Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Development of a Gene Risk Signature for Patients of Pancreatic Cancer

Development of a Gene Risk Signature for Patients of Pancreatic Cancer Hindawi Journal of Healthcare Engineering Volume 2022, Article ID 4136825, 8 pages https://doi.org/10.1155/2022/4136825 Research Article Development of a Gene Risk Signature for Patients of Pancreatic Cancer 1,2 3 4 1 1 1 Tao Liu , Long Chen, Guili Gao, Xing Liang, Junfeng Peng, Minghui Zheng, 1 2 1 Judong Li, Yongqiang Ye , and Chenghao Shao Department of Pancreatic-biliary Surgery, Changzheng Hospital, Navy Medical University, Shanghai, China Department of Hepatobiliary Surgery, Heze Municipal Hospital, No. 2888, Caozhou Road, Mudan District, Heze 274000, Shandong, China Department of Gastrointestinal Surgery, Heze Municipal Hospital, No. 2888, Caozhou Road, Mudan District, Heze 274000, Shandong, China DepartmentofCardiology,HezeMunicipalHospital,No.2888,CaozhouRoad,MudanDistrict,Heze274000,Shandong,China Correspondence should be addressed to Yongqiang Ye; qe2420404@163.com and Chenghao Shao; gq0042402@163.com Received 16 September 2021; Accepted 25 October 2021; Published 7 January 2022 Academic Editor: Kalidoss Rajakani Copyright © 2022 Tao Liu et al. -is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Background. Pancreatic cancer is a highly malignant solid tumor with a high lethality rate, but there is a lack of clinical biomarkers that can assess patient prognosis to optimize treatment. Methods. Gene-expression datasets of pancreatic cancer tissues and normal pancreatic tissues were obtained from the GEO database, and differentially expressed genes analysis and WGCNA analysis were performed after merging and normalizing the datasets. Univariate Cox regression analysis and Lasso Cox regression analysis were used to screen the prognosis-related genes in the modules with the strongest association with pancreatic cancer and construct risk signatures. -e performance of the risk signature was subsequently validated by Kaplan–Meier curves, receiver operating characteristic (ROC), and univariate and multivariate Cox analyses. Result. A three-gene risk signature containing CDKN2A, BRCA1, and UBL3 was established. Based on KM curves, ROC curves, and univariate and multivariate Cox regression analyses in the TRAIN cohort and TEST cohort, it was suggested that the three-gene risk signature had better performance in predicting overall survival. Conclusion. -is study identifies a three-gene risk signature, constructs a nomogram that can be used to predict pancreatic cancer prognosis, and identifies pathways that may be associated with pancreatic cancer prognosis. radiotherapy, chemotherapy, targeted therapy, and immu- 1. Introduction notherapy have been applied in the clinical treatment of Pancreatic cancer is a highly malignant solid tumor [1], and pancreatic cancer with some success. However, for indi- its incidence and mortality rates continue to increase [2]. vidual patients, a model that can effectively predict prog- -e most common symptoms in patients with pancreatic nosis is still needed to guide clinical selection of treatment. cancer are abdominal pain, anorexia, fatigue, and weight loss -e development of high-throughput sequencing technol- [3]; pancreatic cancer lacks specific biomarkers [4], and the ogy has made it possible to discover prognosis-related main serum markers commonly used today are carci- biomarkers. noembryonic antigen and carbohydrate antigen 19-9; Weighted gene co-expression network analysis however, their sensitivity is not ideal [3]. Surgery is the most (WGCNA) has been used to detect correlations between important approach in the treatment of pancreatic cancer. gene modules consisting of highly correlated gene clusters Due to atypical symptoms and the lack of effective screening and specific clinical features [5] and has been widely used to tools, many patients have progressed to an unresectable state identify gene modules associated with clinical features of at the time of diagnosis. With the development of research, various cancers. In the present study, we identified gene 2 Journal of Healthcare Engineering drawn through “rms” package [21] and “regplot” package modules highly correlated with pancreatic cancer tissue by WGCNA. In addition to this, we further identified genes [22] to examine the accuracy of the nomogram by mea- suring the performance of the nomogram by the C-index. associated with prognosis by univariate Cox regression analysis and Lasso Cox regression analysis. Calibration curves at 1, 3, and 5 years survival. Diagonal lines are used as a reference for best prediction. -e R package “timeROC” was used by graph receiver operating 2. Materials and Methods characteristic curves (ROC) to determine the prognostic 2.1. Gene Expression Dataset Collection and Processing. performance of the gene signature and nomogram. Download datasets related to pancreatic cancer gene ex- pression from GEO [6] (https://www.ncbi.nlm.nih.gov/geo/ 3. Results ). -e selection criteria in this article are (1) pancreatic 3.1. Differentially Expressed Genes’ (DEGs) Identification. cancer samples and normal samples were obtained from -e GSE15471, GSE16515, GSE28735, and GSE57495 human samples; (2) the training and validation datasets datasets were merged and normalized by the R package needed to contain survival data; (3) using microarray gene- “sva” [12]. Subsequent differentially expressed gene expression technology or RNA-Seq technology. Datasets analysis using the “limma” package [13] identified 77 GSE15471 [7], GSE16515 [8], GSE28735 [9], and GSE57495 DEGs containing 52 upregulated and 25 downregulated [10] were selected for differential analysis and weighted gene genes. We next performed GO and KEGG enrichment co-expression analysis datasets, in which the GSE28735 analysis of differential genes and plotted the circles. -e dataset containing survival data was used to construct the GO analysis (Figure 1(c)) of their biological process (BP) prognostic gene signature; GSE78229 [11] was selected as the was mainly enriched in extracellular structure organi- external validation dataset. Using the “sva” package [12] ofR zation and extracellular matrix organization; the cellular package, the GSE15471, GSE16515, GSE28735, and component (CC) was mainly enriched in proteinaceous GSE57495 datasets are merged and normalized. extracellular matrix and extracellular matrix; the mo- lecular function (MF) was mainly involved in extracel- 2.2.DifferentiallyExpressedGeneAnalysisandWeightedGene lular matrix structural constituent and platelet-derived Co-Expression Analysis (WGCNA). Identification of differ- growth factor binding. However, no pathways were entially expressed genes (DEGs) by “limma” [13] package in enriched in KEGG analysis. R, setting |log Fold change (logFC)|≥ 1 and adjusted p< 0.05 as standard. And we used “ggplot2” package [14] and 3.2.WeightedGeneCo-ExpressionNetworkConstructionand “pheatmap” package [15] to plot heatmap and volcano map Key Module Identification. Weighted gene co-expression of DEGs. GO and KEGG enrichment analysis of the dif- networks of GSE15471, GSE16515, GSE28735, and ferential genes is carried out by R package “clusterProfiler” GSE57495 were constructed by the “WGCNA” package in R [16] and “GOplot” [17]. (version 4.0). -e samples were clustered, and the sample GSE15471, GSE16515, GSE28735, and GSE57495 data clustering tree was drawn after removing the outliers were merged, and weighted gene co-expression analysis was (Figure 