Integrated transcriptomic analysis reveals hub genes involved in diagnosis and prognosis of pancreatic cancer
The hunt for the molecular markers with specificity and sensitivity has been a hot area for the tumor treatment. Due to the poor diagnosis and prognosis of pancreatic cancer (PC), the excision rate is often low, which makes it more urgent to find the ideal tumor markers.
Robust Rank Aggreg (RRA) methods was firstly applied to identify the differentially expressed genes (DEGs) between PC tissues and normal tissues from GSE28735, GSE15471, GSE16515, and GSE101448. Among these DEGs, the highly correlated genes were clustered using WGCNA analysis. The co-expression networks and molecular complex detection (MCODE) Cytoscape app were then performed to find the sub-clusters and confirm 35 candidate genes. For these genes, least absolute shrinkage and selection operator (lasso) regression model was applied and validated to build a diagnostic risk score model. Cox proportional hazard regression analysis was used and validated to build a prognostic model.
Based on integrated transcriptomic analysis, we identified a 19 gene module (SYCN, PNLIPRP1, CAP2, GNMT, MAT1A, ABAT, GPT2, ADHFE1, PHGDH, PSAT1, ERP27, PDIA2, MT1H, COMP, COL5A2, FN1, COL1A2, FAP and POSTN) as a specific predictive signature for the diagnosis of PC. Based on the two consideration, accuracy and feasibility, we simplified the diagnostic risk model as a four-gene model: 0.3034*log2(MAT1A)-0.1526*log2(MT1H) + 0.4645*log2(FN1) -0.2244*log2(FAP), log2(gene count). Besides, a four-hub gene module was also identified as prognostic model = − 1.400*log2(CEL) + 1.321*log2(CPA1) + 0.454*log2(POSTN) + 1.011*log2(PM20D1), log2(gene count).
Integrated transcriptomic analysis identifies two four-hub gene modules as specific predictive signatures for the diagnosis and prognosis of PC, which may bring new sight for the clinical practice of PC.
KeywordsPancreatic cancer Diagnosis Prognosis Integrated transcriptomic analysis
Area under the curve
Differentially expressed genes
GENE Expression Omnibus
Kyoto Encyclopedia of Genes and Genomes
Least absolute shrinkage and selection operator
Molecular complex detection
Receiver operating characteristic
Robust Rank Aggreg
Serial analysis of gene expression
Cancer Genome Atlas Cancer Genome
Weighted gene co-expression network analysis
Pancreatic cancer (PC) is a common malignant tumor of digestive system and ranks the fourth leading cause of cancer-related death worldwide (Kamisawa et al., 2016). The prognosis of PC is grim, with patients’ displaying the 5-year survival rate of only 8% (Siegel et al., 2016). The high mortality of PC patients mainly attributes to the inability to diagnose the disease early and the cancer being highly resistant to treatment (Ryan et al., 2014). Though recent advances in the diagnosis of PC have being evaluated, PC patients are often diagnosed at a advanced stage, due to non-specific clinical symptoms, the lack of truly effective conventional imageological examinations that will identify early stage, and the absence of specific and sensitive diagnostic biomarkers (Ryan et al., 2014). Hence, it is exceptionally urgent to establish novel diagnostic molecular markers for PC (Resovi et al., 2018; Tempero et al., 2013). In addition, a part of patients diagnosed at early stage also suffer a miserable ending, because of the high grade malignant of PC. It is also necessary to monitor patients at high risk for poor clinical outcome and identify novel prognostic molecular markers as early diagnostic biomarkers.
Technological development largely catalyzed our understanding of cancer genomics. Since the first publication of serial analysis of gene expression (SAGE) technique in 1995 (Velculescu et al., 1995), high-throughput gene expression analysis has revolutionized cancer genetics over the last 15 years (Chibon, 2013). A comprehensive genetic analysis of 24 pancreatic cancers sequenced the coding region of 20, 661 genes and indicated the genetic landscape of PC (Jones et al., 2008). Four frequently mutated genes have been identified in PC, including CDKN2A (p16), SMAD4 (DPC4), and TP53 tumor suppressor genes and KRAS oncogene (Jones et al., 2008). Several candidate cancer genes that alter at low frequency are also identified such as MLL3 and ARID1A (Jones et al., 2008; Balakrishnan et al., 2007). These four frequently mutated genes are well recognized as contributing to the carcinogenesis of PC and regarded as the “driver” genes for this tumor (Iacobuzio-Donahue, 2012), while the diagnostic value of these altered genes for PC need to be further estimated. CA19–9 is the common applied serologic marker for the diagnosis of PC in clinic (Ballehaninna and Chamberlain, 2012). However, CA19–9 has limited performance in detecting early-stage disease (Ballehaninna and Chamberlain, 2012). Hereby, specific and sensitive diagnostic gene models have always been pursued by cancer researchers. But, there has been few gene diagnostic model with high sensitivity and specificity for PC hitherto. The similar predicament has also existed in the study of prognostic biomarkers of PC. Though numerous genes, as ACTN4, LMO2, p16INK4a, have been reported to be involved in the prognosis of PC (Watanabe et al., 2015; Nakata et al., 2009; Gerdes et al., 2002), none of them have been applied in clinic.
