Pan et al. Cancer Cell Int (2018) 18:214 https://doi.org/10.1186/s12935-018-0718-5 Cancer Cell International

PRIMARY RESEARCH Open Access Analysis of dynamic molecular networks for pancreatic ductal adenocarcinoma progression Zongfu Pan1†, Lu Li2†, Qilu Fang1, Yiwen Zhang1, Xiaoping Hu1, Yangyang Qian3 and Ping Huang1*

Abstract Background: Pancreatic ductal adenocarcinoma (PDAC) is one of the deadliest solid tumors. The rapid progression of PDAC results in an advanced stage of patients when diagnosed. However, the dynamic molecular mechanism underlying PDAC progression remains far from clear. Methods: The microarray GSE62165 containing PDAC staging samples was obtained from Expression Omnibus and the diferentially expressed (DEGs) between normal tissue and PDAC of diferent stages were profled using R software, respectively. The software program Short Time-series Expression Miner was applied to cluster, compare, and visualize diferences between PDAC stages. Then, function annotation and pathway enrichment of DEGs were conducted by Database for Annotation Visualization and Integrated Discovery. Further, the Cytoscape plugin DyNetViewer was applied to construct the dynamic –protein interaction networks and to analyze dif- ferent topological variation of nodes and clusters over time. The phosphosite markers of stage-specifc protein kinases were predicted by PhosphoSitePlus database. Moreover, survival analysis of candidate genes and pathways was per- formed by Kaplan–Meier plotter. Finally, candidate genes were validated by immunohistochemistry in PDAC tissues. Results: Compared with normal tissues, the total DEGs number for each PDAC stage were 994 (stage I), 967 (stage IIa), 965 (stage IIb), 1027 (stage III), 925 (stage IV), respectively. The stage-course gene expression analysis showed that 30 distinct expressional models were clustered. Kyoto Encyclopedia of Genes and Genomes analysis indicated that the up-regulated DEGs were commonly enriched in fve fundamental pathways throughout fve stages, including pathways in cancer, small cell lung cancer, ECM-receptor interaction, amoebiasis, focal adhesion. Except for amoebiasis, these pathways were associated with poor PDAC overall survival. Meanwhile, LAMA3, LAMB3, LAMC2, COL4A1 and FN1 were commonly shared by these fve pathways and were unfavorable factors for prognosis. Furthermore, by constructing the stage-course dynamic protein interaction network, 45 functional molecular modules and 19 nodes were identifed as featured regulators for all PDAC stages, among which the collagen family and integrins were considered as two main regulators for facilitating aggressive progression. Additionally, the clinical relevance analysis suggested that the stage IV featured nodes MLF1IP and ITGB4 were signifcantly correlated with shorter overall survival. Moreover, 15 stage-specifc protein kinases were identifed from the dynamic network and CHEK1 was particularly activated at stage IV. Experimental validation showed that MLF1IP, LAMA3 and LAMB3 were progressively increased from tumor initiation to progression. Conclusions: Our study provided a view for a better understanding of the dynamic landscape of molecular interac- tion networks during PDAC progression and ofered potential targets for therapeutic intervention.

*Correspondence: [email protected] †Zongfu Pan and Lu Li contributed equally to this work 1 Department of Pharmacy, Zhejiang Cancer Hospital, Hangzhou 310022, China Full list of author information is available at the end of the article

© The Author(s) 2018. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creat​iveco​mmons​.org/licen​ses/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://creat​iveco​mmons​.org/ publi​cdoma​in/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Pan et al. Cancer Cell Int (2018) 18:214 Page 2 of 18

Keywords: Dynamic molecular networks, Pancreatic ductal adenocarcinoma, Progression, Bioinformatics

Background Te molecular interactions in tumor are varying with Pancreatic ductal adenocarcinoma (PDAC) is one of the cancer staging. Construction and analysis of dynamic most malignant solid tumors arising within the ductal molecular networks provide a view to understand of the pancreas. Te lack of early diagnosis and its rapid dynamic cellular mechanisms of diferent biological pro- progression resulted in an advanced stage of PDAC cess during cancer progression and ofer opportunities patients when diagnosed. In the past several decades, for therapeutic intervention. In the present study, we many eforts have been taken to unveil the molecu- analyzed the microarray from Gene Expression Omnibus lar pathogenesis of PDAC, and to improve the patient and identifed the diferentially expressed genes between prognosis through various therapeutic strategies. How- normal tissue and PDAC of diferent stages, respectively. ever, limited advances have been made to prolong the Further, we constructed the dynamic protein–protein survival and to reduce the mortality. Te cancer stag- interaction networks and analyzed diferent topological ing is one of the most important factors in determining variation of nodes and clusters over time. Several func- treatment strategies and predicting a patient’s outcome tional molecular modules and genes were identifed as [1]. At present, surgical resection is the standard care key regulators for PDAC development. Trough visual- for only 20% of patients with localized disease, while izing the dynamic networks from early to advanced stage, most patients with surgical management will develop our study aimed at improving our understanding of the cancer recurrence and die within 2 years. Moreover, the underlying mechanism of PDAC progression and pro- median survival of advanced inoperable PDAC patient vided novel insights for precise treatment. with systemic chemotherapy is only about 8 months [2]. Terefore, exploring the molecular mechanism under- Methods lying the progression of PDAC may contribute to the Tissue samples development of precise therapeutics for PDAC patients. Human PDAC tissues with four diferent stages and non- PDAC is a complex disease driven by time and con- tumorous tissues were obtained from fve patients who text dependent alterations of multiple genes. Consid- underwent surgical resection at Zhejiang Cancer Hos- erable advances have been made in demonstrating the pital. All specimens had a pathological diagnosis at the genetic alterations involved in the progression PDAC. time of assessment. Studies were approved by the Ethics Frequent mutations in Kirsten Ras (K-RAS), TP53, Committee of Zhejiang Cancer Hospital. CDKN2A and SMAD4 has been identifed as essential drivers for PDAC development [3]. According to the Microarray data integrated genomic analysis, Bailey et al. [4] revealed Microarray dataset GSE62165 containing staging pan- diferent mechanism of molecular evolution underlying creatic ductal adenocarcinoma (PDAC) and non-tumoral four redefned pancreatic cancer subtypes, which pro- pancreatic tissue was obtained from Gene Expression vides a new insight into potential therapeutic relevance Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) and patient selection. Depending on the spatiotempo- database in the National Center for Biotechnology Infor- ral proteomic analysis of pancreas cancer progression, mation (NCBI), which was deposited by Janky and col- Mirus et al. [5] pointed out that dynamic expression league [7]. Te GSE62165 dataset included 118 surgically pattern of serine/threonine stress kinase 4 were asso- resected PDAC varying from stage I to stage IV and 13 ciated with early tumorigenic events. Additionally, control samples. Te detail PDAC patient cohort con- genome-wide transcriptome analysis by Jones et al. [6] sisted of 8 stage I samples, 30 stage IIa samples, 62 stage showed that more than 21,000 genetic altered in PDAC, IIb samples, 5 stage III samples and 13 stage IV samples. which mainly afected 12 core signaling pathways Additionally, the gene expression profle was detected including apoptosis, DNA damage repair, cell adhe- basing on HG-U219 (Afymetrix U219 sion and invasion. Besides, Janky et al. [7] screened the Array) platform. master regulators of transcription involved in PDAC progression and highlighted the HNF1A/B as a puta- Data processing and screening of diferentially expressed tive tumor suppressor in pancreatic cancer. However, genes (DEGs) despite these important advances, the precise dynamic Te CEL fle data of GSE62165 was read using the afy landscape of molecular interaction networks during package in the R language software (version 3.2.3, PDAC progression is incompletely understood. https​://www.r-proje​ct.org/). Background correction, Pan et al. Cancer Cell Int (2018) 18:214 Page 3 of 18

normalization, expression calculation of the original degree centrality (DC), betweenness centrality (BC), array data and log2 transformation were processed by local average connectivity-based method (LAC) were robust multichip average (RMA) algorithm. Empiri- used. Te top 20 DC, BC, and LAC of genes that uniquely cal Bayes method was used to identify signifcant DEGs elevated at each stage were frstly identifed, respectively. between diferent stages of PDAC and normal pancre- Ten, the common genes among the top 20 DC, BC, and atic samples basing on the limma package in R. P values LAC lists at each stage were considered as key nodes for were adjusted for multiple testing depending on the Ben- dynamic network development. For dynamic module jamini–Hochberg False Discovery Rate (FDR) method. analysis, clustering algorithm molecular complex detec- Te strict thresholds for identifying DEGs were set as tion (MCODE) was conducted for analyzing clusters of FDR < 0.01 and |log2 fold change (FC)| ≥ 2. dynamic networks. Te degree cutof was set as 5 and K-core was set as 2. Analysis of DEGs expression manner during PDAC progression Identifcation of PDAC stage‑specifc activated kinases Te software program Short Time-series Expression and phosphosite markers Miner (STEM) is designed for clustering, comparing, and To identify the stage-specifc activated kinase during visualizing gene expression data from short time series PDAC progression, total protein kinase list was acquired microarray experiments [8]. To search the diferences of from the human kinome database (kinase.com) [14] and gene expression between PDAC stages, the STEM tool was intersected with DEGs. Ten, the dynamic DC, BC was applied to profle candidate genes simultaneously and LAC of intersection kinases were retrieved from the from normal status to stage IV. Briefy, the union of DEGs results of dynamic node centrality analysis described from fve stages was input and normalized by STEM. Te above. Furthermore, the substrates of kinases were maximum number of model profles was set as 100, and searched from PhosphoSitePlus (https​://www.phosp​hosit​ FDR was selected as correction method. Te rest param- e.org) online tool. Te sequence characteristics of kinase eters were remained unchanged. Colored clusters indi- substrates were depicted by sequence logo to indicate the cated statistically signifcant number of genes assigned. phosphosite markers.