2(a)). We chose β � 7 (R2 � 0.9) to construct the used to identify co-expressed gene modules using the scale-free network (Figure 2(b)). Eight co-expression “WGCNA” package [5] of R. GO and KEGG analysis was modules were finally identified (which contained a grey then applied to the genes within the module with the highest module composed of genes that could not be categorized) correlation to tumorigenesis. (Figures 2(c) and 2(d)). Next, module-clinical feature cor- relation heat maps were drawn to assess the correlation 2.3. Construction of Risk Signature. -e univariate Cox re- between modules and clinical features (tumor vs. normal). gression and Lasso regression analyses of genes within the -e brown module had the strongest correlation with tumor module were performed by the “survival” package [18] and tissue (r � 0.53 and p � 7e − 22). -erefore, the brown “glmnet” package [19] to screen for prognosis-related genes module was selected as the key module for further analysis. within the module and construct a risk signature. Kaplan–Meier analysis was used to examine the survival 3.3. Construction of a Multigene Signature. Univariate Cox outcomes of the high-risk and low-risk groups, and the regression analysis was performed on 166 genes within the predictive power of the risk signature was assessed using the brown module to screen 30 genes associated with survival at area under the curve (AUC) of the controlled operating p< 0.05, followed by Lasso Cox regression analysis in characteristic (ROC) curve. Prognosis-related genes were GSE28735 to calculate risk scores for pancreatic cancer subsequently calculated in relation to risk score. patients. Risk score � (CDKN2A × 0.672) + (BRCA1 × −0.142)+(UBL3 × −0.185). 2.4. Construction and Valuation of Nomogram. Patients were divided into a high-risk group (n � 21) and Evaluation of prognostic factors are important for stage, a low-risk group (n � 21) according to the median risk score. grade, and risk score in the GSE78229 dataset by uni- -ere was a significant difference in overall survival (OS) variate and multifactor cox regression analysis using the between the high- and low-risk groups (p � 1.385e − 02) “forestplot” package [20] in R. Nomogram which was (Figure 3(c)). -e areas under the curve at 1, 2, and 3 years ster Clu ge) Log (fold chan Journal of Healthcare Engineering 3 -10 -5 -1.5 0 1.5 510 (a) (b) (c) Figure 1: Differentially expressed genes’ (DEGs) identification. Heat map (a) and volcano map (b) of gene-expression profiles of pancreatic cancer tissues and normal tissues after merge of four datasets, GSE15471, GSE16515, GSE28735, and GSE57495. Differentially expressed genes were screened using |logFC|≥ 1 and adjusted using p< 0.05, with red representing upregulated genes and blue representing downregulated genes. (c) GO enrichment analysis of differentially expressed genes. Sample dendrogram and trait heatmap Scale independence Mean connectivity 1.0 18 20 1 0.5 0.0 6 -0.5 2 8 10 12 14 16 18 1 0 20 5 10 15 20 5 10 15 20 Normal Soft Threshold (power) Soft Threshold (power) Tumor (a) (b) Cluster Dendrogram Module-trait relationships 1.0 0.088 -0.19 0.41 0.1 0.48 0.19 -0.37 -0.6 0.066 -0.19 MEpink (0.7) (0.4) (0.07) (0.7) (0.03) (0.4) (0.1) (0.006) (0.8) (0.4) -0.1 -0.16 -0.05 -0.11 -0.69 0.61 -0.21 -0.26 -0.19 -0.24 MEyellow (0.7) (0.5) (0.8) (0.7) (7e-04) (0.004) (0.4) (0.3) (0.4) (0.3) 0.8 0.53 0.25 0.47 0.17 0.047 -0.16 -0.17 -0.39 -0.28 -0.47 MEbrown (0.02) (0.3) (0.04) (0.5) (0.8) (0.5) (0.5) (0.09) (0.2) (0.04) 0.5 -0.1 -0.11 0.68 0.62 -0.049 -0.054 -0.25 -0.24 -0.23 -0.25 MEpurple (0.7) (0.6) (0.001) (0.004) (0.8) (0.8) (0.3) (0.3) (0.3) (0.3) -0.07 -0.18 -0.065 -0.19 -0.26 -0.32 0.73 0.52 -0.027 -0.15 MEgreen 0.6 (0.8) (0.4) (0.8) (0.4) (0.3) (0.2) (2e-04) (0.02) (0.9) (0.5) 0.24 0.17 0.28 0.2 -0.34 -0.35 0.4 0.31 -0.44 -0.47 MEmagenta (0.3) (0.5) (0.2) (0.4) (0.1) (0.1) (0.08) (0.2) (0.05) (0.04) 0.41 0.0049 0.27 -0.12 -0.37 -0.58 0.52 0.052 0.1 -0.28 MEred (0.07) (1) (0.2) (0.6) (0.1) (0.007) (0.02) (0.8) (0.7) (0.2) 0.4 0.21 -0.088 0.35 0.067 -0.14 -0.34 -0.3 -0.54 0.55 0.23 MEblue (0.4) (0.7) (0.1) (0.8) (0.6) (0.1) (0.2) (0.01) (0.01) (0.3) -0.078 -0.21 -0.084 -0.21 -0.25 -0.33 0.02 -0.11 0.74 0.52 MEturquoise (0.7) (0.4) (0.7) (0.4) (0.3) (0.1) (0.9) (0.6) (2e-04) (0.02) -0.5 -0.36 0.33 -0.42 0.32 -0.29 0.23 -0.24 0.43 -0.31 0.3 MEblack (0.1) (0.2) (0.06) (0.2) (0.2) (0.3) (0.3) (0.06) (0.2) (0.2) -0.3 0.0015 -0.25 0.056 -0.24 0.021 -0.24 0.021 0.0082 0.91 Module colors MEgreenyellow (0.2) (1) (0.3) (0.8) (0.3) (0.9) (0.3) (0.9) (1) (2e-08) -0.41 -0.44 0.34 0.35 0.062 0.21 -0.018 -0.0032 0.012 -0.11 MEgrey (0.07) (0.06) (0.1) (0.1) (0.8) (0.4) (0.9) (1) (1) (0.6) -1 (c) (d) Figure 2: Weighted gene co-expression network construction and key module identification. (a) Cluster dendrogram of pancreatic cancer samples and normal samples. (b) According to the scale-free index and the mean connectivity to screen soft threshold. (c) -e cluster dendrogram of co-expression network modules. (d) Relationships between module and trait. was 0.75, 0.893, and 0.733, respectively (Figure 3(d)). -e cohort, respectively. -e results showed that the risk score was significantly associated with OS. Multivariate cox re- prognostic and prognostic accuracy of the three-gene sig- nature was subsequently validated using GSE78229 as a test gression analysis revealed that three-gene signatures were cohort. -ere was also a significant difference in OS between independent predictors of outcome in pancreatic cancer the high-risk and low-risk groups in the test cohort patients. (p � 5.418e − 03) (Figure 3(e)), with areas under the curve at In addition, we created a prognostic nomogram to help 1, 2, and 3 years of 0.773, 0.731, and 0.741, respectively physicians predict overall patient survival in the clinic (Figure 3(f)). (Figure 4(c)). -e calibration curve (Figure 4(d)) of the -e performance of the risk signature was further nomogram and the area under the curve of ROC evaluated by univariate (Figure 4(a)) and multivariate (Figure 4(e)) showed a good concordance between predic- (Figure 4(b)) Cox regression analysis in the train and test tion and observation. Height Log P-value Scale Free Topology Model Fit, signed R 2 Mean Connectivity 4 Journal of Healthcare Engineering 29 28 25 23 13 6 1 pvalue Hazard ratio BEST1 0.001 4.417 (1.769-11.028) CCL2 0.046 1.368 (1.006-1.859) CCL24 0.013 1.492 (1.090-2.042) CD55 <0.001 1.752 (1.296-2.368) EMP3 0.025 1.780 (1.074-2.947) -5 F3 0.008 1.458 (1.105-1.922) GABBR1 0.038 0.746 (0.592-0.986) -7 -6 -5 -4 -3 -2 -1 Log Lambda HAS2 0.039 2.079 (1.037-4.171) IFITM1 0.023 1.278 (1.035-1.578) KCNJ2 0.048 3.116 (1.009-9.662) LY6E 0.030 1.420 (1.035-1.947) NFKB1 0.023 0.271 (0.088-0.836) PLAUR 0.004 1.978 (1.244-3.145) SLC4A4 <0.001 0.364 (0.200-0.662) 0.01 0.1 1 10 100 Hazard ratio (a) (b) 29 30 28 26 25 23 22 23 21 16 12 11 7 6 5 3 2 1 Survival curve (P=1.385e-02) 80 1.0 0.8 0.6 0.4 0.2 0.0 0 123 4 -7 -6 -5 -4 -3 -2 -1 Time (year) Log (λ) high risk low risk (c) (d) Time - dependent ROC curve Survival curve (P=5.418e-03) 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Time (year) False positive rate high risk AUC at 1 year:0.75 low risk AUC at 2 year:0.893 AUC at 3 year:0.733 (e) (f) Figure 3: Continued. True positive rate Survival rate Survival rate Journal of Healthcare Engineering 5 Time - dependent ROC curve 1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate AUC at 1 year:0.773 AUC at 2 year:0.731 AUC at 3 year:0.741 (g) Figure 3: Construction of a multigene