In the recent years, gene expression data from the public database, such as the Cancer Genome Atlas Cancer Genome (TCGA), Gene Expression Omnibus (GEO), offer available information with respect to the molecular mechanism and variety of multiple carcinomas, and are of great value to the diagnosis, prediction of progression in these disease (Chibon, 2013). In this paper, Robust Rank Aggreg (RRA) methods were employed to identify the differentially expressed genes (DEGs) from four PC genome expression datasets. Then, bioinformatics method of weighted gene co-expression network analysis (WGCNA) was applied to identify the gene modules with importance. We next performed the co-expression networks and molecular complex detection (MCODE) of Cytoscape app to find the sub-clusters and confirm the hub genes. Finally, two prediction models, involved in the diagnosis and prognosis, were established.
Materials and methods
Collection of pancreatic cancer related genome expression datasets
Characteristics of gene microarray of this study
Zhang et al
Badea et al
Pei et al
Klett et al
Wang et al
After downloading series matrix files of GSE28735, GSE15471, GSE16515, GSE101448 from GEO, we normalized samples of each matrix files by “normalizeBetweenArrays” (Additional file 1: Figure S1) and identified the initial candidate genes of each dataset by the package “limma” of R (version 3.5.1, http://www.r-project.org/) (Ritchie et al., 2015), setting log2(Fold Change) ≥ 1, adjusted P < 0.05 as standard. The package “impute” was used to complete missing expression data. The DEGs were then identified by R package “Robust Rank Aggreg” and selected to construct a new data frame with log2(Fold Change) ≥ 1, adjusted P < 0.05. RRA method uses a probabilistic model for aggregation to monitor genes that are ranked consistently better than expected under null hypothesis of uncorrelated inputs and allocates a significance score for each gene (Kolde et al., 2012).
WGCNA is a network biology method that is functioned to cluster the highly correlated genes and identify the co-expression modules. The highly correlated genes are used to construct correlation networks, which facilitate gene screening methods that can be used to identify candidate biomarkers (Langfelder and Horvath, 2008). The gene chip of GSE28735 was selected to identify the co-expression modules for having a relatively large number of samples and relatively detailed data of survival index. The cutHeight = 0.8 and minSize = 10 were applied to identify modules.
Gene Ontology (GO) analysis was conducted by the PANTHER classification system for the enrichment analyses (http://pantherdb.org/) (Mi et al., 2019). The statistical test was Fisher’s Exact and False Discovery Rate (FDR) < 0.05 was considered as statistically significant difference. The functional annotation of genes were reflected in cellular component, biological process, and molecular function, three major GO classifications (Ashburner et al., 2000; The Gene Ontology Consortium, 2017). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (Kanehisa et al., 2017) and Reactome pathway (RP) (Fabregat et al., 2018) were performed to analyze related significant pathways. The enrichment analysis was visualized by Graphad Prism 5.0 (La Jolla, CA).
Co-expression network and MCODE analysis
The STRING (https://string-db.org/) was served to identify the pairwise relationships of all genes (Szklarczyk et al., 2017). The co-expression networks of different modules were firstly constructed through the employment of STRING. The cut-off for confidence scores of interactions is 0.4. To further analyze the physical relationships among these distance-related genes, MCODE algorithm was used to select the clusters of the co-expression networks with the default settings: node score cutoff 0.2, K-core: 2 (Bader and Hogue, 2003). The sub-clusters were visualized by Cytoscape (version 3.6.0). The genes in sub-clusters were selected as candidate genes for diagnosis and prognosis analysis of PC.