Function annotation and pathway enrichment of DEGs Kaplan–Meier survival analysis of key genes in PDAC (GO) and Kyoto Encyclopedia of Genes Te online database Kaplan–Meier plotter (www.kmplo​ and Genomes (KEGG) were applied for the functional t.com) is capable to retrieve gene expression data and annotation and pathway enrichment analysis of DEGs clinical information from GEO, European Genome- through using the Database for Annotation Visualiza- phenome Archive (EGA) and Te Cancer Genome tion and Integrated Discovery (DAVID; https​://david​ Atlas (TCGA) [15]. To evaluate the prognostic value .ncifc​rf.gov/) [9–11]. Adjusted P values were calculated of candidate genes, the patient samples were split into by Benjamini–Hochberg FDR method and the thresholds two cohorts according to the best cutof of gene expres- were set as FDR < 0.05 to indicate a statistically signifcant sion computed by Kaplan–Meier plotter. To analyze diference. the relationship between particular pathways and over- all survival, the mean expression of pathway associ- Dynamic protein–protein interaction (PPI) network ated signature genes and best cutof were calculated by construction and variation analysis of node centrality Kaplan–Meier plotter. Besides, the log rank P value and and cluster hazard ratio (HR) with 95% confdence intervals were Te online database Search Tool for the Retrieval of also computed. Interacting Genes (STRING, http://strin​g-db.org/) is GEPIA is a newly developed interactive web server for commonly used to identify the interactions between estimating the RNA sequencing expression data from the known and predict proteins and to construct TCGA and Genotype-Tissue Expression (GTEx) dataset a PPI network [12]. Te DEGs were input into STRING projects [16]. Candidate genes were queried by GEPIA to construct PPI network and further visualized by to explore their expression levels at diferent stages using Cytoscape (version 3.6.1) software. Te Cytoscape plugin TCGA-pancreatic adenocarcinoma data. DyNetViewer was applied to construct the stage-course dynamic protein interaction network [13]. Te time Immunohistochemistry course protein interaction networks (TC-PIN) algo- Parafn-embedded sections were deparafnized and rithm was chosen to constructing sub-networks and rehydrated frstly. Ten, 1 mM EDTA (pH 8.0) was the threshold was set as 2. For dynamic node central- applied to retrieve the antigen. Endogenous peroxide ity analysis, three typical centrality measures including activity was block by 0.3% hydrogen peroxide. Before Pan et al. Cancer Cell Int (2018) 18:214 Page 4 of 18

incubating with primary antibody, 5% goat serum in TBS ontology (GO) analysis was performed using the DAVID was used to avoid non-specifc binding. Te primary online analysis tool. Basing on the results of GO biologi- antibodies contained rabbit anti-MFL1IP (1:100, Protein- cal process (BP) enrichment (Fig. 2a–f), we found up- tech, USA), rabbit anti-LAMA3 (1:200, Abbkine, China) regulated DEGs of fve stages were commonly enriched and rabbit anti-LAMB3 (1:200, Abbkine, China). After in six biological processes, including cell adhesion, extra- incubating with biotinylated secondary antibodies, sec- cellular matrix organization, immune response, colla- tions were developed using DAB (Beyotime, China) and gen fbril organization, collagen catabolic process, and counterstained with hematoxylin. endodermal cell diferentiation. Moreover, extracellular matrix disassembly was shared by stage I, IIa, IIb, and Results IV. Notably, wound healing was enriched at stage IIa, IIb, Identifcation of DEGs at diferent pancreatic ductal III, and IV. Leukocyte migration was found in stage I, IIa, adenocarcinoma (PDAC) stages and analysis of DEGs and III. Besides, type I interferon signaling pathway was expression patterns commonly enriched by stage I, IIb, IV. Angiogenesis was Te microarray dataset GSE62165 was acquired from particularly found in advanced PDAC including stage III GEO. Diferentially expressed genes (DEGs) (FDR < 0.01, and IV. Interestingly, infammatory response was only |log2 FC| ≥ 2) of diferent PDAC stages were screened found at early stage (stage I). out basing on the R analysis, respectively. Compared with Functional annotation of down-regulated DEGs normal tissues, the total DEGs number for each stage indicated that 8 biological processes were commonly were 994 (stage I), 967 (stage IIa), 965 (stage IIb), 1027 enriched at each PDAC stage (Fig. 2a–e, g), including (stage III), 925 (stage IV). Ten, the heatmaps of the top lipid digestion, cellular response to zinc ion, lipid cata- 10 up and down-regulated DEGs at each stage were hier- bolic process, reactive oxygen species metabolic process, archically clustered and displayed, respectively (Fig. 1a– cobalamin metabolic process, proteolysis, cellular amino e). As the results showed in Fig. 1a–e, these top 20 DEGs acid metabolic process, digestion. Transmembrane trans- could clearly distinguish each PDAC stage from normal port was shared by stage I and IIa, while negative regula- pancreatic tissues. tion of growth was conspicuously found at stage IIb, III, To explore the diferences of gene expression between and IV. Additionally, meiotic gene conversion, cellular PDAC stages, the STEM tool was applied to profle stage- amino acid biosynthetic process, amino acid transport, course gene expression patterns. A total of 30 expression cellular response to cadmium ion, and metabolic process manners were enriched among 100 assumed expression were specifcally enriched at each stage, respectively. models (Fig. 1f). Intriguingly, we found some clusters exhibited stage-specifc expression trends throughout Pathway enrichment of DEGs at diferent stages PDAC progression. Genes in the Model 89 (115 genes Subsequently, KEGG pathway analysis demonstrated enriched) were specifcally up-regulated at stage I. Te that the up-regulated DEGs of diferent PDAC stages Model 6 contained 166 genes and particularly elevated at were commonly enriched in fve key pathways including stage IIa. Te Model 3 (113 genes enriched) and Model pathways in cancer, small cell lung cancer, ECM-recep- 2 (45 genes enriched) were stage IIb- and stage III-spe- tor interaction, amoebiasis, focal adhesion (Fig. 3a–f). cifc gene clusters, respectively. Additionally, the Model 1 Staphylococcus aureus infection was enriched at every associated genes specifcally increased at stage IV. Tese stage except stage IV. Phagosome was shared by stage I, dynamic expression patterns suggested that distinct IIa, III, while rheumatoid arthritis was only found at stage molecular signals were required for particular PDAC I and IIa. Protein digestion and absorption occurred at all stages. Notably, we also found six genes that dramati- stages except stage IIa. PI3K-Akt signaling pathway was cally changed from normal status to stage IV (Fig. 1g–h). enriched at stage I, III, and IV. Tree pathways including TNNT1, the troponin T type 1 encoding gene, was up- p53 signaling pathway, platelet activation, and cell cycle regulated increasingly from cancer initiation to advanced were specifcally enriched at stage IV. stage. Other fve genes, including BSPRY, C8ORF47, Te down-regulated DEGs of diferent PDAC stages FAM3B, HOOK1 and REG1A, were clustered in Model were commonly enriched in fve key pathways includ- 16 and signifcantly decreased during cancer progres- ing metabolic pathways, glycine, serine and threonine sion. Tese six genes may act as key regulators for driving metabolism, pancreatic secretion, protein digestion and PDAC progression via their special expression manners. absorption, fat digestion and absorption (Fig. 3a–e, g). Tree pathways including mineral absorption, bile secre- Gene function annotation of DEGs at diferent stage tion, proximal tubule bicarbonate reclamation were To further investigate the biological function of DEGs found in all PDAC stages except stage IV. Biosynthesis between diferent PDAC stages and normal tissues, gene of amino acids occurred at early stage (stage I, IIa, IIb). Pan et al. Cancer Cell Int (2018) 18:214 Page 5 of 18

stage IIa a stage I b

colorkey COL10A1 2 KRT6A COL11A1 1 SDR16C5 CTHRC1 0 COL11A1 GJB2 −1 COL10A1 COL17A1 −2 GJB2 SERPINB5 CTHRC1 CEACAM5 SLC6A14 S100P ANXA8 KRT17 KRT17 CXCL5 S100P CTRL SYCN CELA2A CELA2A SYCN SERPINI2 CPA1 CELP CPA2 PNLIPRP1 CELA3A AMY1A PNLIPRP1 TMED6 ALB SERPINI2 CELA2B PDIA2 CTRL AQP8 TumorNT TumorNT

c stage IIb d stage III e stage IV

CTHRC1 CLPS COL10A1 COL10A1 CPA1 COL11A1 COL11A1 GP2 CTHRC1 GJB2 CELA2A GJB2 SERPINB5 SYCN S100P S100P SERPINI2 COL17A1 KRT17 PNLIPRP2 SDR16C5 TMPRSS4 ALB AHNAK2 CEACAM5 AQP8 IGHG1 COL17A1 PNLIPRP1 IGHM SYCN GREM1 PNLIPRP1 CELA2A MMP11 PDIA2 CELA3A COMP ALB CUZD1 CTHRC1 SERPINI2 CPA2 FN1 CUZD1 SERPINI2 GJB2 SYCN PNLIPRP1 COL11A1 CELA3A AMY1A COL10A1 CPA2 AQP8 LOC100291682 GP2 CELP EPYC CPA1 TumorNT TumorNT TumorNT f g 10 TNNT1

8

ange 6 ch

ld

Fo 4

2

0 (stage)NT IIIa IIbIII IV

h 1.5 BSPRY C8ORF47 FAM3B 1.0 ange HOOK1 ch

ld REG1A

Fo 0.5

0.0 (stage)NT IIIa IIbIII IV Fig. 1 Identifcation and hierarchical clustering analysis of DEGs at diferent PDAC stages and the profling of DEGs expression patterns. a–e Top 10 up and down-regulated DEGs of diferent stages were hierarchically clustered and displayed by heatmaps, respectively. f Gene expression patterns from normal tissue to PDAC stages were clustered, compared, and visualized by the STEM software. Colored clusters indicated statistically signifcant number of genes enriched. The number in bottom left indicated number of genes assigned. The number in the top left represented model identifer. g–h Classic gene expression trends were plotted from normal tissue (NT) to advanced PDAC stage Pan et al. Cancer Cell Int (2018) 18:214 Page 6 of 18

a b

c d

e fg

Fig. 2 Top ten enriched GO (Biological Process) terms of up-regulated and down-regulated DEGs at diferent stages. a–e Gene function annotation of DEGs at stage I, IIa, IIb, III, and IV. f Venn diagram displayed common biological processes of up-regulated DEGs shared by fve stages. g Venn diagram displayed common biological processes of down-regulated DEGs shared by fve stages

Maturity onset diabetes of the young was shared by stage of pathway associated genes and best cutof were calcu- I, IIb, III, IV. Metabolism of xenobiotics by cytochrome lated by Kaplan–Meier plotter. As shown in Fig. 4a–e, P450 was found in advanced PDAC at stage III and IV. pathways in cancer (P = 0.0047), small cell lung cancer For stage IV, three pathways were particularly enriched, (P = 0.0024), ECM-receptor interaction (P = 0.039) and including chemical carcinogenesis, drug metabolism— focal adhesion (P = 0.048) were negatively associated cytochrome P450, and retinol metabolism. with overall survival (OS). Moreover, the intersection of fve pathways associated genes indicated that LAMB3, Kaplan–Meier survival analysis of key pathways LAMA3, COL4A1, LAMC2 and FN1 were commonly and signature genes enriched in these pathways (Fig. 4f). Kaplan–Meier sur- To analyze the prognostic relevance of fve key path- vival analysis suggested that these fve genes were unfa- ways shared by all PDAC stages, mean expression value vorable factors of prognosis (Fig. 4g–k). Pan et al. Cancer Cell Int (2018) 18:214 Page 7 of 18

a b

c d

e f g

Fig. 3 Top ten enriched KEGG pathways of up-regulated and down-regulated DEGs at diferent stages. a–e Pathway enrichment of DEGs at stage I, IIa, IIb, III, and IV, respectively. f Venn diagram displayed common pathways of up-regulated DEGs shared by fve stages. g Venn diagram displayed common pathways of down-regulated DEGs shared by fve stages

