Identification of Biomarkers for the Prognosis of Pancreatic Ductal Adenocarcinoma with miRNA Microarray Data

Abstract

Background

The aim of this study was to explore the mechanism of chemotherapy resistance and to screen biomarkers of pancreatic ductal adenocarcinoma (PDAC).

Methods

MicroRNA (miRNA) expression profile data for GSE38781 were downloaded from the Gene Expression Omnibus database. Differentially expressed miRNAs between short–overall survival (OS) and long-OS patients were screened with the limma package in R. The function and protein–protein interaction (PPI) network of the miRNA target genes were further investigated. Finally, multivariate statistical analysis was performed to verify the significant miRNAs obtained in our work.

Results

In total, 66 miRNAs were identified to be differentially expressed. Gene ontology (GO) and pathway enrichment analysis showed that 163 miRNA target genes were mainly enriched in heart function, cancer development and angiogenesis. Ten nodes, including TGFBR1, TGFBR2, ACVR1 and SHC1, were found to be hub nodes in the PPI network. Multivariate statistical analysis showed 8 of the most significant miRNAs could completely distinguish the 2 groups of samples. Seven target genes (i.e., RET, ETS1, RHOA, NUMB, TIAM, ITGA5 and YY1) of the 8 significant miRNAs were found to be associated with control of cell fate decisions, T-cell lymphoma invasion and angiogenesis enhancement.

Conclusions

The heart function–related pathway, cell cycle, immune system and angiogenesis may be dysregulated in patients with poorer prognosis. The significant nodes (e.g., TGFBR1, TGFBR2, ACVR1 and SHC1) in the PPI network may be potential biomarkers for predicting outcomes for patients with pancreatic cancer. The significant miRNAs and gene targets may be potential biomarkers or therapeutic targets for PDAC.

Keywords

Differentially expressed microRNAs Hierarchical clustering analysis Multivariate statistical analysis Pancreatic adenocarcinoma Protein-protein interaction network

Introduction

Pancreatic cancer is the fourth leading cause of cancer-related death in Western countries (1, 2). Pancreatic adenocarcinomas characterized by glandular architecture accounted for 95% of all pancreatic tumors in 2011 (3). Pancreatic adenocarcinoma can result in various symptoms, such as abdominal pain, significant weight loss, poor appetite and jaundice. Patients with pancreatic adenocarcinoma are commonly found to have a poor prognosis, which partly results from the absence of symptoms in its early stages. It is reported that the median survival time of pancreatic adenocarcinoma patients is less than 6 months, and 5% of patients have a dismal, 5-year survival (4, 5).

At present, the general treatments for pancreatic adenocarcinoma primarily include surgery, radiation and chemotherapy. Radiation and chemotherapy are used for patients not suitable for resection with curative intent, or as adjuvant treatment to increase the survival of patients with surgical resection (6). However, the prognosis of patients after complete resection is poor, with a 3-year disease-free survival rate of 27% and median overall survival (OS) of 15-19 months (7). Therefore, an increasing number of studies have attempted to make contributions to discovering novel biomarkers with good sensitivity for adjuvant therapy and prognosis prediction.

MicroRNAs (miRNAs) are a group of small noncoding RNA molecules, which possess functions related to transcription and posttranscriptional regulation of gene expression (8). The miRNAs are expressed differentially in normal and malignant tissues, and tumor-associated miRNAs are detectable in patient circulation (9). Investigations of miRNAs have provided additional insights for the exploration of new therapies and prognosis information for cancers. It is reported that microRNA-26b targeting ubiquitin-specific peptidase 9 (USP9X) can prevent hepatocellular carcinoma by inhibiting the epithelial-mesenchymal transition (10). MiRNA-15b has been found to be a potential biomarker for predicting the antitumor effect of rosemary in colorectal and pancreatic cancers (11). However, information regarding miRNAs in pancreatic adenocarcinoma is insufficient.

In the present study, we downloaded the miRNA expression profiling of patients with pancreatic ductal adenocarcinoma (PDAC) from the Gene Expression Omnibus (GEO) database. The differentially expressed miRNAs between patients with short OS and long OS were investigated. The purpose of this paper was to explore the relationship of tumor-associated miRNAs with chemotherapy sensitivity and prognosis.

