Abstract
Although autophagy is reported to be involved in tumorigenesis and cancer progression, its correlation with the prognosis of glioma patients remains unclear. Thus, the aim of this study was to identify prognostic autophagy-related genes, analyze their correlation with clinicopathological features of glioma, and further construct a prognostic model for glioma patients. After 139 autophagy-related genes were obtained from the GeneCards database, their expression data in glioma patients were extracted from the Chinese Glioma Genome Atlas database. Univariate and multivariate COX regression analyses were performed to identify prognostic autophagy-related genes. Ten hub autophagy-related genes associated with prognosis were identified. The autophagy risk score (ARS) was only positively correlated with histopathology (P = 0.000) and World Health Organization grade (P = 0.000). Kaplan-Meier analysis showed that the overall survival of patients with a high ARS was significantly worse than that of patients with a low ARS (hazard ratio = 1.59, 95% confidence interval = 1.25-2.03, P = 0.0001). In addition, Gene Ontology and Kyoto Encyclopedia of Genes and Genomes analyses revealed several common biological processes and signaling pathways related to the 10 hub genes in glioblastoma. A prediction model was developed for glioma patients, which demonstrated high prediction efficiency on calibration. Moreover, the area under the receiver operating characteristic curve values for 1-, 3- and 5-year survival probabilities were 0.790, 0.861, and 0.853, respectively. In conclusion, we identified 10 autophagy-related genes that can serve as novel prognostic biomarkers for glioma patients. Our prediction model accurately predicted patient outcomes, and thus, may be a valuable tool in clinical practice.
Keywords: Neoplasms, survival, nomograms
Introduction
Glioma of the brain is a common human malignancy that is harmful to human health globally. With the development of imaging technologies, the overall incidence of gliomas in the past decades has continued to increase [1]. According to an analysis of the Central Brain Tumor Registry of the United States (CBTRUS) data, an estimated 87,240 cases of primary central nervous system (CNS) tumors will be diagnosed in the United States in 2020, including 25,800 malignant tumors and 61430 non-malignant tumors [2]. At present, the main treatment strategy for gliomas worldwide remains surgery followed by postoperative adjuvant radiotherapy and chemotherapy. Although the treatment of gliomas has been improved to some extent, the prognosis of glioma patients continues to not be satisfactory. Notably, overall survival (OS) varies greatly among patients with the same pathological diagnosis [3], which shows that the traditional pathological diagnosis is insufficient for judging patients’ prognosis. In addition, the common clinical prognostic markers (e.g., isocitrate dehydrogenase [IDH] mutation 1p/19q co-deletion) only distinguish two subgroups of patients with the same histopathology. These markers are still not sufficient for more subtle stratification and cannot reflect the individual prognosis of each patient. Therefore, there is an urgent need for a new prediction model that offers greater accuracy for glioma patient prognosis. Moreover, such a model would be valuable for guiding personalized medicine for glioma patients [4].
Research in recent years has demonstrated that autophagy is a process of degradation and reuse of cellular components that plays a key role in tumorigenesis, cancer development, and metastasis [5]. Studies have also shown that tumor cells, especially under stress conditions, can obtain the energy and substances necessary for survival through autophagy, and thus, autophagy is a type of survival mechanism for tumor cells [6]. Thus, the inhibition of autophagy can reduce the tolerance of tumor cells to stress, increase the sensitivity of tumor cells to anti-tumor drugs, and improve the effect of anticancer therapy [7]. Notably, multiple preclinical studies have found that knockout of the key autophagy genes (e.g., beclin-1, Atg12, Atg5) can improve the killing ability of antineoplastic drugs, which indicates that autophagy can protect cancer cells and inhibit the efficacy of antineoplastic drugs [8,9].
These findings further suggest that autophagy-related genes play a key role in tumorigenesis and development and have a significant impact on the prognosis of cancer patients. As such, these genes may also provide prognostic markers. However, autophagy is a complex biological process involving hundreds of genes that is common in the life processes of eukaryotes. The prognostic value of autophagy genes in glioma patients remains unclear, and investigation of which autophagy-related genes have the potential to become prognostic markers is worthwhile.
Toward this end, we obtained the expression profile data for autophagy-related genes and the clinicopathological data of patients from the Chinese Glioma Genome Atlas (CGGA) dataset in order to search for independent prognostic factors by analyzing the correlations between autophagy-related genes and patients’ prognosis. Additionally, from the identified genes, we constructed a new prediction model for glioma patients’ prognosis.
Materials and methods
Screening of autophagy-related genes
The genes related to autophagy were searched in the GeneCards database (https://www.genecards.org/). Genes with a relevance score >8 were screened as autophagy-related genes in this study.
Patient selection
First, we extracted the data for all patients in the dataset 4 cohort of the CGGA (http://cgga.org.cn/index.jsp) [10], which is a database for storing clinical and pathological information on glioma patients in China. Second, patients with missing data were excluded. Finally, we included data for 269 glioma patients in our analysis.