signature. (a) Univariate Cox analysis of genes within the brown module, screening of prognosis- related genes, and forest mapping. (b) -e Lasso coefficient profiles of prognosis-related genes. (c) -e partial likelihood deviance is plotted against log (λ). Kaplan–Meier plot of overall survival of patients is in the high-risk and low-risk groups in the train cohort (d) and test cohort (f). ROC curves for three-gene signatures in train cohort (e) and test cohort (g). [28]. Li et al. also found that CDKN2A knockdown inhibited 4. Discussion proliferation, migration, invasion, and epithelial mesen- Pancreatic cancer is highly malignant, lacks reliable early chymal transition in glioblastoma cells [29]. Bioinformatics screening methods, and has a poor prognosis, with an ex- studies have found that CDKN2A is also associated with pected 5-year survival rate of approximately 9% [1]. breast cancer [30], prostate cancer [31], and colorectal -erefore, there is an urgent need to find biomarkers that cancer [32]. BRCA1 has been reported to play an oncogenic affect the prognosis of pancreatic cancer in clinical treat- role in bladder cancer, with significantly lower expression ment, which will facilitate the assessment of patient prog- levels in cancer tissues than in normal tissues, and over- nosis and will help to improve the prognosis by tailoring the expression of BRCA1 reduced cell proliferation, migration, treatment to the individual patient. and colony-forming ability [33]. In contrast, in bladder Tumor development is the result of multigene interac- cancer, upregulation of BRCA1 was able to resist oxidative tions, and therefore, an increasing number of risk signatures stress, thereby promoting bladder cancer cell growth [34]. are used to predict prognosis [23–26]. In this study, we Pitt et al. also found mutations in BRCA1 in thyroid cancer. proposed a three-gene (CDKN2A, BRCA1, and UBL3) risk In addition to this, bioinformatics studies have found that signature by WGCNA and Lasso Cox regression analysis for BRCA1 is also associated with ovarian [35], colorectal [36], predicting overall survival in pancreatic cancer patients, with and gastric [37] cancers. Consistent with our speculation, statistically significant differences in overall survival between Zhao et al. found that, in NSCLC, UBL3 acts as a tumor high-and low-risk groups in the train cohort and test cohort. suppressor gene to inhibit cancer cell proliferation [38]. We then evaluated the prognostic performance of risk GSEA analysis revealed differences in 2 key signaling signature with the AUC of ROC, and the results showed that pathways between high- and low-risk groups. Base excision the risk signature could predict overall survival of pancreatic repair (BER) removes endogenous DNA damages that occur cancer patients accurately. Subsequent univariate and at all times in human cells, and its defects are associated with multivariate Cox analyses showed that the risk score could tumorigenesis [39], but cancer cells are also able to tolerate predict prognosis as an independent prognostic factor. In oxidative stress through increased BER activity, and tar- addition, we combined clinical characteristics to construct geting BER can improve the efficacy of radio/chemotherapy nomogram that can be used in the clinic to guide person- [40]. Our results show that the BER pathway is enriched in alized treatment. the high-risk group, suggesting that the BER pathway is Among the risk genes we identified, CDKN2A (cyto- active in high-risk patients, possibly leading to shorter skeleton-associated protein 2-like) was significantly highly survival by affecting their sensitivity to clinical treatment. An expressed in the high-risk group and positively correlated increasing number of studies have found that abnormal with risk score; BRCA1 (glutathione S-transferase Mu 5) and metabolism affects patient prognosis [41, 42], and the en- UBL3 (Ubiquitin-like 3) were significantly down regulated richment of propanoate metabolism pathway in the low-risk in the high-risk group and negatively correlated with the risk group suggests that the risk signature may affect patient score. CDKN2A has been reported to promote lung ade- prognosis through tumor metabolism. nocarcinoma invasion and is correlated with poor prognosis In summary, our study identified a 3-gene risk signature [27]. Monteverde et al. found that CDKN2A could promote for predicting prognosis, and the value of this risk signature nonsmall cell lung cancer (NSCLC) progression by regu- was validated in an external test cohort. By combining this lating transcriptional elongation, and targeting CDKN2A risk signature with clinical tumor pathology staging, a visual could enhance therapeutic response in patients with NSCLC nomogram was created to facilitate the prediction of survival True positive rate 6 Journal of Healthcare Engineering pvalue Hazard ratio pvalue Hazard ratio grade 0.046 1.702 (1.009-2.871) grade 0.113 1.586 (0.896-2.808) stage 0.477 0.682 (0.237-1.960) stage 0.066 0.350 (0.114-1.070) riskScore 0.008 4.028 (1.433-11.325) riskScore 0.013 3.974 (1.340-11.783) 0.25 0.50 1.0 2.0 4.0 8.0 0 12 3456789 10 1112 Hazard ratio Hazard ratio (a) (b) Points 1.0 0 102030405060708090 100 stage* 0.8 low risk** 0.6 high G2 grade G4 0.4 G1 G3 Total points 0.2 160 180 200 220 240 260 280 0.0 0.859 Pr (futime<3) 0.006 0.035 0.2 0.7 0.998 0.0 0.2 0.4 0.6 0.8 1.0 0.746 Pr (futime<2) Nomogram-predicted OS (%) 0.006 0.035 0.2 0.7 0.998 0.451 Pr (futime<1) 1-year 0.002 0.015 0.08 0.5 0.98 2-year 3-year (c) (d) 1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 1 – specificity AUC at 1 years: 0.769 AUC at 2 years: 0.834 AUC at 3 years: 0.865 (e) Figure 4: Evaluation of the predictive value of three-gene signatures and the creation of nomogram. (a) Test cohort risk signature univariate. (b) Multivariate Cox regression analysis forest plot. (c) Nomogram. (d) Calibration plot. (e) ROC curve for test cohort. Sensitivity Observed OS (%) Journal of Healthcare Engineering 7 [12] L. C. Chu, M. G. Goggins, and E. K. Fishman, “Diagnosis and outcomes. We further analyzed risk signature-related sig- detection of pancreatic cancer,” ?e Cancer Journal, vol. 23, naling pathways and risk genes’ expression and survival no. 6, pp. 333–342, 2017. analysis in a variety of cancers, providing a research di- [13] I. Diboun, L. Wernisch, C. A. Orengo, and M. Koltzenburg, rection for exploring prognosis-related genes in in vivo and “Microarray analysis after RNA amplification can detect in vitro experiments. We further analyzed risk signature- pronounced differences in gene expression using limma,” related signaling pathways and risk gene expression and BMC Genomics, vol. 7, no. 1, p. 252, 2006. survival analysis in a variety of cancers, providing a research [14] L. Wilkinson, “ggplot2: elegant graphics for data analysis by direction for exploring prognosis-related genes in cytolog- WICKHAM, H,” Biometrics, vol. 67, no. 2, pp. 678-679, 2011. ical and animal experiments. [15] H. C. Ho, A. Knudby, and W. A. Huang, “Spatial framework to map heat health risks at multiple scales,” International Journal of Environmental Research and Public Health, vol. 12, Data Availability no. 12, pp. 16110–23, 2015. [16] G. Yu, L.-G. Wang, Y. Han, and Q.