An eight-miRNA signature expression-based risk scoring system for prediction of survival in pancreatic adenocarcinoma

Abstract

PURPOSE:

The purpose of this study was to establish a risk scoring system based on miRNAs to evaluate the prognosis in pancreatic adenocarcinoma.

METHODS:

Using a miRNA microarray dataset (179 pancreatic adenocarcinoma specimens and 4 normal control specimens) from TCGA, differentially expressed miRNAs were identified. Cox proportional hazards regression analysis was used to identify significant prognostic miRNAs, with which a risk scoring system was established and tested on a validation set. Cox regression analysis was performed to identify independent predictors of survival from clinical characteristics. Stratified Cox regression analyses were conducted to unravel the associations of clinical characteristics with survival. Differentially expressed genes (DEGs) were screened followed by functional annotation of the DEGs.

RESULTS:

Eight miRNAs (miR-1301, miR-598, miR-1180, miR-155, miR-496, miR-203, miR-193b, miR-135b) were independent predictors for survival. A risk scoring system was established with the 8 signature miRNAs. Upon Cox multivariate regression analysis, risk score, new tumor and targeted molecular therapy were independent predictors of prognosis. Stratified Cox regression analyses found that targeted molecular therapy and new tumor are associated with survival of patients. Survival-related DEGs were significantly enriched with regulation of transforming growth factor beta receptor, potassium ion transport and MAPK signaling pathway.

CONCLUSIONS:

The study proposes 8-miRNA expression-based risk scoring system to predict prognosis in pancreatic adenocarcinoma. New tumor and targeted molecular therapy were independent predictors of prognosis. Transforming growth factor beta receptor, potassium ion transport and MAPK signaling pathway may be related to prognosis in pancreatic adenocarcinoma.

Keywords

miRNA prognosis pathway risk score stratified analyses

1. Introduction

Pancreatic cancer, one of leading cancers, is estimated to cause 41,780 deaths in the United States in 2016 [1]. Its most common type is pancreatic adenocarcinoma. Early diagnosis of pancreatic cancer remains a challenge, and patients have a very poor prognosis with a five-year survival rate of 5–6% [2]. Surgical removal as the only cure is possible in only 20% of new cases [3]. In order to improve outcome of patients, identification of molecular prognostic markers have been an active area to investigate.

MicroRNA (miRNA) is a small non-binding RNA molecule (19–25 nucleotides) binding to 3’-UTR of target messenger RNAs (mRNAs) to regulate human genes [4]. They are involved in gene regulation, development, pathogenesis of various malignances and other biological processes [5]. MiRNAs are very exclusive to different types of tissues or cells, thus raising the prospects that they are of prognostic value for predicting outcome of cancer patients. Increasing studies have recommended panels of miRNA in serum as biomarkers for early diagnosis of pancreatic cancer [6, 7, 8]. A meta-analysis provides evidence that tumoural miR-21 is a promising predictor of prognosis after pancreatic ductal adenocarcinoma resection [9]. However, correlations of miRNAs with prognosis of pancreatic cancer have not been fully elucidated. Although Zhou et al. [10] have suggested a 13-miRNA signature as an independent prognostic marker of PC. However, prognostic value of the 13-miRNA signature was not tested in a validation set, and the underlying mechanisms were not further studied.

Comparing to the study by Zhou et al., this study not only developed a risk score system based on survival-related miRNAs from The Cancer Genome Atlas (TCGA), but also validate the risk score system on a Gene Expression Omnibus (GEO) dataset. Besides, stratified Cox regression analyses were conducted to identify significant clinical variables associated with survival. In addition, survival-related differentially expressed genes (DEGs) were identified. Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis were performed for function annotation of the DEGs identified. This study would provide novel insights into the associations of miRNAs with survival of patients with pancreatic cancer.

2. Materials and methods

2.1 Datasets

A miRNA-seq dataset (platform: Illumina HiSeq 2000 RNA Sequencing) and a mRNA-seq dataset (platform: Illumina HiSeq 2000 RNA Sequencing) were downloaded from TCGA (https://gdc-portal.nci.nih.gov/) data portal. Susequently, 183 panceatic adenocarcinoma specimens with both miRNA-seq and mRNA-seq were retrieved, including 179 pancereatic adenocartinoma specimens and 4 normal control specimens. The end point of the study was survival ratio defined as a percentage of patients surviving for a specific length of time after diagnosis. Data was processed according to the TCGA human project protection and data access polices.

As a validation dataset, a gene expression dataset (GSE62498) [11] was downloaded from GEO database (http://www.ncbi.nlm.nih.gov/geo/) based on GPL16 142 platform, consisting of miRNA expression profiling of 69 pancreatic ductual adenocartinoma samples. Clinical characteristics of the miRNA-seq dataset, mRNA-seq dataset and GSE62498 dataset were shown in Table 1.