Identification of prognostic genes
First, univariate Cox regression analysis was performed to screen autophagy-related genes significantly associated with the prognosis of glioma patients. The autophagy-related genes for which P<0.05 were further included in multivariate regression analysis, and independent prognostic autophagy-related genes were identified.
Oncomine database and human protein atlas (HPA) analysis
The Oncomine gene expression array database (www.oncomine.org) was used to assess the mRNA expression levels of the identified prognostic genes. In addition, the expression levels of the corresponding proteins in gliomas and normal tissues were reviewed in the HPA (http://www.proteinatlas.org/) [11].
Pathway analysis
From the prognostic autophagy-related genes, we determined the nine most relevant coexpression genes in the TCGA-CNS/Brain database (http://www.cbioportal.org/). Gene Ontology (GO) analysis of all prognostic genes and coexpression genes was performed using the DAVID [12] database (https://david.ncifcrf.gov/), including biological process, cellular component, and molecular function. The Cytoscape tool [13] was used to implement the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. We used the String database [14,15] to observe protein-protein interaction between the proteins encoded by the prognostic genes (https://string-db.org/).
Correlations between autophagy risk score (ARS) and clinicopathological features of patients
The ARS was constructed according to the expression of a prognostic gene and a correlation risk coefficient for the gene. The formula for the ARS is as follows: ARS = gene1*β1 + gene2*β2 + gene3*β3…….gene (n)*β (n). Chi-square test was performed to explore correlations between clinical parameters and the ARS. The relationship of the ARS to the prognosis of glioma patients was analyzed by Kaplan-Meier survival curve analysis.
Development and validation of nomogram
Based on clinical indicators and the ARS, a predictive model was constructed using the “Survival” package and “rms” package. A calibration curve and time-dependent receive operating characteristic (ROC) curve were generated to test the accuracy of the prognostic model. In addition, the “survivalROC” package was used to draw the ROC curve to compare the predictive efficiencies among the nomogram, traditional predictive models, and common clinical prognostic indicators.
Statistical analysis
All data were analyzed using SPSS software (version 23.0, Corp., Armonk, NY, USA) and R software version 3.6.1. Chi-square test was used to identify differences in classified variables. Differences in continuous variables were analyzed by the independent sample t-test. Univariate and multivariate Cox regression analyses were used to identify independent prognostic factors. The Kaplan-Meier method was applied to draw the survival curves, and the log-rank test was used to compare the survival curves. Time-dependent ROC curve analysis was performed using R package “survivalROC”. All tests were two-sided, and P-values <0.05 were considered to be statistically significant. For multiple testing, we use the Bonferroni correction method to correct the P-value.
Results
Glioma patients’ characteristics
A total of 269 patients with glioma were included in this study. The clinicopathological data of patients are summarized in Table S1. The average age of all patients was 42.58±11.83 years. The incidence of glioma was higher in males than in females, and the median survival time of all patients was 31.2 months (interquartile range, 0.7-138.1 months). The 1-, 3-, and 5-year survival rates for all glioma patients were 77.32%, 46.84%, and 34.94%, respectively.
Screening of prognostic autophagy-related genes in 269 glioma patients
To search for prognostic autophagy-related genes of glioma patients, all candidate genes related to autophagy were searched through the GeneCards database. A total of 139 autophagy-related genes with a relevance score >8 were selected for further investigation. The IRGM and BECN2 genes were subsequently excluded, because expression of these genes was not found by mRNA expression microarray analysis for glioma patients. Finally, the remaining 137 genes were entered in the univariate Cox regression analysis. On this analysis, 86 autophagy-related genes were significantly associated with the prognosis of glioma patients (Table S2). These 86 genes were further incorporated into the multivariate analysis, and 10 hub autophagy-relate genes associated with prognosis were finally identified (Table 1). The detailed information for the 10 hub autophagy-related genes is presented in Table 2. The expression data for the 10 autophagy-related genes were obtained from the mRNA array data of dataset 4. The differential expression levels of the autophagy-related genes between high- and low-grade glioma cases were compared (Figure 1). Furthermore, we also analyzed the mRNA levels of autophagy-related genes based on data from the Oncomine database and the protein expression levels based on data from the HPA database (Table 3; Figure S1).
Table 1.
