Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/hindawi-publishing-corporation/development-of-an-autophagy-related-gene-prognostic-model-and-nomogram-rQHhNe0joW
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 © 2021 Ying Wang 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
- 1687-8450
- eISSN
- 1687-8469
- DOI
- 10.1155/2021/8810849
- Publisher site
- See Article on Publisher Site
Abstract
Hindawi Journal of Oncology Volume 2021, Article ID 8810849, 13 pages https://doi.org/10.1155/2021/8810849 Research Article Development of an Autophagy-Related Gene Prognostic Model and Nomogram for Estimating Renal Clear Cell Carcinoma Survival 1 1 2 3 1 Ying Wang , Yinhui Yao , Jingyi Zhao , Chunhua Cai, Junhui Hu, and Yanwu Zhao Department of Pharmacy, e Affiliated Hospital of Chengde Medical College, Chengde 067000, China Department of Functional Center, Chengde Medical College, Chengde 067000, China Department of Medical Insurance, e Affiliated Hospital of Chengde Medical College, Chengde 067000, China Correspondence should be addressed to Yanwu Zhao; cyfyzyw@163.com Received 14 September 2020; Revised 29 December 2020; Accepted 24 January 2021; Published 19 February 2021 Academic Editor: Raffaele Palmirotta Copyright © 2021 Ying Wang 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. Kidney renal clear cell carcinoma (KIRC) is a fatal malignancy of the urinary system. Autophagy is implicated in KIRC occurrence and development. Here, we evaluated the prognostic value of autophagy-related genes (ARGs) in kidney renal clear cell carcinoma. Materials and Methods. We analyzed RNA sequencing and clinical KIRC patient data obtained from TCGA and ICGC to develop an ARG prognostic signature. Differentially expressed ARGs were further evaluated by functional as- sessment and bioinformatic analysis. Next, ARG score was determined in 215 KIRC patients using univariable Cox and LASSO regression analyses. An ARG nomogram was built based on multivariable Cox analysis. /e prognosis nomogram model based on the ARG signatures and clinicopathological information was evaluated for discrimination, calibration, and clinical usefulness. Results. A total of 47 differentially expressed ARGs were identified. Of these, 8 candidates that significantly correlated with KIRC overall survival were subjected to LASSO analysis and an ARG score built. Functional enrichment and bioinformatic analysis were used to reveal the differentially expressed ARGs in cancer-related biological processes and pathways. Multivariate Cox analysis was used to integrate the ARG nomogram with the ARG signature and clinicopathological information. /e nomogram exhibited proper calibration and discrimination (C-index � 0.75, AUC �>0.7). Decision curve analysis also showed that the nomogram was clinically useful. Conclusions. KIRC patients and doctors could benefit from ARG nomogram use in clinical practice. Autophagy is an evolutionarily conserved process that 1.Introduction influences cellular homeostasis by degrading damaged or- Renal cell carcinoma (RCC) accounts for 2–3% of all adult ganelles and intracellular content [9, 10]. Recent studies have malignancies [1] and is one of the most lethal urologic implicated autophagy in cancer occurrence and progression cancers [2]. RCC incidence is rising in the US and most [11–15]. However, whether an autophagy signature derived Western countries [3]. Kidney renal clear cell carcinoma from one or more autophagy-related genes (ARGs) can (KIRC) is the most common type of RCC [4]. Despite recent predict long-term KIRC survival is unclear. Here, we used treatment advances, KIRC survival is poor [5]. TMN staging TCGA data to compare ARG expression profiles in KIRC vs. is a method of determining cancer prognosis and suggesting noncancer control tissue and assessed the ARG prognostic treatment strategies. However, TMN does not consider value. An eight-ARG prognostic signature whose prognostic genetic features as its classification is based on clinico- value was independent of clinical factors was developed by pathologic information [6–8]. /us, new markers for early multivariate Cox regression analyses. Next, we constructed KIRC detection are needed for better outcomes. and validated an eight-ARG prognostic model by integrating 2 Journal of Oncology regression analyses were used to evaluate correlation be- our newly established eight-ARG signature with classical clinicopathological risk factors for survival prediction in tween prognosis and the clinicopathological features in KIRC patients. KIRC patients. An ARG-clinicopathologic nomogram based on multi- variate analysis results was used to predict 1-, 3-, and 5-year OS. 2.Materials and Methods /e nomogram was subjected to 1000 bootstrap resamples for 2.1. Autophagy-Related Genes (ARGs). /e ARGs used in this internal validation of the analyzed database. Analysis of no- study were obtained from the Human Autophagy Database mogram discrimination performance was determined by (HADb, http://www.autophagy.lu/index.html), which in- concordance index (C-index) analysis, which predicts the cludes information on the 232 known autophagy genes. model’s prognostic value. Calibration plots were also used to determine the nomogram’s prognostic value. Nomogram calibration for 1-, 3-, and 5-year OS was done by comparing 2.2. Patient Database. ARG expression data (mRNA) and observed survival with the