Table 1
Baseline and clinical characteristics of patients in TCGA and GSE26498 datasets

Clinical characteristics	TCGA	GSE62498
	( $n=$ 178)	( $n=$ 69)
Age (mean $\pm$ SD)	64.69 $\pm$ 11.09	–
Gender (Male/Female/–)	91/74/13	–
Chronic pancreatitis history	13/117/48	–
(Yes/No/–)
Diabetes history (Yes/No/–)	36/99/43	–
Alcohol (Yes/No/–)	97/57/24	–
Tobacco (Never/Reform/Current/–)	59/56/19/44	–
New tumor (Yes/No/–)	55/101/22	–
Pathologic_M (M0/M1/–)	75/3/100	–
Pathologic_N (N0/N1/–)	44/117/17	–
Pathologic_T (T1/T2/T3/T4/–)	8/19/134/3/14	–
Pathologic_stage (I/II/III/IV/–)	20/137/4/3/	–
Radiation therapy (Yes/No/–)	38/107/33	–
Targeted molecular therapy	102/48/28	–
(Yes/No/–)
Dead (Death/Alive/–)	83/8213	49/16/4
Overall survival months	17.11 $\pm$ 15.35	20.36 $\pm$ 16.5
(mean $\pm$ SD)

“–” means information not available; SD, standard deviation; TCGA, The Cancer Genome Atlas.

2.2 Identification of differentially expressed miRNAs

Of the 179 pancreatic cancer specimens, 15 did not have survival results and 19 were alive with survival record shorter than 6 months, both of which thus were deleted from the study. The remaining 146 specimens were selected as training set and divided into two groups: bad prognosis group in which patients died in 6 months after diagnosis, and good prognosis group in which patients lived longer than 24 months. Differentially expressed miRNAs were screened between the bad prognosis group, and the good prognosis group by using Limma package [12] in R3.1.0 language, with false discovery rate (FDR) $<$ 0.05 as a strict cutoff.

2.3 Prognosis-related miRNA screening

In order to identify significant miRNAs associated with overall survival of pancreatic adenocarcinoma patients, univariate Cox proportional hazards regression analysis [13] was performed on the differentially expressed miRNAs screened using Survival package in R3.1.0 language. The log-rank test $P$ -value $<$ 0.05 was set as a strict cutoff value. The signficant miRNAs derived from univariate Cox regression analysis were then subjected to multivariate Cox regression analysis with log-rank test $P$ -value $<$ 0.05 as a threshold.

2.4 Risk scoring system

A risk score for a patient corresponded to a patient’s risk of death and was calculated as a linear combination of miRNAs expression levels weighted by regression coefficient ( $\beta$ ) extracted from above Cox regression analysis. Based on expression levels of the significant miRNAs extracted from multivariate Cox regression analysis, a risk scoring system was established by using the following formula [10, 14]:

$\displaystyle\text{Risk score}=\beta\text{gene1}\times\text{exprgene1}+\beta% \text{gene2}\times\text{exprgene2}+\ldots+\beta\text{genen}\times\text{exprgenen}$

$\beta$ value derived from the training set was applied to the GSE62498 dataset (test set) for predicting the prognosis of patients. According to the risk score which was assigned to each patient, patients in the training set and the validation set were stratified into two groups with the median risk score as the cutoff point: high risk group ( $\leqslant$ median risk score) and low risk group ( $>$ median risk score). Kaplan-Meier survival analysis was performed for the high and low risk groups in the training set and the validation set, respectively.

2.5 Analysis of correlations of patients’ clinical characteristics with survival

Cox univariate and multivariate regression analysis was performed to assess the associations of survival of patients in two sets with their clinical variables including age, gender, pathologic_M, alcohol, tobacco, chronic pancreatitis history, diabetes history, pathologic_N, pathologic_T, pathologic_stage, radiation therapy, new tumor, targeted molecular therapy, and risk score.

2.6 Stratified analyses

Correlations of each clinical variable with expression of each signature miRNA included in the risk scoring formula was evaluated by using Cox univariate regression analysis with survival package in R language [15]. Stratified Cox regression analyses were performed in high and low risk patients according to each clinical variable as well as comparison between patients stratifed by each clinical variable in high and low risk groups. Survival curves were made by the Kaplan-Meier and log-rank method. $P$ -value $<$ 0.05 suggests statistical significance unless specifically indicated.

2.7 GO and KEGG enrichment analyses

Following identification of mRNAs corresponding to the miRNAs in the training set based on individual specimen number, DEGs between the high risk group and the low risk group were identified using Limma package (2) in R3.1.0 language using FDR $<$ 0.05 as a cutoff value. Pearson’s correlation coefficient ( $r$ ) between expression of the screened DEGs and risk scores were calculated by using “cor” function in R language. According to the correlation efficient, the DEGs were separated into positive DEGs and negative DEGs. Next, for unveiling the functional involvement of these DEGs in the pancreatic cancer, Gene ontology (GO) [16] and Kyoto Encyclopedia of Genes and Genomes (KEGG) [17] enrichment analysis enrichment analysis were conducted for positive DEGs and negative DEGs by using DAVID softwhere [18], respectively.

3. Results

3.1 Identification of differentially expressed miRNAs

After cutting one miRNA with averaged expression less than 1 reads per million (RPM), there were total 318 miRNAs left in the training set. Of remaining 146 specimens in the training set, 18 were included in the bad prognosis group and 19 were included in the good prognosis group. Consequentially, a total of 36 differentially expressed miRNAs were selected between the good prognosis group and the bad prognosis group.

Figure 1.

Survival of patients in different groups stratified by low and high risk in TCGA (A) and GSE62498 (B).

Figure 2.

Comparison of 8 prognostic miRNAs expression between high and low risk patients in TCGA (A) and GSE62498 dataset (B). *0.01 $\leqslant p<$ 0.05; **0.005 $\leqslant p<$ 0.01; *** $p<$ 0.005.