Protein–protein interaction (PPI) network construction Cluster analysis of dynamic networks Te PPI networks of the DEGs between PDAC and DyNetViewer is a novel Cytoscape app for constructing, normal pancreatic tissues at diferent stages were con- analyzing, and visualizing dynamic molecular interac- structed by the online database STRING. Te typical PPI tion networks. To analyze the dynamic cluster attribu- network of stage IV was shown in Fig. 5, and the net- tion for PDAC progression, the networks from fve stages work consisted of 868 nodes interacting via 4717 edges. were input into DyNetViewer and calculated by MCODE Expression level of up-regulated DEGs and down-reg- algorithm over time. Te clusters at each stage which ulated DEGs in the PPI network was shown in red and exhibited highly interconnected regions were considered blue. as key regulators for promoting networks development. Pan et al. Cancer Cell Int (2018) 18:214 Page 8 of 18

abPathways in cancer Small cell lung cancer c Amoebiasis .0 .0 HR = 2.09 (1.24 − 3.52) .0 HR = 2.08 (1.28 − 3.39) HR = 1.5 (0.97 − 2.3) logrank P = 0.0047 logrank P = 0.0024 logrank P = 0.065 0. 81 0. 81 0. 81 y y y .6 .6 .6 Probabilit Probabilit Probabilit 0. 40 0. 40 0. 40 .2 .2 .2

Expression Expression Expression low low low high high high 0. 00 0. 00 0. 00

020406080 020406080 020406080 Time (months) Time (months) Time (months) Number at risk Number at risk Number at risk low 51 20 951 low 59 23 10 61 low 74 28 11 61 high 126 38 830high 11835 720 high 10330 620 d ECM-receptor interaction e Focal adhesion f .0 .0 Pathways in cancer HR = 1.63 (1.02 − 2.59) HR = 1.54 (1 − 2.38) logrank P = 0.039 logrank P = 0.048

16 0. 81 0. 81 Focal adhesion Small cell y y 2 1 .6 .6 lung cancer 0 0 0 6 0 4 3 0 LAMB3 Probabilit Probabilit 6 0 0. 40 0. 40 LAMA3 1 0 COL4A1 0 0 0 .2 .2 LAMC2 2 0 FN1 Expression Expression 0 0 low low 0 0 0 high high 0. 00 0. 00 0 0 0 5 020406080 020406080 2 Time (months) Time (months) Number at risk Number at risk Amoebiasis low 59 23 10 61 low 73 28 11 61 ECM-receptor interaction high 118 35 720high 10430 620

ghLAMB3 LAMA3 i COL4A1 .0 .0 .0 HR = 2.17 (1.42 − 3.31) HR = 3.86 (2.09 − 7.11) HR = 1.65 (1.06 − 2.58) logrank P = 0.00023 logrank P = 3.6e−06 logrank P = 0.026 0. 81 0. 81 0. 81 y y y .6 .6 .6 Probabilit Probabilit Probabilit 0. 40 0. 40 0. 40 .2 .2 .2

Expression Expression Expression low low low high high high 0. 00 0. 00 0. 00

020406080 020406080 020406080 Time (months) Time (months) Time (months) Number at risk Number at risk Number at risk low 91 36 11 61 low 51 22 951 low 13 146158 1 high 86 22 620 high 126368 30 high 46 12 200 j LAMC2 k FN1 .0 .0 HR = 3.21 (1.74 − 5.94) HR = 1.58 (1.04 − 2.4) logrank P = 8.7e−05 logrank P = 0.029 0. 81 0. 81 y y .6 .6 Probabilit Probabilit 0. 40 0. 40 .2 .2

Expression Expression low low high high 0. 00 0. 00

020406080 020406080 Time (months) Time (months) Number at risk Number at risk low 47 18 951 low 93 34 12 71 high 130 40 830 high 84 24 510 Fig. 4 The clinical relevance of key pathways and signature genes. a–e Overall survival analysis of fve fundamental pathways by Kaplan–Meier plotter. Log-rank test was used to evaluate signifcance. f Intersection of fve fundamental pathways associated genes. g–k Overall survival analysis of fve candidate genes by Kaplan–Meier plotter. Log-rank test was used to evaluate signifcance Pan et al. Cancer Cell Int (2018) 18:214 Page 9 of 18

APOBEC1 S100A16 MALL ANTXR1 PM20D1 ADAM19 PHLDA1 CELSR1 A1CF C1QTNF3 HAS2 RARRES3 S100A14 TUBB3 MLPH SLC41A1 ANTXR2 E2F8 FAIM2 BCL2A1 ESCO2 FANCD2 RTEL1 CTRL BRSK2 EPSTI1 FAM83D OSBP2 PSAT1 TNFRSF10A TAP2 STIL ONECUT1 SYTL2 TFE3 HELLS CRABP2 UBE2T GMNN FOXL1 RTP4 NR5A2 PMAIP1 CDC6 DLGAP5 NDC80TUBG2 NEK6 GBP4 NMI CASC5 MREG IFIT3 FAS RARB FANCI ANO1 PTF1A PRC1 CKAP2 CLDN3 NCAPG DEPDC1 PRICKLE1 PHGDH CLDN23 TRIM14 TRIM22 KIAA0101 MKNK1 PPP1R3B GJB2 CSTA BID CDCA5 CCNA2 CENPK CLDN18 XAF1 BUB1 HIST1H2BD OASL NR0B2 PARP9 ZWINT ASPM DTL RAB23 USP18 ISG20CTSK MKI67 CDKN3 AURKA GLI2 CLDN10 SERPINB9 MX2 FRZB SAMD9 CHEK1 CDC25B CLDN11 CENPE RBPJL CTSL2 BST2 SPDL1 SMC4 NUSAP1 SHCBP1 HIST2H2BE GJB1 DDX60 CENPF GLI3 TOP2A ISG15 IFI6 SEC11C PGC MX1 CCNB2 KIF20A GLRX OAS1 IFI44L PRDM9 HMMR DKK3 PCSK6 PPP2R2C CCNB1 KIF11 ANLN TPX2 PARP14 IFIT2 IFI27 CDKN2B TAP1 TNFSF13B KIF23 CENPA RSAD2 MLF1IP BACE1 CTSE PSMB8 TTK MAD2L1 KIF26B CDK1 CTSB SMC2 CDKN2A RPL3L PDX1 TRIM31 OAS2 NEK2 PSMB9 OAS3 CEP55 LMNB2 KRT6A PYCARD HIST1H2AC RRM2 CDC20 TYMS CTHRC1 GBP5 GBP2 PTTG1 KIF1A ECT2 DKK2 PLA2G7 IFI44 KCNN4 TRIM29 HOXB2 CKS2 OIP5 DUOXA2 CD58 HLA-F UBE2C UBE2S SFN KRT17 PCSK5 MAP4K4 GBP7 HLA-DMB RACGAP1 DKK1 ABLIM3 PLA2G1B MAFB KIF2C KRT6B GBP1 CD74 CASP1 KIF26A FBXW12 SFRP5 FMN1 CCL18 ODAM BIRC3 FBXO2 HIST1H4J ROR2 TBX3 KRT7 CYP4F2 HLA-DPB1HIST2H2AA3 TRIM50 GBP3 CCNE2 FBXO32 PDZK1 DUOX2 MT1H AMT PTGS1 ZBTB16 FZD1 GREM1 MRC2 SSH1 HRH1 FCGR1A E2F3 EGR2 TFDP1 PARD6A SFRP2 FOXF1 PELI2 KCNJ16 MT1F FBXL7 AREG FZD7 WNT2 MT1X BAG2 IL1R2 CCL11 AP1S2 HOXB3 PRR16 PMP22 CXCL10 CCL20 APCS TNFSF11 CHN1 BMP4 CEACAM5 IAPP ICAM1 LMO7 AMHR2 GPSM2 WFDC1 C6 HCAR2 PDE4B CENPW RHOF ABR IL1A GABBR1 CFTR NEDD4L CRIP1 MT1G CCL13 AK4 TBXAS1 WNT5A ARHGAP26 ADAM9 MRC1 SAA1 GNA15 DEPDC1B CAPG CXCL14 ADM ENTPD3 LYZ STON1 GATA4 PPAPDC1A IL1RAP CXCL12 KLK3 RAC2 SEL1L NOX4 LEF1 EREG TNFRSF11B LY6E CHRM3 PTGS2 ENTPD1 KIRREL2 VDR MBOAT2 LY75 ADCY7 DNM1 SPON2 RAMP1 CXCR4 PDE1A SOX4 WT1 NHS HSPA1A SCTR NMU PTGES PTGIS TWIST1 SFRP4 KRT19 HSPA6 IL1RN OPN3 ADAMTSL3 PTPN12 TM7SF2 KCNJ5 ANXA1 GNB4 HTR2B F2RL2 RUNX2 DAPP1 PSCA MST1 ZFPM2 EDNRA GRPR LIMS1 ADAM10 SQLE ATP6V1A RASSF2 PTGER2 NT5E SVIL ADAM12 OLR1 ADORA2B FPR3 FPR2 LY96 SEC23A LMAN1 NCF2 ERBB4 DR1 IRAK2 GNG7 SRPX2 ANXA8 GRP PTK7 FGA NRG4 TGFB1I1 LILRB4 TF ADAMTS4 PROS1 PAK3SH3KBP1 NPY1R CXCL9 CCKBR CD86 NPHS1 KCND2 ALB ERO1LB FHL2 SDK1 COCH AHNAK S100A11 IRAK3 HEPH IL8 CRP TGFBI SLCO1B3 TGM2 DPP10 CFH GPR68 FGL1 IGFBP3 EGF SERPINE1SERPINA4 DPYSL3 C5 CXCL5 F3 PLAU PAPPA ANP32E MYOF CP EDIL3 ITIH4 STXBP6 ANXA2 C1R S100A9 PKM CD14 F13A1 SPARC PLXNC1 S100A6 FCGR2A ITGAM EFNB2 THBS2 IGFBP5 SEMA3C F5 ACTN4 SERPINA1 PLXNA2 RAET1L HPN THY1 S100A4 PLAT MSN ITGB2 SERPINA5 CD55 ARPC1B MET THBS1 LTBP1 ITGB6 VAV3 RAET1G C2 C1QC EFNA5 TOR4A ACTN1 ECM1 SEMA3A INHBA PLXDC1 FCGR2B FCER1G ISLR SNAI2 ASPN PI3 RYR2 EPHA4 FN1 P4HA1 ZIC2 CD109 GZMA FLNA FGG ITGB4 ITGA2 NRP2 LGALS1 HPGD C1QB C4BPB PDGFRB IBSP CEACAM1 SLC16A3 PTPRC TIMP1 COL8A1 NRCAM SLC16A1 CCDC109B ANGPT2 VCL APOH FLNC KLK1 WIPF1 SERPINB2 ITGAV COL1A2 RAP1GAP PON3 VSIG4 SCD5 PCDH7 PDIA2 MMP1 ITGA3 SDC1 CSGALNACT2 NRTN CD248 MCU CD36 PLOD2 FBN1 SH3PXD2A FCGR3B CALD1 ITGA9 COL7A1 CYBRD1 TFF2 CD163 MYH9 FGFR3 TGFB3 COL3A1 P4HA3 COL1A1 C14orf105 SPOCK1 TFF1 AMBP COL4A2 PPARG FCGR3A CH25H ACACB TIMP3 VEGFC PLN LCP2 ITGB5 LAMA3 COL12A1 TFF3 SCD CD68 FOXA1 FGF7 LAMC2 COL11A1 SLC39A8 CILP C14orf37 SERPINB5 PON2 IL7R TREM1 TYROBP TPM4 COL4A1 CD53 ITGA1 COL24A1 FOXA3 AXL ENPP1 VASP MMP9 COL6A2 MS4A7 CD38 COL6A3 COL17A1 PLIN2 TREM2 TPM2 TNC COL6A1 ITGB1 SLC30A2 RHBDL2 PLAUR NID1 COL16A1 ZAK SEMA6D ACTG2 COL5A2 SERPINH1 P4HB MFAP5 FUT3 ELOVL5 RNASE1 MMP2 ITGA4 FBLN2 KLF5 APOE MYL9 LAMA4 COL5A1 ACTR3C BGN CHST11 MUC15 SLC7A7 F11 LAPTM5 CEACAM6 MDK DST LOXL2 MFAP2 ARHGAP27 LIF MUC17 PLS3 EPHA2 COL4A5 LEPRE1 SLC39A5 CLEC5A MYH11 ROBO1 EFEMP2 FUT9 FGFBP1 CDH2 DSECHSY3 PCOLCE PIWIL2 LIFR FUT1 MUC4 DHRS12 ACSM3 ACADL FLT3 ACTA2 MSR1 COL8A2 BST1 LAMB3 KDELC1 B3GNT3 TRIB2 FGF12 MYLK DMD LOXL1 LOX MUC13 MMP14 PDPN COL10A1 VCAN MUC20 PFKP UBASH3B CDH11 CDH6 PRELP TTN SV2B POSTN DDR2 APOL1 CTRB1 MMP3 HENMT1 GCNT3 GPT2 PFKFB4 PGM2 PROX1 ADAM28 EMILIN1 FRAS1 MUC16 NNMT IRS1 FYB CDH13 FREM2 TNNT1 MMP11 FKBP10 ST3GAL2 ACSS1 PYGL GPC6 CDH17 ARNTL2 OMD GALNT10 HSD11B1 CHL1 C1GALT1 SDR16C5 HK1 MMP10 PLCE1 HUNK GFPT2 CNN1 COMP CDH3 LCN2 ZEB1 ST6GALNAC5 TFAP2A TCN1 PRSS1 MEGF6 GALNT14 HSD17B8 GDA HS3ST2 HK2 RBP1 CTRB2 XYLT1 CPA1 ST6GALNAC1 PRDM1 GLIPR1 FBP2 FERMT2 DTNA FSCN1 CTNND2 ABAT AOX1 XDH CPB1 GLIPR2 MSLN PRLR HS3ST1 UGT2B15 SERPINI1 PRSS3 SNTB1 REG4 AMY2B ACSL5 CYP2C18 TAGLN TMPRSS2 DHRS9 REG1A HS6ST2 IMPA2 PDZD2 UGT2A3 UGT2B7 UGT1A1 CPA2 AMY1A NTRK2 IL4R CTRC ID1 ALDH1A1 GPX8 IGJ TMPRSS4 GNMT CYP2S1 CYP26A1 STRA6 AKR1B10 CEL HTRA1 GATM CYP1B1 ADH1B ANPEP SPINK1 CELA3A CELA2A INPP4B GSTA1 ALOX5AP SLPI GSTT1 GCAT AHR GSTA4 MGLL PNLIP REG1B ABCG1 EPHX1 GK CORO2A HPSE LXN GSTA2 MME RUNX1 GAMT ADH1A GGT6 APOC1 CMTM3 IDO1 MATN3 GP2 DDC PNLIPRP1 FABP3 GPX2 SKAP2 CKLF PIFO KYNU STS TDO2 FXYD2 SLC3A1 G6PC2 CMTM8 CLPS AOC3 AQP9 MZB1 AQP1 SYNC KAL1 HN1 ANO5 RUNX1T1 AQP8 COTL1 CTH ARSE NQO1 NPC1 SCEL BCAT1 CST1 CBS NELL1 MAT1A ARL4C SDSL AQP12A

