observed when ectopically expressing the miR-204 ( Figure 6C-6D). Interestingly, silencing of KIF15 did not affect and rather increased the colony forming ability of both the transfected cell lines, despite silencing of the latter elicited increased sub-G1 cells in both cell lines ( Figure 6C-6D). Altogether, these observations suggested that increasing the miR-204 intracellular levels triggered effects on the cell cycle and clonogenic ability of the cells which could be partially recapitulated by silencing each of the single miR-204 targets,

DISCUSSION
In this experimental work we have detailed the modulation of microRNA-204 in both gastric carcinoma and cholangiocarcinoma samples, which exhibited common downregulation of this microRNA in the tumor tissues as opposed to normal tissues. We have identified, in silico, a 37-gene target signature whose levels anticorrelated with those of the miR-204 and shown that 20 out of 37 genes exhibited prognostic potential in GC, with some of the genes of the signature exhibiting independent prognostic potential, in Kaplan-Meier analysis. We have also shown that altering the miR-204 levels in both gastric carcinoma and cholangiocarcinoma cells changed the levels of the miRNA-204 targets genes in a direction compatible with a reversal of the miR-204dependent effect. Last but not least, exogenously added miR-204 affected the cell cycle progression and inhibited the clonogenicity of the transfected cells. The miR-204 is a well-characterized microRNA, already shown to be downregulated in many solid tumors [15,18,19]. Despite the fact that its mode of action and its effector targets may differ according to the tissue of origin and the stage of the tumor, there is indication that it may represent a general tumor suppressor [15,[18][19][20][21][22][23][24]. Thus, the data collected here for both GC and CC specimens and the experiments performed in the GTL-16 and HUCCT1 cells are in line with the literature data. This is significant since the collections of specimens analyzed highly differed for their source (FFPE vs frozen) and time of collection (TCGA vs home-made casuistries) and came from different centers. All these factors could strongly influence the interpatient heterogeneity and contribute significant biases to the analysis.
Notably, 36 out of the identified 37 genes were identically modulated in gastric cancer, cholangiocarcinoma and esophageal cancers tissues as well, thus highlighting the potential relevance of the low levels of the miR-204 observed in GI tumors types. In facts, it is not very common that unrelated tumors exhibit a high degree of similarity in their gene expression profile and this may further suggest a relevant effect of the miR-204 downregulation on the progression of the diseases.
The primary sclerosing cholangitis (PSC) is a chronic liver disease characterized by cholestasis, inflammation and fibrosis and it is known to progress to cholangiocarcinoma (CC) [25,26]. Recently, a distinct set of microRNAs was identified from sera of patients affected by PSC and CC [27]. Notably, no upregulated miR-204 was observed by Bernuzzi et al. both in the PSC and the CC patient sera. However, it remains to be addressed whether reduced expression of miR-204 would be observed in PSC tissues rather than in patient sera and whether expression of miR-204 may represent a biomarker of progression to CC.
On this line, we found here that the intestinal type of gastric tumors in the TGCA cohorts exhibited a deeper miR-204 downregulation and, consequently, higher levels of its gene target signature. Although the limited data available here do not allow deriving conclusions, it would be tempting in the future to investigate whether our observation may possess patient stratification potential.
Here we partially addressed the specific contribution of each miR-204 gene target to the cell cycle progression and clonogenicity of the GTL-16 and HUCCT1 cells. We found that SHCBP1, RAD51, CENP-A, CENP-E and NOTCH1 were mainly responsible for the "mimic-204 phenotype". More in detail, while silencing of SHCBP1, RAD51, NOTCH1 and FOXM1 significantly increased the cells in the sub-G1 phase, all of the tested miR-204 targets (except for KIF15) affected the clonogenicity of the transfected cell lines. Downregulation of NOTCH1, among the targets, resulted in effective increase of sub-G1 cells, in line with the NOTCH signaling being a major player in cholangiocarcinoma progression with translational potential for cholangiocarcinoma [28][29][30][31].
On the other hand, we did not observe significant induction of sub-G1 cells when transfecting the mimic-204 in both GTL-16 and HUCCT1 cells. Thus, targeting of the single targets did not fully mimic the effect of miR-204 expression. This apparent discrepancy may be explained by the observation that, with regard to the action of the microRNA, it is the coordinated increase of the levels of all or some of its targets that impinges significantly on the overall phenotype. Additionally, in vitro experiments revealed that the seven representative target genes exhibited different kinetics of modulation in time. This may be accounted by differences in the steady-state levels of the target mRNAs, different accessibility of the 3'-UTR, or cell subpopulation-specific expression of some www.impactjournals.com/oncotarget targets, within the cell lines examined. With regard to this, we also observed different kinetics between the two cell lines analyzed (GTL-16 and HUCCT1, respectively, with the GTL-16 exhibiting a earlier modulation of the targets when transfected with the mimic-204 (as compared with the HUCCT1 cells, transfected in very similar conditions). This would be, possibly, due to different proliferation rates or intracellular processing of the mimic molecules between the GC and the CC cells.
Last but not least, all of the samples included in our casuistries (both gastric cancer and cholangiocarcinoma) derived from patients not previously treated with chemotherapy. This is an important point. In fact, our and others published work indicate that miR-204 may play a role in determining chemoresistance, at least partially through the modulation of the anti-apoptotic BCL2 protein, in gastric [14] and breast cancer cells [32]. We aim to perform more studies to investigate how the levels of miR-204 are modulated in chemotherapy-treated gastric cancer and cholangiocarcinoma cell lines and, importantly, whether the chemotherapy-induced stress may change the spatial and/or temporal modulation of its targets.