predicted probability. Additionally, associated clinical information for KIRC patients were a nomogram and calibration curve were developed on R using downloaded from TCGA. /ese included data on 539 KIRC the package rms. Decision curve analyses (DCAs) were used to tissues and 72 nontumor control tissue. Additional mRNA determine the nomogram’s clinical utility by quantifying net data on nontumor tissues from 47 patients were downloaded benefit at various threshold probabilities in KIRC patients. from the International Cancer Genome Consortium DCA for 1-, 3-, and 5-year OS was done using stdca and dca (ICGC). /e following patient cases were excluded from the packages. /ese analyses were done on R (version 3.5.3). analysis: (a) non-KIRC cases, (b) cases lacking mRNA data, (c) cases with missing data, (d) cases with survival time<30 3.Results days, and (e) the race was white. Ultimately, 215 KIRC patients were selected for further analysis. 3.1. Differentially Expressed Autophagy-Related Genes (ARGs). A total of 220 ARGs were extracted and were identified to represent between 119 nontumor KIRC tissues 2.3. Bioinformatic Analysis. To identify differentially and 539 KIRC tissues. Using FDR �<0.05 and |log (FC) |>1 expressed ARGs between KIRC and nontumor samples, we as cutoffs, we identified 47 differentially expressed ARGs used edgeR package on R with false discovery rate (FDR) � (tumor vs. normal tissues). Of these, 40 were upregulated <0.05 and |log fold change (logFC)|>1 as cutoffs. Functional and 7 were downregulated and were visualized on a scatter and pathway enrichment analyses were done using clus- plot (Figures 1(a) and 1(b)). terProfiler package. KEGG functional pathway analysis data were visualized using the GOplot package. GO terms and KEGG pathways with p �<0.05 were considered statistically 3.2. Functional Annotation and Protein-Protein Interaction significant. Next, STRING (http://string-db.org/) and pro- (PPI) Analysis. /e 47 differentially expressed ARGs were tein-protein interaction (PPI) network analyses of ARGs subjected to GO and KEGG pathway analyses to determine were done and results with a score (median confidence)> 0.4 their biological functions. /is analysis identified the top were visualized. enriched terms in biological processes (BPs) as regulation of Principal component analysis (PCA) was used to cluster endopeptidase activity, regulation of peptidase activity, and KIRC patients into different groups using Consensu- regulation of cysteine-type endopeptidase activity involved sClusterPlus package. in apoptotic process. /e most enriched terms for cellular To estimate the prognostic value of ARGs, we performed components (CCs) were autophagosome, autophagosome univariate Cox regression analysis on the 215 KIRC patients membrane, and inflammasome complex. /e most enriched using the survival package, with p �<0.05 indicating sta- terms for molecular function (MF) were ubiquitin protein tistical significance. Next, least absolute shrinkage and se- ligase binding, ubiquitin-like protein ligase binding, and lection operator (LASSO) Cox regression analysis was used peptidase regulator activity (Figure 2(a)). KEGG analysis to select potential ARGs from all significantly differentially found the 47 differentially expressed ARGs to be highly expressed ARGs identified by univariate Cox regression associated with human cytomegalovirus infection, auto- analysis. LASSO Cox analysis was done using the glmnet phagy-animal, and HIF-1 signaling, among other pathways. package. Risk score was calculated based on a linear com- Furthermore, the z-score of enriched pathways more than bination of ARG expression values after weighting regres- zero showed that most pathways were likely to be increase sion coefficients. Patients were classified into low- and high- (Figures 2(b) and 2(c)). risk groups using median risk score as cutoff. Protein-protein interaction (PPI) network analysis be- tween the 47 differentially expressed ARGs was done using STRING (Figure 3). 2.4. Statistical Analysis. Kaplan–Meier (KM) analysis and a two-sided log-rank test were used to determine overall survival in different clusters or in the high- and low-risk 3.3. Consensus Clustering. To comprehend the distinct groups. Receiver operating curve (ROC) analyses using clusters of ARGs with KIRC patients, consensus clustering survivalROC package evaluated the specificity and sensitivity was performed to identify selection of adequate groups. We of prognosis prediction. Univariate and multivariate Cox found that k � 2 was up to the mustard of clustering stability Journal of Oncology 3 Volcano Type VEGFA 5.0 CDKN1A CX3CL1 APOL1 CXCR4 HSPB8 GABARAP GABARAPL1 EIF4EBP1 MYC 0 2.5 EGFR BAX ERO1A GAPDH −5 SERPINA1 P4HB RACK1 BCL2 0.0 BAG1 ERBB2 ITGB4 RAB24 FAS RGS19 CASP1 CASP4 ATG16L1 −2.5 BID EEF2K DIRAS3 ATG16L2 GNAI3 PRKCQ NKX2−3 IFNG −5.0 TP73 NRG3 IL24 SPNS1 020 40 60 CCR2 NLRC4 -log10 (FDR) BIRC5 SPHK1 Sig CDKN2A Down PTK6 ATG9B Not GRID1 Up Type (a) (b) 12.5 10.0 7.5 5.0 2.5 0.0 Type (c) Figure 1: Differentially expressed autophagy-related genes (ARGs) between 119 nontumor and 539 kidney renal clear cell carcinoma (KIRC) samples. (a) /e volcano plot of the 47 differentially expressed ARGs (tumor (T) vs. normal tissues (N). Red and green indicate high and low expression, respectively. (b) Hierarchical clustering of differentially expressed ARG expression