Materials and Methods

Affymetrix Microarray data and Differentially Expressed miRNA Analysis

The miRNA expression profiling of GSE38781 was downloaded from the publicly available GEO database (12), which had been deposited by Giovannetti and his colleagues (7). Patients with PDAC who underwent radical surgical resection with curative intent (pancreaticoduodenectomy, total pancreatectomy and distal pancreatectomy) and who were treated with 3 cycles of standard gemcitabine adjuvant regimen were carefully selected. A total of 26 tumor tissues were collected from patients with PDAC, among which, 19 samples were unsuitable for miRNA analysis and were used for the microarray development. The 19 samples were divided into 2 groups including a short-OS group (from PDAC patients dying within 1 year of diagnosis) and long-OS group (from patients who had survived more than 30 months). The raw data were downloaded for further studies based on the platform of GPL14903 (3D-Gene Human miRNA V16_1.0.0).

After the raw data were preprocessed, the differentially expressed miRNAs between the short-OS and long-OS group were analyzed with the limma package in R (13). A multiple testing correction was performed using the Benjamini-Hochberg false discovery rate (HB FDR) (14). An FDR <0.05 and |log FC (fold change)| >1 were used as the threshold for identifying differentially expressed miRNAs.

Hierarchical Clustering Analysis

Hierarchical clustering has been widely applied in analyzing gene expression patterns (15, 16). With hierarchical clustering, the most similar expression patterns can be clustered in a hierarchy of nested subsets. The expression values of differentially expressed miRNAs were selected according to the probe information from the downloaded files. The hierarchical clustering analysis for differentially expressed miRNAs was performed with Cluster software (17), and the results were visualized on a heatmap with Treeview software (18).

MiRNA Target Gene Prediction and Function Enrichment Analysis

To analyze the role of miRNAs in biological mechanism in response to chemotherapy, a gene ontology (GO) function analysis was performed for target genes of the differentially expressed miRNAs. The target genes for differentially expressed miRNAs were collected based on the databases of miRecord (19), miRTarBase (20) and Tarbase 6.0 (21). GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis for the miRNA target genes were performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) tool (22). A p value <0.05 was defined as the cutoff value.

Construction of Protein–Protein Interaction Network

To further analyze the function of miRNA target genes in PDAC patients, we performed a protein–protein interaction (PPI) network analysis. The miRNA target genes which were enriched in at least 1 pathway or GO term were input into the Human Protein Reference Database (HPRD; http://www.hprd.org/) (23), and the protein pairs involved with the target genes were obtained. The miRNA target genes and those normal genes that were interacting with at least 3 target genes were mapped into the PPI network using the Python program (24). The PPI network was visualized with the Cytoscape package (25).

Furthermore, to identify the significant nodes, topological characteristics of the PPI network were examined by a Cytoscape plug-in, including for node degree, average shortest path length (ASPL), clustering coefficient (CC) and centrality. The significant node genes were further analyzed based on the automatically selected weighting method with the GeneMANIA tool (26).

Multivariate Statistical Analysis

To verify the significance of the miRNAs identified in this paper, we performed logistic regression analysis for the differentially expressed miRNAs, using 2 multivariate statistical methods: logistic regression analysis and discriminant analysis.

Logistic regression analysis was performed on the miRNA expression values in 19 patients to identify the miRNAs that could effectively distinguish the samples from the 2 groups (short-OS vs. long-OS groups). Discriminant analysis, also known as “resolution,” is a multivariate statistical analysis to identify the type ascription according to various characteristic values of a study in certain conditions. Two kinds of multivariate statistical analysis were performed using SPSS 19.0 software with a threshold of p<0.05.

Results

Identification of Differentially Expressed miRNAs and Hierarchical Clustering Analysis

To find the differentially expressed miRNAs between the short-OS group and long-OS group, we obtained the publicly available microarray dataset GSE38781 from the GEO database. Total 66 miRNAs were identified to be differentially expressed between the short-OS and long-OS groups based on the cutoff value of FDR <0.05 and |log FC| >1.

After the expression values of the 66 differentially expressed miRNAs were preprocessed, hierarchical clustering of 63 miRNAs in 19 samples was performed. As shown in Figure 1, the samples from the short-OS group and long-OS group are clearly distinguished based on the expression profiling of the differentially expressed miRNAs. Meanwhile, the expression patterns of miRNAs are clustered into 2 subclusters. In subcluster 1, 33 miRNAs show high expression levels in the short-OS group and low expression levels in the long-OS group. A total of 30 miRNAs show low expression levels in the short-OS group and high expression levels in the long-OS group (subcluster 2).