3.2 Development of a 8-miRNA risk scoring system

Univariate Cox regression analysis found that 16 out of the 36 differentially expressed miRNAs were significantly related to prognosis of patients. With the 16 significant miRNAs, multivariate Cox regression analysis was employed and found 8 miRNAs were independent predictors of survival, including miR-1301, miR-598, miR-1180, miR-155, miR-496, miR-203, miR-193b, miR-135b (Table 2). With the 8-miRNA prognostic signature, a risk scoring system for prediction of prognosis was formulated as follow:

$\displaystyle\text{Risk score}=(-1.167)*\text{Exp miR-1301}+(-0.369)*\text{Exp% miR-598}+0.529*\text{Exp miR-1180}+0.344*\text{Exp miR-155}+(-0.417)*\text{% Exp miR-469}+0.166*\text{Exp miR-203}+0.183*\text{Exp miR-193b}+(-0.151)*\text% {Exp miR-135b}$

Table 2
Eight signature miRNAs in the risk scoring system

miRNA	Beta	HR	Lower 95%	Upper 95%	$P$ -value
miR-1301	$-$ 1.1668	0.3114	0.1644	0.5895	0.00003
miR-598	$-$ 0.3693	0.6912	0.4962	0.9629	0.0029
miR-1180	0.5293	1.6978	1.0358	2.7829	0.0036
miR-155	0.3439	1.4105	0.9848	2.0203	0.0061
miR-496	$-$ 0.4167	0.6592	0.4032	1.0779	0.0097
miR-203	0.1660	1.1806	0.9471	1.4717	0.0140
miR-193b	0.1826	1.2004	0.8992	1.6024	0.0215
miR-135b	$-$ 0.1511	0.8598	0.6636	1.1140	0.0253

HR, hazard ratio.

Figure 3.

Kaplan-Meier survival curves for high and low risk patients in different ages. A, Kaplan-Meier curves for patients with age $<$ 65 years. B, Kaplan-Meier curves for patients with age $\geqslant$ 65 years. C, a combined image of A and B. L, low risk; H, high risk.

Risk score calculated for each individual patient in the training set (TCGA dataset) and the median risk score was 0.944. All patients in the training set were thus grouped into high risk patients (risk score $\geqslant$ 0.944, $n=$ 73) and low risk patients (risk score $<$ 0.944, $n=$ 73). As shown in a Kaplan-Meier survival curve (Fig. 1A), survival ratio of low risk patients was markedly different from high risk patients ( $P$ -value $=$ 3.32e-06). Average survival time of the low risk group was markedly longer compared to that of the high risk group (22.29 $\pm$ 17.07 months vs 14.87 $\pm$ 12.28 months). Prognostic value of the risk scoring system was validated in GSE42498 dataset. Similarly, difference of survival ratio between high (risk score $\geqslant$ 1.82) and low risk (risk score $<$ 1.82) patients reached remarkable significance ( $P$ -value $=$ 1.56e-05) in the validate set, and the low risk patients had an obviously longer average survival time than high risk patients(risk score $<$ 1.82) (25.46 $\pm$ 17.93 months vs 14.19 $\pm$ 8.68 months, Fig. 1B).

As shown in Fig. 3, in both TCGA and GSE62498 dataset, expression levels of the 8 prognostic miRNAs were obviously different between the high risk group and the low risk group.

Table 3

Results of Cox regression analyses on clinical characteristics and risk score

Clinical characteristics	Univariable Cox	Multivariable Cox
	$P$ -value	$P$ -value
Age (Above/Below median)	0.078
Gender (Male/Female)	0.563
Pathologic_M (M0/M1)	0.302
Alcohol (Yes/No)	0.819
Tobacco (Never/Reform/	0.149
Current)
Chronic pancreatitis history	0.699
(Yes/No)
Diabetes history (Yes/No)	0.683
Pathologic_N (N0/N1)	0.013	0.207
Pathologic_T (T1/T2/T3/T4)	0.034	0.566
Pathologic_stage (I/II/III/IV)	0.040	0.758
Radiation therapy (Yes/No)	0.037	0.388
New tumor (Yes/No)	0.011	0.041
Targeted molecular therapy	0.001	0.0002
(Yes/No)
Risk score	0.00001	0.0002

Table 4

Univariate Cox regression analysis for each signature miRNA expression in patients stratified by clinical features

Clinical characteristics	miR-1301	miR-598	miR-1180	miR-155	miR-496	miR-203	miR-193b	miR-135b
Age (Above VS. Below 65)	0.758	0.348	0.654	0.098	0.679	0.027	0.350	0.214
Gender (Male VS. Female)	0.944	0.497	0.382	0.615	0.293	0.898	0.631	0.231
Pathologic_M (M0 VS. M1)	0.101	0.576	0.078	0.038	0.004	0.808	0.117	0.706
Pathologic_N (N0 VS. N1)	0.077	0.302	0.272	0.038	0.715	0.228	0.081	0.575
Pathologic_T (T1 VS. T2 VS. T3 VS. T4)	0.365	0.603	0.856	0.176	0.237	0.050	0.012	0.007
Pathologic_stage (I VS. II VS. III VS. IV)	0.089	0.106	0.323	0.044	0.089	0.062	0.010	0.007
Alcohol (Yes VS. No)	0.421	0.759	0.810	0.289	0.567	0.437	0.664	0.298
Tobacco (Never VS. Reform VS. Current)	0.598	0.392	0.559	0.950	0.959	0.972	0.447	0.552
Chronic pancreatitis history (Yes VS. No)	0.434	0.214	0.428	0.192	0.751	0.187	0.662	0.647
Diabetes history (Yes VS. No)	0.293	0.66	0.629	0.269	0.187	0.759	0.306	0.875
New tumor (Yes VS. No)	0.026	0.100	0.088	0.094	0.138	0.009	0.015	0.137
Radiation therapy (Yes VS. No)	0.137	0.785	0.112	0.158	0.591	0.236	0.516	0.934
Targeted molecular therapy (Yes VS. No)	0.210	0.511	0.001	0.313	0.331	0.189	0.058	0.989