Study population
In the present study 40 gastric surgical specimens from patients who underwent resection for cancer at the Regina Elena Cancer Institute and 11 specimens from the Sant' Andrea Hospital, Rome, were included MiR-204 levels were also analyzed from 26 fresh cholangiocarcinoma tissues of patients who underwent resection for tumor at Regina Elena Cancer Institute and from 19 cholangiocarcinoma formalin embedded tissues (FFPE) from the Sant' Orsola Hospital, Bologna. Case series included also additional hepatic fresh frozen tissues samples, corresponding to 23 HCC and to 11 liver metastases, from Regina Elena Cancer Institute. For each patient, we collected tissue samples from the tumor lesion and from the uninvolved mucosa, defined as the portion of tissue located at 2 cm from the safe resection margin of the tumor lesion. Data regarding patients' demographics, clinical and pathological features as well as the clinical outcomes were obtained from prospectively collected databases. The ethical committees of each of the mentioned institutes have approved the study.

RNA extraction
Fresh frozen tissues samples were obtained immediately after surgical resection and stored in RNA later (THERMOFISHER, MA USA) to stabilize and protect RNA in fresh specimens. After disruption and homogenization in QIAzol Lysis Reagent (QIAGEN, Chatsworth, CA) RNA was extracted using miRNeasy Kit (QIAGEN Chatsworth, CA) following the manufacturer's instructions. The concentration, purity and quality control of RNA extracted were assessed using Agilent 2100 Bioanalyzer. RNA from FFPE samples was extracted using the miRneasy ® FFPE kit (QIAGEN Chatsworth, CA) following the manufacturer's instructions. Total RNA from cell lines was extracted using the TRIZOL Reagents (GIBCO). Small amounts of RNA (10 ng) were reverse transcribed using the TaqMan microRNA Reverse Transcription Kit (Applied Biosystems) in a final volume of 10ul using a ABI Prism 7000 Sequence Detection System (Applied Biosystems). The PCR Reactions were initiated with a 10 minute incubation at 95°C followed by 40 cycles of 95°C for 15 seconds and 60°C for 60 seconds. Q-PCR quantification of miRNA expression was performed using TaqMan MicroRNA ® Assays (Applied