Fig. 1

Hierarchical clustering heatmaps of the differentially expressed miRNAs. The gradient of colors from blue to orange represents expression values from low to high. *represents the miRNA expressed from the mature miRNA or the miRNA with relatively low-abundance.

Function Enrichment Analysis

After searching in the databases of miRecord, miRTarBase and Tarbase 6.0, we obtained 163 target genes for 11 differentially expressed miRNAs. GO enrichment and KEGG pathway analysis were performed for these genes with DAVID.

As shown in Table I, the significantly enriched pathways were mainly related to cancer processes and cardiac function, such as colorectal cancer, pancreatic cancer, arrhythmogenic right ventricular and hypertrophic cardiomyopathy. The most significant pathway was the hsa05210:colorectal cancer pathway (p = 2.47E-04). The overexpressed GO terms were relevant for regulation of transcription, regulation of RNA metabolic process, cell motion, cell migration and blood vessel development (Tab. I).

TABLE I

Function enrichment analysis of target genes for differentially expressed miRNAs

Term	Count	%	p Value
GO enrichment analysis
GO:0006355: regulation of transcription, DNA-dependent	43	26.875	7.56E-07
GO:0006357: regulation of transcription from RNA	25	15.625	1.27E-06
GO:0051252: regulation of RNA metabolic process	43	26.875	1.37E-06
GO:0045449: regulation of transcription	51	31.875	2.27E-05
GO:0006350: transcription	40	25	5.32E-04
GO:0006928: cell motion	20	12.5	1.14E-06
GO:0048870: cell motility	16	10	1.47E-06
GO:0051674: localization of cell	16	10	1.47E-06
GO:0016477: cell migration	15	9.375	2.16E-06
GO:0001568: blood vessel development	11	6.875	3.82E-04
GO:0001944: vasculature development	11	6.875	4.62E-04
GO:0048514: blood vessel morphogenesis	10	6.25	5.46E-04
Pathway enrichment analysis
hsa05210: Colorectal cancer	8	5	2.47E-04
hsa04520: Adherens junction	7	4.375	0.001001
hsa05220: Chronic myeloid leukemia	6	3.75	0.00528
hsa04350: TGF-beta signaling pathway	6	3.75	0.009825
hsa05212: Pancreatic cancer	5	3.125	0.023013
hsa05412: Arrhythmogenic right ventricular	5	3.125	0.027435
hsa04512: ECM-receptor interaction	4	2.5	0.01338
hsa05410: Hypertrophic cardiomyopathy	4	2.5	0.013726
hsa05414: Dilated cardiomyopathy	4	2.5	0.016233

GO = gene ontology.

PPI Network of Target Genes

Based on the information from the HPRD database, we collected the protein pairs involved with the 163 miRNA target genes. A PPI network with 222 nodes and 562 edges was constructed using the Cytoscape software (Fig. 2). After further analyzing the topological characteristics of the PPI network, we obtained 10 hub nodes, including transforming growth factor, beta receptor 1 (TGFBR1), SMAD family member 4 (SMAD4), Src homology 2 domain containing transforming protein 1 (SHC1), activin A receptor type I (ACVR1) and transforming growth factor beta receptor II (TGFBR2). The topological characteristics for the 10 significant nodes are listed in Table II. Then the biological functions of the 10 genes were further analyzed with the GeneMANIA tool. According to the annotation information recorded in GeneMANIA, a PPI network involved with the 10 target genes and 20 normal human genes was established (Fig. 3). GeneMANIA functional annotation for the 10 hub genes showed that the 10 genes were mainly involved in the pathway-restricted SMAD protein phosphorylation, the transforming growth factor beta receptor signaling pathway and transforming growth factor beta binding (Tab. III).