Table 5

Stratified Cox regression analyses for survival ratio between high and low risk patients among different subgroups

Clinical characteristics	$P$ -value
Age ( $>=$ 65, $N=$ 75)	0.005
Age ( $<$ 65, $N=$ 76)	0.001
Gender (Male, $N=$ 85)	0.182
Gender (Female, $N=$ 64)	0.298
Chronic pancreatitis history (Yes, $N=$ 13)	0.202
Chronic pancreatitis history (No, $N=$ 109)	0.000
Diabetes history (Yes, $N=$ 32)	0.238
Diabetes history (No, $N=$ 94)	0.000003
Alcohol (Yes, $N=$ 90)	0.00006
Alcohol (No, $N=$ 51)	0.085
Tobacco (Never, $N=$ 54)	0.092
Tobacco (Reform, $N=$ 53)	0.158
Tobacco (Current, $N=$ 17)	0.052
Pathologic_M (M0, $N=$ 68)	0.081
Pathologic_M (M1, $N=$ 2)	–
Pathologic_N (N0, $N=$ 40)	0.027
Pathologic_N (N1, $N=$ 105)	0.0002
Pathologic_T (T1 $+$ T2, $N=$ 24)	0.051
Pathologic_T (T3 $+$ T4, $N=$ 124)	0.205
Pathologic_stage (I $+$ II, $N=$ 143)	0.163
Pathologic_stage (III $+$ Iv, $N=$ 5)	0.064
Radiation therapy (Yes, $N=$ 37)	0.130
Radiation therapy (No, $N=$ 102)	0.063
Targeted molecular therapy (Yes, $N=$ 98)	0.001
Targeted molecular therapy (No, $N=$ 45)	0.007
New tumor (Yes, $N=$ 54)	0.034
New tumor (No, $N=$ 87)	0.0001

3.3 Correlations of clinical characteristics and risk score with prognosis of patients

Upon univariate and multivariate Cox regression analyses, the study found that risk score, new tumor and targeted molecular therapy were independent predictors of prognosis of pancreatic cancer patients (Table 3).

Univariate Cox regression analysis was performed on correlations of each signature miRNA expression with each clinical characteristic. Table 4 showed that miR-203 expression was significantly different between patients $\geqslant$ 65 years old and patients $<$ 65 years old ( $P=$ 0.0268). Pathologic_M/N/T and pathologic stage appear to be significantly associated with expression of miR-155, miR-496, miR-193b and miR-135b ( $P<$ 0.05). Differences in expression of miR-1301, miR-203 and miR-193b achieved significance between patients with or without new tumor ( $P<$ 0.05). In addition, miR-1180 expression was remarkably linked to targeted molecular therapy ( $P=$ 0.00126).

As shown in Table 5, patients in both training and validation sets were stratified according to each clinical characteristic, separately. Stratified cox regression analyses found that survival ratio was significantly different between the high risk patients and the low risk patients regardless of age, pathologic_N, targeted molecular therapy, and new tumor ( $P$ -value $<$ 0.05). Besides, differences of survival ratio were significant in patients who did not have chronic pancreatitis history or diabetes history, or drink alcohol ( $P<$ 0.05). Moreover, Kaplan-Meier survival analysis was conducted for patients stratified by age (Fig. 3), chronic pancreatitis history (Fig. 4), diabetes history (Fig. 5), alcohol (Fig. 6), pathologic_N (Fig. 7), targeted molecular therapy (Fig. 8), and new tumor (Fig. 9), separately.

Figure 4.

Kaplan-Meier survival curves for patients stratified by chronic pancreatitis history. A, Kaplan-Meier curves for the patients without chronic pancreatitis history. B, Kaplan-Meier curves in the patients with chronic pancreatitis history. C, a combined image of A and B. L, low risk; H, high risk.

Figure 5.

Kaplan-Meier survival curves for patients stratified by diabetes history. A, Kaplan-Meier curves for the patients without diabetes history. B, Kaplan-Meier curves for the patients with diabetes history. C, a combined image of A and B. L, low risk; H, high risk.

Figure 6.

Kaplan-Meier survival curves for patients stratified by alcohol. A, Kaplan-Meier curves for the patients not drinking alcohol. B, Kaplan-Meier curves in the patients drinking alcohol. C, a combined image of A and B. L, low risk; H, high risk.

Figure 7.

Kaplan-Meier survival curves for patients stratified by pathologic_N. A, Kaplan-Meier curves for pathologic N0 patients. B, Kaplan-Meier curves for pathologic N1 patients. C, a combined image of A and B. L, low risk; H, high risk.

Figure 8.