Construction and validation of diagnostic risk model
The gene expression profile of GSE28735 (n = 90) were served as training cohort to build least absolute shrinkage and selection operator (Lasso) regression model for diagnosis. Lasso regression is a kind of penalized regression method, which identifies regression coefficients for genes to shrink a weighted average of mean squared prediction error for cases (Zhao and Simon, 2010; Cai et al., 2018). The risk score model of Lasso regression was built by the package “LARS” of R (Xiao et al., 2015). To further determine the superiority of risk score model, we assessed the sensitivity and specificity of genes that formed risk model and the risk score model respectively, receiver operating characteristic (ROC) analysis was performed and the area under the curve (AUC) value was calculated to compare the prognostic performance. Besides, the gene expression profile of GSE16515 (n = 52) was used as validation cohort to verify the ability of diagnostic risk score model. Since the optimum model contained too many mRNAs to diagnose PC, we deleted mRNAs with low weight step by step and then rebuilt lasso regression model. We set AUC > 0.90 as the cut-off value to get the simplified diagnostic risk model with minimum quantity of gene count.
Construction and validation of prognostic risk model
The gene expression profile of GSE78229 (n = 49) was served as training cohort to build prognostic model. Cox proportional hazard regression model was completed by SPSS (version 20.0). Additionally, the gene expression profile of GSE28735 (n = 41) and TCGA (https://cancergenome.nih.gov/) were used as validation cohort to verify the prognostic ability of risk score model. The results were displayed with hazard ratios (HRs) and coefficient with 95% confidence intervals (95%CI). The package “survminer” of R was applied to visualize the survival curves.
Identification of the DEGs by integrated analysis
WGCNA and co-expression analysis
List of 35 key genes in four gene chips
Construction and validation of diagnostic risk model
Construction and validation of prognostic risk model
Application of the public archives is a powerful weapon to understand three fundamental questions of cancers: from exploring cancer biology, to prediction of progression, and treatments to which it will respond (Rung and Brazma, 2013). However, many difficulties are encountered in data collection, analysis and annotation for the rather noisy data from one individual research. Thus, integrating different databases can generate valuable resources and overcome the rather noise from different individual study (Dai et al., 2017). In this study, we adopted an integrated analysis, the RRA method, to select significant DEGs from four independent datasets of PC gene chips, which could provide more convincing research results.
Ultimately, 138 up-regulated and 165 down-regulated DEGs were selected by RRA methods from the four independent datasets of PC. Part of them have been documented to be tumor promoter genes of PC, such as GABRP, CEACAM5, CEACAM6 and CST1 (Takehara et al., 2007; Govindan et al., 2009; Riley et al., 2009; Jiang et al., 2015). Some of them are considered as the tumor suppressor genes of PC, such as PLA2G1B, SERPINI2 and NR5A2 (Goonesekere et al., 2018; Bailey et al., 2016; Murtaugh, 2014). Several genes have been proven to be the prognostic or diagnostic biomarkers of PC, such as LCN2, CLDN18, LAMC2 and SULF1(Bartsch et al., 2018; Ito et al., 2011; Kosanam et al., 2013; Lyu et al., 2018). GO enrichment analysis of up-regulated genes revealed that these significant genes were highly related to extracellular matrix (ECM) regulation, which was consistent with the clinical features of PC: early local invasion/distant metastasis (Wray et al., 2005). Enrichment analysis of down-regulated genes is also indicative of the close connection between these significant genes and another crucial pathway in PC (Halbrook and Lyssiotis, 2017; Michalski et al., 2017). Furthermore, we also discovered many DEGs, whose roles in PC are still ill-defined, such as IFI27, KRT17, COMP and COL8A1. Their functions need to be further researched in PC.
Next, WGCNA and co-expression networks were used to identify the hub genes of PC. The significant modules of WGCNA were involved in ECM regulation, metabolism correction pathways, PI3K-Akt signaling pathway and platelet derived growth factor signaling pathway, which have been widely studied in PC (Vaquero et al., 2003; Stoll et al., 2005; Weissmueller et al., 2014). Prognostic and diagnostic predictive models in multiple cancers could be identified according to the information of clinical indicator, pathological classification and related gene expression (van ’t Veer et al., 2002a; van ’t Veer et al., 2002b). Finally, a set of robust prognostic signatures including SYCN, PNLIPRP1, CAP2, GNMT, MAT1A, ABAT, GPT2, ADHFE1, PHGDH, PSAT1, ERP27, PDIA2, MT1H, COMP, COL5A2, FN1, COL1A2, FAP and POSTN were identified by lasso regression analysis from DEGs and could diagnose the PC. Then, a four-gene simplified diagnostic risk model was built. A four-gene prognostic signature composing CEL, CPA1, POSTN and PM20D1 was established by Cox proportional hazards regression model combined with Kaplan-Meier survival analysis and could predict the overall survival of PC.