TABLE II

Top 10 gene nodes with high node degrees in the network and the corresponding topology parameters

Node	Node degree	Centrality	ASPL	CC
TGFBR1	49	0.421756	2.371041	0.058693
SMAD4	48	0.426641	2.343891	0.045404
SHC1	31	0.37585	2.660633	0.054187
ACVR1	24	0.359935	2.778281	0.065217
TGFBR2	22	0.378425	2.642534	0.137255
RET	20	0.350238	2.855204	0.1
ITGB3	17	0.345313	2.895928	0.054945
CDKN1A	17	0.324047	3.085973	0
EZR	16	0.325	3.076923	0.038462
BCL2	16	0.342636	2.918552	0
AVE (mean value)	5.06306	0.28866	3.537157	0.119178

ACVR1 = activin A receptor type I; ASPL = average shortest path length; CC = clustering coefficient; SMAD4 = SMAD family member 4; SHC1 = Src homology 2 domain containing transforming protein 1; TGFBR1 = transforming growth factor, beta receptor 1; TGFBR2 = transforming growth factor, beta receptor II.

TABLE III

GeneMANIA functional annotation

Feature	FDR	Count	Total
Pathway-restricted SMAD protein phosphorylation	1.54E-09	7	36
Transforming growth factor beta receptor signaling pathway	3.48E-09	9	141
Cytokine receptor binding	3.48E-09	9	139
Regulation of SMAD protein phosphorylation	5.50E-09	6	25
Cellular response to transforming growth factor beta stimulus	5.50E-09	9	160
Response to transforming growth factor beta stimulus	5.50E-09	9	160
Transforming growth factor beta binding	1.34E-08	5	12
Regulation of pathway-restricted SMAD protein phosphorylation	1.69E-08	6	32
Transmembrane receptor protein serine	4.45E-08	9	211
Transforming growth factor beta receptor binding	5.16E-08	5	16

FDR = false discovery rate.

Fig. 2

Protein–protein interaction (PPI) network for miRNA target genes and some normal interactional proteins. The size of the nodes represents the node degree in the network – the higher the degree, the bigger the node. The color of the nodes represent the centrality, the higher the centrality, the deeper the color (red) and the lower the centrality, the lighter the color (green).

Fig. 3

A network was constructed from GeneMANIA annotation information for 10 important genes, including physical interaction, genetic interaction, co-expression and shared pathways and protein structure domain. Black nodes represent 10 important genes, and gray nodes represent gene interactions with the black nodes. The network also contains 20 normal human proteins.

Multivariate Statistical Analysis

To verify the significance of the differentially expressed miRNAs through a statistical analysis, we used 2 kinds of multivariate statistical methods to analyze these miRNAs.

A binary logistic regression model (27) was applied to calculate the expression values of 66 significant miRNAs and the correlations of the 2 classified groups of patients. As shown in Table IV, 40 miRNAs were found to effectively distinguish the short-OS and long-OS patients in the binary logistic regression model (p<0.05). A discriminant analysis model (28) was also applied to analyze the significance of 66 miRNAs. Based on the expression values of the 66 miRNAs in 19 patients, 8 miRNAs (i.e., hsa-miR-1914, hsa-miR-4281, hsa-miR-1274a, hsa-miR-1249, hsa-miR-1207-3p, hsa-miR-466, hsa-miR-1290 and hsa-miR-31) were identified to be significant (p<0.05; Tab. V), which was consistent with the results of the binary logistic regression model. Discriminant analysis suggested that the 8 miRNAs could completely distinguish the short-OS and long-OS patients (accuracy 100%).

TABLE IV

MiRNAs identified by binary logistic regression

miRNA	Score	df	p Value
hsa-miR-1197	16.548	1	0.000
hsa-miR-1914	14.674	1	0.000
hsa-miR-4321	11.989	1	0.001
hsa-miR-940	11.791	1	0.001
hsa-miR-4281	11.670	1	0.001
hsa-miR-1207-3p	11.556	1	0.001
hsa-miR-1914	11.279	1	0.001
hsa-miR-1246	11.157	1	0.001
hsa-miR-371-5p	10.821	1	0.001
hsa-miR-181a-2	10.469	1	0.001
hsa-miR-887	9.840	1	0.002
hsa-miR-3610	9.747	1	0.002
hsa-miR-532-5p	9.591	1	0.002
hsa-miR-483-3p	9.520	1	0.002
hsa-miR-1249	9.348	1	0.002
hsa-miR-92a-2	9.048	1	0.003
hsa-miR-25	9.028	1	0.003
hsa-miR-466	8.633	1	0.003
hsa-miR-1915	8.321	1	0.004
hsa-miR-1200	8.280	1	0.004
hsa-miR-1274a	8.270	1	0.004
hsa-miR-149	7.781	1	0.005
hsa-miR-619	7.756	1	0.005
hsa-miR-4284	7.528	1	0.006
hsa-miR-1911	6.723	1	0.010
hsa-miR-4286	6.679	1	0.010
hsa-miR-1290	6.668	1	0.010
hsa-miR-3681	6.643	1	0.010
hsa-miR-3196	6.632	1	0.010
hsa-miR-4271	6.606	1	0.010
hsa-miR-197	6.213	1	0.013
hsa-miR-1236	6.165	1	0.013
hsa-miR-671-3p	6.086	1	0.014
hsa-miR-326	5.456	1	0.020
hsa-miR-31	4.981	1	0.026
hsa-miR-934	4.815	1	0.028
hsa-miR-412	4.329	1	0.037
hsa-miR-4251	4.003	1	0.045
hsa-miR-204	3.952	1	0.047
hsa-miR-4310	3.946	1	0.047