Results of univariate and multivariate Cox analyses of prognostic genes for OS in glioma
| Features | Univariate Cox analysis | Multivariate Cox analysis | ||||
|---|---|---|---|---|---|---|
|
|
|
|||||
| HR | 95% CI | P-value | HR | 95% CI | P-value | |
| ULK1 | 0.67 | 0.52-0.87 | 0.002 | 2.27 | 1.14-4.54 | 0.020 |
| ATG10 | 0.51 | 0.34-0.77 | 0.001 | 2.59 | 1.31-5.13 | 0.006 |
| ATG16L2 | 0.57 | 0.45-0.73 | 0.000 | 0.48 | 0.27-0.86 | 0.013 |
| RB1CC1 | 0.65 | 0.53-0.79 | 0.000 | 2.03 | 1.06-3.88 | 0.032 |
| RUBCNL | 1.12 | 1.01-1.24 | 0.038 | 0.71 | 0.54-0.94 | 0.016 |
| PRKN | 0.46 | 0.34-0.62 | 0.000 | 0.45 | 0.23-0.85 | 0.015 |
| GSK3B | 0.49 | 0.35-0.69 | 0.000 | 3.20 | 1.42-7.24 | 0.005 |
| TBC1D5 | 0.30 | 0.23-0.41 | 0.000 | 0.27 | 0.13-0.56 | 0.000 |
| PIK3CB | 0.70 | 0.6-0.82 | 0.000 | 0.57 | 0.34-0.96 | 0.034 |
| RAB33B | 1.40 | 1.19-1.64 | 0.000 | 1.54 | 1.14-2.07 | 0.004 |
HR, hazard ratio; OS, overall survival; CI, confidence interval.
Table 2.
Functional details of the 10 independently prognostic autophagy-related genes
| Gene symbol | Full name | Category | Relevance score |
|---|---|---|---|
| ULK1 | Unc-51 Like Autophagy Activating Kinase 1 | Protein Coding | 37.03 |
| ATG10 | Autophagy Related 10 | Protein Coding | 28.09 |
| ATG16L2 | Autophagy Related 16 Like 2 | Protein Coding | 21.84 |
| RB1CC1 | RB1 Inducible Coiled-Coil 1 | Protein Coding | 20.16 |
| RUBCNL | Rubicon Like Autophagy Enhancer | Protein Coding | 16.07 |
| PRKN | Parkin RBR E3 Ubiquitin Protein Ligase | Protein Coding | 11.71 |
| GSK3B | Glycogen Synthase Kinase 3 Beta | Protein Coding | 11.21 |
| TBC1D5 | TBC1 Domain Family Member 5 | Protein Coding | 9.99 |
| PIK3CB | Phosphatidylinositol-4,5-Bisphosphate 3-Kinase Catalytic Subunit Beta | Protein Coding | 9.16 |
| RAB33B | RAB33B, Member RAS Oncogene Family | Protein Coding | 8.48 |
Figure 1.
Differential mRNA expression of autophagy-related genes between low-grade glioma and high-grade glioma groups. ***P<0.001, ****P<0.0001.
Table 3.
Significant differences in transcription levels of autophagy-related genes between glioma and normal brain tissues (Oncomine)
| Gene | Type of Glioma vs. Normal | T-test | P-value | Reference |
|---|---|---|---|---|
| ULK1 | Glioblastoma | -7.45 | 5.94E-06 | Bredel Brain 2 |
| ATG10 | Glioblastoma | 3.725 | 0.007 | Bredel Brain 2 |
| RB1CC1 | Glioblastoma | -13.189 | 1.34E-09 | Bredel Brain 2 |
| RUBCNL | Glioblastoma | 4.789 | 9.51E-05 | Bredel Brain 2 |
| GSK3B | Glioblastoma | -3.857 | 0.001 | Bredel Brain 2 |
| PIK3CB | Glioblastoma | -7.286 | 4.28E-04 | Bredel Brain 2 |
| PRKN | Anaplastic Astrocytoma | -4.988 | 6.61E-06 | Sun Brain |
| TBC1D5 | Glioblastoma | -7.22 | 1.12E-09 | Sun Brain |
| RAB33B | Glioblastoma | 5.097 | 2.52E-04 | TCGA Brain |
| ATG16L2 | Glioblastoma | -2.568 | 0.005 | TCGA Brain 2 |
Biological processes related to hub autophagy-related genes
To identify the biological functions of these genes in glioma, we searched for the nine most relevant coexpression genes in the TCGA-CNS/Brain (http://www.cbioportal.org/). Then, GO enrichment analysis and KEGG analysis were conducted. GO enrichment analysis revealed that these genes were mostly enriched in autophagy, macroautophagy, regulation of MAP kinase activity, and intracellular protein transport (Figure 2A-C). From the pathway analysis, these genes were mainly involved in autophagy signal, the B cell receptor signaling pathway, and the prolactin signaling pathway (Figure 2D). In addition, a protein-protein interaction network of the proteins encoded by these 10 autophagy-related genes was constructed (Figure S2).
Figure 2.
Functional enrichment analysis and significantly enriched terms of 10 prognostic autophagy-related genes in gliomas. A. GO biological process analysis. B. Cellular component analysis. C. Molecular function analysis. D. KEGG pathway analysis of prognostic autophagy-related genes. The size represents the gene number in this KEGG term.
Correlations between ARS and clinicopathological features
To identify any correlations between autophagy-related genes and clinicopathological features, the ARS was calculated for each patient according to the formula based on the expression of autophagy-related genes and their correlation coefficients. The formula for calculating the ARS was as follows: Autophagy Risk Score (ARS) = ULK1* (0.820) + ATG10* (0.952) + ATG16L2* (-0.734) + RB1CC1* (0.708) + RUBCNL* (-0.342) + PRKN* (-0.799) + GSK3B* (1.163) + TBC1D5* (-1.309) + PIK3CB* (-0.562) + RAB33B* (0.432).