levels. (c) Expression of the 47 differentially expressed ARGs. (Figures 4(a)–4(e)). /us, KIRC patients could be grouped (Figure 4(g)). Kaplan–Meier survival analysis of the 2 into 2 clusters (cluster1 and cluster2). Comparison of the 2 subgroups found significant prognostic differences between clusters based on KIRC patient clinicopathological features KIRC patients (p< 0.0001), and cluster1 significantly cor- found no significant correlation between KIRC molecular related with better OS relative to cluster2 (Figure 4(f)). clusters and clinicopathological factors such as age, gender, smoking, pharmaceutical, and pathological N or M. Notably, 3.4. Correlation between ARGs and KIRC. Spearman analysis cluster1 significantly correlated with lower grade (p< 0.0001), stage (p< 0.05), or pathological T (p< 0.001) of the correlation between the 47 differentially expressed logFC Gene expression APOL1 ATG16L1 ATG16L2 ATG9B BAG1 BAX BCL2 BID BIRC5 CASP1 CASP4 CCR2 CDKN1A CDKN2A CX3CL1 CXCR4 DIRAS3 EEF2K EGFR EIF4EBP1 ERBB2 ERO1A FAS GABARAP GABARAPL1 GAPDH GNAI3 GRID1 HSPB8 IFNG IL24 ITGB4 MYC NKX2−3 NLRC4 NRG3 P4HB PRKCQ PTK6 RAB24 RACK1 RGS19 SERPINA1 SPHK1 SPNS1 TP73 VEGFA GO:0052547 GO:0052548 BP CC MF GO:2000116 GO:0001558 GO:0043281 4 Journal of Oncology Regulation of cell growth Regulation of endopeptidase activity Regulation of peptidase activity Cell growth Response to oxygen levels Regulation of cysteine-type endopeptidase activity involved in apoptotic process Regulation of cysteine-type endopeptidase activity Intrinsic apoptotic signaling pathway Positive regulation of protein localization to membrane Count Positive regulation of cysteine-type endopeptidase activity involved in apoptotic process Vacuolar membrane Autophagosome Autophagosome membrane p.adjust Inflammasome complex 0.01 Pore complex 0.02 0.03 Ubiquitin protein ligase binding 0.04 Ubiquitin-like protein ligase binding Protein phosphatase binding Phosphatase binding Peptidase regulator activity BH domain binding Death domain binding Protein phosphatase 2A binding Peptidase activator activity Cysteine-type endopeptidase regulator activity involved in apoptotic process 0.05 0.10 0.15 0.20 Gene ratio (a) ID Description GO : 0043281 Regulation of cysteine-type endopeptidase activity involved in apoptotic process GO : 0001558 Regulation of cell growth GO : 2000116 Regulation of cysteine-type endopeptidase activity GO : 0052548 Regulation of endopeptidase activity GO : 0052547 Regulation of peptidase activity GO : 0016049 Cell growth GO : 1905477 Positive regulation of protein localization to membrane GO : 0097193 Intrinsic apoptotic signaling pathway GO : 0070482 Response to oxygen levels GO : 0043280 Positive regulation of cysteine-type endopeptidase activity involved in apoptotic process logFC z-score Downregulated Decreasing Increasing Upregulated (b) Response to oxygen levels Regulation of protein localization to membrane Regulation of peptidase activity Regulation of endopeptidase activity Regulation of cysteine-type endopeptidase activity involved in apoptotic process Regulation of cysteine-type endopeptidase activity Regulation of cell growth Regulation of apoptotic signaling pathway logFC Protein insertion into membrane Positive regulation of protein localization to membrane Positive regulation of peptidase activity Positive regulation of endopeptidase activity Positive regulation of cysteine-type endopeptidase activity involved in apoptotic process Positive regulation of cysteine-type endopeptidase activity Peptidyl-serine phosphorylation Macroautophagy Intrinsic apoptotic signaling pathway G1/S transition of mitotic cell cycle Cellular response to oxygen levels Cell growth (c) Figure 2: Functional annotation of the 47 differentially expressed ARGs. (a) Gene ontology analysis predicted relevant biological processes. (b) Outer circle shows a scatter plot for each term’s logFC of the ARGs. (c) Heatmap of the relationship between ARGs and KEGG pathways. GO:0043280 GO:0070482 GO:0097193 GO:1905477 GO:0016049 Journal of Oncology 5 ERO1L SPNS1 APOL1 EEF2K P4HB SERPINA1 IL24 RGS19 EIF4EBP1 ITGB4 CX3CL1 GNAI3 GRID1 CXCR4 NRG3 SPHK1 VEGFA EGFR CCR2 ERBB2 IFNG PTK6 GAPDH PRKCQ MYC CDKN1A GNB2L1 BIRC5 CASP4 CASP1 CDKN2A FAS BAG1 NLRC4 TP73 HSPB8 BAX NKX2-3 BCL2 GABARAPL1 ATG16L1 BID GABARAP DIRAS3 ATG16L2 RAB24 Figure 3: Protein-protein interaction (PPI) network of ARGs that are differentially expressed in KIRC. Consensus CDF Delta area Tracking plot 1.0 0.5 0.8 0.4 0.6 0.3 5 0.4 3 0.2 0.2 0.1 0.0 Samples 0.0 0.2 0.4 0.6 0.8 1.0 23456 789 Consensus index (a) (b) (c) Consensus matrix k = 2 Consensus matrix k = 3 1.00 0.75 0.50 0.25 p = 3.877e − 04 0.00 0123456789 10 Time (years) Cluster1 385 321 248 206 149 92 50 25 12 3 1 Cluster2 122 93 74 57 41 24 12 5 1 1 0 0123456789 10 Time (years) Cluster 1 1 Cluster1 2 2 Cluster2 (d) (e) (f) Figure 4: Continued. CDF Relative change in area under CDF curve Cluster Survival probability 6 Journal of Oncology Cluster Cluster Cluster1 Cluster2 ∗∗ ∗ N Stage ∗∗∗ Grade N0 Pharmaceutical N1 Smoking 0 Gender Age M0 Status −5 M1 VEGFA ∗∗ SPNS1 T1 −10 ATG16L2 T2 RAB24 T3 BAG1 T4 GABARAPL1 Stage NKX2−3 Stage I ERBB2 Stage II BCL2 Stage III CX3CL1 Stage IV ∗∗∗ CDKN1A Grade MYC G1 HSPB8 G2 EIF4EBP1 G3 G4 GAPDH P4HB Pharmaceutical GABARAP No RACK1 Yes ATG16L1 Smoking PRKCQ FAS GNAI3 NRG3 1 EEF2K Gender EGFR FEMALE APOL1 MALE ITGB4 Age TP73 ≤65 BAX >65 BID