TABLE V

MiRNAs identified by discriminant analysis

miRNA	Short OS	Long OS
hsa-miR-1914	-2.352	-8.595
hsa-miR-4281	0.194	0.685
hsa-miR-1274a	-0.211	-0.802
hsa-miR-1249	-2.940	-10.798
hsa-miR-1207-3p	11.548	34.288
hsa-miR-466	20.182	69.342
hsa-miR-1290	-3.835	-12.264
hsa-miR-31	0.214	0.760
(Constant)	-198.728	-1,710.581

OS = overall survival.

From the miRecord, miRTarBase and Tarbase 6.0, we obtained 26 target genes of the 8 significant miRNAs. Based on the information from the HPRD, 19 target genes were mapped into the PPI network, among which 7 target genes were found to have high node degrees and were mainly involved in control of cell fate decisions, E26 oncogene homolog 1 and enhancing angiogenesis (Tab. VI).

TABLE VI

Functional annotations of 7 important genes regulated by 8 miRNAs identified by discriminant analysis

Gene symbol	Node degree	Function annotation
RET	20	Ret proto-oncogene
RHOA	14	Ras homolog gene family
ETS1	11	E26 oncogene homolog 1
NUMB	9	Control of cell fate decisions
TIAM1	7	T-cell lymphoma invasion and metastasis 1
ITGA5	6	Enhance angiogenesis
YY1	6	Multifunctional transcription factor

Discussion

As the mortality rate for pancreatic cancer patients is almost 0.99 (29), there is an urgent need for understanding the molecular mechanisms underlying this disease to develop better detection, diagnostic and prognostic markers as well as therapeutic targets that could improve clinical management and therapeutic outcomes for these patients. In this study, we downloaded the miRNA expression profiles from PDAC patients with similar clinicopathological characteristics and chemotherapy but different clinical outcomes (short-OS vs. long-OS groups). Poor prognosis had an association with chemotherapy resistance (30). The miRNAs that showed different expression levels between the short-OS and long-OS groups were selected. The relevant miRNA target genes were collected, and the functions for significant genes were further investigated. The purpose of this paper was to explore the chemotherapy sensitivity–related mechanism and the potential biomarkers for prognosis prediction.

Our results showed that 66 miRNAs were differentially expressed, and hierarchical cluster analysis for these miRNAs showed a good separation of the samples from the short-OS and long-OS groups. A total of 163 target genes for the differentially expressed miRNAs were collected. Pathway enrichment analysis for the 163 miRNA target genes revealed that the significant pathways were mainly involved with cancer development and heart function such as arrhythmogenic right ventricular, hypertrophic cardiomyopathy and dilated cardiomyopathy. As outlined in previous reports, the cardiovascular disease can lead to various pathophysiological changes which may affect pharmacokinetics such as drug absorption, distribution, metabolism and excretion (31). The cardiac dysfunctions may result in hypoperfusion to the drug clearance site, a decline in the volume of drug distributed and impairment of drug clearance (32). The limited distribution of anticancer drug in tumor tissue is related to the resistance of cancers to chemotherapy (33). In addition, compelling evidence shows that the arachidonic acid (AA) cascade, which is important in cardiovascular disease, is involved in the development of pancreatic adenocarcinomas (34). The treatment and prevention of cardiovascular disease with therapies such as β-blockers, inhibitors of AA-metabolizing enzymes and a low-fat diet for cardiovascular disease are also effective for PDAC. Although there is no direct association between heart disease and PDAC, heart functions may be related to drug resistance and may influence the survival of PDAC patients.