According to the median ARS value, the 269 patients were divided into a high-risk group (n = 135) and a low-risk group (n = 134), and the correlations between the ARS and the individual clinicopathological features were analyzed by chi-square test. The results showed that the ARS was significantly correlated with World Health Organization (WHO) grade (P = 0.000) and histopathology (P = 0.000), but not with age (P = 0.668), gender (P = 0.055), IDH mutation (P = 0.672), radiotherapy (P = 0.239), chemotherapy (P = 0.059), or recurrence (P = 0.362) (Table 4). We also further assessed the distribution of ARS values among the 269 patients stratified by WHO grade and histopathology. Consistently, the results revealed that the ARS was significantly associated with WHO grade and histopathology (Figure 3).
Table 4.
Correlations between autophagy risk score (ARS) and clinicopathological features in patients with glioma
| Characteristics | Autophagy Risk Score | P-value | |
|---|---|---|---|
|
| |||
| Low-risk Score | High-risk Score | ||
| Age, years | |||
| ≥42 | 71 | 68 | 0.668 |
| <42 | 63 | 67 | |
| Gender | |||
| Male | 85 | 69 | 0.055 |
| Female | 49 | 66 | |
| WHO grade | |||
| II | 68 | 32 | 0.000 |
| III | 15 | 38 | |
| IV | 51 | 65 | |
| Histopathology | |||
| O | 9 | 7 | 0.000 |
| OA | 18 | 16 | |
| A | 41 | 9 | |
| AO | 2 | 11 | |
| AOA | 9 | 17 | |
| AA | 4 | 10 | |
| GBM | 51 | 65 | |
| IDH | |||
| Mutation | 62 | 59 | 0.672 |
| Wild-type | 72 | 76 | |
| Radiotherapy | |||
| Yes | 123 | 118 | 0.239 |
| No | 11 | 17 | |
| Chemotherapy | |||
| Yes | 65 | 81 | 0.059 |
| No | 69 | 54 | |
| Recurrence | |||
| Yes | 8 | 12 | 0.362 |
| No | 126 | 123 | |
WHO, World Health Organization; O, oligodendroglioma; OA, oligoastrocytoma; A, astrocytoma; AO, anaplastic oligodendroglioma; AOA, anaplastic oligoastrocytoma; AA, anaplastic astrocytoma; GBM, glioblastoma; IDH, isocitrate dehydrogenase.
Figure 3.
Correlations between autophagy risk score (ARS) and clinicopathological features in gliomas. A and B. Differential distribution of ARS in glioma patients stratified by grade and histopathology, separately. ****P<0.0001.
Correlation between ARS and OS of patients
To assess the prognostic value of autophagy-related genes in glioma patients, Kaplan-Meier analysis was performed to compare patients’ OS between the high- (n = 135) and low-risk score groups (n = 134). The results indicated that the OS of patients with a high ARS was significantly worse than that of patients with a low ARS (hazard ratio [HR] = 1.59, 95% confidence interval [CI] = 1.25-2.03, P = 0.0001), and the median survival times of the two groups were 53.43 months and 18.20 months, respectively (Figure 4). To further assess the prognostic value of ARS in subgroups of glioma patients, all patients were further stratified by age, gender, WHO grade, IDH mutation, and recurrence. The results similarly showed that the OS of patients with a high ARS was significantly worse than that of patients with a low ARS, except in the recurrence subgroup (Figure S3A-J).
Figure 4.
Kaplan-Meier survival analysis of the predictive ability of ARS for OS in glioma. The OS of patients with a low-risk score was significantly longer than that of patients with a high-risk score (P = 0.0001).
In addition, we also analyzed the effect of each autophagy gene on the prognosis of glioma patients in the GEPIA database. The results are presented in Figure S4A-J.
To clarify the role of the hub autophagy-related genes in glioblastoma, we analyzed the expression changes of these 10 genes through the TCGA dataset of Cbioportal database. The glioblastoma dataset (TCGA, cell 2013) [16] included 543 cases, of which 152 cases had mRNA expression profile data, that could be used to analyze genetic changes and their correlation with pathological features. Changes in the expression of autophagy-related genes we-re found in 40 (26.3%) of the 152 cases (Figure S5). High-frequency genetic changes often suggested that these genes could play an important role in tumorigenesis and the development of gliomas.
A personalized prognostic prediction model
To filter the clinicopathological factors for use in developing the prediction model, univariate and multivariate Cox regression analyses were performed. Multivariate analysis revealed five independent prognostic risk factors, of which age (P = 0.010), ARS (P = 0.004), histopathology (P = 0.012), and recurrence (P = 0.004) predicted a poor prognostic and radiotherapy (P = 0.003) predicted a good prognosis (Table 5).
Table 5.