Status CXCR4 Alive IL24 Dead NLRC4 CASP1 CCR2 IFNG CASP4 RGS19 BIRC5 SPHK1 CDKN2A DIRAS3 SERPINA1 ERO1A GRID1 ATG9B PTK6 (g) Figure 4: Differential clinicopathological characteristics and overall survival (OS) in KIRC clusters 1 and 2. (a) Consensus clustering cumulative distribution function (CDF) for k � 2 to 10. (b) Relative change in area under CDF curve for k � 2 to 10. (c) Tracking plot for k � 2 to 10. (d-e) Consensus clustering matrix for k � 2 (d) and k � 3 (e). (f) Heatmap and clinicopathological characteristics of the 2 clusters (cluster1 and cluster2) defined by ARG consensus expression. (g) Kaplan–Meier curves for KIRC patients. ARGs and KIRC and principal component analysis (PCA) ERBB2 + (−0.00099) × expression level of HSPB8 + 0.047 × expression level of SPHK1. Next, risk scores were revealed a clear-cut distinction between cluster1 and cluster2 (see Supplementary 1). used to group the 215 patients into high- and low-risk groups based on median risk score. KM analysis revealed significant OS differences between the 2 groups (p �<0.001; 3.5. Autophagy-Related Gene Score Building. Here, we Figure 5(c)). To further explore utility of risk scores based on assessed the prognostic value of the 47 differentially ARG signature, the 215 patients were classified into 10 expressed ARGs in KIRC using univariate Cox regression subgroups based on different patient clinicopathological analysis. /is analysis indicated that 20 of the forty-seven features. KM analysis showed that KIRC patients in the low- genes were strongly significantly associated with survival risk group had significantly better OS relative to those in the (p�<0.05). Of these 20 ARGs, 12 ARGs were associated with high-risk group in the 10 subgroups (p �<0.001, Figure 6). poor OS (hazard ratio �>1). /e rest were associated with favorable OS (hazard ratio �<1) (Figure 5(a)). Finally, 3.6. Evaluation of the Predictive Performance of the Auto- LASSO Cox regression analysis identified 8 ARGs (ATG16L2, ATG9B, BID, BIRC5, CX3CL1, ERBB2, HSPB8, phagy-Related Gene (ARG) Signature Using ROC Analysis. and SPHK1) as capable of predicting KIRC clinical outcomes Receiver operating characteristic (ROC) curve analysis was (Figure 5(b)). /e selected 8 ARGs were then used to create a used to evaluate the predictive accuracy of 1-, 3-, and 5-year risk assessment model and risk score determined as follows: survival in KIRC patients. /e AUC values for ROC curve risk score � 0.028 × expression level of ATG16L2 + analysis of 1-, 3-, and 5-year ARG-based OS were 0.728, 0.032 × expression level of ATG9B + 0.047 × expression level 0.729, and 0.784, respectively (Figures 5(d)–5(f)), indicating of BID + 0.044 × expression level of BIRC5 + (−0.0036) × that ARG risk scores outperform conventional clinical expression level of CX3CL1 + (−0.012) × expression level of prognostic factors in predicting long-term (5-year) but not Journal of Oncology 7 20 20 20 19 19 18 17 17 16 14 10 8 5 2 p value Hazard ratio ATG16L2 0.006 1.416 (1.103 − 1.819) 1.508 (1.181 − 1.924) ATG9B <0.001 12.0 0.552 (0.346 − 0.881) BAG1 0.013 0.660 (0.529 − 0.823) BCL2 <0.001 2.811 (1.756 − 4.499) BID <0.001 11.8 2.033 (1.589 − 2.603) BIRC5 <0.001 2.048 (1.264 − 3.316) CASP4 0.004 1.384 (1.020 − 1.879) CDKN2A 0.037 11.6 0.578 (0.463 − 0.721) CX3CL1 <0.001 1.399 (1.129 − 1.733) EIF4EBP1 0.002 0.607 (0.458 − 0.805) ERBB2 <0.001 11.4 0.540 (0.313 − 0.932) GABARAP 0.027 0.770 (0.662 − 0.896) HSPB8 <0.001 0.607 (0.371 − 0.993) NRG3 0.047 11.2 PRKCQ 0.018 0.666 (0.475 − 0.934) 0.028 1.328 (1.032 − 1.710) PTK6 0.030 1.517 (1.041 − 2.211) RGS19 11.0 SPHK1 <0.001 1.934 (1.527 − 2.449) SPNS1 0.020 2.534 (1.157 − 5.550) TP73 0.012 2.924 (1.268 − 6.746) 10.8 Hazard ratio −6 −5 −4 −3 −2 Log (λ) (a) (b) 1.00 1-year OS 1.0 0.75 0.8 0.50 0.25 0.6 p = 3.571e − 06 0.00 0.4 0123456789 10 Time (years) 0.2 High risk 107 81 62 46 31 17 10 6 3 2 0 Low risk 108 89 72 62 50 31 21 10 5 2 1 0.0 0123456789 10 0.0 0.2 0.4 0.6 0.8 1.0 Time (years) False positive rate Risk Risk score (AUC = 0.728) Grade (AUC = 0.744) High risk Age (AUC = 0.559) Stage (AUC = 0.874) Gender (AUC = 0.507) T (AUC = 0.825) Low risk Smoking (AUC = 0.447) M (AUC = 0.775) Pharmaceutical (AUC = 0.557) N (AUC = 0.533) (c) (d) Figure 5: Continued. Risk Survival probability Partial likelihood deviance True positive rate 8 Journal of Oncology 3-year OS 5-year OS 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 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate False positive rate Risk score (AUC = 0.729) Grade (AUC = 0.706) Risk score (AUC = 0.784) Grade (AUC = 0.670) Age (AUC = 0.562) Stage (AUC = 0.793) Gender (AUC = 0.496) T (AUC = 0.757) Age (AUC = 0.597) Stage (AUC = 0.732) Smoking (AUC = 0.473) M (AUC = 0.669) Gender (AUC = 0.482) T (AUC = 0.698) Smoking (AUC = 0.438) M (AUC = 0.632) Pharmaceutical (AUC = 0.599) N (AUC = 0.557) Pharmaceutical (AUC = 0.640) N (AUC = 0.555) (e) (f) Figure 5: Determination of risk scores for 215 KIRC patients using the 8-ARG risk signature. (a) Identification of a 20-ARG risk signature. (b) Least absolute shrinkage and selection operator (LASSO) coefficient profiles of the 20 ARGs. (c) Kaplan–Meier survival analysis based on the 8-ARG signature risk scores in KIRC patients. (d-f) Receiver operating characteristic (ROC) curves reveal the predictive accuracy of 1-, 3-, and 5-year survival in KIRC patients. 