Kaplan-Meier survival curves for patients stratified by targeted molecular therapy. A, Kaplan-Meier curves for patients not receiving targeted molecular therapy. B, Kaplan-Meier curves for patients receiving targeted molecular therapy. C, a combined image of A and B. L, low risk; H, high risk.

Figure 9.

Kaplan-Meier survival curves for patients stratified by new tumor. A, Kaplan-Meier curves for patients without new tumor. B, Kaplan-Meier curves for patients with new tumor. C, a combined image of A and B. L, low risk; H, high risk.

Moreover, stratified analyses found that survival ratio was significantly different (Table 6) between low risk patients with or without new tumor ( $P$ -value $=$ 0.00167), between high risk patients with or without radiation therapy ( $P$ -value $=$ 0.0324), and between high risk patients with or without targeted molecular therapy ( $P$ -value $=$ 7.72E-04). Furthermore, Kaplan-Meier survival curves for high and low risk patients stratified by targeted molecular therapy, radiation therapy or new tumor were exhibited in Figs 9–11, separately. These results suggest that targeted molecular therapy and new tumor are associated with survival of patients.

Besides a heatmap displaying the expression of the 6 prognostic miRNAs, distribution of risk score and survival time of each patient in the training set and the validate set was shown in Fig. 6.

Table 6

Stratified cox regression analyses for survival ratio in high and low risk patients

Clinical characteristics	Low risk	High risk
Age (Above/Below 65)	0.209	0.476
Gender (Male/Female)	0.774	0.498
Pathologic_M (M0/M1)	0.304	0.284
Pathologic_N (N0/N1)	0.0646	0.159
Pathologic_T (T1/T2/T3/T4)	0.083	0.493
Pathologic_stage (I/II/III/IV)	0.121	0.449
New tumor (Yes/No)	0.002	0.808
Tobacco (Never/Reform/Current)	0.192	0.836
Alcohol (Yes/No)	0.978	0.649
Chronic pancreatitis history (Yes/No)	0.306	0.755
Diabetes history (Yes/No)	0.799	0.080
Radiation therapy (Yes/No)	0.617	0.032
Targeted molecular therapy (Yes/No)	0.301	0.001

3.4 Functional annotation of DEGs

The study identified a total of 143 DEGs between high risk and low risk patients in the training set (FDR $<$ 0.05). Furthermore, of the 143 DEGs identified, 48 were positively associated with risk score whereas 98 were negatively correlated with risk score. The DEGs were ranked according to their correlation coefficients, and thus top 20 positive and negative DEGs were selected. As shown in Fig. 7A, expression of the top 20 positive and negative DEGs were displayed in a heatmap. On GO functional annotation analysis (Table 7), down DEGs were related to negative regulation of cell proliferation, metal ion transport, cation transport, regulation of transforming growth factor and other biological processes. Up DEGs were significantly associated with neurological system process, cognition, sensory perception and other biological processes (Fig. 7B). Moreover, KEGG pathway enrichment analysis showed that the DEGs were significantly linked to Mitogen-activated protein kinase (MAPK), ribosome, olfactory transduction, hypertrophic cardiomyopathy and other signaling pathways (Fig. 7C, Table 7).

4. Discussion

Pancreatic cancer ranks fourth in the list cancer-related deaths in the western world with a poor prognosis [19]. The current study defined a panel of 8 signature miRNAs using data from TCGA, and established a risk scoring system based on the 8-miRNA signature, which could strongly predict the survival of patients with pancreatic adenocarcinoma. Prognostic value of the risk scoring system was validated on a GEO dataset. Moreover, multivariate Cox regression analysis identfied risk score as an independent risk predictor of prognosis.

Figure 10.

Kaplan-Meier survival curves for high or low risk patients stratified by targeted molecular therapy. A, Kaplan-Meier curves for low risk patients stratified by targeted new therapy. B, Kaplan-Meier curves for high risk patients stratified by targeted new therapy. C, a combined image of A and B. L, low risk; H, high risk.

Figure 11.

Kaplan-Meier survival curves for high or low risk patients stratified by new tumor. A, Kaplan-Meier curves for low risk patients stratified by new tumor. B, Kaplan-Meier curves for high risk patients stratified by new tumor. C, a combined image of A and B. L, low risk; H, high risk.

Figure 12.

Kaplan-Meier survival curves for high or low risk patients stratified by radiation therapy. A, Kaplan-Meier curves for low risk patients stratified by radiation therapy. B, Kaplan-Meier curves for high risk patients stratified by radiation therapy. C, a combined image of A and B. L, low risk; H, high risk.