Results of univariate and multivariate Cox analyses identifying clinicopathological factors associated with the prognosis of glioma patients
| Characteristic | Univariate analysis | Multivariate analysis | ||||
|---|---|---|---|---|---|---|
|
|
|
|||||
| HR | 95% CI | P-value | HR | 95% CI | P-value | |
| Age | 1.03 | 1.02-1.04 | 0.000* | 1.02 | 1.01-1.03 | 0.010* |
| Gender | 0.75 | 0.59-0.96 | 0.022* | 0.77 | 0.60-1.00 | 0.050 |
| WHO grade | 2.03 | 1.75-2.36 | 0.000* | 1.04 | 0.62-1.75 | 0.891 |
| Histopathology | 1.35 | 1.27-1.44 | 0.000* | 1.31 | 1.06-1.62 | 0.012* |
| IDH mutation | 0.53 | 0.42-0.68 | 0.000* | 0.84 | 0.64-1.11 | 0.211 |
| Radiotherapy | 0.54 | 0.36-0.80 | 0.002* | 0.53 | 0.35-0.80 | 0.003* |
| Chemotherapy | 1.39 | 1.09-1.78 | 0.008* | 0.80 | 0.60-1.07 | 0.131 |
| Recurrence | 1.98 | 1.25-3.14 | 0.004* | 2.11 | 1.28-3.47 | 0.004* |
| ARS | 1.22 | 1.08-1.36 | 0.001* | 1.19 | 1.06-1.34 | 0.004* |
significant variables.
HR, hazard ratio; CI, confidence interval; WHO, World Health Organization; IDH, isocitrate dehydrogenase; ARS, autophagy risk score.
Using the five identified prognostic indicators, a prediction model was constructed using R software. As shown in Figure 5, the first row includes the score information, totaling 100 points. The second to eighth rows are age, histopathology, recurrence, radiotherapy, and ARS. The scores for individual indicators were obtained by comparing the first row, and the total points were calculated. Then, the 1-, 3-, and 5-year survival rates of the patients were obtained by comparing the survival rates in rows 8-10. A higher point total indicated a worse prognosis.
Figure 5.
Nomogram for predicting 1-, 3-, and 5-year survival rates in glioma patients.
Evaluation of prediction model by calibration and time-dependent ROC curve analyses
A calibration curve and time-dependent ROC curve were prepared to test the accuracy of the prognostic model. We could see from the calibration curves that the 1-, 3- and 5-year survival curves predicted by the model were very close to the real survival curves, which indicated that the prediction power of the model was high (Figure 6A-C).
Figure 6.
Calibration curves and time-dependent ROC curves for validation of the nomogram. A-C. Calibration curves for predicting OS at 1, 3, and 5 years show the plots for each model in terms of the agreement between predicted (blue line) and observed outcomes (the 45-degree line). D. Time-dependent ROC curves for the ability of the nomogram to predict the 1-, 3-, and 5-year survival rates in glioma patients.
Ultimately, we further generated the time-dependent ROC curve to evaluate the accuracy of the prediction model. The area under the ROC curve (AUC) values for 1-, 3-, and 5-year survival were 0.790, 0.861, and 0.853, respectively (Figure 6D).
Comparison of predictive efficiency among different prognostic models
To further evaluate the superiority of the clinical application value of the nomogram developed in this study, we compared the prediction accuracies among the nomogram, the traditional prediction model (i.e., age, WHO classification and histopathology), a gene-model based on 10 autophagy-related genes, the IDH mutation clinical biomarker, and the WHO grade through ROC curve analysis. The results showed that the accuracy of the nomogram for predicting the prognosis of patients was significantly better than that of the traditional predictive model, IDH mutation, and WHO grade (Figure 7).
Figure 7.
Comparison of the predictive efficiencies of different prognostic models and common markers by ROC curve analysis. We evaluated and compared the predictive performances of different models by calculating the AUC values.
Discussion
Glioma is a common malignant disease with high morbidity and mortality, and the poor prognosis of glioma patients has always been the most intractable problem faced by clinicians. Thus, a better understanding of the molecular pathological mechanisms of glioma is urgently needed, along with the identification of potential therapeutic targets and new prognostic biomarkers.
In recent years, autophagy of tumor cells has gradually attracted considerable attention as a potential mechanism that could be employed in novel treatments. So far, it has been accepted that during autophagy, a common biological process in the life activities of eukaryotes, large amounts of cytoplasmic macromolecules and organelles are degraded in membrane vesicles [17]. Overall, it is a complex biological process involving hundreds of molecules, and each step is regulated by different autophagy-related genes [18]. As described, mounting evidence demonstrates that many autophagy-related genes (e.g., ULK1, ATG10, GSK3B, and ATG16L2) are up-regulated in tumor tissues where they act as oncogenes to regulate tumor proliferation, apoptosis, and metastasis [19-21]. For example, Zhu et al. confirmed that autophagy is necessary for the regulation of epithelial-mesenchymal transition (EMT) by hypoxia-inducible factor 1α (HIF-1α) as well as the metastatic ability of prostate cancer stem cells [22]. Likewise, endometrial carcinomas harbor frequent alterations in components of the autophagy pathway, including changes in gene copy number and mutations, in particular in the oncogene PIK3CA, the gene that encodes the PI3K catalytic subunit p110α, and the tumor suppressor PTEN. PIK3CB, which encodes the other ubiquitously expressed class I isoform p110β, is less frequently altered, but the few mutations identified to date have all been shown to be oncogenic [23].