1.00 1.00 1.00 1.00 0.75 0.75 0.75 0.75 0.50 0.50 0.50 0.50 0.25 0.25 0.25 0.25 p = 1.971e − 06 p = 1.617e − 01 p = 2.126e − 03 p = 2.618e − 04 0.00 0.00 0.00 0.00 0123456789 10 0123456789 10 0123456789 10 0123456789 10 Time (years) Time (years) Time (years) Time (years) High risk 71 54 40 30 19 10 3 2 1 0 0 High risk 36 27 22 16 12 7 6 4 2 2 0 High risk 77 58 45 34 22 14 75220 High risk 30 23 17 12 9321100 Low risk 62 51 44 39 34 24 17 9 4 1 1 Low risk 46 38 28 23 16 751110 Low risk 57 48 41 33 24 15 10 4 1 0 0 Low risk 51 41 31 29 26 16 12 6 4 2 1 0123456789 10 0123456789 10 0123456789 10 0123456789 10 Time (years) Time (years) Time (years) Time (years) Risk (age≤65) Risk (age>65) Risk (male) Risk (female) High risk High risk High risk High risk Low risk Low risk Low risk Low risk (a) (b) (c) (d) 1.00 1.00 1.00 1.00 0.75 0.75 0.75 0.75 0.50 0.50 0.50 0.50 0.25 0.25 0.25 0.25 p = 1.025e − 04 p = 5.074e − 03 p = 8.524e − 01 p = 5.114e − 06 0.00 0.00 0.00 0.00 0123456789 10 0123456789 10 0123456 0123456789 10 Time (years) Time (years) Time (years) Time (years) High risk High risk 23 18 14 9 5 2 1 84 63 48 37 26 15 8 6 3 2 0 High risk 62 46 34 24 15 7 3 1 1 1 0 High risk 45 35 28 22 16 10 6 5 2 1 0 Low risk 13 11 9 6 4 2 0 Low risk 95 78 63 56 46 29 22 10 5 2 1 Low risk 66 55 45 38 30 17 12 5 2 2 1 Low risk 42 34 27 24 20 14 10 5300 0123456789 10 0123456789 10 0123456 0123456789 10 Time (years) Time (years) Time (years) Time (years) Risk (smoking 1 + 2) Risk (smoking 3 + 4 + 5) Risk (pharmaceutical-yes) Risk (pharmaceutical-no) High risk High risk High risk High risk Low risk Low risk Low risk Low risk (e) (f) (g) (h) Figure 6: Continued. Risk (smoking 1 + 2) Risk (age≤65) Survival probability Survival probability True positive rate Risk (smoking 3 + 4 + 5) Risk (age>65) Survival probability Survival probability Risk (pharmaceutical-yes) Risk (male) Survival probability Survival probability True positive rate Risk (pharmaceutical-no) Risk (female) Survival probability Survival probability Journal of Oncology 9 1.00 1.00 1.00 1.00 0.75 0.75 0.75 0.75 0.50 0.50 0.50 0.50 0.25 0.25 0.25 0.25 p = 3.823e − 01 p = 8.195e − 05 p = 4.007e − 02 p = 9.585e − 03 0.00 0.00 0.00 0.00 0123456789 10 0123456789 10 012345 6 7 8 9 10 012345678 Time (years) Time (years) Time (years) Time (years) High risk 32 29 23 18 13 8 6 5 3 2 0 High risk 75 52 39 28 18 9 3 1 0 0 0 High risk 41 39 32 29 19 12 8 5 3 2 0 High risk 66 42 30 17 12 5 1 1 0 Low risk 61 50 39 35 29 20 16 6 4 1 1 Low risk 47 39 33 27 21 11 64110 Low risk 72 63 52 46 37 23 16 8 5 2 1 Low risk 36 26 20 16 13 8 6 2 0 0123456789 10 0123456789 10 012345 6 7 8 9 10 012345678 Time (years) Time (years) Time (years) Time (years) Risk (G1 + 2) Risk (G3 + 4) Risk (Stage I + II) Risk (Stage III + IV) High risk High risk High risk High risk Low risk Low risk Low risk Low risk (i) (j) (k) (l) 1.00 1.00 1.00 1.00 0.75 0.75 0.75 0.75 0.50 0.50 0.50 0.50 0.25 0.25 0.25 0.25 p = 9.316e − 03 p = 1.617e − 02 p = 6.351e − 04 p = 2.159e − 02 0.00 0.00 0.00 0.00 0123456789 10 012345678 012345 6 7 8 9 10 012345678 Time (years) Time (years) Time (years) Time (years) High risk 46 43 35 31 21 13 8 5 3 2 0 High risk 61 38 27 15 10 4 1 1 0 High risk 81 70 56 42 28 15 9 6 3 2 0 High risk 26 11 6 4 3 2 0 0 0 Low risk 75 66 54 47 38 24 17 8 5 2 1 Low risk 33 23 18 15 12 7 5 2 0 Low risk 96 80 66 58 47 29 20 9 5 2 1 Low risk 12 96432210 0123456789 10 012345678 012345 6 7 8 9 10 012345678 Time (years) Time (years) Time (years) Time (years) Risk (T1 + 2) Risk (T3 + 4) Risk (M0) Risk (M1) High risk High risk High risk High risk Low risk Low risk Low risk Low risk (m) (n) (o) (p) 1.00 1.00 1.00 1.00 0.75 0.75 0.75 0.75 0.50 0.50 0.50 0.50 0.25 0.25 0.25 0.25 p = 1.198e − 05 p = 6.958e − 01 p = 2.233e − 06 p = 4.056e − 01 0.00 0.00 0.00 0.00 0123456789 10 012345678 012345 6 7 8 9 10 01234567 8 9 10 Time (years) Time (years) Time (years) Time (years) High risk 97 75 58 44 29 17 9 6 3 2 0 High risk 10 6 4 2 2 0 0 0 0 High risk 72 56 41 32 22 11 5 3 2 1 0 High risk 35 25 21 14 9643110 Low risk 105 87 71 61 49 30 21 10 5 2 1 Low risk 321111100 Low risk 92 76 64 56 44 30 21 9 5 2 1 Low risk 16 13 866111000 0123456789 10 012345678 012345 6 7 8 9 10 01234567 8 9 10 Time (years) Time (years) Time (years) Time (years) Risk (N0) Risk (N1) Risk (cluster1) Risk (cluster2) High risk High risk High risk High risk Low risk Low risk Low risk Low risk (q) (r) (s) (t) Figure 6: Survival differences between high- and low-risk KIRC patients stratified by clinicopathological characteristics. (a), (b) Age; (c), (d) gender; (e), (f) smoking; (g), (h) pharmaceutical; (i), (j) pathological grade; (k), (l) pathological stage; (m), (n) pathological T; (o), (p) pathological M; (q), (r) pathological N; (s), (t) cluster. short-term survival (1- and 3-year) in KIRC patient OS, regression analyses of the 8-ARG signature as an inde- highlighting ARG risk score as a novel KIRC prognosis pendent KIRC prognostic factor showed that age, phar- indicator. maceutical, pathological grade, pathological stage, pathological T, pathological M, pathological N, cluster, and risk score correlated with significant OS differences in KIRC 3.7. Establishment of the Nomogram. Heatmap analysis of patients (Figure 7(b)). Multivariate analysis using the factors the expression of the 8 ARGs in high- vs. low-risk groups mentioned earlier revealed that age, pharmaceutical, path- revealed significant differences in