Table 7

Significant GO terms and KEGG pathways

GO terms for positive DEGs	Gene count	$P$ -value	Genes
Cognition	7	0.016	DFNB31, KLK8, OR56B4, KRT12, OR1N2, OR5H6, CALB1
Neurological system process	8	0.018	DFNB31, KLK8, OR56B4, KRT12, OR1N2, OR5H6, PDYN,CALB1
Sensory perception	5	0.021	DFNB31, OR56B4, KRT12, OR1N2, OR5H6
Synaptic transmission	3	0.038	DFNB31, KLK8, PDYN
Cartilage development	2	0.042	MSX1, PKD1
Transmission of nerve impulse	3	0.049	DFNB31, KLK8, PDYN
Neuropeptide signaling pathway	2	0.049	PKD1, PDYN
GO terms for negative DEGs	Gene count	$P$ -value	Genes
Regulation of transforming growth factor beta receptor	3	0.014	MEN1, CHST11, ZEB1
Metal ion transport	7	0.019	ABCC9, KCNA2, SLC39A8, ATP1A2, CACNA2D3,SLC30A10, KCNA7
Negative regulation of cell proliferation	6	0.025	MEN1, PTPRM, LBX1, NDN, MTBP, ZEB1
Embryonic skeletal system morphogenesis	3	0.028	MEN1, CHST11, ZEB1
Potassium ion transport	4	0.037	ABCC9, KCNA2, ATP1A2, KCNA7
Cation transport	7	0.040	ABCC9, KCNA2, SLC39A8, ATP1A2, CACNA2D3,SLC30A10, KCNA7
Embryonic skeletal system development	3	0.048	MEN1, CHST11, ZEB1
KEGG pathways for all DEGs	Gene count	$P$ -value	Genes
Cardiac muscle contraction	2	0.030	ATP1A2, CACNA2D3
Hypertrophic cardiomyopathy (HCM)	2	0.032	PRKAG3, CACNA2D3
Ribosome	2	0.033	RPS3A, RPL8
Olfactory transduction	3	0.045	OR56B4, OR1N2, OR5H6
MAPK signaling pathway	2	0.041	RELB, CACNA2D3

GO, gene ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

The study suggest that the 8 prognostic miRNAs, including miR-1301, miR-598, miR-1180, miR-155, miR-496, miR-203, miR-193b and miR-135b were significantly associated with prognosis of pancreatic adenocarcinoma. Unsurprisingly, previous researchers have found that some of the 8 miRNAs are involved in pancreatic cancer.

Figure 13.

Risk score (upper), survival time (middle) and expression of 8 signature miRNAs (lower) in the training set (A) and the validation set (B). Red points stand for dead specimens; black points stand for alive specimens.

Figure 14.

Functional annotation of DEGs. A, a heat map of top 20 DEGs. Vertical axis represents genes, and horizontal axis represents samples with low to high risk. B, GO terms enriched with negative (upper pie chart) or positive DEGs (lower pie chart). C, significant signaling pathways. In a pie chart, each sector corresponds to the number of genes enriched with each GO term or pathway.

For instance, it has long been reported that miR-155 is overexpressed in pancreatic cancer [20]. Besides, miR-155 plays a part in regulating pancreatic cancer invasion and migration [21], and promotes cellular oxidative stress [22]. Moreover, Liu et al. provide evidence that miR-155 is related to prognosis of pancreatic cancer [23], which is in accordance with the finding of the present study. Similarly, Greither et al. have reported that up-regulated expression of miR-155 and miR-203 is correlated with poor survival of patients with pancreatic tumor [24]. Furthermore, increasing studies have found that miR-203 is involved in modulating tumor cells proliferation, apoptosis, cell cycle progression, migration and invasion in pancreatic cancers [25, 26]. Previous studies have established that miR-193b plays a role in tumor growth and metastasis in pancreatic cancer [27, 28]. Munding et al. conducted global microRNA expression profiling of tumor tissues and suggest that miR-135b is a promising biomarker for pancreatic ductal adenocarcinoma [29]. Nonetheless, few previous studies have reported the associations of miR-1301, miR-598, miR-1180 and miR-496 with prognosis of pancreatic adenocarcinoma.

In the present study, new tumor and targeted molecular therapy were defined as independent predictors of prognosis. Similarly, results of stratified cox regression analyses showed that targeted molecular therapy and new tumor are associated with survival of patients. In line with these findings, recent studies have established that individualized molecular therapy is particular pertinent for pancreatic cancer [30, 31]. Furthermore, the study found that presence of new tumor was significantly associated with miR-1301, miR-203 and miR-193b expression, while targeted molecular therapy was significantly related to miR-1180 expression. These observations indicate that new tumor and targeted molecular therapy have a bearing on survival of patients with pancreatic adenocarcinoma partly through regulating miR-1301, miR-203 and miR-193b expression.

In the present study, DEGs between high and low risk patients were significantly enriched for regulation of transforming growth factor beta receptor, potassium ion transport and MAPK signaling pathway, suggesting that transforming growth factor beta receptor, potassium ion transport and MAPK signaling pathway may play a role in prognosis. Previous studies have reported that expression of transforming growth factor beta receptor correlates with survival of patients with pancreatic cancer [32] and participate in regulating pancreatic cancer cell metastasis [33]. Recently, Sauter et al. reports that two-pore potassium channels are key drivers of tumor progression and thus recommended as novel therapeutic target in pancreatic cancer [34]. A rich body of evidence have proved that MAPK pathway is implicated in pancreatic cancer growth, invasiveness and progression [34, 35]. Noticeably, in a study by Ikeda et al. [36]. MAPK signaling pathway activation induced deregulated expression of 183 miRNAs, of which miR193b was most remarkably inhibited. It indicates that MAPK activation may be partly responsible for the aberrant expression of miRNAs in pancreatic cancer. MAPK pathway might affect prognosis by regulating expression of miRNAs.

There are some limitations in the study. Firstly, its sample size is limited. Further studies with a larger number of patients are warranted to verify the prognostic value of the risk scoring system. Secondly, this study does not include experiments that are needed to validate the results of the study extracted from analysis of miRNA expression profiles from TCGA. Further validation experiment would provide exact information about pathogenesis of pancreatic adenocarcinoma.