-Y. He, “clusterProfiler: an -e analyzed datasets generated during the present study are R package for comparing biological themes among gene available from the corresponding author upon reasonable clusters,” OMICS: A Journal of Integrative Biology, vol. 16, request. no. 5, pp. 284–287, 2012. [17] W. Walter, F. Sanchez-Cabo, ´ and M. Ricote, “GOplot: an R Conflicts of Interest package for visually combining expression data with func- tional analysis: Figure 1,” Bioinformatics, vol. 31, no. 17, -e authors declare that they have no conflicts of interest in pp. 2912–2914, 2015. the present study. [18] C. S. Groeneveld, V. S. Chagas, S. J. M. Jones et al., “RTNsurvival: an R/Bioconductor package for regulatory network survival analysis,” Bioinformatics, vol. 35, no. 21, References pp. 4488-4489, 2019. [19] N. Simon, J. Friedman, T. Hastie, and R. Tibshirani, “Reg- [1] R. L. Siegel, K. D. Miller, and A. Jemal, “Cancer statistics, ularization paths for cox’s proportional hazards model via 2020,” CA: A Cancer Journal for Clinicians, vol. 70, no. 1, coordinate descent,” Journal of Statistical Software, vol. 39, pp. 7–30, 2020. no. 5, pp. 1–13, 2011. [2] L. Rahib, B. D. Smith, R. Aizenberg, A. B. Rosenzweig, [20] J. Kim, K. S. Kaufman, and H. Bang, “Graphing ratio measures J. M. Fleshman, and L. M. Matrisian, “Projecting cancer in- on forest plot,” Journal of the American College of Cardiology, cidence and deaths to 2030: the unexpected burden of thyroid, vol. 71, no. 5, pp. 585-586, 2018. liver, and pancreas cancers in the United States,” Cancer [21] B. J. Wells, K. M. Lenoir, J. F. Diaz-Garelli et al., “Predicting Research, vol. 74, no. 11, pp. 2913–2921, 2014. current glycated hemoglobin values in adults: development of [3] H. Zhu, T. Li, Y. Du, and M. Li, “Pancreatic cancer: challenges an algorithm from the electronic health record,” JMIR and opportunities,”BMCMedicine, vol. 16, no. 1, p. 214, 2018. Medical Informatics, vol. 6, no. 4, p. e10780, 2018. [4] L. Zhang, S. Sanagapalli, and A. Stoita, “Challenges in diag- [22] R. Marshall, Regplot: Enhanced Regression Nomogram Plot, nosis of pancreatic cancer,” World Journal of Gastroenterol- ogy, vol. 24, no. 19, pp. 2047–2060, 2018. [23] Z. Meng, D. Ren, K. Zhang, J. Zhao, X. Jin, and H. Wu, “Using [5] P. Langfelder and S. Horvath, “WGCNA: an R package for ESTIMATE algorithm to establish an 8-mRNA signature weighted correlation network analysis,” BMC Bioinformatics, prognosis prediction system and identify immunocyte infil- vol. 9, no. 1, p. 559, 2008. tration-related genes in Pancreatic adenocarcinoma,” Aging, [6] E. Clough and T. Barrett, “-e gene expression omnibus vol. 12, no. 6, pp. 5048–5070, 2020. database,” Methods in Molecular Biology, vol. 1418, pp. 93– [24] X.-H. Shi, X. Li, H. Zhang et al., “A five-microRNA signature 110, 2016. for survival prognosis in pancreatic adenocarcinoma based on [7] L. Badea, V. Herlea, S. O. Dima, T. Dumitrascu, and TCGA data,” Scientific Reports, vol. 8, no. 1, p. 7638, 2018. I. Popescu, “Combined gene expression analysis of whole- [25] G.-W. Huang, Y.-J. Xue, Z.-Y. Wu et al., “A three-lncRNA tissue and microdissected pancreatic ductal adenocarcinoma signature predicts overall survival and disease-free survival in identifies genes specifically overexpressed in tumor epithelia,” patients with esophageal squamous cell carcinoma,” BMC Hepato-gastroenterology, vol. 55, no. 88, pp. 2016–2027, 2008. Cancer, vol. 18, no. 1, p. 147, 2018. [8] H. Pei, L. Li, B. L. Fridley et al., “FKBP51 affects cancer cell [26] Z. Meng, Q. Yuan, J. Zhao et al., “-e m6A-related mRNA response to chemotherapy by negatively regulating Akt,” signature predicts the prognosis of pancreatic cancer pa- Cancer Cell, vol. 16, no. 3, pp. 259–266, 2009. tients,” Molecular ?erapy-Oncolytics, vol. 17, pp. 460–470, [9] G. Zhang, A. Schetter, P. He et al., “DPEP1 inhibits tumor cell invasiveness, enhances chemosensitivity and predicts clinical [27] G. Xiong, L. Li, X. Chen et al., “Up-regulation of CKAP2L outcome in pancreatic ductal adenocarcinoma,” PLoS One, vol. 7, no. 2, Article ID e31507, 2012. expression promotes lung adenocarcinoma invasion and is associated with poor prognosis,” OncoTargets and ?erapy, [10] D.-T. Chen, A. H. Davis-Yadley, P.-Y. Huang et al., “Prog- nostic fifteen-gene signature for early stage pancreatic ductal vol. 12, pp. 1171–1180, 2019. [28] T. Monteverde, S. Sahoo, M. La Montagna et al., “CKAP2L adenocarcinoma,” PLoS One, vol. 10, no. 8, Article ID e0133562, 2015. promotes non-small cell lung cancer progression through regulation of transcription elongation,” Cancer Research, [11] J. Wang, S. Yang, P. He et al., “Endothelial nitric oxide synthase traffic inducer (NOSTRIN) is a negative regulator of vol. 81, no. 7, pp. 1719–1731, 2021. [29] Y. F. Li, W. C. Tsai, C. H. Chou et al., “CDKN2A knockdown disease aggressiveness in pancreatic cancer,” Clinical Cancer Research, vol. 22, no. 24, pp. 5992–6001, 2016. exerts antitumor effects by increasing miR-4496 in 8 Journal of Healthcare Engineering glioblastoma cell lines,” International Journal of Molecular Sciences, vol. 22, no. 1, 2020. [30] L. Fan, J. Hou, and G. Qin, “Prediction of disease genes based on stage-specific gene regulatory networks in breast cancer,” Frontiers in Genetics, vol. 12, Article ID 717557, 2021. [31] T. Feng, D. Wei, Q. Li et al., “Four novel prognostic genes related to prostate cancer identified using Co-expression structure network analysis,” Frontiers in Genetics, vol. 12, Article ID 584164, 2021. [32] Y. Zhang, J. Luo, Z. Liu et al., “Identification of hub genes in colorectal cancer based on weighted gene co-expression network analysis,”Bioscience Reports, vol. 41, no. 7, Article ID BSR20211280, 2021. [33] Y. C. Jou, S. C. Wang, Y. C. Dia et al., “Anti-cancer effects and tumor marker role of glutathione S-transferase Mu 5 in human bladder cancer,” International Journal of Molecular Sciences, vol. 22, no. 6, Article ID 3056, 2021. [34] T. Li, D. Jiang, and K. Wu, “p62 promotes bladder cancer cell growth by activating KEAP1/NRF2-dependent antioxidative response,”CancerScience, vol. 111, no. 4, pp. 1156–1164, 2020. [35] Y. Wang, L. Lei, Y. G. Chi, L. B. Liu, and B. P. Yang, “A comprehensive understanding of ovarian carcinoma survival prognosis by novel biomarkers,”EuropeanReviewforMedical and Pharmacological Sciences, vol. 23, no. 19, pp. 8257–8264, [36] Y. L. Sun, Y. Zhang, Y. C. Guo, Z. H. Yang, and Y. C. Xu, “A prognostic model based on six metabolism-related genes in colorectal cancer,” BioMed Research International, vol. 2020, Article ID 5974350, 2020. [37] Y. Chen, B. Li, J. Wang et al., “Identification and verification of the prognostic value of the glutathione S-transferase Mu genes in gastric cancer,”OncologyLetters, vol. 20, no. 4, p. 100, [38] X. Zhao, Y. Zhou, Q. Hu et al., “Identification of a potential tumor suppressor gene, UBL3, in non-small cell lung cancer,” Cancer Biology and Medicine, vol. 17, no. 1, pp. 76–87, 2020. [39] S. S. Wallace, D. L. Murphy, and J. B. Sweasy, “Base excision repair and cancer,”CancerLetters, vol. 327, no. 1-2, pp. 73–89, [40] G. J. Grundy and J. L. Parsons, “Base excision repair and its implications to cancer therapy,” Essays in Biochemistry, vol. 64, no. 5, pp. 831–843, 2020. [41] T. E. Oni, G. Biffi, L. A. Baker et al., “SOAT1 promotes mevalonate pathway dependency in pancreatic cancer,” ?e Journal of Experimental Medicine, vol. 217, no. 9, 2020. [42] C. Li, X. Li, G. Li et al., “Identification of a prognosis‑asso- ciated signature associated with energy metabolism in tri- ple‑negative breast cancer,” Oncology Reports, vol. 44, no. 3, pp. 819–837, 2020. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Healthcare Engineering Hindawi Publishing Corporation

Development of a Gene Risk Signature for Patients of Pancreatic Cancer

Loading next page...