status (p �<0.001), gender ological N, and risk score remained significantly associated (p �<0.001), pathological grade (p �<0.001), pathological with the OS (Figure 7(c)). stage (p �<0.001), pathological T (p � p< 0.001), patho- A prognostic nomogram to predict 1-, 3-, and 5-year OS logical M (p �<0.05), pathological N (p �<0.05), and cluster was established using multivariate analysis results. Total (p �<0.01) (Figure 7(a)). Moreover, univariate Cox points were calculated by integrating risk score, age, and Risk (N0) Risk (T1 + 2) Risk (G1 + 2) Survival probability Survival probability Survival probability Risk (N1) Risk (T3 + 4) Risk (G3 + 4) Survival probability Survival probability Survival probability Risk (cluster1) Risk (M0) Risk (Stage I + II) Survival probability Survival probability Survival probability Risk (cluster2) Risk (M1) Risk (Stage III + IV) Survival probability Survival probability Survival probability 10 Journal of Oncology Risk Risk ∗∗ Cluster High Low ∗∗∗ ∗∗ Cluster ∗∗∗ Stage cluster1 ∗∗∗ Grade cluster2 Pharmaceutical Smoking ∗ ∗∗∗ Gender N0 Age ∗∗∗ N1 Status −5 M0 M1 HSPB8 ∗∗∗ T1 T2 T3 T4 CX3CL1 ∗∗∗ Stage Stage I Stage II Stage III Stage IV ERBB2 ∗∗∗ Grade G1 G2 G3 G4 BID Pharmaceutical No Yes Smoking BIRC5 ∗∗∗ Gender Female SPHK1 Male Age ≤65 >65 ∗∗∗ Status ATG16L2 Alive Dead ATG9B (a) p value Hazard ratio p value Hazard ratio Age Age 1.021 (1.003 − 1.040) 1.026 (1.005 − 1.047) 0.022 0.017 Gender 1.053 (0.678 − 1.635) Gender 1.186 (0.739 − 1.904) 0.818 0.479 Smoking 0.243 0.884 (0.719 − 1.087) Smoking 0.137 1.230 (0.936 − 1.615) Pharmaceutical 3.881 (2.486 − 6.060) Pharmaceutical 4.319 (2.339 − 7.975) <0.001 <0.001 Grade 2.234 (1.661 − 3.005) Grade 1.174 (0.817 − 1.685) <0.001 0.386 Stage <0.001 1.975 (1.613 − 2.417) Stage 0.452 1.256 (0.693 − 2.276) 2.063 (1.611 − 2.641) 1.138 (0.651 − 1.988) T <0.001 T 0.650 4.547 (2.885 − 7.164) 1.790 (0.745 − 4.298) M <0.001 M 0.193 N <0.001 3.126 (1.612 − 6.063) N 0.038 2.255 (1.046 − 4.863) 1.652 (1.040 − 2.622) 1.065 (0.635 − 1.784) Cluster 0.033 Cluster 0.812 Riskscore <0.001 1.417 (1.282 − 1.568) Riskscore <0.001 1.366 (1.181 − 1.580) 01234567 0246 Hazard ratio Hazard ratio (b) (c) Figure 7: Relationship between the risk score, clinicopathological features, and cluster1/2 subgroups in 215 KIRC patients. (a) /e heatmap of the 8-ARG expression in low- and high-risk KIRC. Clinicopathological feature distribution was compared in low- vs. high-risk groups. ∗∗ ∗∗∗ p �<0.01; p �<0.001. Forest plot of univariate (b) and multivariate (c) Cox regression analyses in KIRC. Journal of Oncology 11 0 102030405060708090 100 Points Age 30 45 60 75 90 Yes Pharmaceutical No N1 N0 Riskscore 0123456789 10 11 12 Total Points 0 102030405060708090 100 110 1-year survival 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 3-year survival 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.05 5-year survival 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.05 (a) ROC curve (AUC = 0.742) ROC curve (AUC = 0.792) ROC curve (AUC = 0.856) 1.0 1.0 1.0 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate False positive rate False positive rate (b) (c) (d) 1.0 1.0 1.0 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.0 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Nomogram-predicted probability of 1-year survival Nomogram-predicted probability of 3-year survival Nomogram-predicted probability of 5-year survival (e) (f) (g) 0.30 0.10 0.4 0.25 0.20 0.3 0.05 0.15 0.2 0.10 0.00 0.1 0.05 0.00 0.0 −0.05 −0.05 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 reshold probability reshold probability reshold probability None None None All All All Nomogram Nomogram Nomogram (h) (i) (j) Figure 8: Nomogram prediction of overall survival (OS) in KIRC patients. (a) /e prognostic nomogram for predicting 1-, 3-, and 5-year OS. (b–d) ROC curve based on the prognostic nomogram for 1-, 3-, and 5-year OS. (e–g) Calibration plots for predicting patient 1-, 3-, and 5-year OS. (h–j) Decision curve analyses (DCAs) of the prognostic nomogram for 1-, 3-, and 5-year risk. Net benefit Actual 1-year survival True positive rate Net benefit Actual 3-year survival True positive rate Net benefit Actual 5-year survival True positive rate 12 Journal of Oncology pharmaceutical (Figure 8(a)). Considering the discrimina- creating a relatively accurate tool for predicting KIRC pa- tion ability of the prognostic nomogram, ROC analysis was tients’ OS. Our internal validation findings indicated the nomo- conducted. /e results indicated that AUC for 1-, 3-, and 5- year survival were 0.742, 0.792, and 0.856, suggesting that gram’s convincing discrimination and calibration power. the prognostic nomogram has higher prediction efficacy Furthermore, an interval validation C-index �>0.7 con- (Figures 8(b)–8(d)). Moreover, the prognostic nomogram’s firmed the nomogram’s clinical prognostic accuracy. C-index value of 0.75 (95% CI, 0.69–0.80) in all KIRC pa- However, it is still hard to determine when to use the tients also indicated good discrimination. prognostic nomogram. DCA selects the best treatment Calibration curve analysis of the nomogram’s 1-, 3-, and approach by analyzing various potential strategies, thereby 5-year survival prediction revealed satisfactory predictive guiding clinical decisions [29, 30]. Here, we evaluated if the accuracy by the nomograms relative to actual observations prognostic nomogram