Differentially expressed ARGs
As shown in Figure 1, 36 differentially expressed ARGs with a false discovery rate (FDR)<0.05 and |logFC|>1 were identified from 232 ARGs. A volcano map (Figure 1A), boxplots (Figure 1B), and a heatmap (Figure 1C) indicated that 20 ARGs (BCL2, CAPN2, CCR2, CDKN1A, FAS, FKBP1B, GABARAP, HSPB8, ITPR1, MAP1LC3C, NKX2-3, NRG1, NRG2, NRG3, PINK1, PRKN, SESN2, TMEM74, TNFSF10, and TP53INP2) were down-regulated while 16 ARGs (ATG9B, ATIC, BCL2L1, BID, BIRC5, CAPN10, CD46, CDKN2A, EIF4EBP1, ERO1A, HSP90AB1, IFNG, MYC, SPHK1, TP73, and VEGFA) were overexpressed in colon tumor tissues. Functional enrichment analysis identified numerous GO and KEGG enrichment pathways (Figure 2). The 36 genes were primarily related to the molecular functions of autophagy, process utilizing autophagic mechanism, and intrinsic apoptotic signaling pathways. As seen in Figure 2A, these 36 genes are mainly related to molecular functions (MF) of autophagy, process utilizing autophagic mechanism and intrinsic apoptotic signaling pathway, they are correlated with cellular components (CC) of autophagosome, vacuolar membrane and autophagosome membrane, the genes are also involved in biological processes (BP) of ubiquitin protein ligase binding, ubiquitin-like protein ligase binding and protein kinase regulator activity. These ARGs participate in the pathways of p53 signaling pathway, apoptosis, and human cytomegalovirus infection (Figure 2B).
A forest map identified 20 ARGs that were associated with prognosis in colon cancer patients (Figure 3A). Of these 20 prognosis-related genes, six genes were determined to be protective and 14 ARGs were associated with increased risk. Results of the KEGG analysis indicated that prognostic ARGs were mainly involved in pathways of autophagy, spinocerebellar ataxia, and Huntington’s disease (Figure 3B). Prognosis-related genes were correlated with macro-autophagy, autophagy and process utilizing autophagic mechanism (Figure 3C). Mutations in these 20 genes, examined using the cBioPortal website, showed that missense mutations, amplifications and deep detection were the most common mutation types (Figure 4). Five ARGs (DAPK1, ULK1, PELP1, TSC1 and CASP3) had a mutation rate ≥3%, among which DAPK1 had the highest mutation rate.
Development of a prognosis model
Using multivariate analysis to develop a risk score for colon cancer patients, 8 ARGs were significantly related to prognosis. The risk score was defined as [Expression level of SERPINA1*(-0.11979)] + [Expression level of DAPK1*(-0.29697)] + [Expression level of MAP1LC3C*(1.50543)] + [Expression level of MAPK9*(-0.62080)] + [Expression level of TSC1*(-0.64199)] + [Expression level of ULK3*(-0.31259)] + [Expression level of CASP3*(-0.44136)] + [Expression level of WIPI1*(-0.27200)]. Based on the risk score, patients were divided into high-risk and low-risk groups using the median risk score as the cut-off point between groups (Figure 5A). Patients with higher risk scores were more likely to be deceased (Figure 5B). A heatmap was used to show differences in expression for these 8 prognosis-related ARGs between groups (Figure 5C).
Clinicopathologic characteristics of TCGA colon cancer patients were downloaded from TCGA database (Additional Table 1). Examination of the survival curves for the low-risk and high-risk patient groups, drawn using the Kaplan-Meier method (Figure 5D), showed that high-risk patients had a significantly lower probability of survival (p<0.01). Univariate and multivariate analyses were performed to identify prognosis-related factors in colon cancer patients (Figure 6A and Figure 6B). Factors with a p-value <0.05 in the univariate analysis were included in the multivariate analysis. Forest maps showed that age, pharmaceutical use, tumor invasion depth, lymph node metastasis, distant metastasis and the risk score were still significant after multivariate analysis. Therefore, the risk score was independently associated with prognosis of patients [Hazard ratio (HR)=1.537, 95% CI=1.354-1.745, p<0.001; Figure 6B]. AUC of the ROC were used to test the prediction efficiency of the prognosis model (Figure 6C). AUC of the risk score (0.701) was greater than that for any other clinicopathologic characteristics, including American Joint Committee on Cancer stage, which showed that the risk score could be reliable predictor of prognosis in colon cancer patients.
To better understand the influence of these factors on patient survival, a nomogram was drawn to predict 3- and 5-year survival rates of colon cancer cases (Figure 7A). The score obtained from the multivariate analysis was used to predict survival. Accordingly, if a 55-year old colon cancer patient with tumor of T2N0M0 stage has a high calculated risk score, his or her estimated 5-year survival rate is 40 percent according to the predicted result of nomogram model. Moreover, calibration graphs depicting the differences between nomogram-predicted and actual survival rates of colon cancer patients showed that predicted 3- and 5-year survival rates were close to the actual survival rates (Figure 7B and Figure 7C), indicating that this nomogram model accurately predicted survival. Interestingly, the nomogram model was made into a web page at https://doctorwang.shinyapps.io/DynNomapp, which could be easily accessed using desktops, tablets and smartphones (Additional Figure 2). The prognostic nomogram with online webserver is more effective for providing accurate and individualized survival prediction in colon cancer patients.
The calculated risk score was associated with other clinicopathological characteristics, including tumor infiltration depth (Figure 8A) and distant metastasis (Figure 8B), suggesting that this model may also be predictive of tumor growth and metastasis.