 
/lp/hindawi-publishing-corporation/development-of-a-gene-risk-signature-for-patients-of-pancreatic-cancer-6pKWwUanSx

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2022 Tao Liu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
2040-2295
eISSN
2040-2309
DOI
10.1155/2022/4136825
Publisher site
See Article on Publisher Site

Abstract

Hindawi Journal of Healthcare Engineering Volume 2022, Article ID 4136825, 8 pages https://doi.org/10.1155/2022/4136825 Research Article Development of a Gene Risk Signature for Patients of Pancreatic Cancer 1,2 3 4 1 1 1 Tao Liu , Long Chen, Guili Gao, Xing Liang, Junfeng Peng, Minghui Zheng, 1 2 1 Judong Li, Yongqiang Ye , and Chenghao Shao Department of Pancreatic-biliary Surgery, Changzheng Hospital, Navy Medical University, Shanghai, China Department of Hepatobiliary Surgery, Heze Municipal Hospital, No. 2888, Caozhou Road, Mudan District, Heze 274000, Shandong, China Department of Gastrointestinal Surgery, Heze Municipal Hospital, No. 2888, Caozhou Road, Mudan District, Heze 274000, Shandong, China DepartmentofCardiology,HezeMunicipalHospital,No.2888,CaozhouRoad,MudanDistrict,Heze274000,Shandong,China Correspondence should be addressed to Yongqiang Ye; qe2420404@163.com and Chenghao Shao; gq0042402@163.com Received 16 September 2021; Accepted 25 October 2021; Published 7 January 2022 Academic Editor: Kalidoss Rajakani Copyright © 2022 Tao Liu et al. -is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Background. Pancreatic cancer is a highly malignant solid tumor with a high lethality rate, but there is a lack of clinical biomarkers that can assess patient prognosis to optimize treatment. Methods. Gene-expression datasets of pancreatic cancer tissues and normal pancreatic tissues were obtained from the GEO database, and differentially expressed genes analysis and WGCNA analysis were performed after merging and normalizing the datasets. Univariate Cox regression analysis and Lasso Cox regression analysis were used to screen the prognosis-related genes in the modules with the strongest association with pancreatic cancer and construct risk signatures. -e performance of the risk signature was subsequently validated by Kaplan–Meier curves, receiver operating characteristic (ROC), and univariate and multivariate Cox analyses. Result. A three-gene risk signature containing CDKN2A, BRCA1, and UBL3 was established. Based on KM curves, ROC curves, and univariate and multivariate Cox regression analyses in the TRAIN cohort and TEST cohort, it was suggested that the three-gene risk signature had better performance in predicting overall survival. Conclusion. -is study identifies a three-gene risk signature, constructs a nomogram that can be used to predict pancreatic cancer prognosis, and identifies pathways that may be associated with pancreatic cancer prognosis. radiotherapy, chemotherapy, targeted therapy, and immu- 1. Introduction notherapy have been applied in the clinical treatment of Pancreatic cancer is a highly malignant solid tumor [1], and pancreatic cancer with some success. However, for indi- its incidence and mortality rates continue to increase [2]. vidual patients, a model that can effectively predict prog- -e most common symptoms in patients with pancreatic nosis is still needed to guide clinical selection of treatment. cancer are abdominal pain, anorexia, fatigue, and weight loss -e development of high-throughput sequencing technol- [3]; pancreatic cancer lacks specific biomarkers [4], and the ogy has made it possible to discover prognosis-related main serum markers commonly used today are carci- biomarkers. noembryonic antigen and carbohydrate antigen 19-9; Weighted gene co-expression network analysis however, their sensitivity is not ideal [3]. Surgery is the most (WGCNA) has been used to detect correlations between important approach in the treatment of pancreatic cancer. gene modules consisting of highly correlated gene clusters Due to atypical symptoms and the lack of effective screening and specific clinical features [5] and has been widely used to tools, many patients have progressed to an unresectable state identify gene modules associated with clinical features of at the time of diagnosis. With the development of research, various cancers. In the present study, we identified gene 2 Journal of Healthcare Engineering drawn through “rms” package [21] and “regplot” package modules highly correlated with pancreatic cancer tissue by WGCNA. In addition to this, we further identified genes [22] to examine the accuracy of the nomogram by mea- suring the performance of the nomogram by the C-index. associated with prognosis by univariate Cox regression analysis and Lasso Cox regression analysis. Calibration curves at 1, 3, and 5 years survival. Diagonal lines are used as a reference for best prediction. -e R package “timeROC” was used by graph receiver operating 2. Materials and Methods characteristic curves (ROC) to determine the prognostic 2.1. Gene Expression Dataset Collection and Processing. performance of the gene signature and nomogram. Download datasets related to pancreatic cancer gene ex- pression from GEO [6] (https://www.ncbi.nlm.nih.gov/geo/ 3. Results ). -e selection criteria in this article are (1) pancreatic 3.1. Differentially Expressed Genes’ (DEGs) Identification. cancer samples and normal samples were obtained from -e GSE15471, GSE16515, GSE28735, and GSE57495 human samples; (2) the training and validation datasets datasets were merged and normalized by the R package needed to contain survival data; (3) using microarray gene- “sva” [12]. Subsequent differentially expressed gene expression technology or RNA-Seq technology. Datasets analysis using the “limma” package [13] identified 77 GSE15471 [7], GSE16515 [8], GSE28735 [9], and GSE57495 DEGs containing 52 upregulated and 25 downregulated [10] were selected for differential analysis and weighted gene genes. We next performed GO and KEGG enrichment co-expression analysis datasets, in which the GSE28735 analysis of differential genes and plotted the circles. -e dataset containing survival data was used to construct the GO analysis (Figure 1(c)) of their biological process (BP) prognostic gene signature; GSE78229 [11] was selected as the was mainly enriched in extracellular structure organi- external validation dataset. Using the “sva” package [12] ofR zation and extracellular matrix organization; the cellular package, the GSE15471, GSE16515, GSE28735, and component (CC) was mainly enriched in proteinaceous GSE57495 datasets are merged and normalized. extracellular matrix and extracellular matrix; the mo- lecular function (MF) was mainly involved in extracel- 2.2.DifferentiallyExpressedGeneAnalysisandWeightedGene lular matrix structural constituent and platelet-derived Co-Expression Analysis (WGCNA). Identification of differ- growth factor binding. However, no pathways were entially expressed genes (DEGs) by “limma” [13] package in enriched in KEGG analysis. R, setting |log Fold change (logFC)|≥ 1 and adjusted p< 0.05 as standard. And we used “ggplot2” package [14] and 3.2.WeightedGeneCo-ExpressionNetworkConstructionand “pheatmap” package [15] to plot heatmap and volcano map Key Module Identification. Weighted gene co-expression of DEGs. GO and KEGG enrichment analysis of the dif- networks of GSE15471, GSE16515, GSE28735, and ferential genes is carried out by R package “clusterProfiler” GSE57495 were constructed by the “WGCNA” package in R [16] and “GOplot” [17]. (version 4.0). -e samples were clustered, and the sample GSE15471, GSE16515, GSE28735, and