5. Conclusions

This study suggests a risk scoring system based on 8-miRNA signature for assessment of prognosis of patients with pancreatic adenocarcinoma. New tumor and targeted molecular therapy were defined as independent predictors of prognosis. Transforming growth factor beta receptor, potassium ion transport and MAPK signaling pathway may be associated with prognosis in pancreatic adenocarcinoma. This study proposes a novel method to predict the survival of patients, which would pave a way for the determination of a suitable individualized therapy for patients. It is necessary to establish the clinical application of this risk scoring system for pancreatic adenocarcinoma.

Footnotes

Conflict of interest

None.

References

Siegel

R.L.

Miller

K.D.

and Jemal

, Cancer statistics, 2016, Ca A Cancer Journal for Clinicians 66 (2016), 7.

Stewart

and Wild

, World Cancer Report 2014, International Agency for Research on Cancer, 2014.

Bondsmith

Banga

Hammond

T.M.

and Imber

C.J.

, Pancreatic adenocarcinoma, BMJ 344 (2012), e2476.

Cheng

, Circulating miRNAs: Roles in cancer diagnosis, prognosis and therapy, Advanced Drug Delivery Reviews 81 (2015), 75–93.

Croce

C.M.

and Calin

G.A.

, miRNAs, cancer, and stem cell division, Cell 122 (2005), 6–7.

Lliu

Chen

Yao

Shen

Wang

Zhuang

Ning

Zhang

Yuan

Zen

and Zhang

C.Y.

, Serum microRNA expression profile as a biomarker in the diagnosis and prognosis of pancreatic cancer, Clinical Chemistry 58 (2012), 610–618.

Schultz

N.A.

Dehlendorff

Jensen

B.V.

Bjerregaard

J.K.

Nielsen

K.R.

Bojesen

S.E.

Dan

Nielsen

S.E.

Yilmaz

and Holländer

N.H.

, MicroRNA Biomarkers for Detection of Pancreatic Cancer, Jama The Journal of the American Medical Association 311 (2014), 392–404.

Ganepola

G.A.

Rutledge

J.R.

Suman

Yiengpruksawan

and Chang

D.H.

, Novel blood-based microRNA biomarker panel for early diagnosis of pancreatic cancer, World Journal of Gastrointestinal Oncology 6 (2014), 22.

Frampton

A.E.

Krell

Jamieson

N.B.

Gall

T.M.H.

Giovannetti

Funel

Prado

M.M.

Krell

Habib

N.A.

and Castellano

, microRNAs with prognostic significance in pancreatic ductal adenocarcinoma: A meta-analysis, European Journal of Cancer 51 (2015), 1389–1404.

10.

Zhou

Huang

Zhu

Zhang

Wang

Zhu

and Shu

, A panel of 13-miRNA signature as a potential biomarker for predicting survival in pancreatic cancer, Oncotarget 7 (2016), 69616–69624.

11.

Yang

Wang

Schetter

Tang

Funamizu

Yanaga

Uwagawa

Satoskar

A.R.

and Gaedcke

, A novel MIF signaling pathway drives the malignant character of pancreatic cancer by targeting NR3C2, Cancer Research 76 (2016), 3838.

12.

Smyth

, Limma: Linear models for microarray data, Bioinformatics and Computational Biology Solutions Using R and Bioconductor (2005), 397–420.

13.

Wang

Hang

Zou

and Mao

J.H.

, A novel gene expression-based prognostic scoring system to predict survival in gastric cancer, Oncotarget 7 (2016), 55343–55351.

14.

Zhao

and Sun

, Cox survival analysis of microarray gene expression data using correlation principal component regression, Statistical Applications in Genetics and Molecular Biology 6 (2007), Article16.

15.

Ahmed

Brennan

Eppig

Price

C.C.

Lamar

Delanowood

Bangen

K.J.

Edmonds

E.C.

Clark

and Nation

D.A.

, Visuoconstructional impairment in subtypes of mild cognitive impairment, Alzheimers and Dementia 23 (2015), 43.

16.

Camon

Magrane

Barrell

Lee

Dimmer

Maslen

Binns

Harte

Lopez

and Apweiler

, The gene ontology annotation (goa) database: Sharing knowledge in uniprot with gene ontology, Nucleic Acids Research 32 (2004), D262–D266.

17.

Kanehisa

Sato

Kawashima

Furumichi

and Tanabe

, KEGG as a reference resource for gene and protein annotation, Nucleic Acids Research 44 (2016), D457–D462.

18.

Dennis

Jr. Sherman

B.T.

Hosack

D.A.

Yang

Gao

Lane

H.C.

and Lempicki

R.A.

, DAVID: Database for annotation, visualization, and integrated discovery, Genome Biol 4 (2003), P3.

19.

Ferlay

Steliarova-Foucher

Lortet-Tieulent

Rosso

Coebergh

Comber

Forman

and Bray

, Cancer incidence and mortality patterns in Europe: Estimates for 40 countries in 2012, Eur J Cancer (2013).

20.

Lee

E.J.

Gusev

Jiang

Nuovo

G.J.

Lerner

M.R.

Frankel

W.L.

Morgan

D.L.

Postier

R.G.

Brackett

D.J.

and Schmittgen

T.D.

, Expression profiling identifies microRNA signature in pancreatic cancer, International Journal of Cancer 120 (2007), 1046.