Early diagnosis of PC has always been the challenge in cancer field (Keiji Hanada et al. 2017). CA19–9 is perhaps the widely evaluated tumor marker in PC patients, while its universal applicability in the diagnosis of PC was severely limited for the non-specific expression in several benign and malignant diseases (Balakrishnan et al., 2007). A great deal of effort has been made for the early detection of PC, and put forward kinds of diagnostic biomarkers for PC, such as ICAM-1, OPG, TIMP-1 (Brand et al., 2011). However, these biomarkers have also not broken through the dilemma of difficult detection of early PC. Endoscopic ultrasound-guided fine needle aspiration (EUS-FNA) is a new development technique for forecasting the quality of pancreatic neoplasm in recent years (Puli et al., 2013). Previous analysis showed that EUS-FNA displayed a high specificity, but lower sensitivity (Puli et al., 2013). Though EUS-FNA has shortcomings in PC diagnosis, it is an efficient method to obtain tissues of pancreatic neoplasm besides surgery. Hence, building an effective diagnostic risk model based on gene detection in pancreatic neoplasm is of great importance in PC diagnosis and may break the state quo. In this study, we firstly confirmed a 19-gene prognostic model through integrated transcriptomic analysis. Among these genes, SYCN and POSTN have been reported as the diagnostic biomarkers in PC (Makawita et al., 2013; Dong et al., 2018). PNLIPRP1 and GNMT have been documented as the tumor suppressor genes of PC (Goonesekere et al., 2018; Zhang et al., 2013) and PHGDH been certified as the tumor promoter gene of PC (Song et al., 2018). This risk model could accurately diagnose PC in our subsequent verification, while many variables in this model need to be controlled, resulting in limited application of these prognostic signatures in clinic in the future. In addition, this model was based on the DEGs from surgical specimen of PC. It remains unclear whether this model displayed an excellent accurate diagnosis in the tissues from EUS-FNA. The expression of these biomarkers should be detected in a large number of tissues from EUS-FNA to confirm the high specificity and sensitivity of the diagnostic model. Thus, further study is necessary to the clinical application of the model for PC diagnosis.
PC is one of the worst prognosis cancers, making the prognostic biomarkers becoming especially important in PC. Recently, it has been reported that several predicted models for risk estimation, such as S100P, ERO1LB, SULF1, ITGA2, GPRC5A, ACTN4, LMO2, p16INK4a and CLPS (Watanabe et al., 2015; Nakata et al., 2009; Gerdes et al., 2002; Lyu et al., 2018; Zhang et al., 2013; Ji et al., 2014; Zhu et al., 2017; Li et al., 2018; Liu et al., 2018). A study recently identified three genes COL11A1, GJB2 and CTRL as prognostic biomarkers for PC by using integrated whole genome microarray analysis and immunohistochemical assay (Sun et al., 2018). Most existing prognosis models of PC involve only one gene or mRNA, which have their limitations. Because the expression and crosstalk of multiple genes jointly account for the outcome of PC. That’s the reason why we paid far more attention in identifying co-expression networks and hub genes. The prognostic model that we built involves multi-hub genes that interact in different modules and pathways, which improves the specificity and reliability of the model. Testing of serum miRNAs has been a novel method for predicting the outcome of PC patients. Researchers from Nanjing, PR China identified a six-miRNA (miR-19a-3p, miR-192-5p, miR-19b-3p, let-7b-5p, miR-223-3p, and miR-25-3p) signature in the serum for PC early and noninvasive diagnosis (Zou et al., 2019). Besides, a study was to identify a prognostic model that combined the clinical factors-distance from common hepatic artery or superior mesenteric artery and biomarker CA19–9 to predict the outcome, which also indicated that one gene or mRNA could not fully forecast the outcome (Suzuki et al., 2019). Anyway, using clinical factors related models for predicting survival of PC are intuitive. Although a mass of clinical prediction models for PC have been reported, most succumb to bias and have not been validated externally (Strijker et al., 2019).