GSE57495 data clustering tree was drawn after removing the outliers were merged, and weighted gene co-expression analysis was (Figure 2(a)). We chose β � 7 (R2 � 0.9) to construct the used to identify co-expressed gene modules using the scale-free network (Figure 2(b)). Eight co-expression “WGCNA” package [5] of R. GO and KEGG analysis was modules were finally identified (which contained a grey then applied to the genes within the module with the highest module composed of genes that could not be categorized) correlation to tumorigenesis. (Figures 2(c) and 2(d)). Next, module-clinical feature cor- relation heat maps were drawn to assess the correlation 2.3. Construction of Risk Signature. -e univariate Cox re- between modules and clinical features (tumor vs. normal). gression and Lasso regression analyses of genes within the -e brown module had the strongest correlation with tumor module were performed by the “survival” package [18] and tissue (r � 0.53 and p � 7e − 22). -erefore, the brown “glmnet” package [19] to screen for prognosis-related genes module was selected as the key module for further analysis. within the module and construct a risk signature. Kaplan–Meier analysis was used to examine the survival 3.3. Construction of a Multigene Signature. Univariate Cox outcomes of the high-risk and low-risk groups, and the regression analysis was performed on 166 genes within the predictive power of the risk signature was assessed using the brown module to screen 30 genes associated with survival at area under the curve (AUC) of the controlled operating p< 0.05, followed by Lasso Cox regression analysis in characteristic (ROC) curve. Prognosis-related genes were GSE28735 to calculate risk scores for pancreatic cancer subsequently calculated in relation to risk score. patients. Risk score � (CDKN2A × 0.672) + (BRCA1 × −0.142)+(UBL3 × −0.185). 2.4. Construction and Valuation of Nomogram. Patients were divided into a high-risk group (n � 21) and Evaluation of prognostic factors are important for stage, a low-risk group (n � 21) according to the median risk score. grade, and risk score in the GSE78229 dataset by uni- -ere was a significant difference in overall survival (OS) variate and multifactor cox regression analysis using the between the high- and low-risk groups (p � 1.385e − 02) “forestplot” package [20] in R. Nomogram which was (Figure 3(c)). -e areas under the curve at 1, 2, and 3 years ster Clu ge) Log (fold chan Journal of Healthcare Engineering 3 -10 -5 -1.5 0 1.5 510 (a) (b) (c) Figure 1: Differentially expressed genes’ (DEGs) identification. Heat map (a) and volcano map (b) of gene-expression profiles of pancreatic cancer tissues and normal tissues after merge of four datasets, GSE15471, GSE16515, GSE28735, and GSE57495. Differentially expressed genes were screened using |logFC|≥ 1 and adjusted using p< 0.05, with red representing upregulated genes and blue representing downregulated genes. (c) GO enrichment analysis of differentially expressed genes. Sample dendrogram and trait heatmap Scale independence Mean connectivity 1.0 18 20 1 0.5 0.0 6 -0.5 2 8 10 12 14 16 18 1 0 20 5 10 15 20 5 10 15 20 Normal Soft Threshold (power) Soft Threshold (power) Tumor (a) (b) Cluster Dendrogram Module-trait relationships 1.0 0.088 -0.19 0.41 0.1 0.48 0.19 -0.37 -0.6 0.066 -0.19 MEpink (0.7) (0.4) (0.07) (0.7) (0.03) (0.4) (0.1) (0.006) (0.8) (0.4) -0.1 -0.16 -0.05 -0.11 -0.69 0.61 -0.21 -0.26 -0.19 -0.24 MEyellow (0.7) (0.5) (0.8) (0.7) (7e-04) (0.004) (0.4) (0.3) (0.4) (0.3) 0.8 0.53 0.25 0.47 0.17 0.047 -0.16 -0.17 -0.39 -0.28 -0.47 MEbrown (0.02) (0.3) (0.04) (0.5) (0.8) (0.5) (0.5) (0.09) (0.2) (0.04) 0.5 -0.1 -0.11 0.68 0.62 -0.049 -0.054 -0.25 -0.24 -0.23 -0.25 MEpurple (0.7) (0.6) (0.001) (0.004) (0.8) (0.8) (0.3) (0.3) (0.3) (0.3) -0.07 -0.18 -0.065 -0.19 -0.26 -0.32 0.73 0.52 -0.027 -0.15 MEgreen 0.6 (0.8) (0.4) (0.8) (0.4) (0.3) (0.2) (2e-04) (0.02) (0.9) (0.5) 0.24 0.17 0.28 0.2 -0.34 -0.35 0.4 0.31 -0.44 -0.47 MEmagenta (0.3) (0.5) (0.2) (0.4) (0.1) (0.1) (0.08) (0.2) (0.05) (0.04) 0.41 0.0049 0.27 -0.12 -0.37 -0.58 0.52 0.052 0.1 -0.28 MEred (0.07) (1) (0.2) (0.6) (0.1) (0.007) (0.02) (0.8) (0.7) (0.2) 0.4 0.21 -0.088 0.35 0.067 -0.14 -0.34 -0.3 -0.54 0.55 0.23 MEblue (0.4) (0.7) (0.1) (0.8) (0.6) (0.1) (0.2) (0.01) (0.01) (0.3) -0.078 -0.21 -0.084 -0.21 -0.25 -0.33 0.02 -0.11 0.74 0.52 MEturquoise (0.7) (0.4) (0.7) (0.4) (0.3) (0.1) (0.9) (0.6) (2e-04) (0.02) -0.5 -0.36 0.33 -0.42 0.32 -0.29 0.23 -0.24 0.43 -0.31 0.3 MEblack (0.1) (0.2) (0.06) (0.2) (0.2) (0.3) (0.3) (0.06) (0.2) (0.2) -0.3 0.0015 -0.25 0.056 -0.24 0.021 -0.24 0.021 0.0082 0.91 Module colors MEgreenyellow (0.2) (1) (0.3) (0.8) (0.3) (0.9) (0.3) (0.9) (1) (2e-08) -0.41 -0.44 0.34 0.35 0.062 0.21 -0.018 -0.0032 0.012 -0.11 MEgrey (0.07) (0.06) (0.1) (0.1) (0.8) (0.4) (0.9) (1) (1) (0.6) -1 (c) (d) Figure 2: Weighted gene co-expression network construction and key module identification. (a) Cluster dendrogram of pancreatic cancer samples and normal samples. (b) According to the scale-free index and the mean connectivity to screen soft threshold. (c) -e cluster dendrogram of co-expression network modules. (d) Relationships between module and trait. was 0.75, 0.893, and 0.733, respectively (Figure 3(d)). -e cohort, respectively. -e results showed that the risk score was significantly associated with OS. Multivariate cox re- prognostic and prognostic accuracy of the three-gene sig- nature was subsequently validated using GSE78229 as a test gression analysis revealed that three-gene signatures were cohort. -ere was also a significant difference in OS between independent predictors of outcome in pancreatic cancer the high-risk and low-risk groups in the test cohort patients. (p � 5.418e − 03) (Figure 3(e)), with areas under the curve at In addition, we created a prognostic nomogram to help 1, 2, and 3 years of 0.773, 0.731, and 0.741, respectively physicians predict overall patient survival in the clinic (Figure 3(f)). (Figure 4(c)). -e calibration curve (Figure 4(d)) of the -e performance of the risk signature was further nomogram and the area under the curve of ROC evaluated by univariate (Figure 4(a)) and multivariate (Figure 4(e)) showed a good concordance between predic- (Figure 4(b)) Cox regression analysis in the train and test tion and observation. Height Log P-value Scale Free Topology Model Fit, signed R 2 Mean Connectivity 4 Journal of Healthcare Engineering 29 28 25 23 13 6 1 pvalue Hazard ratio BEST1 0.001 4.417 (1.769-11.028) CCL2 0.046 1.368 (1.006-1.859) CCL24 0.013 1.492 (1.090-2.042) CD55 <0.001 1.752 (1.296-2.368) EMP3 0.025 1.780 (1.074-2.947) -5 F3 0.008 1.458 (1.105-1.922) GABBR1 0.038 0.746 (0.592-0.986) -7 -6 -5 -4 -3 -2 -1 Log Lambda HAS2 0.039 2.079 (1.037-4.171) IFITM1 0.023 1.278 (1.035-1.578) KCNJ2 0.048 3.116 (1.009-9.662) LY6E 0.030 1.420 (1.035-1.947) NFKB1 0.023 0.271 (0.088-0.836) PLAUR 0.004 1.978 (1.244-3.145) SLC4A4 <0.001 0.364 (0.200-0.662) 0.01 0.1 1 10 100 Hazard ratio (a) (b) 29 30 28 26 25 23 22 23 21 16 12 11 7 6 5 3 2 1 Survival curve (P=1.385e-02) 80 1.0 0.8 0.6 0.4 0.2 0.0 0 123 4 -7 -6 -5 -4 -3 -2 -1 Time (year) Log (λ) high risk low risk (c) (d) Time - dependent ROC curve Survival curve (P=5.418e-03) 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Time (year) False positive rate high risk AUC at 1 year:0.75 low risk AUC at 2 year:0.893 AUC at 3 year:0.733 (e) (f) Figure 3: Continued. True positive rate Survival rate Survival rate Journal of Healthcare Engineering 5 Time - dependent ROC curve 1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate AUC at 1 year:0.773 AUC at 2 year:0.731 AUC at 3 year:0.741 (g) Figure 3: Construction of a multigene signature. (a) Univariate Cox analysis of genes within the brown module, screening of prognosis- related genes, and forest mapping. (b) -e Lasso coefficient profiles of prognosis-related genes. (c) -e partial likelihood deviance is plotted against log (λ). Kaplan–Meier plot of overall survival of patients is in the high-risk and low-risk groups in the train cohort (d) and test cohort (f). ROC curves for three-gene signatures in train cohort (e) and test cohort (g). [28]. Li et al. also found that CDKN2A knockdown inhibited 4. Discussion proliferation, migration, invasion, and epithelial mesen- Pancreatic cancer is highly malignant, lacks reliable early chymal transition in glioblastoma cells [29]. Bioinformatics screening methods, and has a poor prognosis, with an ex- studies have found that CDKN2A is also associated with pected 5-year survival rate of approximately 9% [1]. breast cancer [30], prostate cancer [31], and colorectal -erefore, there is an urgent need to find biomarkers that cancer [32]. BRCA1 has been reported to play an oncogenic affect the prognosis of pancreatic cancer in clinical treat- role in bladder cancer, with significantly lower expression ment, which will facilitate the assessment of patient prog- levels in cancer tissues than in normal tissues, and over- nosis and will help to improve the prognosis by tailoring the expression of BRCA1 reduced cell proliferation, migration, treatment to the individual patient. and colony-forming ability [33]. In contrast, in bladder Tumor development is the result of multigene interac- cancer, upregulation of BRCA1 was able to resist oxidative tions, and therefore, an increasing number of risk signatures stress, thereby promoting bladder cancer cell growth [34]. are used to predict prognosis [23–26]. In this study, we Pitt et al. also found mutations in BRCA1 in thyroid cancer. proposed a three-gene (CDKN2A, BRCA1, and UBL3) risk In addition to this, bioinformatics studies have found that signature by WGCNA and Lasso Cox regression analysis for BRCA1 is also associated with ovarian [35], colorectal [36], predicting overall survival in pancreatic cancer patients, with and gastric [37] cancers. Consistent with our speculation, statistically significant differences in overall survival between Zhao et al. found that, in NSCLC, UBL3 acts as a tumor high-and low-risk groups in the train cohort and test cohort. suppressor gene to inhibit cancer cell proliferation [38]. We then evaluated the prognostic performance of risk GSEA analysis revealed differences in 2 key signaling signature with the AUC of ROC, and the results showed that pathways between high- and low-risk groups. Base excision the risk signature could predict overall survival of pancreatic repair (BER) removes endogenous DNA damages that occur cancer patients accurately. Subsequent univariate and at all times in human cells, and its defects are associated with multivariate Cox analyses showed that the risk score could tumorigenesis [39], but cancer cells are also able to tolerate predict prognosis as an independent prognostic factor. In oxidative stress through increased BER activity, and tar- addition, we combined clinical characteristics to construct geting BER can improve the efficacy of radio/chemotherapy nomogram that can be used in the clinic to guide person- [40]. Our results show that the BER pathway is enriched in alized treatment. the high-risk group, suggesting that the BER pathway is Among the risk genes we identified, CDKN2A (cyto- active in high-risk patients, possibly leading to shorter skeleton-associated protein 2-like) was significantly highly survival by affecting their sensitivity to clinical treatment. An expressed in the high-risk group and positively correlated increasing number of studies have found that abnormal with risk score; BRCA1 (glutathione S-transferase Mu 5) and metabolism affects patient prognosis [41, 42], and the en- UBL3 (Ubiquitin-like 3) were significantly down regulated richment of propanoate metabolism pathway in the low-risk in the high-risk group and negatively correlated with the risk group suggests that the risk signature may affect patient score. CDKN2A has been reported to promote lung ade- prognosis through tumor metabolism. nocarcinoma invasion and is correlated with poor prognosis In summary, our study identified a 3-gene risk signature [27]. Monteverde et al. found that CDKN2A could promote for predicting prognosis, and the value of this risk signature nonsmall cell lung cancer (NSCLC) progression by regu- was validated in an external test cohort. By combining this lating transcriptional elongation, and targeting CDKN2A risk signature with clinical tumor pathology staging, a visual could enhance therapeutic response in patients with NSCLC nomogram was created to facilitate the prediction of survival True positive rate 6 Journal of Healthcare Engineering pvalue Hazard ratio pvalue Hazard ratio grade 0.046 1.702 (1.009-2.871) grade 0.113 1.586 (0.896-2.808) stage 0.477 0.682 (0.237-1.960) stage 0.066 0.350 (0.114-1.070) riskScore 0.008 4.028 (1.433-11.325) riskScore 0.013 3.974 (1.340-11.783) 0.25 0.50 1.0 2.0 4.0 8.0 0 12 3456789 10 1112 Hazard ratio Hazard ratio (a) (b) Points 1.0 0 102030405060708090 100 stage* 0.8 low risk** 0.6 high G2 grade G4 0.4 G1 G3 Total points 0.2 160 180 200 220 240 260 280 0.0 0.859 Pr (futime<3) 0.006 0.035 0.2 0.7 0.998 0.0 0.2 0.4 0.6 0.8 1.0 0.746 Pr (futime<2) Nomogram-predicted OS (%) 0.006 0.035 0.2 0.7 0.998 0.451 Pr (futime<1) 1-year 0.002 0.015 0.08 0.5 0.98 2-year 3-year (c) (d) 1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 1 – specificity AUC at 1 years: 0.769 AUC at 2 years: 0.834 AUC at 3 years: 0.865 (e) Figure 4: Evaluation of the predictive value of three-gene signatures and the creation of nomogram. (a) Test cohort risk signature univariate. (b) Multivariate Cox regression analysis forest plot. (c) Nomogram. (d) Calibration plot. (e) ROC curve for test cohort. Sensitivity Observed OS (%) Journal of Healthcare Engineering 7 [12] L. C. Chu, M. G. Goggins, and E. K. Fishman, “Diagnosis and outcomes. We further analyzed risk signature-related sig- detection of pancreatic cancer,” ?e Cancer Journal, vol. 23, naling pathways and risk genes’ expression and survival no. 6, pp. 333–342, 2017. analysis in a variety of cancers, providing a research di- [13] I. Diboun, L. Wernisch, C. A. Orengo, and M. Koltzenburg, rection for exploring prognosis-related genes in in vivo and “Microarray analysis after RNA amplification can detect in vitro experiments. We further analyzed risk signature- pronounced differences in gene expression using limma,” related signaling pathways and risk gene expression and BMC Genomics, vol. 7, no. 1, p. 252, 2006. survival analysis in a variety of cancers, providing a research [14] L. Wilkinson, “ggplot2: elegant graphics for data analysis by direction for exploring prognosis-related genes in cytolog- WICKHAM, H,” Biometrics, vol. 67, no. 2, pp. 678-679, 2011. ical and animal experiments. [15] H. C. Ho, A. Knudby, and W. A. Huang, “Spatial framework to map heat health risks at multiple scales,” International Journal of Environmental Research and Public Health, vol. 12, Data Availability no. 12, pp. 16110–23, 2015. [16] G. Yu, L.-G. Wang, Y. Han, and Q.