21.

Huang

Jiang

and Qiu

, Regulation of miR-155 affects pancreatic cancer cell invasiveness and migration by modulating the STAT3 signaling pathway through SOCS1, Oncology Reports 30 (2013), 1223–30.

22.

Wang

Zhu

Chen

Song

Zeng

Yang

Wen

and Chiao

P.J.

, Micro-RNA-155 is induced by K-Ras oncogenic signal and promotes ROS stress in pancreatic cancer, Oncotarget 6 (2015), 21148–21158.

23.

Liu

W.J.

Zhao

Y.P.

Zhang

T.P.

Zhou

Cui

Q.C.

Zhou

W.X.

You

Chen

and Shu

, MLH1 as a direct target of MiR-155 and a potential predictor of favorable prognosis in pancreatic cancer, Journal of Gastrointestinal Surgery 17 (2013), 1399–1405.

24.

Greither

Grochola

L.F.

Udelnow

Lautenschläger

Würl

and Taubert

, Elevated expression of microRNAs 155, 203, 210 and 222 in pancreatic tumors is associated with poorer survival, International Journal of Cancer Journal International Du Cancer 126 (2010), 73–80.

25.

Wang

and Xu

, MiR203 regulates the proliferation, apoptosis and cell cycle progression of pancreatic cancer cells by targeting Survivin, Mol Med Rep 8 (2013), 379–84.

26.

Miao

Xiong

Lin

Cheng

Jiong

L.U.

Zhang

and Cheng

, miR-203 inhibits tumor cell migration and invasion via caveolin-1 in pancreatic cancer cells, Oncology Letters 7 (2014), 658.

27.

Jin

Sun

Yang

Chang

and Chen

, Deregulation of the miR-193b-KRAS axis contributes to impaired cell growth in pancreatic cancer, Plos One 10 (2015), e0125515.

28.

Kong

Song

and Sun

, miR-193b directly targets STMN1 and uPA genes and suppresses tumor growth and metastasis in pancreatic cancer, Molecular Medicine Reports 10 (2014), 2613.

29.

Munding

J.B.

Adai

A.T.

Maghnouj

Urbanik

Zöllner

Liffers

S.T.

Chromik

A.M.

Uhl

Szafranskaschwarzbach

A.E.

and Tannapfel

, Global microRNA expression profiling of microdissected tissues identifies miR-135b as a novel biomarker for pancreatic ductal adenocarcinoma, International Journal of Cancer 131 (2012), E86.

30.

Chantrill

L.A.

Nagrial

A.M.

Watson

Johns

A.L.

Martyn-Smith

Simpson

Mead

Jones

M.D.

Samra

J.S.

and Gill

A.J.

, Precision medicine for advanced pancreas cancer: The individualized molecular pancreatic cancer therapy (IMPaCT) trial, Clinical Cancer Research An Official Journal of the American Association for Cancer Research 21 (2015), 2029.

31.

Garrido-Laguna

and Hidalgo

, Pancreatic cancer: From state-of-the-art treatments to promising novel therapies, Nature Reviews Clinical Oncology 12 (2015), 319–34.

32.

Friess

Yamanaka

Büchler

Ebert

Beger

H.G.

Gold

L.I.

and Korc

, Enhanced expression of transforming growth factor β isoforms in pancreatic cancer correlates with decreased survival, Gastroenterology 105 (1993), 1846–56.

33.

Rowlandgoldsmith

M.A.

Maruyama

Matsuda

Idezawa

Ralli

and Korc

, Soluble type II transforming growth factor-beta receptor attenuates expression of metastasis-associated genes and suppresses pancreatic cancer cell metastasis, Molecular Cancer Therapeutics 1 (2002), 161.

34.

Sauter

D.R.P.

Sørensen

C.E.

Rapedius

Brüggemann

and Novak

, pH-sensitive K + channel TREK-1 is a novel target in pancreatic cancer, Biochimica Et Biophysica Acta 1862 (2016), 1994–2003.

35.

Zhang

Yan

Collins

M.A.

Bednar

Rakshit

Zetter

B.R.

Stanger

B.Z.

Chung

Rhim

A.D.

and di Magliano

M.P.

, Interleukin-6 is required for pancreatic cancer progression by promoting MAPK signaling activation and oxidative stress resistance, Cancer Research 73 (2013), 6359–6374.

36.

Ikeda

Tanji

Makino

Kawata

and Furukawa

, MicroRNAs associated with mitogen-activated protein kinase in human pancreatic cancer, Molecular Cancer Research 10 (2012), 259–269.

An eight-miRNA signature expression-based risk scoring system for prediction of survival in pancreatic adenocarcinoma

Abstract

PURPOSE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Materials and methods

2.1 Datasets

Table 1 Baseline and clinical characteristics of patients in TCGA and GSE26498 datasets

2.3 Prognosis-related miRNA screening

2.4 Risk scoring system

2.5 Analysis of correlations of patients’ clinical characteristics with survival

2.6 Stratified analyses

2.7 GO and KEGG enrichment analyses

3. Results

3.1 Identification of differentially expressed miRNAs

Table 2 Eight signature miRNAs in the risk scoring system

4. Discussion

Footnotes

Conflict of interest

References

Table 1
Baseline and clinical characteristics of patients in TCGA and GSE26498 datasets

Table 2
Eight signature miRNAs in the risk scoring system