CPPED1 SULF1 TMEM97 Fig. 5 The protein–protein interaction network of DEGs at stage IV. The PPI network was constructed by STRING and visualized by Cytoscape software. Red nodes represented up-regulated DEGs. Blue nodes represented down-regulated DEGs

A total of 45 featured molecular modules were identi- ITGAM, COL1A1 and COL1A2 participated in as much fed for all PDAC stages (Additional fle 1: Figures S1– as four pathways at stage I, respectively (Fig. 6f). Ten, S5). At stage I, 11 clusters that contributed to network we found that ten clusters were particularly involved were specifcally screened out, among which the cluster in the network of stage IIa, and the cluster 13 was also 11 also contributed to the network of stage IIa and IIb identifed at stage IIb (Fig. 6b). Pathway enrichment (Fig. 6a). Tese 11 gene clusters were enriched in path- showed that these clusters were mainly enriched in cell ways of infuenza A, cell cycle, rheumatoid arthritis, cycle, chemokine signaling pathway, infuenza A, vascu- Staphylococcus aureus infection, amoebiasis, protein lar smooth muscle contraction, focal adhesion, leishma- digestion and absorption, ECM-receptor interaction niasis, Staphylococcus aureus infection, protein digestion and platelet activation (Fig. 6f). Furthermore, we found and absorption, ECM-receptor interaction, complement that key genes including IL1A, CXCL8, ICAM1, ITGB2, and coagulation cascades, basal cell carcinoma and Pan et al. Cancer Cell Int (2018) 18:214 Page 10 of 18 y) it y) y) it it a b c ar Cluster21 ar Cluster1 ar Cluster12

odul Cluster13 Cluster2 Cluster13 6 odul 21.0 odul 6 Cluster3 (M Cluster22 (M

(M Cluster14 20.5 e Cluster23 e Cluster4 e Cluster15 but i 4 Cluster24 but but tr i 20.0 Cluster5 i 4 Cluster16 tr tr Cluster17 at Cluster25 r at Cluster6 at 19.5 r r Cluster7 Cluster18 te Cluster26 te te 3 2 Cluster19 us 2 Cluster27 cl us Cluster8 us

cl cl Cluster20 Cluster28

ic 2 Cluster10 ic ic Cluster11 Cluster11 1 Cluster11 0 nam 0 nam nam

0 (stage) IIIa IIb III IV Dy (stage) I IIa IIb III IV Dy (stage) I IIa IIbIII IV Dy

CDC20 OocyteOoOocytoc e mmeiosis BUB1 TTK f AURKA GBP5 GBP1 CDK1 d e CellC ll cyclcyccyy e MAD2L1 GBP2 OAS3 CCNB2 TPM2 Progesterone-mediatede e- oocyte OAS1 maturationa TPM4

y) NOD-likeODD-like receptor signalingalin pathwayp way CCNB1 y) it it

ar OAS2 HLA-F

ar ITGA1 Cellular senescencsen e Cluster29 Cluster38 CCNA2Hypertrophicphhicc cardiomyopatc yopatatthhy (HCM(HC ) odul OASL odul 4 Cluster30 5 Cluster39 RSAD2 DilatedDilat cardiomyopathyyopayoop hyy (DCM(D(DCCMM) (M

(M InfluenzaIn n A IL1A MX1 ITGB6

Cluster31 e Cluster40 ITGAV e 4 CXCL8 MYLK ut Cluster41 ITGA2 ArrhythmoAr hythmomoogenicogogeggenicenien rightrightg ventricularveventntr cularcuulal 3 Cluster32 but i CXCL10 CXCL5 cardiomcardioommyopathymyopmym y (ARVC)(A(A C)C) ISG15 ib

tr Cluster42 tr Cluster33 3 at ITGA3 at Cluster43 PertussiPertus iss Reeegggulationgulagulatguulatula ion ofof actinin cytoskeletonytoso etontoon Cluster34 r CCL20 HumaHumanummmaann papillopappapillomllomaviruslolomommaavavirvviruiruruus infectioninfeinfectiof tiion r 2 AGE-RAGEAGE-RARRAGEE sisignasignalinggnanaalilingngng pathwaypaathwway in te diabeticdiabetabeticettitiicc complicationsco icccaaations te Cluster35 2 Cluster44 Rheumatoidumam toididd aarthritiss

us IL-17IL 17 signalings pathwayp hway TNC SmallSmalll cellcec lung cancerceHMMRHr us Cluster45 Cluster36 cl MMP3 ICAM1