-Y. He, “clusterProfiler: an -e analyzed datasets generated during the present study are R package for comparing biological themes among gene available from the corresponding author upon reasonable clusters,” OMICS: A Journal of Integrative Biology, vol. 16, request. no. 5, pp. 284–287, 2012. [17] W. Walter, F. Sanchez-Cabo, ´ and M. Ricote, “GOplot: an R Conflicts of Interest package for visually combining expression data with func- tional analysis: Figure 1,” Bioinformatics, vol. 31, no. 17, -e authors declare that they have no conflicts of interest in pp. 2912–2914, 2015. the present study. [18] C. S. Groeneveld, V. S. Chagas, S. J. M. Jones et al., “RTNsurvival: an R/Bioconductor package for regulatory network survival analysis,” Bioinformatics, vol. 35, no. 21, References pp. 4488-4489, 2019. [19] N. Simon, J. Friedman, T. Hastie, and R. Tibshirani, “Reg- [1] R. L. Siegel, K. D. Miller, and A. Jemal, “Cancer statistics, ularization paths for cox’s proportional hazards model via 2020,” CA: A Cancer Journal for Clinicians, vol. 70, no. 1, coordinate descent,” Journal of Statistical Software, vol. 39, pp. 7–30, 2020. no. 5, pp. 1–13, 2011. [2] L. Rahib, B. D. Smith, R. Aizenberg, A. B. Rosenzweig, [20] J. Kim, K. S. Kaufman, and H. Bang, “Graphing ratio measures J. M. Fleshman, and L. M. Matrisian, “Projecting cancer in- on forest plot,” Journal of the American College of Cardiology, cidence and deaths to 2030: the unexpected burden of thyroid, vol. 71, no. 5, pp. 585-586, 2018. liver, and pancreas cancers in the United States,” Cancer [21] B. J. Wells, K. M. Lenoir, J. F. Diaz-Garelli et al., “Predicting Research, vol. 74, no. 11, pp. 2913–2921, 2014. current glycated hemoglobin values in adults: development of [3] H. Zhu, T. Li, Y. Du, and M. Li, “Pancreatic cancer: challenges an algorithm from the electronic health record,” JMIR and opportunities,”BMCMedicine, vol. 16, no. 1, p. 214, 2018. Medical Informatics, vol. 6, no. 4, p. e10780, 2018. [4] L. Zhang, S. Sanagapalli, and A. Stoita, “Challenges in diag- [22] R. Marshall, Regplot: Enhanced Regression Nomogram Plot, nosis of pancreatic cancer,” World Journal of Gastroenterol- ogy, vol. 24, no. 19, pp. 2047–2060, 2018. [23] Z. Meng, D. Ren, K. Zhang, J. Zhao, X. Jin, and H. Wu, “Using [5] P. Langfelder and S. Horvath, “WGCNA: an R package for ESTIMATE algorithm to establish an 8-mRNA signature weighted correlation network analysis,” BMC Bioinformatics, prognosis prediction system and identify immunocyte infil- vol. 9, no. 1, p. 559, 2008. tration-related genes in Pancreatic adenocarcinoma,” Aging, [6] E. Clough and T. Barrett, “-e gene expression omnibus vol. 12, no. 6, pp. 5048–5070, 2020. database,” Methods in Molecular Biology, vol. 1418, pp. 93– [24] X.-H. Shi, X. Li, H. Zhang et al., “A five-microRNA signature 110, 2016. for survival prognosis in pancreatic adenocarcinoma based on [7] L. Badea, V. Herlea, S. O. Dima, T. Dumitrascu, and TCGA data,” Scientific Reports, vol. 8, no. 1, p. 7638, 2018. I. Popescu, “Combined gene expression analysis of whole- [25] G.-W. Huang, Y.-J. Xue, Z.-Y. Wu et al., “A three-lncRNA tissue and microdissected pancreatic ductal adenocarcinoma signature predicts overall survival and disease-free survival in identifies genes specifically overexpressed in tumor epithelia,” patients with esophageal squamous cell carcinoma,” BMC Hepato-gastroenterology, vol. 55, no. 88, pp. 2016–2027, 2008. Cancer, vol. 18, no. 1, p. 147, 2018. [8] H. Pei, L. Li, B. L. Fridley et al., “FKBP51 affects cancer cell [26] Z. Meng, Q. Yuan, J. Zhao et al., “-e m6A-related mRNA response to chemotherapy by negatively regulating Akt,” signature predicts the prognosis of pancreatic cancer pa- Cancer Cell, vol. 16, no. 3, pp. 259–266, 2009. tients,” Molecular ?erapy-Oncolytics, vol. 17, pp. 460–470, [9] G. Zhang, A. Schetter, P. He et al., “DPEP1 inhibits tumor cell invasiveness, enhances chemosensitivity and predicts clinical [27] G. Xiong, L. Li, X. Chen et al., “Up-regulation of CKAP2L outcome in pancreatic ductal adenocarcinoma,” PLoS One, vol. 7, no. 2, Article ID e31507, 2012. expression promotes lung adenocarcinoma invasion and is associated with poor prognosis,” OncoTargets and ?erapy, [10] D.-T. Chen, A. H. Davis-Yadley, P.-Y. Huang et al., “Prog- nostic fifteen-gene signature for early stage pancreatic ductal vol. 12, pp. 1171–1180, 2019. [28] T. Monteverde, S. Sahoo, M. La Montagna et al., “CKAP2L adenocarcinoma,” PLoS One, vol. 10, no. 8, Article ID e0133562, 2015. promotes non-small cell lung cancer progression through regulation of transcription elongation,” Cancer Research, [11] J. Wang, S. Yang, P. He et al., “Endothelial nitric oxide synthase traffic inducer (NOSTRIN) is a negative regulator of vol. 81, no. 7, pp. 1719–1731, 2021. [29] Y. F. Li, W. C. Tsai, C. H. Chou et al., “CDKN2A knockdown disease aggressiveness in pancreatic cancer,” Clinical Cancer Research, vol. 22, no. 24, pp. 5992–6001, 2016. exerts antitumor effects by increasing miR-4496 in 8 Journal of Healthcare Engineering glioblastoma cell lines,” International Journal of Molecular Sciences, vol. 22, no. 1, 2020. [30] L. Fan, J. Hou, and G. Qin, “Prediction of disease genes based on stage-specific gene regulatory networks in breast cancer,” Frontiers in Genetics, vol. 12, Article ID 717557, 2021. [31] T. Feng, D. Wei, Q. Li et al., “Four novel prognostic genes related to prostate cancer identified using Co-expression structure network analysis,” Frontiers in Genetics, vol. 12, Article ID 584164, 2021. [32] Y. Zhang, J. Luo, Z. Liu et al., “Identification of hub genes in colorectal cancer based on weighted gene co-expression network analysis,”Bioscience Reports, vol. 41, no. 7, Article ID BSR20211280, 2021. [33] Y. C. Jou, S. C. Wang, Y. C. Dia et al., “Anti-cancer effects and tumor marker role of glutathione S-transferase Mu 5 in human bladder cancer,” International Journal of Molecular Sciences, vol. 22, no. 6, Article ID 3056, 2021. [34] T. Li, D. Jiang, and K. Wu, “p62 promotes bladder cancer cell growth by activating KEAP1/NRF2-dependent antioxidative response,”CancerScience, vol. 111, no. 4, pp. 1156–1164, 2020. [35] Y. Wang, L. Lei, Y. G. Chi, L. B. Liu, and B. P. Yang, “A comprehensive understanding of ovarian carcinoma survival prognosis by novel biomarkers,”EuropeanReviewforMedical and Pharmacological Sciences, vol. 23, no. 19, pp. 8257–8264, [36] Y. L. Sun, Y. Zhang, Y. C. Guo, Z. H. Yang, and Y. C. Xu, “A prognostic model based on six metabolism-related genes in colorectal cancer,” BioMed Research International, vol. 2020, Article ID 5974350, 2020. [37] Y. Chen, B. Li, J. Wang et al., “Identification and verification of the prognostic value of the glutathione S-transferase Mu genes in gastric cancer,”OncologyLetters, vol. 20, no. 4, p. 100, [38] X. Zhao, Y. Zhou, Q. Hu et al., “Identification of a potential tumor suppressor gene, UBL3, in non-small cell lung cancer,” Cancer Biology and Medicine, vol. 17, no. 1, pp. 76–87, 2020. [39] S. S. Wallace, D. L. Murphy, and J. B. Sweasy, “Base excision repair and cancer,”CancerLetters, vol. 327, no. 1-2, pp. 73–89, [40] G. J. Grundy and J. L. Parsons, “Base excision repair and its implications to cancer therapy,” Essays in Biochemistry, vol. 64, no. 5, pp. 831–843, 2020. [41] T. E. Oni, G. Biffi, L. A. Baker et al., “SOAT1 promotes mevalonate pathway dependency in pancreatic cancer,” ?e Journal of Experimental Medicine, vol. 217, no. 9, 2020. [42] C. Li, X. Li, G. Li et al., “Identification of a prognosis‑asso- ciated signature associated with energy metabolism in tri- ple‑negative breast cancer,” Oncology Reports, vol. 44, no. 3, pp. 819–837, 2020.

Journal

Journal of Healthcare EngineeringHindawi Publishing Corporation

Published: Jan 7, 2022

References