In addition, the target genes of differentially expressed miRNAs in PDAC patients were mapped into a PPI network. Ten genes were identified to be the hub nodes, including TGFBR1, TGFBR2, SMAD4, ACVR1 and SHC1. TGFBR1 and TGFBR2 encode transforming growth factor, beta receptor I and transforming growth factor, beta receptor II, respectively, which are members of the transforming growth factor beta (TGFB) receptor subfamily. TGFBR1 and TGFBR2 are TGF-beta receptors and play roles in transducing the TGF-beta signal. TGF-beta modulates many aspects of cellular function, and the disruption of the activity of TGFB super members (such as TGFBR1 and TGFBR2) contributes to a variety of diseases such as various cancers and pulmonary hypertension (35, 36). SMAD family member 4 (SMAD4), encoded by the SMAD4 gene, is a member of the Smad signal transduction protein family. Smad proteins are phosphorylated and activated by transmembrane serine-threonine receptor kinases in response to TGF-beta signaling (37). Previous evidence has implied that the mutation of the SMAD4 gene is correlated with the outcomes of patients with pancreatic cancer (38). SMAD4 inactivation has close associations with shorter OS in pancreatic cancer patients with resected surgeries (39). ACVR1 encodes the Activin A receptor type I, which is a signaling protein belonging to the TGF beta superfamily. Mutations of ACVR1 have been found in patients with pancreatic cancer (40). Src homology 2 domain containing transforming protein 1 (SHC1), encoded by SHC1, is reported to have functions in regulation of apoptosis and drug resistance in mammalian cells (41). Recent evidence shows that SHC1 regulated by mi-365 is associated with gemcitabine resistance in patients with pancreatic cancer (42). Therefore, the significant gene nodes in the PPI network may play key roles in the mechanism of chemotherapy resistance and may be potential genetic biomarkers for the prognosis of patients with pancreatic cancer. However, a large number of further studies are warranted.

In addition, by multivariate statistical methods, 8 of 66 differentially expressed miRNAs were found to be the most significant and could completely distinguish the samples from the 2 groups (short-OS and long-OS). Seven target genes of the 8 significant miRNAs were proven to have higher node degrees in the PPI network (i.e., RET, ETS1, RHOA, NUMB, TIAM, ITGA5 and YY1).

RET and ETS1 are important proto-oncogenes associated with tumorigenesis (43, 44). RHOA and NUMB are involved in the important process of cell cycle and regulate cell proliferation and apoptosis (45, 46). TIAM1 and ITGA5 regulate invasion and metastasis of immune cells and angiogenesis, which are closely associated with tumorigenesis, metastasis and chemotherapy resistance (47, 48). YY1 is a kind of multifunctional transcription factor controlling complex biological processes (49). The 7 important genes regulated by 8 miRNAs affect the biological processes of cell cycle, immune system and angiogenesis, and drug resistance of pancreatic cancer. So the 8 significant miRNAs (hsa-miR-1914, hsa-miR-4281, hsa-miR-1274a, hsa-miR-1249, hsa-miR-1207-3p, hsa-miR-466, hsa-miR-1290 and hsa-miR-31) may play key roles in regulating the resistance of patients with pancreatic cancer.

In conclusion, the differentially expressed miRNAs between PDAC patients with short-OS and long-OS affected their resistance by regulating their target genes. Heart function related pathway, cell cycle, immune system and angiogenesis may be dysregulated in patients with poorer prognosis. The significant nodes (such as TGFBR1, TGFBR2, ACVR1 and SHC1) in the PPI network may be potential markers for predicted the outcomes of patients with pancreatic cancer. Although some significant genes and miRNAs were found to be related to prognosis of pancreatic cancer, lack of the experimental evidence was a limitation of our work. Our findings may prompt further prospective studies and research into the predicted and therapeutic role of the significant genes and differentially expressed miRNAs. However, extensive further studies should be conducted to verify the findings of our work.

Footnotes

Financial support: No grants or funding have been received for this study.

Conflict of interest: All authors declare that they have no conflict of interest to state.

References

Hezel

Kimmelman

Stanger

Bardeesy

Depinho

Genetics and biology of pancreatic ductal adenocarcinoma.