cl MYL9 1 Focacalall adaadhesiondh ssionioo 1 Cluster37 ic ic ECM-receptoEECCCMM-reMM--recececepeptoeptptor CKS2 TNF signaling pathwayway MMP1 ITGB2 Amoebiasim biaassisiss 0 interaiinnteractionntnteractteeerraraaccctioctttionon 0 nam COL6A1 nam MMP9 ITGAM COL4A5 (stage) IIIa IIbIII IV CXCR4 (stage) IIIa IIb III IV Dy COL1A2 Dy COL4A1 COL6A3 LLeukocyte transendothelialt a migrationm COL4A2 COL3A1 Relaxin signalinga g pathwaya y COL12A1 FPR3 Staphylococcuhy s GNA15 COL1A1 Complementen anda d coagulationcoagagu Proteinin ddigestioiggestion aureuseu infectiotionncascades sc ACTA2 COL17A1 FGG and aabsorptios rp ion COL5A2 g CXCL9CCL20 CXCL10 OAS3 FPR2 COL10A1 CXCR4 Platelet activation RSAD2 F2RL2 COL5A1 COL7A1 VCAN CD86 h SERPINA1 TTK CXCL5 COL11A1 BUB1 Chemokinmok e InfluenzaInI f ueenz A OAS1 MAD2L1 CCNB2 signalingli g papathwayt y CCNB1 CDC20 Rheumatoidmatoi arthritisa hrh RRM2 MAD2L1 EDNRA MMP1 IRAK2 p53 signalingna pathway MYH11 CellCeCele l cycleyycclle CCNBC 2 ACTG2 Progesterone-mediatedProgPrPror m dia oocyte CXCL8 IL1A HLA-DPB1 CDK1 maturatior n Cell addhdhesiondhesdhesionn moleculesmolecoollel ees (CAMs)(CAM CALD1 TuberculosiT sis FCGR3A TTK BUB1 PertussiPeertusssis ICAM1 CXCL9 CCNB1 Oocyte meiosis GNB4 MX1 CCL20 Malariaariaa LeishmaniasiLeLeishmL shhmhmamaaniasin a isis CellCe cyclyyc e Vascularscularar smooths RSAD2 SDC1 ITGAIT M FPR3 CCNA2 ADCY7 FCGR2A CXCL5 OAS1 FCGR1A CDK1 musclecllee contractioncont onn CXCL10 StaphylococcusStaphyttap yllococclococcccccuscucus Progesterone-mediatedPro on d oocyteo ACTA2 ITGB2 aureusauu s infectiofec ioonn Oocyte meiosiio s maturation CDC20 CXCR4 OAS3 EDNRA GNA15 PlateletPPlllaatatelet aactivationaccttiivavvatata HematopoieticHeHematoHiemomaatantopo t c cellcee lineagelinneage InfluenzaInfl nz A AGE-RAGEAGGEE-EER-RRARAAGGEE signsignalingnnalin pathwaypatth a inn FCGR3B RelaxinRelaxinaxin signalingsignsis g a pathwapathwayy ITGB1 diabeticdiabdidiaaabbebetbeticeticc complicoo plicicatioc ns ITGA2 GNB4 Chemokineh kine signalingna in pathway FGG BMP4 CCNA2 TNC ACTA2 MYH11 LAMC2 THBS2 WNT5A COL17A1 COL3A1 ITGB5 PhagosomePhP ACTG2 COL4A2 Rheumumatoiummatoiidd arthritis OAS2 PI3KPPII3K-Akt3K3KK-Akt-Aktktt signals gngnaaalinalliniinngg pathpataathwththwwwaay ITGA1 GLI2 GNA15 VascularV la smoothm h LAMB3 FN1 AmoebAmAAmommoeboebioeebbiaiaass COL5A2 ADCY7 FN1 MYLK FGA Plateleteet activationactctivationiv onn MYLK Leukocyteocyteccy transerannsseendothelialen migration musclele contractionra o TPM2 GLI3 MX1 COL1A1 COL4A1 DilaDDilaatedateedd cacardiomyoomyoopopathypapathyy (DCM) BasalBa celle carcinoma COL7A1 LAMA3 FPR3 MYL9 HumanHuHHumumamaann papillomapapillomavppapillopappipilloloommaaviravvirururus iinfenffectiofe onon COL1A2 RegulationReRegulateggultatiioonn ofof actinactintintiin cytcytosccyyyttotoosskeletokelekee etonn CXCL8 MalaMaM l rir a LEF1 PI3K-AktPI33KK-K--Akt-AAAktAkkkttst ssisignasignalingsignalinsignalignaiggnalgnalinna iinngnggp pathwaathwathwthway CALD1 FocaFFocalFoocalcacala aadadheadhesiondhdhehesesionssiioionon ArrhythmogA rhy ggeenicc rright ventricularr COL11A1 COL6A3 cardiomycar myyopaty aattthhhy (ARVC) ProteinP ote n didigestioigigegests n MYL9 StapStS aphylococcuphhhy o s ComplementC lemem n and HumanH an papillomaviruspaap lllooommavirusmamaavviruuuss ininfeinfectionnfectioonn F5 COL5A1 andan aabsorptioorrppttioonn ICAM1 HypertHypperterrtrtrrophicophphichicc cardiomyopathcaardiomommyoopp ththythyy (HHCM)HCCCM) coagulatiog la ion COL5A1 COMP auaureusurreu infection n ProteinProteinn ddigestion SmallSmaalll cecellell lunglunungungn cacanceana ceeerr F2RL2 ECM-receptoECE MMreececepcepcepeptorpptttooror TPM4 cascadess d AmAmoebmomoeebbiab assiss COL17A1 andnd abaabsorptions Small cellce lung cancecer interactinintn er ctctionon SERPINA1 ITGB2 ITGA3 COL12A1 HMMR VASP PLAUR COL10A1 RelaxinRRelaxiReelaxinellaaaxxixin signalingsigsigignaliininginngg pathwapathwpaththwathhhwwaw y F13A1 LAMB3 AGE-AAGGEEER-R-RAGERARAGGEE signsignalisiggnnaalliinginngg papathpatththwayhwaay ini COL5A2 F2RL2 ReReegggulationguultl titionn ofof aactinac cytoskeletonyt Complementom me and COL10A1 diabeticdiaabeticabetibbeeettiicc ccoomomplicationmpplicpllilicicaticacatatiotit nnss ITGAM CKS2 ISG15 coagulatioc ti n COL11A1 FLNA OASL LAMA4 COL12A1 HMMR ITGB6 cascadesc s COL7A1 LAMA3 ITGB6 ISG15 ECM-receptoECECM-receptoCMCM-receptoMM--rrreeceptoeceecececepeptptooorr FoFocFococalcacalal adadhdhesionhesisiooonn interactioiinterainntnterraacttitioion DilatedDilateatedatted cardiomyopathyrdiomyopathypap hyhy (DCM(D( ) FCGR1A PLAU Hematopoieticma cell lineage COL1A2 ITGA1 ITGA2 FPR3 CKS2 MAD2L1 COL1A1 ITGB5 ITGA3 CXCL5 COL6A3 HypertrHyppeeerrrtrophict c cardiomyopathypat y (HCM) TPM4 CCNB2 COL4A2COL3A1 i FCGR1A CDK1 LAMC2 Arrhythmogenic right ventricular TPM2 CCNB1 Staphylococcusccus aureuseus infection COL4A1 cardiomyopathy (ARVC) BUB1 OocyteOocyocyoc t mmeiosis FLNA LeishmaniasiL h s Rheumata ooiid arthritisa Hematopoieticm opo et c cellll lineaglineneage TPM2 Progesterone-mediatedProgestero atedd oocyte Malariaala a Inteteestinalestis immunemmu networktwworkwo foror IgA maturation CellC l cyccycley productionppr ction Cellularar senescences cecenccee CDC20 LeukocyteLeeuuk cytete transendothelialann ndooth all migrationmigrmim g CKS2 ICAM1 LAMC2HMMR HLA-DPB1 j LAMB3 COL12A1 p53 signalingi pathwahwwaway TPM4 COL6A3 HypertHypeypertrerrtrophicrtrt oopphhiic cardiomyopathyc pathy (HCM)HC LAMA3 COL10A1 TGFB3 F13A1 Regueggulationulu aatioattiiioononno ofof actinaaccttin cytoskeletocytcyyty oskeeleton MYH11 CCNA2 CXCL8 CALD1 ITGA3 COL5A1 Dilated cardiomyopathyrdiooomyomyooppapataththhyy (DCM(DDCM) CXCR4 Small cell lunglun cancerncernc ECM-receptoCMM-rrececepepeptoepptortorr COL17A1 ToxoplasmosiTooxox s s ITGA1 ITGA4 interactioinntnteratereeracra tiononn COL4A1 SERPINE1 THBS1 ArrhythmogeniAArArrrrhyt c Complementmp and ISG15 ITGB6 ACTG2 ProteinPrroteein didigestioges onn COL7A1 PROS1 rightrriigg veventricular MYLK COL1A1 coagulatioc n ITGA9 Vascular smoothsmoo andana d absorptioo tion cardiomyopathycara my FocalF adhesion LAMC2 HuHumaumanmaananp papaapppillpillllooomavommamaaviruavirusavauaiavivir ssi innnfnfectionnfecfeecttiioon muscle contractiotr tion cascades ITGA2 AmoebiasisAmoeA eb as COL1A2 AGE-RAGEAGE-RAGAGGEE-RAGEERE-RE AAGEEs signalinsignsignalingignaligng aallilingiiningnngggp pathwaypathpaathwwwaaayy inn ITGAV (ARVCC) FoFococalcacaall adhdhheesioes onon COL11A1 diabdidiaabeticabbeticbeticticcc cocoomplicationsomplicm icatioata onnnss ACTA2 SERPINA1 LAMB3 ITGB5 Platelet activationation COL3A1 EDNRA CDK1 ADCY7 TPM4COL5A2 LAMA4 VCL MYL9 RelaxinReRelaxR signalinggnalinAGE-RAGEAGE-RA pathwayGE signalinsignaling pathway in PI3K-APIP 3K-A-AkAktAkA t ssiiggnalinggng a inngg papathwata hway CCNB2 MMP1 COL7A1 ECECMM-r-receptorreeeccceepeptptotoror diabetic complications FN1 ACTN1 OocyteOoO cy mmeiosisis CXCL8 LAMA3 iinntntteeraereraaaccctioctttiioionon ACTG2 TPM2 AmAmomooebiasioeebiasiia s Dilated COL1A2 Vascularsccu a smooth RRM2 MAD2L1 COL6A3 PrProrogesterone-mediatedogo e e mee oocytococytcyty e cardiomyopathy COL12A1 ProteinPrP o eii didigestioestistiooonn VEGFC musclemuscs contractionc maturationma attio CHEK1 p53535 signalingsignalin pathwaypathw y (DCM) andd aabsorptionb or n COL6ASmallSm2mallmam cellcelellll lungl g ccancercaaanc PlaP atelett activatioct n COL5A2 MYLK BUB1 RheumatoidRheumR aat arthritiarthr IL-17sL-17 ssignaling COL6A1 RIG-I-like receptorpt signalingalin pathwath y RelaxinRRel xinn signalingsi pathwathw y CCNB1 pathwap h y COL3A1 GNA15 MYH11 COL11A1 COL5A1 COL4A2 Cellell cyclcle CDC20 COL4A1 TTK CXCL5 COL1A1 ACTA2 AURKACCNA2 COL17A1 ISG15 CCL20 CXCL10 COL24A1 COL10A1 CKS2 Fig. 6 Clusters analysis of dynamic network during PDAC progression. a–e The charts of dynamic cluster attributes identifed from dynamic networks at diferent stages. The dynamic cluster attributes for fve stages were calculated by DyNetViewer (MCODE algorithm) over time. f–j Network of pathway analysis at each stage by ClueGO. Gene clusters of each stage were enriched for KEGG pathways and the interaction of pathways at fve stages were further constructed by ClueGO, respectively Pan et al. Cancer Cell Int (2018) 18:214 Page 11 of 18