In the present study, we first screened 139 genes significantly associated with autophagy. Of these, univariate following by multivariate Cox analysis identified 10 autophagy-related genes (ULK1, ATG10, RB1CC1, GSK3B, PIK3CB, RUBCNL, PRKN, TBC1D5, and RAB33B) as potentially independent prognostic risk biomarkers for glioma patients. We calculated an ARS for each patient and found that the ARS was significantly correlated with WHO grade and histopathology, making it useful for distinguishing the prognosis of patients in different subgroups.
Accumulating research has continued to suggest that autophagy-related genes can act as oncogenes and result in poor prognosis of tumor patients. In one example, Chen et al. discovered that ULK1 is upregulated in gastric cancer and associated with tumor T stage and recurrence [19]. In 2015, Ruvolo et al. first reported that glycogen synthase kinase (GSK)-3β is regulated by phosphorylation of AKT kinase, and phosphorylated GSK-3β is an important poor prognostic factor in patients with acute myeloid leukemia [24]. Similarly, ATG10 is reported to be involved in the proliferation and migration of lung cancer cells, and its overexpression is a poor prognostic biomarker in patients with lung cancer [25].
Furthermore, in the present study, GO and KEGG analyses further found that the 10 prognostic genes were mainly involved in tumor autophagy, regulation of MAP kinase activity and intracellular protein transport, and Erk tumor signaling. Previous research has confirmed that activated ULK1 forms a complex with p38MAPK, which is found at significantly increased levels under a tumor burden and participates in the activation of autophagy [26]. Jiang et al. provided evidence that autophagy is involved in the MEK1/ERK1/2 signaling pathway [27]. Additionally, several recently identified autophagy substrates reveal novel functions of autophagy early in the metastatic cascade in the direct regulation of the EMT as well as tumor cell migration and invasion [28]. Our results suggest that these autophagy-related oncogenes are involved in the biological processes of tumorigenesis and development, which is consistent with the previously reported results [29].
With the development of genomics technology, greater attention has been given to the molecular markers, such as IDH mutation and 1p/19q co-deletion, related to the pathological features of tumors in order to judge the prognosis of patients more accurately. Admittedly, we often also find that the prognosis of some patients with the same pathological diagnosis can vary greatly, and even some glioblastoma patients with long-term survival have a better prognosis than patients with diffuse astrocytoma [30]. Thus, it is evident that the traditional histopathology and existing prognostic markers are still insufficient for judging the prognosis of glioma patients. Accurate prognosis is the key to personalized medicine. Thus, we combined autophagy-related prognostic markers with pathological features to construct a visual prognostic prediction model for glioma patients.
The presented nomogram is a graphical prediction model based on statistical algorithms, which confer points to each variable (i.e., age, histopathology, recurrence, radiotherapy, and ARS). By summarizing all the points, it allows for the survival probability of each patient to be read from the 1-, 3-, and 5-year survival rates. Through a comparison of ROC curves, we found that the developed nomogram is more accurate than the traditional clinical model and autophagy gene model. Thus, we proposed that the ARS derived from autophagy-related gene expression could be incorporated into a predictive nomogram model to better predict clinical outcomes. This model supports the value of integrating gene signature and traditional prognostic factors to more accurately predict the prognosis of glioma patients.
There are some limitations in this study. The nomogram included 10 autophagy-related genes, and thus, could likely be further optimized and made more suitable for clinical practice. Additionally, future research should seek to elucidate the underlying mechanisms of the genes linked to poor prognosis in glioma patients in vivo and in vitro. The next steps include studying the effects of each gene on the occurrence and development of glioma disease, to determine the potential role of each gene as an oncogene. Such research will provide insight into the underlying mechanisms for the poor prognosis of glioma patients and further provide a theoretical basis for the development of targeted therapies.
In conclusion, we identified 10 autophagy-related genes with prognostic value in glioma cohorts. These genes were independent indicators of prognosis, and a nomogram based on the 10 autophagy-related genes as well as clinicopathological features showed good efficacy for predicting the 1-, 3-, and 5-year survival probabilities for glioma patients. Our prediction model accurately predicted survival outcomes in the included patients, indicating its potential utility in clinical practice.
Acknowledgements
This paper was supported by Zhejiang Province welfare technology applied research project (grant number: 2017C37111).
Disclosure of conflict of interest
None.