Genes Dev 2006;20(10):1218–1249

Hariharan

Saied

Kocher

Analysis of mortality rates for pancreatic cancer across the world.

HPB (Oxford) 2008;10(1):58–62

Jemal

Bray

Center

Ferlay

Ward

Forman

Global cancer statistics.

CA Cancer J Clin 2011;61(2):69–90

Hezel

Kimmelman

Stanger

Bardeesy

Depinho

Genetics and biology of pancreatic ductal adenocarcinoma.

Genes Dev 2006;20(10):1218–1249

Sperti

Pasquali

Piccoli

Pedrazzoli

Survival after resection for ductal adenocarcinoma of the pancreas.

Br J Surg 1996;83(5):625–631

Childs

Szymonifka

Delaying chemoradiation until after completion of adjuvant chemotherapy for pancreatic cancer may not impact local control.

Pract Radiat Oncol 2014;4(2):e117–e123

Giovannetti

van der Velde

Funel

High-throughput microRNA (miRNAs) arrays unravel the prognostic role of MiR-211 in pancreatic cancer.

PLoS ONE 2012;7(11):e49145

Chen

Rajewsky

The evolution of gene regulation by transcription factors and microRNAs.

Nat Rev Genet 2007;8(2):93–103

Ellinger

Muller

MicroRNAs: a novel non-invasive biomarker for patients with urological malignancies.

Curr Pharm Biotechnol 2014;15(5):486–491

10.

Shen

Lin

Yang

Zhang

Jia

MicroRNA-26b inhibits epithelial-mesenchymal transition in hepatocellular carcinoma by targeting USP9X.

BMC Cancer 2014;14(1):393

11.

González-Vallinas

Molina

Vicente

Expression of microRNA-15b and the glycosyltransferase GCNT3 correlates with antitumor efficacy of Rosemary diterpenes in colon and pancreatic cancer.

PLoS ONE 2014;9(6):e98556

12.

Barrett

Troup

Wilhite

NCBI GEO: mining tens of millions of expression profiles—database and tools update.

Nucleic Acids Res 2007;35Database issue:D760–D765

13.

Chapter V in book Smyth

Limma: linear models for microarray data.

In: Robert

Gentleman

Carey

Vincent J

Wolfgang

Huber

Irizarry

Rafael A

Sandrine

Dudoit

ed. Bioinformatics and computational biology solutions using R and Bioconductor New York

Springer

2005397–420

14.

Benjamini

Hochberg

Controlling the false discovery rate: a practical and powerful approach to multiple testing.

J R Stat Soc, B 1995;57:289–300

15.

Eisen

Spellman

Brown

Botstein

Cluster analysis and display of genome-wide expression patterns.

Proc Natl Acad Sci USA 1998;95(25):14863–14868

16.

Iyer

Eisen

Ross

The transcriptional program in the response of human fibroblasts to serum.

Science 1999;283(5398):83–87

17.

Bouman

Shapiro

Cook

Atkins

Cheng

Cluster: an unsupervised algorithm for modeling Gaussian mixtures.

School of Electrical Engineering

Purdue University

1997

18.

Page

Visualizing phylogenetic trees using TreeView.

Curr Protoc Bioinformatics 2002;6.21-6:2–15

19.

Xiao

Zuo

Cai

Kang

Gao

miRecords: an integrated resource for microRNA-target interactions.

Nucleic Acids Res 2009;37Database issue:D105–D110

20.

Hsu

Lin

miRTarBase: a database curates experimentally validated microRNA-target interactions.

Nucleic Acids Res 2011;39Database issue:D163–D169

21.

Vergoulis

Vlachos

Alexiou

TarBase 6.0: capturing the exponential growth of miRNA targets with experimental support.

Nucleic Acids Res 2012;40Database issue:D222–D229

22.

Huang

Sherman

Lempicki

Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources.

Nat Protoc 2009;4(1):44–57

23.

Keshava Prasad

Goel

Kandasamy

Human Protein Reference Database: 2009 update.

Nucleic Acids Res 2009;37Database issue:D767–D772

24.

Sanner

Python: a programming language for software integration and development.

J Mol Graph Model 1999;17(1):57–61

25.

Shannon

Markiel

Ozier

Cytoscape: a software environment for integrated models of biomolecular interaction networks.

Genome Res 2003;13(11):2498–2504

26.