platelet activation (Fig. 6g). CXCL8, ADCY7, ITGAM, RACGAP1, ITGB5 and AURKA were also dynamically ITGB2, ITGB1, IL1A, ICAM1, ITGA2, THBS2, SDC1, changed at particular stage (Fig. 7d). COL3A1, COL1A2 and COL1A1 were commonly shared by four or more pathways, respectively (Fig. 6g). In the Kaplan–Meier survival analysis of key nodes network of stage IIb, ten clusters were found acting as According to the node centrality analysis above, we chose key modules for the protein interaction network (Fig. 6c). the featured nodes of stage IV to determine their clinical Tese clusters were associated with cell cycle, vascular relevance. Kaplan–Meier survival analysis showed that smooth muscle contraction, infuenza A, Staphylococ- high expression level of MLF1IP (also known as CENPU) cus aureus infection, focal adhesion, complement and and ITGB4 were signifcantly correlated with shorter coagulation cascades, protein digestion and absorp- overall survival, respectively (Fig. 8a). Furthermore, we tion and ECM-receptor interaction (Fig. 6h). ADCY7, also analyzed the expression manner of featured nodes ITGAM, ITGB2, MYL9 were shared by four or more (MLF1IP/CDC25B/ITGB4/TRIM22) during PDAC pro- pathways, respectively (Fig. 6h). For stage III, we found gression (Fig. 8b). Te expression level of these four genes 9 unique clusters were essential for its network (Fig. 6d). at stage IV showed no signifcant changes compared with Tese gene clusters mainly involved in cell cycle, vascular other stages, which indicated that these nodes potentially smooth muscle contraction, arrhythmogenic right ven- promoted cancer progression by directly maintaining the tricular cardiomyopathy, amoebiasis, complement and molecular network without depending on their expres- coagulation cascades, protein digestion and absorption sion level (Fig. 8b). and ECM-receptor interaction (Fig. 6i). Two key genes including CXCL8 and TGFB3 were shared by as much as Identifcation of PDAC stage‑specifc activated protein four pathways, respectively (Fig. 6i). Moreover, a total of kinases and the phosphosite markers 8 clusters specifcally participated in the network of stage Considerable studies have shown a causal role of protein IV (Fig. 6e). Tese clusters mainly enriched in cell cycle, kinase mutations or dysregulations in tumorigenesis and vascular smooth muscle contraction, dilated cardiomyo- cancer progression. Cancer research has been trying to pathy, IL-17 signaling pathway, protein digestion and turn these molecules into valid drug candidates for emer- absorption and ECM-receptor interaction (Fig. 6j). No gence targeted therapies. Depending on the dynamic genes were found shared by more than two pathways at networks, the PDAC stage-specifc activated protein stage IV (Fig. 6j). kinases were also identifed. As shown in Fig. 9a–f, a total of 15 kinases involved in the dynamic networks with dis- Node centrality analysis of dynamic networks tinct patterns. TTK, AURKA, BUB1, CDK1 and NEK2 To elucidated the key genes that participated in the were fundamental kinases with high node degree, which dynamic networks, the node centrality at each PDAC may be required for maintaining the molecular signals stage were calculated. Tree typical centrality measures underlying tumor progression (Fig. 9a). Besides, we also including Betweenness Centrality (BC), Degree Cen- found CHEK1, the checkpoint kinase 1 coding gene trality (DC), Local Average Connectivity-based method (degree = 32), specifcally activated at stage IV (Fig. 9d), (LAC) were applied in our study. We screened out 19 key which may indicate a potential drug target for advanced genes that potentially involved in the dynamic network PDAC. Te kinase ABR was the featured node for stage development (Fig. 7a–c). Five genes including NDC80, III, though getting low node degree (Fig. 9d). Addition- KIF2C, KIF20A, OIP5, ZWINT were specifcally found ally, PDGFRB was particularly found at stage IIa, IIb and at stage I with highest DC, BC, and LAC. FGA, CRP III in the dynamic network (Fig. 9d). Finally, we also pre- and ITGB1 were featured nodes for stage II. WNT5A dicted the phospho-targets of eight candidate kinases. was essential for maintaining the network at stage IIa Te protein kinases AurA (encoded by AURKA), CDK1, and IIb. F5 was uniquely found at stage IIb. Addition- Chk1 (encoded by CHEK1), NEK2 and TTK phosphoryl- ally, fve genes including ITGA4, COL6A2, ITGA9, ated their substrates by targeting serine, threonine and THBS1 and SERPINE1 were characteristic nodes at stage tyrosine, while BUB1 and skMLCK (encoded by MYLK) III. Four nodes including ITGB4, MLF1P, TRIM22 and targeted the serine and threonine for phosphorylation CDC25B were potential key genes for promoting the (Fig. 9g). Tyrosine was the only phosphorylation site of advanced status of PDAC. Finally, we also sorted the top PDGFRB substrates (Fig. 9g). Moreover, the consensus ten nodes ranked by standard deviation of DC, among sequences of kinase substrates were depicted by sequence which NCD80, KIF20A, ITGB1 and KIF2C also existed logo to indicate the phosphosite markers (Fig. 9g). in the 19 key nodes mentioned above (Fig. 7d). Te remaining genes including NCAPG, CENPE, KIAA0101, Pan et al. Cancer Cell Int (2018) 18:214 Page 12 of 18

NDC80 FGA ITGA4 ITGB4 KIF20A FGA COL6A2 ITGB4 a b KIF2C CRP COL6A2 MLF1IP NDC80 CRP ITGA9 MLF1IP

KIF20A ITGB1 ITGA9 TRIM22 KIF2C ITGB1 ITGA4 TRIM22

OIP5 WNT5A THBS1 CDC25B ZWINT WNT5A SERPINE1 CDC25B

ZWINT F5 SERPINE1 OIP5 F5 THBS1 60 40 e) ) eg re AC

(D 30 (L

e 40 e ut ib but i tr tr at

at 20 ode ode n 20 n c

ic 10 na mi nam Dy Dy 0 0 (stage) I IIa IIb III IV (stage) IIIa IIb IIIIV

c ZWINT FGA ITGA4 ITGB4 d NCAPG KIAA0101 KIF2C CRP COL6A2 MLF1IP ITGB1 KIF2C CENPE KIF20A OIP5 ITGB1 ITGA9 TRIM22 ITGB5 AURKA 15000 60 NDC80 RA CGAP1 KIF20A WNT5A THBS1 CDC25B e) NDC80 F5 SERPINE1 eenness) eg re w (D et e

(B 10000 40 ut e ib tr but i at tr at

5000 node 20 c node c na mi na mi Dy 0 Dy 0 (stage) I IIa IIb III IV (stage) I IIa IIb IIIIV Fig. 7 Node centrality of dynamic network during PDAC progression. a–c Featured nodes with values of DC, LAC and BC from dynamic networks at diferent stages. b The list of top ten nodes ranked by standard deviation of DC from dynamic network

Validation of gene expression patterns in PDAC tissues special expression manner may indicate the pathologic To investigate the expression patterns of MLF1IP, molecular basis of tumorigenesis and progression. LAMA3 and LAMB3 during PDAC progression, we per- formed the immunohistochemistry (IHC) analysis. As Discussion a featured node of stage IV, MLF1IP expressed weekly PDAC is one of the most lethal tumors with limited sur- in normal pancreatic tissue (Fig. 10). With the progres- vival improvement over the last decades. Te rapid pro- sion of PDAC, MLF1IP dramatically increased in PDAC gression of PDAC results in an advanced stage of patients tissues and was cytoplasmic and nuclear localization. when diagnosed. Kong et al. [17] utilized infammation- LAMA3 and LAMB3 were common genes shared by accelerated ­KrasG12D-driven PDAC mouse model to fve fundamental pathways throughout diferent PDAC illustrate the dynamic landscape of pancreatic carcino- stages. Te IHC analysis showed that LAMA3 and genesis. Teir high temporal resolution transcriptional LAMB3 were positively stained in PDAC tissues with a data defned a transcriptional signature of early pancre- progressive increase manner (Fig. 10). Interestingly, we atic carcinogenesis and a molecular network driving for- found that LAMB3 positively expressed in normal pan- mation of preneoplastic lesions. However, the dynamic creatic connective tissue. With the initiation of PDAC, molecular mechanism underlying PDAC progression LAMB3 mainly existed in stroma and cytoplasm of can- remains far from clear. In this study, we analyzed the cer cells. However, LAMB3 gradually translocated into microarray from GEO and identifed the DEGs between the nucleus of cancer cells at advanced PDAC stages. Tis normal tissue and diferent staging PDAC, respectively. Pan et al. Cancer Cell Int (2018) 18:214 Page 13 of 18

a MLF1IP CDC25B ITGB4 TRIM22 .0 .0 .0 HR = 2.73 (1.57 − 4.74) .0 HR = 0.77 (0.5 − 1.19) HR = 2.5 (1.39 − 4.52) HR = 1.43 (0.87 − 2.34) logrank P = 0.15 logrank P = 0.00021 logrank P = 0.24 logrank P = 0.0016 0. 81 0. 81 0. 81 0. 81 y y y y .6 .6 .6 .6 Probabilit Probabilit Probabilit Probabilit 0. 40 0. 40 0. 40 0. 40 .2 .2 .2 .2 low low low low high high

high high 0. 00 0. 00 0. 00 0. 00 020406080 020406080 020406080 020406080 Time (months) Time (months) Time (months) Time (months) Number at risk Number at risk Number at risk Number at risk low 49 23 12 51 low 50 15 300 low 44 16 841 low 43 19 10 61 high 128 35 530 high 127 43 14 81 high 133 42 940 high 134 39 720 b

MLF1IP CDC25B ITGB4 8 TRIM22 7 5 10 2468 1234 234567 123456

(stage) I II III IV (stage) I II III IV (stage) I II III IV (stage) I II III IV Fig. 8 Kaplan–Meier curves for overall survival of key nodes. a Overall survival curves stratifed by expression of MLF1IP, CDC25B, ITGB4 and TRIM22, respectively. Patients were divided into high-expression group and low-expression group, according to expression of four genes above. Log-rank test was used to evaluate signifcance. b Expression manners of MLF1IP, CDC25B, ITGB4 and TRIM22 at diferent stages were determined by GEPIA using TCGA-pancreatic adenocarcinoma data

Further, we constructed a dynamic molecular interaction that LAMA3, LAMB3, LAMC2, COL4A1 and FN1 were networks and identifed the functional modules and fea- commonly shared by these fve key pathways and nega- tured nodes for each PDAC stage, which may be respon- tively correlated with overall survival. LAMA3, LAMB3 sible for PDAC progression. and LAMC2 encoded the subunits of laminin, which was Te molecular basis of diferent PDAC stages were component the basement membrane and involved in cell complex and dynamic. Te stage-course gene expression migration. Recently study demonstrated that combina- patterns profled landscape of diferences between PDAC tion of serum LAMC2, CA19.9 and CA125 was able to stages. TNNT1, encoding slow skeletal muscle troponin signifcantly improve upon the performance of CA19.9 T, kept increasing signifcantly from early to advanced alone in detecting PDAC [22], while the exact function of stage. TNNT1 was up-regulated in human induced pluri- LAMC2 need to be fully elucidated. Te role of LAMA3 potent stem cells and immortalized retinal pigment epi- and LAMB3 in PDAC were rarely studied, too. We found thelial [18]. Studies demonstrated that TNNT1 involved that LAMA3 and LAMB3 were robustly expressed at in breast cancer cell proliferation and highly expressed each PDAC stage when compared with normal tis- in leiomyosarcoma metastases [19, 20], while its role in sue, which should be essential for tumorigenesis and PDAC remained mystic. Li et al. [21] found that HOOK1 progression. negatively regulated epithelial–mesenchymal transi- Tere were also some pathways enriched at particu- tion by inhibiting the activity of SHP2. Te dramatically lar stage. Rheumatoid arthritis was specifcally found at decrease of HOOK1 in PDAC may suggest a molecular stage I and IIa, which indicated that immune dysregula- basis of aberrant EMT during cancer progression. Path- tion occurred early during tumorigenesis. A recent study way enrichment showed that pathways in cancer, small from Ikeura et al. [23] also showed that autoimmune cell lung cancer, ECM-receptor interaction, amoebiasis, pancreatitis has the increased risk for pancreatic cancer focal adhesion were commonly enriched from early to after 62.4 months of mean follow-up period. Tus, the advanced stage, which should be essential for maintain- relationship between autoimmune and pancreatic cancer ing the pathological status of PDAC. Except for amoe- need to be closely concerned. For stage IV, three pathways biasis, these pathways were signifcantly associated with including p53 signaling pathway, platelet activation, and poor PDAC overall survival. Further study suggested cell cycle were specifcally identifed. Missense mutations Pan et al. Cancer Cell Int (2018) 18:214 Page 14 of 18

abMAP4K4 AURKA NEK2 MAP4K4 AURKA NEK2 c MAP4K4 AURKA NEK2 EPHA2 BUB1 IRAK2 EPHA2 BUB1 IRAK2 EPHA2 BUB1 IRAK2

) TTK CDK1 EPHA4 TTK CDK1 EPHA4 ee) 80 TTK CDK1 EPHA4 40 8000 enness ) AC egr (L 60 30 we (D e et

e 6000 (B

40 but

i 20 but e tr i tr 20 at but i at 8 3 4000 tr ode at

6 n ode 2 n c 2000

4 ode c

1 n

2 c na mi na mi

0 Dy 0 0 Dy (stage) IIIa IIbIII IV (stage) I IIa IIbIII IV na mi (stage) IIIa IIbIII IV Dy d MYLK CHEK1 e MYLK CHEK1 f MYLK CHEK1 ) ee ) 35 PDGFRB ABR PDGFRB ABR 30 3000 PDGFRB ABR ROR2 IRAK3 AC egr 34 29 ROR2 IRAK3 eenness)