In this paper, we identified a four-gene prognostic signature for PC, containing CEL, CPA1, POSTN and PM20D1. Compared to the study of Defeng Sun et al., we included more quantified datasets of PC. Poor prognosis of PC may be due to the hallmarks of easy migration and resistance. POSTN has been reported to be related to the resistance and invasion in cancers (Park et al., 2016; Landré et al., 2016). In PC, periostin, encoded by POSTN, could enhance the invasiveness and resistance ability of PC cells via activation of the PI3 kinase pathway (Baril et al., 2007). CPA1 could promote the development of PC via ER stress (Tamura et al., 2018) and CEL has also been reported as the risk factor of PC (Dalva et al., 2017). PM20D1 is related to the metabolism pathway (Long et al., 2016), and it may be involved in cancer via influencing tumor metabolism (data no shown). These previous documents have also highlighted the potential role of the four genes in PC. Here, the results of survival analysis cross-checked the accuracy of this prognostic risk model in different cohort and indicated that these four genes could serve as predictive biomarkers for PC.
Rapid development of technology platforms, free access to many published experimental datasets and different statistic values account for the diversity of methods to treat the same question. RRA method is a rigorous way to integrate their results in an unbiased manner for getting rid of noise and error (Kolde et al., 2012). The candidate genes were obtained from RRA analysis of four independent gene chips with great statistic difference. The molecular biology experiments discussed above indicated the functional role of predictors in cancers, while there is little experimental evidence to demonstrate their role in PC. Biological systems are holistic and complicated. Bioinformatic findings provide theoretical guide for basic experiments. Biostatistical and bioinformatics approaches to biological systems will definitely require experimental validation to define their biological relevance.
We have to admit some limitations of this study. Firstly, a major issue is that we can’t collect enough cases of PC in our own institute due to the characteristic of PC. Secondly, due to the lack of the details on TNM stage, symptoms, complications, metastasis, treatment, etc., we can’t make sure that the diagnostic risk model could be used in any manner to diagnose PC and whether the biomarkers could be further tested as serum markers for surveillance purposes. But the candidate genes were selected from four independent gene profiles and diagnostic and prognostic risk model were both validated in other cohort, which could make up for it slightly. We have already collected specimen of PC in different centre with adequate information and the finding would be further verified not long in the future.
Integrated transcriptomic analysis identifies two four-hub gene modules as specific predictive signatures for the diagnosis and prognosis of PC, respectively. Further study of these hub genes may improve the clinic status of pancreatic cancer therapy.
We appreciate TCGA and GEO project team.
YYZ and LPC searched the public database and collected data. YZ, SKH and ZJD were in charge of the figure preparation. MW and QXC analyzed data. ZZZ and XJD provided ideas, wrote manuscript and submitted paper. All authors read and approved the final manuscript.
This work was supported by Wenzhou Science & Technology Bureau Project (grant number Y20180075), Zhejiang Province Natural Science Fund of China (grant number LY14H030001).
Ethics approval and consent to participate
Consent for publication
The authors declare that they have no competing interests.
- Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, et al. Gene ontology: tool for the unification of biology. Gene Ontol Consortium Nat Genet. 2000;25:25–9.Google Scholar
- Ballehaninna UK, Chamberlain RS. The clinical utility of serum CA 19-9 in the diagnosis, prognosis and management of pancreatic adenocarcinoma: an evidence based appraisal. J Gastrointest Oncol. 2012;3:105–19. https://doi.org/10.3978/j.issn.2078-6891.2011.021.CrossRefPubMedPubMedCentralGoogle Scholar