Zuberi

Franz

Rodriguez

GeneMANIA prediction server 2013 update.

Nucleic Acids Res 2013;41Web Server issue:W115–22

27.

Bihrmann

Toft

Nielsen

Ersbøll

Spatial correlation in Bayesian logistic regression with misclassification.

Spat Spatio-Temporal Epidemiol 2014;9:1–12

28.

Liu

Thung

Shen

Multi-task linear programming discriminant analysis for the identification of progressive MCI individuals.

PLoS ONE 2014;9(5):e96458

29.

Hayat

Howlader

Reichman

Edwards

Cancer statistics, trends, and multiple primary cancer analyses from the Surveillance, Epidemiology, and End Results (SEER) Program.

Oncologist 2007;12(1):20–37

30.

Virappane

Gale

Hills

Mutation of the Wilms’ tumor 1 gene is a poor prognostic factor associated with chemotherapy resistance in normal karyotype acute myeloid leukemia: the United Kingdom Medical Research Council Adult Leukaemia Working Party.

J Clin Oncol 2008;26(33):5429–5435

31.

Rodighiero

Effects of cardiovascular disease on pharmacokinetics.

Cardiovasc Drugs Ther 1989;3(5):711–730

32.

Shammas

Dickstein

Clinical pharmacokinetics in heart failure: an updated review.

Clin Pharmacokinet 1988;15(2):94–113

33.

Minchinton

Tannock

Drug penetration in solid tumours.

Nat Rev Cancer 2006;6(8):583–592

34.

Schuller

Mechanisms of smoking-related lung and pancreatic adenocarcinoma development.

Nat Rev Cancer 2002;2(6):455–463

35.

Massagué

TGF-β signal transduction.

Annu Rev Biochem 1998;67(1):753–791

36.

Attisano

Wrana

Signal transduction by the TGF-β superfamily.

Science 2002;296(5573):1646–1647

37.

Heldin

C-H

Miyazono

ten Dijke

TGF-β signalling from cell membrane to nucleus through SMAD proteins.

Nature 1997;390(6659):465–471

38.

Tascilar

Skinner

Rosty

The SMAD4 protein and prognosis of pancreatic ductal adenocarcinoma.

Clin Cancer Res 2001;7(12):4115–4121

39.

Blackford

Serrano

Wolfgang

SMAD4 gene mutations are associated with poor prognosis in pancreatic cancer.

Clin Cancer Res 2009;15(14):4674–4679

40.

Jung

Beck

Cabral

Activin type 2 receptor restoration in MSI-H colon cancer suppresses growth and enhances migration with activin.

Gastroenterology 2007;132(2):633–644

41.

Pelicci

Lanfrancone

Grignani

A novel transforming protein (SHC) with an SH2 domain is implicated in mitogenic signal transduction.

Cell 1992;70(1):93–104

42.

Hamada

Masamune

Miura

Satoh

Shimosegawa

MiR-365 induces gemcitabine resistance in pancreatic cancer cells by targeting the adaptor protein SHC1 and pro-apoptotic regulator BAX.

Cell Signal 2014;26(2):179–185

43.

Nakayama

Ito

Ohtsuru

Expression of the Ets-1 proto-oncogene in human gastric carcinoma: correlation with tumor invasion.

Am J Pathol 1996;149(6):1931–1939

44.

Alberti

Carniti

Miranda

Roccato

Pierotti

RET and NTRK1 proto-oncogenes in human diseases.

J Cell Physiol 2003;195(2):168–186

45.

Sahai

Alberts

Treisman

RhoA effector mutants reveal distinct effector pathways for cytoskeletal reorganization, SRF activation and transformation.

EMBO J 1998;17(5): 1350–1361

46.

Wirtz-Peitz

Nishimura

Knoblich

Linking cell cycle to asymmetric division: Aurora-A phosphorylates the Par complex to regulate Numb localization.

Cell 2008;135(1):161–173

47.

Bissell

Radisky

Putting tumours in context.

Nat Rev Cancer 2001;1(1):46–54

48.

Bonauer

Carmona

Iwasaki

MicroRNA-92a controls angiogenesis and functional recovery of ischemic tissues in mice.

Science 2009;324(5935):1710–1713

49.

Bushmeyer

Park

Atchison

Characterization of functional domains within the multifunctional transcription factor, YY1.

J Biol Chem 1995;270(50):30213–30220