(L ROR2 IRAK3 w (D 33 28

e 2000 et e

32 ut 27 ut (B 31 ib

ib 26 1000 e ttr tr

30 a 25 ut at ib

10 8 ttr

ode 200 a n

node 6

150 ic ic 5 4 100 node

nam 2 c

nam 50 0

Dy 0 Dy (stage) I IIa IIbIII IV (stage) 0 I IIa IIb III IV na mi (stage) IIIa IIbIII IV g Dy

protein kinase sequence logo phospho-target protein kinase sequence logo phospho-target

serine (S) serine (S) AurA threonine (T) NEK2 threonine (T) tyrosine (Y) tyrosine (Y)

serine (S) BUB1 PDGFRB threonine (T) tyrosine (Y)

serine (S) serine (S) CDK1 threonine (T) skMLCK threonine (T) tyrosine (Y)

serine (S) serine (S) Chk1 threonine (T) TTK threonine (T) tyrosine (Y) tyrosine (Y)

Fig. 9 Identifcation of PDAC stage-specifc activated kinases and their phosphosite markers. a–f Featured kinases with values of DC, LAC and BC from dynamic networks at diferent stages. g Phospho-target and sequence logo of protein kinases were predicted and listed

in the p53 tumor suppressor inactivated its antiprolif- and stroma were associated with tumor metastasis [27]. erative properties but could also promote metastasis Inhibition of platelet activation prevented the P-selec- through a gain-of-function activity [24]. Aberrant p53 tin and integrin-dependent accumulation of cancer cell signaling were predominately seen in some in situ lesions microparticles and reduced tumor growth and metasta- as well as invasive PDAC, indicating this signaling may sis [28]. Moreover, activated platelet that interacted with occur mid-to-late stage in the pathogenesis of this dis- cancer cells was also sufcient to prime cisplatin insen- ease [25]. Deregulation of cell cycle has also been impli- sitivity in pancreatic cancer cells [29]. In our study, we cated in PDAC progression. Six genes including CHEK1, found a total of 12 genes were enriched in platelet acti- CCNB1, CCNB2, CDK1, CDKN2A and SFN were shared vation, among which half of them were collagen fam- by p53 signaling pathway and cell cycle. Due to the poor ily members (COL1A1, COL1A2, COL3A1, COL5A2, outcome of advanced PDAC, promising therapies like COL11A1). Tese results may indicate a novel role of cell cycle inhibitors are currently under development collagens in facilitating PDAC development. However, [26]. Extravasated platelet activation in pancreatic cancer Pan et al. Cancer Cell Int (2018) 18:214 Page 15 of 18

MLF1IP LAMA3 LAMB3 NT Stage I Stage II Stage II I Stage IV

100 μm 100 μm 100 μm Fig. 10 Experimental validation of gene expression patterns in PDAC tissues. Expression manners of MLF1IP, LAMA3 and LAMB3 in diferent PDAC staging tissues and normal pancreatic tissues were performed by immunohistochemistry

collagen-mediated platelet activation during PDAC pro- digestion and absorption and ECM-receptor interaction. gression still need to be fully demonstrated. PDAC is characterized by the excessive deposition of Te progression of PDAC exhibited dynamic molecu- extracellular matrix (ECM), which is thought to contrib- lar interaction networks from early to advanced stage, ute to its malignant behavior. Duan et al. [30] found that among which highly interconnected regions were con- type I collagen could promote epithelial–mesenchymal sidered as key regulators for maintaining molecular transition in pancreatic cancer by activating β1-integrin networks. We identifed a total of 45 unique clusters for coupling with the Hedgehog pathway. In our study, we fve stages, and these clusters showed special expression found high expression of integrin subunit beta 4 (ITGB4) pattern at diferent stages. Pathways enrichment indi- in PDAC was correlated with poor prognosis. Tis result cated that cell cycle, protein digestion and absorption was in consistent with Masugi’s study [31], whose group and ECM-receptor interaction were fundamental signal- noted that upregulation of integrin β4 promoted epithe- ing for all fve PDAC stages. Moreover, the collagen fam- lial–mesenchymal transition and was a novel prognostic ily and integrins were two main regulators for protein marker in PDAC. Intriguingly, we also found that ITGB4 Pan et al. Cancer Cell Int (2018) 18:214 Page 16 of 18

typically contributed to molecular network of stage IV, those without HRR-related gene mutations after oxali- which suggested a temporal dependent manner of ITGB4 platin-based chemotherapy [37]. In our study, we found for promoting PDAC progression. Due to the essential that CHEK1 was highly activated in stage IV, which may role of ITGB4 in advanced PDAC, the underlying mecha- reveal the malignant evolution of PDAC and potential nism need to be further elucidated. therapeutic target. In the dynamic molecular interaction network, topo- logical variation of nodes was essential for network progress. Basing on the typical centrality measures, we Conclusions found total 19 key nodes uniquely contributed to fve In summary, we screened the DEGs of diferent PDAC stages, respectively. At early stage, aberrant cell mitosis stages and constructed dynamic molecular interac- and motility were frequently required for tumorigen- tion network to illustrate the underlying mechanism of esis. Featured nodes including NDC80, ZWINT, OIP5, PDAC progression. Five genes that commonly shared KIF2C and KIF20A were mainly associated with chromo- by fve fundamental pathways may act as key regula- some segregation and spindle checkpoint activity. Over- tors throughout all PDAC stages. Meanwhile, collagen expression of NDC80 was correlated with prognosis of family and integrins were identifed as potential regula- pancreatic cancer and regulated cell cycle and prolifera- tors for driving PDAC progression. Additionally, PDAC tion [32]. With the progression of PDAC, tumor stroma stage-specifc protein kinases were also identifed for became more and more abundant. As mentioned above, potentially targeted therapy. Tese timing and context cancer cells actively involved in the production of extra- dependent nodes and pathways could be pivotal mecha- cellular matrix proteins and interacted with ECM by nism for promoting the dynamic progression of PDAC. integrins. Integrins played a role in cell migration, mor- Our study provided a view for a better understanding of phologic development, diferentiation, and metastasis. the dynamic landscape of molecular interaction networks Node centrality analysis showed that ITGB1 (stage IIa), during PDAC progression and ofered potential opportu- ITGA4 (stage III), ITGA9 (stage III), ITGB4 (stage IV) nities for therapeutic intervention. Further studies such specifcally functioned at particular stage, which indi- as single-cell sequencing or other multi-omics analy- cated an essential role of integrins during PDAC progres- sis are needed to comprehensively reveal the intricate sion. Moreover, a total of four key nodes were identifed mechanism of cancer cells and tumor microenvironment at stage IV, which may facilitate the aggressiveness of changes during PDAC progression. PDAC. Te clinical relevance analysis suggested that MLF1IP (also known as CENPU) and ITGB4 were sig- nifcantly correlated with shorter overall survival. Role of Additional fle MLF1IP had been preliminarily elucidated in some can- cers including bladder cancer, ovarian cancer and pros- Additional fle 1: Figures S1–S5. Featured clusters at diferent PDAC tate cancer, while little is known in PDAC progression stages. [33–35]. Te precise function of MLF1IP in PDAC need to be further investigated. Abbreviations PDAC: pancreatic ductal adenocarcinoma; DEGs: diferentially expressed Protein kinases have been widely investigated in can- genes; logFC: log2 fold change; OS: overall survival; IHC: immunohistochem- cers, since they are promising molecular targets for istry; GEO: Gene Expression Omnibus; GO: gene ontology; KEGG: Kyoto cancer treatment. Depending on the dynamic molecu- Encyclopedia of Genes and Genomes; PPI: protein–protein interaction; FDR: False Discovery Rate; TCGA:​ the Cancer Genome Atlas; DC: degree centrality; lar interaction networks, several PDAC stage-specifc BC: betweenness centrality; LAC: local average connectivity-based method; kinases were identifed. TTK, AURKA, BUB1, CDK1 ECM: extracellular matrix. and NEK2 were fundamental kinases that participated Authors’ contributions in cell cycle and mitosis. Te AURKA selective inhibi- ZFP and PH conceived and designed the study. LL, QLF, and YWZ collected, tor alisertib could induced cell cycle arrest and facili- processed and validated data. YYQ and XPH analyzed data and prepared tated autophagic cell death in pancreatic cancer cells fgures. ZFP and LL drafted the manuscript. PH revised the manuscript. All [36]. Te phase I trial (NCT01924260) was carrying out authors read and approved the fnal manuscript. to investigate the safety and efcacy of alisertib in pan- Author details creatic cancer patients when given in combination with 1 Department of Pharmacy, Zhejiang Cancer Hospital, Hangzhou 310022, China. 2 Department of Pharmacy, The First Afliated Hospital, College gemcitabine. CHEK1 was required for checkpoint-medi- of Medicine, Zhejiang University, Hangzhou 310003, China. 3 Key Laboratory ated cell cycle arrest and preserving the integrity of the of Head & Neck Cancer Translational Research of Zhejiang Province, Zhejiang genome. Recent study pointed that patients with inacti- Cancer Hospital, Hangzhou 310022, China. vating homologous recombination repair (HRR) related Acknowledgements gene mutations showed signifcantly longer PFS than Not applicable. Pan et al. Cancer Cell Int (2018) 18:214 Page 17 of 18