could guide clinical decisions and improve patient outcomes. /e 5-year decision curve (Figures 8(e)–8(g)). /e nomogram’s 1, 3, or 5-year de- cision curve analyses (DCA) showed that the more clini- analysis showed high tolerance and threshold probability (up to 83%), indicating that using the prognostic nomo- cally useful nomogram constructs predicted long-term survival, especially 5-year survival, suggesting that if a gram to predict long-term survival enhanced patient patient or doctor’s threshold probability was less than 83%, benefits. using the nomogram to predict 3- to 5-year prognosis has Although the prognostic nomogram performs well in more benefit than completely ignoring the scheme for all predicting KIRC prognosis, this study has several limita- programs (Figures 7(h)–7(j)). However, 1-year DCA tions. First, the patients in this cohort were not represen- showed a limited threshold probability range of about 18% tative of all races affected by KIRC as the data were only, indicating that the prognostic nomogram was clini- exclusively obtained from TCGA and ICGC databases. cally useful. Secondly, because publicly available data are limited, clin- icopathological characteristics were not analyzed compre- hensively. /us, while the utility of the prognostic 4.Discussion nomogram was assessed comprehensively by an internal Patient prognosis influences treatment decisions [16, 17]. validation using a bootstrap test, external validation was not applied. Hence, our findings should be evaluated in pro- ARGs have been implicated in numerous cancers, including KIRC. In past studies, some ARGs have emerged as potential spective clinical studies. KIRC prognostic factors [18–20]. For instance, BIRC5 is a crucial antiapoptotic protein that positively correlates with 5.Conclusions KIRC pathological grade and clinical stage [18]. As a mo- In conclusion, our study not only uncovered a novel 8-gene lecular marker of tumor behavior and prognosis, ATG16L2 signature as a potential biomarker of KIRC prognosis but is associated with KIRC risk and patient outcome [19]. BID also provided a risk assessment model for KIRC prognosis. is located on chromosome 22q11.21 and is an apoptosis- related protein. CASP4 is reported to promote cell migration Data Availability by influencing actin cytoskeleton remodeling [20]. SPHK1 upregulation in renal cell carcinoma may promote cancer Data underlying this study are provided in Supplementary progression, and its silencing may suppress cell proliferation Materials (Supplementary 2 and 3) and are available on via reduced HIF-2α expression [21]. ATG9B expression TCGA (https://gdc.cancer.gov/) and the International significantly correlates with TNM staging, distant metastasis, Cancer Genome Consortium (ICGC) (https://icgc.org/). and survival time of clear cell renal cell carcinoma patients [22]. However, there is no consensus regarding its satis- Conflicts of Interest factory predictive performance due to limited sample size or lack of data validating candidate ARGs as diagnostic and /e authors declare no conflicts of interest. prognostic biomarkers. Numerous studies based on TCGA datasets show that Acknowledgments ARGs can predict OS in various cancers, including glioma [23, 24], ovarian [25], breast [26], bladder [27], and colo- /is project was sponsored by the Chengde Science and rectal [28] cancer. Here, high-throughput RNA-seq data Technology Planning Project (grant nos. 201701A086, from TCGA were analyzed to investigate the role of ARGs in 202006A088 and 202006A049). KIRC. In this study, we have constructed a useful nomogram associated with the prognostic significance of ARG scores Supplementary Materials and clinicopathologic information that can predict KIRC patient survival. In differentially expressed ARGs, many Supplementary 1 (Figure S1): correlation between ARGs and potential confounding factors were identified and estab- KIRC patients. (A) Spearman correlation analysis of the 47 lished high-risk and low-risk groups, which were signifi- differentially expressed ARGs. (B) Principal component cantly related to OS of KIRC. Additionally, the analysis of total RNA expression profiles in KIRC patients. clinicopathologic factors of age, pharmaceutical, and path- Supplementary 2: based on the primary filter criteria ological N were integrated into the prognostic nomogram, mentioned in materials and methods, we identified 220 Journal of Oncology 13 vivo by inducing autophagy and cell cycle arrest,” Biomedicine ARGs. Supplementary 3: based on the primary filter criteria & Pharmacotherapy, vol. 120, Article ID 109469, 2019. mentioned in materials and methods, the clinicopathological [16] R. Bejar, K. E. Stevenson, B. A. Caughey et al., “Validation of a features of 215 KIRC patients were extracted. (Supplemen- prognostic model and the impact of mutations in patients tary Materials) with lower-risk myelodysplastic syndromes,” Journal of Clinical Oncology, vol. 30, no. 27, pp. 3376–3382, 2012. [17] N. Yamamoto, T. Watanabe, N. Katsumata et al., “Con- References struction and validation of a practical prognostic index for [1] K. D. Miller, L. Nogueira, A. B. Mariotto et al., “Cancer patients with metastatic breast cancer,” Journal of Clinical treatment and survivorship statistics, 2019,” CA: A Cancer Oncology, vol. 16, no. 7, pp. 2401–2408, 1998. Journal for Clinicians, vol. 69, no. 5, pp. 363–385, 2019. [18] Z.