- Baril P, Gangeswaran R, Mahon PC, Caulee K, Kocher HM, Harada T, et al. Periostin promotes invasiveness and resistance of pancreatic cancer cells to hypoxia-induced cell death: role of the beta4 integrin and the PI3k pathway. Periostin promotes invasiveness and resistance of pancreatic cancer cells to hypoxia-induced cell death: role of the beta4 integrin and the PI3k pathway. Oncogene. 2007;26:2082–94. https://doi.org/10.1038/sj.onc.1210009.CrossRefPubMedGoogle Scholar
- Bartsch DK, Gercke N, Strauch K, Wieboldt R, Matthäi E, Wagner V, et al. The combination of MiRNA-196b, LCN2, and TIMP1 is a potential set of circulating biomarkers for screening individuals at risk for familial pancreatic Cancer. J Clin Med. 2018;7:295. https://doi.org/10.3390/jcm7100295.CrossRefPubMedCentralGoogle Scholar
- Dai M, Chen X, Mo S, Li J, Huang Z, Huang S, et al. Meta-signature LncRNAs serve as novel biomarkers for colorectal cancer: integrated bioinformatics analysis, experimental validation and diagnostic evaluation. Sci Rep. 2017;7:46572. https://doi.org/10.1038/srep46572.CrossRefPubMedPubMedCentralGoogle Scholar
- Dalva M, El Jellas K, Steine SJ, Johansson BB, Ringdal M, Torsvik J, et al. Copy number variants and VNTR length polymorphisms of the carboxyl-ester lipase (CEL) gene as risk factors in pancreatic cancer. Pancreatology. 2017;17:83–8. https://doi.org/10.1016/j.pan.2016.10.006.CrossRefPubMedGoogle Scholar
- Goonesekere NCW, Andersen W, Smith A, Wang X. Identification of genes highly downregulated in pancreatic cancer through a meta-analysis of microarray datasets: implications for discovery of novel tumor-suppressor genes and therapeutic targets. J Cancer Res Clin Oncol. 2018;144:309–20. https://doi.org/10.1007/s00432-017-2558-4.CrossRefPubMedGoogle Scholar
- Govindan SV, Cardillo TM, Moon SJ, Hansen HJ, Goldenberg DM. CEACAM5-targeted therapy of human colonic and pancreatic cancer xenografts with potent labetuzumab-SN-38 immunoconjugates. Clin Cancer Res. 2009;15:6052–61. https://doi.org/10.1158/1078-0432.CCR-09-0586.CrossRefPubMedPubMedCentralGoogle Scholar
- Klett H, Fuellgraf H, Levit-Zerdoun E, Hussung S, Kowar S, Küsters S, et al. Identification and validation of a diagnostic and prognostic multi-gene biomarker panel for pancreatic ductal adenocarcinoma. Front Genet. 2018;9:108. https://doi.org/10.3389/fgene.2018.00108.CrossRefPubMedPubMedCentralGoogle Scholar
- Kosanam H, Prassas I, Chrystoja CC, Soleas I, Chan A, Dimitromanolakis A, et al. Laminin, gamma 2 (LAMC2): a promising new putative pancreatic cancer biomarker identified by proteomic analysis of pancreatic adenocarcinoma tissues. Mol Cell Proteomics. 2013;12:2820–32. https://doi.org/10.1074/mcp.M112.023507.CrossRefPubMedPubMedCentralGoogle Scholar
- Makawita S, Dimitromanolakis A, Soosaipillai A, Soleas I, Chan A, Gallinger S, et al. Validation of four candidate pancreatic cancer serological biomarkers that improve the performance of CA19.9. BMC Cancer. 2013;13:404. https://doi.org/10.1186/1471-2407-13-404.CrossRefPubMedPubMedCentralGoogle Scholar
- Puli SR, Bechtold ML, Buxbaum JL, Eloubeidi MA. How good is endoscopic ultrasound-guided fine-needle aspiration in diagnosing the correct etiology for a solid pancreatic mass?: a meta-analysis and systematic review. Pancreas. 2013;42(1):20–6. https://doi.org/10.1097/MPA.0b013e3182546e79.CrossRefPubMedGoogle Scholar
- Riley CJ, Engelhardt KP, Saldanha JW, Qi W, Cooke LS, Zhu Y, et al. Design and activity of a murine and humanized anti-CEACAM6 single-chain variable fragment in the treatment of pancreatic cancer. Cancer Res. 2009;69:1933–40. https://doi.org/10.1158/0008-5472.CAN-08-2707.CrossRefPubMedPubMedCentralGoogle Scholar
- Suzuki F, Fujiwara Y, Hamura R, Haruki K, Sakamoto T, Shiba H, et al. Combination of Distance from Superior Mesenteric Artery and Serum CA19–9 as a Novel Prediction of Local Recurrence in Patients With Pancreatic Cancer Following Resection. Anticancer Res. 2019;39:1469–78. https://doi.org/10.21873/anticanres.13264.CrossRefPubMedGoogle Scholar
- Zhang G, He P, Tan H, Budhu A, Gaedcke J, Ghadimi BM, et al. Integration of metabolomics and transcriptomics revealed a fatty acid network exerting growth inhibitory effects in human pancreatic cancer. Clin Cancer Res. 2013;19:4983–93. https://doi.org/10.1158/1078-0432.CCR-13-0209.CrossRefPubMedPubMedCentralGoogle Scholar
- Zou X, Wei J, Huang Z, Zhou X, Lu Z, Zhu W, et al. Identification of a six-miRNA panel in serum benefiting pancreatic cancer diagnosis. Cancer Med. 2019. https://doi.org/10.1002/cam4.2145.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.