Competing interests interaction networks, with increased coverage and integration. Nucleic The authors declare that they have no competing interests. Acids Res. 2013;41(Database issue):D808–15. 13. Li M, Yang J, Wu FX, Pan Y, Wang J. DyNetViewer: a Cytoscape app for Availability of data and materials dynamic network construction, analysis and visualization. Bioinformatics. All data generated and analyzed during this study are included in this pub- 2018;34(9):1597–9. lished article and its Additional fles. 14. Manning G, Whyte DB, Martinez R, Hunter T, Sudarsanam S. The protein kinase complement of the human genome. Science. Consent for publication 2002;298(5600):1912–34. Not applicable. 15. Lanczky A, Nagy A, Bottai G, Munkacsy G, Szabo A, Santarpia L, Gyorfy B. miRpower: a web-tool to validate survival-associated miRNAs utilizing Ethics approval and consent to participate expression data from 2178 breast cancer patients. Breast Cancer Res Not applicable. Treat. 2016;160(3):439–46. 16. Tang Z, Li C, Kang B, Gao G, Li C, Zhang Z. GEPIA: a web server for cancer Funding and normal gene expression profling and interactive analyses. Nucleic This work was supported by National Natural Science Foundation, People’s Acids Res. 2017;45(W1):W98–102. Republic of China (Nos. 81703578, 81802673), Zhejiang Provincial Natural Sci- 17. Kong B, Bruns P, Behler NA, Chang L, Schlitter AM, Cao J, Gewies A, Ruland ence Foundation of China, Zhejiang Province, People’s Republic of China (No. J, Fritzsche S, Valkovskaya N, et al. Dynamic landscape of pancreatic LQ18H160017), Zhejiang Provincial Program for the Cultivation of 151 Talents carcinogenesis reveals early molecular networks of malignancy. Gut. (to Ping Huang), and Zhejiang Provincial Program for the Cultivation of High- 2018;67(1):146–56. level Innovative Heath Talents (to Ping Huang). 18. Kuroda T, Yasuda S, Nakashima H, Takada N, Matsuyama S, Kusakawa S, Umezawa A, Matsuyama A, Kawamata S, Sato Y. Identifcation of a gene encoding slow skeletal muscle troponin T as a novel marker for immor- Publisher’s Note talization of retinal pigment epithelial cells. Sci Rep. 2017;7(1):8163. Springer Nature remains neutral with regard to jurisdictional claims in pub- 19. Davidson B, Abeler VM, Forsund M, Holth A, Yang Y, Kobayashi Y, Chen L, lished maps and institutional afliations. Kristensen GB, Shih Ie M, Wang TL. Gene expression signatures of primary and metastatic uterine leiomyosarcoma. Hum Pathol. 2014;45(4):691–700. Received: 10 August 2018 Accepted: 18 December 2018 20. Shi Y, Zhao Y, Zhang Y, AiErken N, Shao N, Ye R, Lin Y, Wang S. TNNT1 facilitates proliferation of breast cancer cells by promoting G1/S phase transition. Life Sci. 2018;208:161–6. 21. Li S, Wang L, Zhao Q, Liu Y, He L, Xu Q, Sun X, Teng L, Cheng H, Ke Y. SHP2 positively regulates TGFbeta1-induced epithelial–mesenchymal References transition modulated by its novel interacting protein Hook1. J Biol Chem. 1. Adsay NV, Bagci P, Tajiri T, Oliva I, Ohike N, Balci S, Gonzalez RS, Basturk O, 2014;289(49):34152–60. Jang KT, Roa JC. Pathologic staging of pancreatic, ampullary, biliary, and 22. Chan A, Prassas I, Dimitromanolakis A, Brand RE, Serra S, Diamandis EP, gallbladder cancers: pitfalls and practical limitations of the current AJCC/ Blasutig IM. Validation of biomarkers that complement CA19.9 in detect- UICC TNM staging system and opportunities for improvement. Semin ing early pancreatic cancer. Clin Cancer Res. 2014;20(22):5787–95. Diagn Pathol. 2012;29(3):127–41. 23. Ikeura T, Miyoshi H, Uchida K, Fukui T, Shimatani M, Fukui Y, Sumimoto K, 2. Von Hof DD, Ervin T, Arena FP, Chiorean EG, Infante J, Moore M, Seay Matsushita M, Takaoka M, Okazaki K. Relationship between autoimmune T, Tjulandin SA, Ma WW, Saleh MN, et al. Increased survival in pan- pancreatitis and pancreatic cancer: a single-center experience. Pancrea- creatic cancer with nab-paclitaxel plus gemcitabine. N Engl J Med. tology. 2014;14(5):373–9. 2013;369(18):1691–703. 24. Weissmueller S, Manchado E, Saborowski M, Morris JP, Wagenblast E, 3. Dunne RF, Hezel AF. Genetics and biology of pancreatic ductal adenocar- Davis CA, Moon SH, Pfster NT, Tschaharganeh DF, Kitzing T, et al. Mutant cinoma. Hematol Oncol Clin North Am. 2015;29(4):595–608. p53 drives pancreatic cancer metastasis through cell-autonomous PDGF 4. Bailey P, Chang DK, Nones K, Johns AL, Patch AM, Gingras MC, Miller DK, receptor beta signaling. Cell. 2014;157(2):382–94. Christ AN, Bruxner TJ, Quinn MC, et al. Genomic analyses identify molecu- 25. DiGiuseppe JA, Hruban RH, Goodman SN, Polak M, van den Berg FM, lar subtypes of pancreatic cancer. Nature. 2016;531(7592):47–52. Allison DC, Cameron JL, Oferhaus GJ. Overexpression of p53 protein in 5. Mirus JE, Zhang Y, Hollingsworth MA, Solan JL, Lampe PD, Hingorani SR. adenocarcinoma of the pancreas. Am J Clin Pathol. 1994;101(6):684–8. Spatiotemporal proteomic analyses during pancreas cancer progression 26. Matera R, Saif MW. New therapeutic directions for advanced pancreatic identifes serine/threonine stress kinase 4 (STK4) as a novel candidate bio- cancer: cell cycle inhibitors, stromal modifers and conjugated therapies. marker for early stage disease. Mol Cell Proteomics. 2014;13(12):3484–96. Expert Opin Emerg Drugs. 2017;22(3):223–33. 6. Jones S, Zhang X, Parsons DW, Lin JC, Leary RJ, Angenendt P, Mankoo 27. Miyashita T, Tajima H, Makino I, Nakagawara H, Kitagawa H, Fushida S, P, Carter H, Kamiyama H, Jimeno A, et al. Core signaling pathways in Harmon JW, Ohta T. Metastasis-promoting role of extravasated platelet human pancreatic cancers revealed by global genomic analyses. Science. activation in tumor. J Surg Res. 2015;193(1):289–94. 2008;321(5897):1801–6. 28. Mezouar S, Darbousset R, Dignat-George F, Panicot-Dubois L, Dubois 7. Janky R, Binda MM, Allemeersch J, Van den Broeck A, Govaere O, Swinnen C. Inhibition of platelet activation prevents the P-selectin and integrin- JV, Roskams T, Aerts S, Topal B. Prognostic relevance of molecular sub- dependent accumulation of cancer cell microparticles and reduces types and master regulators in pancreatic ductal adenocarcinoma. BMC tumor growth and metastasis in vivo. Int J Cancer. 2015;136(2):462–75. Cancer. 2016;16:632. 29. Chen HX, Lan X, Liu MG, Zhou B, Wang BL, Chen P. Direct TGF-beta 1 8. Ernst J, Bar-Joseph Z. STEM: a tool for the analysis of short time series signaling between activated platelets and pancreatic cancer cells primes gene expression data. BMC Bioinform. 2006;7:191. cisplatin insensitivity. Cell Biol Int. 2013;37(5):478–84. 9. Arakawa K, Kono N, Yamada Y, Mori H, Tomita M. KEGG-based pathway 30. Duan W, Ma J, Ma Q, Xu Q, Lei J, Han L, Li X, Wang Z, Wu Z, Lv S, et al. visualization tool for complex omics data. Silico Biol. 2005;5(4):419–23. The activation of beta1-integrin by type I collagen coupling with the 10. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis hedgehog pathway promotes the epithelial–mesenchymal transition in AP, Dolinski K, Dwight SS, Eppig JT, et al. Gene ontology: tool for the pancreatic cancer. Curr Cancer Drug Targets. 2014;14(5):446–57. unifcation of biology. The Gene Ontology Consortium. Nat Genet. 31. Masugi Y, Yamazaki K, Emoto K, Efendi K, Tsujikawa H, Kitago M, Itano 2000;25(1):25–9. O, Kitagawa Y, Sakamoto M. Upregulation of integrin beta 4 promotes 11. Dennis G Jr, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki epithelial–mesenchymal transition and is a novel prognostic marker in RA. DAVID: database for annotation, visualization, and integrated discov- pancreatic ductal adenocarcinoma. Lab Invest. 2015;95(3):308–19. ery. Genome Biol. 2003;4(5):P3. 32. Meng QC, Wang HC, Song ZL, Shan ZZ, Yuan Z, Zheng Q, Huang XY. Over- 12. Franceschini A, Szklarczyk D, Frankild S, Kuhn M, Simonovic M, Roth A, expression of NDC80 is correlated with prognosis of pancreatic cancer Lin J, Minguez P, Bork P, von Mering C, et al. STRING v91: protein–protein and regulates cell proliferation. Am J Cancer Res. 2015;5(5):1730–40. Pan et al. Cancer Cell Int (2018) 18:214 Page 18 of 18

33. Li H, Zhang H, Wang Y. Centromere protein U facilitates metastasis of 36. Wang F, Li H, Yan XG, Zhou ZW, Yi ZG, He ZX, Pan ST, Yang YX, Wang ZZ, ovarian cancer cells by targeting high mobility group box 2 expression. Zhang X, et al. Alisertib induces cell cycle arrest and autophagy and sup- Am J Cancer Res. 2018;8(5):835–51. presses epithelial-to-mesenchymal transition involving PI3K/Akt/mTOR 34. Winter JM, Gildea DE, Andreas JP, Gatti DM, Williams KA, Lee M, Hu Y, and sirtuin 1-mediated signaling pathways in human pancreatic cancer Zhang SY, Mullikin JC, Wolfsberg TG, et al. Mapping complex traits in a cells. Drug Des Dev Ther. 2015;9:575–601. diversity outbred F1 mouse population identifes germline modifers of 37. Kondo T, Kanai M, Kou T, Sakuma T, Mochizuki H, Kamada M, Nakatsui metastasis in human prostate cancer. Cell Syst. 2017;4(1):31–45. M, Uza N, Kodama Y, Masui T, et al. Association between homologous 35. Wang S, Liu BB, Zhang JJ, Sun W, Dai CY, Sun WY, Li QW. Centromere recombination repair gene mutations and response to oxaliplatin in protein U is a potential target for gene therapy of human bladder cancer. pancreatic cancer. Oncotarget. 2018;9(28):19817–25. Oncol Rep. 2017;38(2):735–44.

Ready to submit your research ? Choose BMC and benefit from:

• fast, convenient online submission • thorough peer review by experienced researchers in your field • rapid publication on acceptance • support for research data, including large and complex data types • gold Open Access which fosters wider collaboration and increased citations • maximum visibility for your research: over 100M website views per year

At BMC, research is always in progress.

Learn more biomedcentral.com/submissions