Supporting Information
References
- 1.Nomura E, Ioka A, Tsukuma H. Trends in the incidence of primary intracranial tumors in Osaka, Japan. Jpn J Clin Oncol. 2011;41:291–294. doi: 10.1093/jjco/hyq204. [DOI] [PubMed] [Google Scholar]
- 2.Ostrom QT, Cioffi G, Gittleman H, Patil N, Waite K, Kruchko C, Barnholtz-Sloan JS. CBTRUS statistical report: primary brain and other central nervous system tumors diagnosed in the United States in 2012-2016. Neuro Oncol. 2019;21:v1–v100. doi: 10.1093/neuonc/noz150. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Field KM, Rosenthal MA, Yilmaz M, Tacey M, Drummond K. Comparison between poor and long-term survivors with glioblastoma: review of an Australian dataset. Asia Pac J Clin Oncol. 2014;10:153–161. doi: 10.1111/ajco.12076. [DOI] [PubMed] [Google Scholar]
- 4.Kamel HFM, Al-Amodi HSAB. Exploitation of gene expression and cancer biomarkers in paving the path to era of personalized medicine. Genomics Proteomics Bioinformatics. 2017;15:220–235. doi: 10.1016/j.gpb.2016.11.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Gewirtz DA. The four faces of autophagy: implications for cancer therapy. Cancer Res. 2014;74:647–651. doi: 10.1158/0008-5472.CAN-13-2966. [DOI] [PubMed] [Google Scholar]
- 6.Strohecker AM, White E. Autophagy promotes BrafV600E-driven lung tumorigenesis by preserving mitochondrial metabolism. Autophagy. 2014;10:384–385. doi: 10.4161/auto.27320. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Heydt Q, Larrue C, Saland E, Bertoli S, Sarry JE, Besson A, Manenti S, Joffre C, Mansat-De Mas V. Oncogenic FLT3-ITD supports autophagy via ATF4 in acute myeloid leukemia. Oncogene. 2018;37:787–797. doi: 10.1038/onc.2017.376. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Apel A, Herr I, Schwarz H, Rodemann HP, Mayer A. Blocked autophagy sensitizes resistant carcinoma cells to radiation therapy. Cancer Res. 2008;68:1485–1494. doi: 10.1158/0008-5472.CAN-07-0562. [DOI] [PubMed] [Google Scholar]
- 9.Carew JS, Nawrocki ST, Kahue CN, Zhang H, Yang C, Chung L, Houghton JA, Huang P, Giles FJ, Cleveland JL. Targeting autophagy augments the anticancer activity of the histone deacetylase inhibitor SAHA to overcome Bcr-Abl-mediated drug resistance. Blood. 2007;110:313–322. doi: 10.1182/blood-2006-10-050260. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Zhao Z, Meng F, Wang W, Wang Z, Zhang C, Jiang T. Comprehensive RNA-seq transcriptomic profiling in the malignant progression of gliomas. Sci Data. 2017;4:170024. doi: 10.1038/sdata.2017.24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Pontén F, Jirström K, Uhlen M. The human protein Atlas--a tool for pathology. J Pathol. 2008;216:387–393. doi: 10.1002/path.2440. [DOI] [PubMed] [Google Scholar]
- 12.Huang DW, Sherman BT, Tan Q, Kir J, Liu D, Bryant D, Guo Y, Stephens R, Baseler MW, Lane HC, Lempicki RA. DAVID bioinformatics resources: expanded annotation database and novel algorithms to better extract biology from large gene lists. Nucleic Acids Res. 2007;35:W169–175. doi: 10.1093/nar/gkm415. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003;13:2498–2504. doi: 10.1101/gr.1239303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Szklarczyk D, Gable AL, Lyon D, Junge A, Wyder S, Huerta-Cepas J, Simonovic M, Doncheva NT, Morris JH, Bork P, Jensen LJ, Mering CV. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res. 2019;47:D607–D613. doi: 10.1093/nar/gky1131. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Szklarczyk D, Morris JH, Cook H, Kuhn M, Wyder S, Simonovic M, Santos A, Doncheva NT, Roth A, Bork P, Jensen LJ, von Mering C. The STRING database in 2017: quality-controlled protein-protein association networks, made broadly accessible. Nucleic Acids Res. 2017;45:D362–D368. doi: 10.1093/nar/gkw937. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Brennan CW, Verhaak RG, McKenna A, Campos B, Noushmehr H, Salama SR, Zheng S, Chakravarty D, Sanborn JZ, Berman SH, Beroukhim R, Bernard B, Wu CJ, Genovese G, Shmulevich I, Barnholtz-Sloan J, Zou L, Vegesna R, Shukla SA, Ciriello G, Yung WK, Zhang W, Sougnez C, Mikkelsen T, Aldape K, Bigner DD, Van Meir EG, Prados M, Sloan A, Black KL, Eschbacher J, Finocchiaro G, Friedman W, Andrews DW, Guha A, Iacocca M, O’Neill BP, Foltz G, Myers J, Weisenberger DJ, Penny R, Kucherlapati R, Perou CM, Hayes DN, Gibbs R, Marra M, Mills GB, Lander E, Spellman P, Wilson R, Sander C, Weinstein J, Meyerson M, Gabriel S, Laird PW, Haussler D, Getz G, Chin L. The somatic genomic landscape of glioblastoma. Cell. 2013;155:462–477. doi: 10.1016/j.cell.2013.09.034. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Boya P, Reggiori F, Codogno P. Emerging regulation and functions of autophagy. Nat Cell Biol. 2013;15:713–720. doi: 10.1038/ncb2788. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Pyo JO, Nah J, Jung YK. Molecules and their functions in autophagy. Exp Mol Med. 2012;44:73–80. doi: 10.3858/emm.2012.44.2.029. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Chen MB, Ji XZ, Liu YY, Zeng P, Xu XY, Ma R, Guo ZD, Lu JW, Feng JF. Ulk1 over-expression in human gastric cancer is correlated with patients’ T classification and cancer relapse. Oncotarget. 2017;8:33704–33712. doi: 10.18632/oncotarget.16734. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Liao W, Liu F, Zhang H, Liao W, Xu N, Xie W, Zhang Y. Upregulation of GPNCA is associated with poor prognosis through enhancement of tumor growth via regulating GSK3B. Sci Rep. 2020;10:2044. doi: 10.1038/s41598-020-58729-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Li QX, Zhou X, Huang TT, Tang Y, Liu B, Peng P, Sun L, Wang YH, Yuan XL. The Thr300Ala variant of ATG16L1 is associated with decreased risk of brain metastasis in patients with non-small cell lung cancer. Autophagy. 2017;13:1053–1063. doi: 10.1080/15548627.2017.1308997. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Zhu H, Wang D, Zhang L, Xie X, Wu Y, Liu Y, Shao G, Su Z. Upregulation of autophagy by hypoxia-inducible factor-1alpha promotes EMT and metastatic ability of CD133+ pancreatic cancer stem-like cells during intermittent hypoxia. Oncol Rep. 2014;32:935–942. doi: 10.3892/or.2014.3298. [DOI] [PubMed] [Google Scholar]
- 23.Mazloumi Gavgani F, Smith Arnesen V, Jacobsen RG, Krakstad C, Hoivik EA, Lewis AE. Class I phosphoinositide 3-kinase PIK3CA/p110alpha and PIK3CB/p110beta isoforms in endometrial cancer. Int J Mol Sci. 2018;19:3931. doi: 10.3390/ijms19123931. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Ruvolo PP, Qiu Y, Coombes KR, Zhang N, Neeley ES, Ruvolo VR, Hail N Jr, Borthakur G, Konopleva M, Andreeff M, Kornblau SM. Phosphorylation of GSK3alpha/beta correlates with activation of AKT and is prognostic for poor overall survival in acute myeloid leukemia patients. BBA Clin. 2015;4:59–68. doi: 10.1016/j.bbacli.2015.07.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Xie K, Liang C, Li Q, Yan C, Wang C, Gu Y, Zhu M, Du F, Wang H, Dai J, Liu X, Jin G, Shen H, Ma H, Hu Z. Role of ATG10 expression quantitative trait loci in non-small cell lung cancer survival. Int J Cancer. 2016;139:1564–1573. doi: 10.1002/ijc.30205. [DOI] [PubMed] [Google Scholar]
- 26.Liu Z, Sin KWT, Ding H, Doan HA, Gao S, Miao H, Wei Y, Wang Y, Zhang G, Li YP. p38beta MAPK mediates ULK1-dependent induction of autophagy in skeletal muscle of tumor-bearing mice. Cell Stress. 2018;2:311–324. doi: 10.15698/cst2018.11.163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Jiang Y, Zhang Y, Chu F, Xu L, Wu H. Circ_0032821 acts as an oncogene in cell proliferation, metastasis and autophagy in human gastric cancer cells in vitro and in vivo through activating MEK1/ERK1/2 signaling pathway. Cancer Cell Int. 2020;20:74. doi: 10.1186/s12935-020-1151-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Kenific CM, Stehbens SJ, Goldsmith J, Leidal AM, Faure N, Ye J, Wittmann T, Debnath J. NBR1 enables autophagy-dependent focal adhesion turnover. J Cell Biol. 2016;212:577–590. doi: 10.1083/jcb.201503075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Liu J, Long S, Wang H, Liu N, Zhang C, Zhang L, Zhang Y. Blocking AMPK/ULK1-dependent autophagy promoted apoptosis and suppressed colon cancer growth. Cancer Cell Int. 2019;19:336. doi: 10.1186/s12935-019-1054-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Das P, Puri T, Jha P, Pathak P, Joshi N, Suri V, Sharma MC, Sharma BS, Mahapatra AK, Suri A, Sarkar C. A clinicopathological and molecular analysis of glioblastoma multiforme with long-term survival. J Clin Neurosci. 2011;18:66–70. doi: 10.1016/j.jocn.2010.04.050. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.