-G. Shi, S.-Q. Li, Z.-J. Li, X.-J. Zhu, P. Xu, and G. Liu, [2] R. Arezoo, M. Mitra, M. Zahra et al., “Co-expression of cancer “Expression of vimentin and survivin in clear cell renal cell stem cell markers OCT4 and NANOG predicts poor prog- carcinoma and correlation with p53,” Clinical and Transla- nosis in renal cell carcinomas,” Scientific Reports, vol. 8, tional Oncology, vol. 17, no. 1, pp. 65–73, 2014. p. 11739, 2018. [19] T. Xu, H. Ruan, Z. Song et al., “Identification of CXCL13 as a [3] R. L. Siegel, K. D. Miller, and A. Jemal, “Cancer statistics, potential biomarker in clear cell renal cell carcinoma via 2019,” CA: A Cancer Journal for Clinicians, vol. 69, no. 1, comprehensive bioinformatics analysis,” Biomedicine & pp. 7–34, 2019. Pharmacotherapy, vol. 118, p. 109264, 2019. [4] B. Zhang, Q. Wu, Z. Wang et al., “/e promising novel [20] G. Papoff, D. Presutti, C. Lalli et al., “CASP4 gene silencing in biomarkers and candidate small molecule drugs in kidney epithelial cancer cells leads to impairment of cell migration, renal clear cell carcinoma: evidence from bioinformatics cell-matrix adhesion and tissue invasion,” Scientific Reports, analysis of high-throughput data,” Molecular Genetics & vol. 8, no. 1, p. 17705, 2018. Genomic Medicine, vol. 7, no. 5, p. E607, 2019. [21] P. Bouquerel, C. Gstalder, D. Muller ¨ et al., “Essential role for [5] B. Shuch, A. Amin, A. J. Armstrong et al., “Understanding SphK1/S1P signaling to regulate hypoxia-inducible factor 2α pathologic variants of renal cell carcinoma: distilling thera- expression and activity in cancer,” Oncogenesis, vol. 5, no. 3, peutic opportunities from biologic complexity,” European p. e209, Article ID e209, 2016. Urology, vol. 67, no. 1, pp. 85–97, 2015. [22] Z. Ma, Z. Qi, Z. Shan, J. Li, J. Yang, and Z. Xu, “/e role of [6] S. Y. Lee, I. Hwang, Y. S. Park, J. Gardner, and J. Y. Ro, CRP and ATG9B expression in clear cell renal cell carci- “Metastatic lymph node ratio in advanced gastric carcinoma: noma,” Bioscience Reports, vol. 37, no. 6, Article ID a better prognostic factor than number of metastatic lymph BSR20171082, 2017. nodes?” International Journal of Oncology, vol. 36, pp. 1461– [23] F. Luan, W. Chen, M. Chen et al., “An autophagy-related long 1467, 2010. non-coding RNA signature for glioma,” FEBS Open Bio, vol. 9, [7] K. H. Choi, B. S. Kim, S. T. Oh, J. H. Yook, and B. S. Kim, no. 4, pp. 653–667, 2019. “Comparison the sixth and seventh editions of the AJCC [24] H. Zhang, X. Lu, N. Wang et al., “Autophagy-related gene staging system for t1 gastric cancer: a long-term follow-up expression is an independent prognostic indicator of glioma,” study of 2124 patients,” Gastric Cancer, vol. 20, no. 1, Oncotarget, vol. 8, no. 37, pp. 60987–61000, 2017. pp. 43–48, 2017. [25] Y. An, F. Bi, Y. You, X. Liu, and Q. Yang, “Development of a [8] P. Cheng, “A prognostic 3-long noncoding RNA signature for novel autophagy-related prognostic signature for serous patients with gastric cancer,” Journal of Cellular Biochemistry, ovarian cancer,” Journal of Cancer, vol. 9, no. 21, vol. 119, no. 11, pp. 9261–9269, 2018. pp. 4058–4071, 2018. [9] T. Johansen and T. Lamark, “Selective autophagy mediated by [26] J. Zhou, D. Hang, Y. Jiang et al., “Evaluation of genetic autophagic adapter proteins,” Autophagy, vol. 7, no. 3, variants in autophagy pathway genes as prognostic bio- pp. 279–296, 2011. markers for breast cancer,” Gene, vol. 627, pp. 549–555, 2017. [10] J. Y. Li, Y. H. Lei, N. Yao et al., “Autophagy in cancer cell [27] S.-s. Wang, G. Chen, S.-h. Li et al., “Identification and vali- death,” Chinese Journal of Cancer, vol. 36, p. 52, 2017. dation of an individualized autophagy-clinical prognostic [11] Y. Zhou, M. Jing, L. Xia et al., “Autophagy defects and related index in bladder cancer patients,” OncoTargets and erapy, genetic variations in renal cell carcinoma with eosinophilic vol. 12, pp. 3695–3712, 2019. cytoplasmic inclusions,” Scientific Reports, vol. 8, p. 9972, [28] Z. Zhou, S. Mo, W. Dai et al., “Development and validation of an autophagy score signature for the prediction of post-op- [12] L. Serrano-Oviedo, M. Ortega-Muelas, J. GarcÃ-a-Cano et al., erative survival in colorectal cancer,” Frontiers in Oncology, “Autophagic cell death associated to Sorafenib in renal cell vol. 9, p. 878, 2019. carcinoma is mediated through Akt inhibition in an ERK1/2 [29] M. Fitzgerald, B. R. Saville, and R. J. Lewis, “Decision curve independent fashion,” PLoS One, vol. 13, Article ID e0200878, analysis,” JAMA, vol. 313, no. 4, pp. 409-410, 2015. 2018. [30] K. F. Kerr, M. D. Brown, K. Zhu, and H. Janes, “Assessing the [13] S. N. Devenport and Y. M. Shah, “Functions and Implications clinical impact of risk prediction models with decision curves: of autophagy in colon cancer,” Cells, vol. 8, no. 11, p. 1349, guidance for correct interpretation and appropriate use,” Journal of Clinical Oncology, vol. 34, no. 21, pp. 2534–2540, [14] B. Wang, Y. D. Tang, Y. Tang, B. Yu, D. Gui, and H. Xu, “Expression of autophagy-related factor p62 for lung cancer diagnosis and prognosis: a systematic review and meta- analysis,” Mathematical Biosciences and Engineering, vol. 16, no. 6, pp. 6805–6821, 2019. [15] X. Gao, Y. Wang, Y. Li et al., “Huganpian, a traditional Chinese medicine, inhibits liver cancer growth in vitro and in
Journal
Journal of Oncology – Hindawi Publishing Corporation
Published: Feb 19, 2021