Integrative analysis of somatic mutations and differential expression profiles in glioblastoma based on aging acceleration

Abstract

Background: Glioblastoma (GBM) is an aggressive brain tumor and the mechanisms of progression are very complex. Accelerated aging is a driving factor of GBM. However, there has not been thorough research about the mechanisms of GBM progression based on aging acceleration. Methods: The aging predictor was modeled based on normal brain samples. Then an aging acceleration background network was constructed to explore GBM mechanisms. Results: The accelerated aging-related mechanisms provided an innovative way to study GBM, wherein integrative analysis of somatic mutations and differential expression revealed key pathologic characteristics. Moreover, the influence of the immune system, the nervous system and other critical factors on GBM were identified. The survival analysis also disclosed crucial GBM markers. Conclusion: An integrative analysis of multi-omics data based on aging acceleration identified new driving factors for GBM.

Introduction

Much research has indicated that there are relationships between cancers and aging; further, it also has been held that cancers cannot be isolated from the effects of accumulated genetic mutations without accelerated aging [1,2]. Fortunately, integrating genomics and transcriptomics has been deeply rooted in the public mind [3]. Therefore, exploring the mechanisms between aging and traditional cancer markers (i.e. mutations and differential expression) is a powerful approach to study the progression of GBM.

Actually, cancers may be promoted by a decline in cellular functions, (i.e. a dysregulated immune system), often induced by accelerated aging, which has been identified as an intrinsic risk factor for cancers [3]. According to statistics, the majority of cancer victims are elderly. This highlighted the necessity for the clinical practice of geriatric oncology.

Among past achievements, a series of GBM markers has been identified (i.e. through differential expression) [4]. Fortunately, The Cancer Genome Atlas (TCGA) database generated a large amount of required cancer data, including somatic mutations, gene expression, and clinical follow-up profiles; the GEO platform also provided relative profiles (i.e. healthy persons). However, an integrative analysis of GBM based on aging acceleration has not been studied thoroughly.

In this manuscript, a series of works were represented by the following flow chart (Figure 1): (1) modelling the aging predictor; (2) exploring crucial accelerated aging-related somatic mutations based on aging acceleration; (3) integrating the accelerated aging-related mutations and differential expression based on the aging acceleration background network; (4) investigating the biologic functions of the “accelerated aging-related SNPs to differential expression” modules. In short, the mechanisms of GBM progression associated with aging acceleration were explored in this paper.

Figure 1.

The computational flow chart.

Materials and methods

Modelling the aging predictor using normal brain data

We obtained the transcriptional profiles of healthy persons (584 samples) from the Gene Expression Omnibus (GEO) platform (GSE15745). The brain tissues included the cerebellum, frontal cortex, pons, and temporal cortex. We then divided all samples into two groups (old age samples ≥50 and young samples < 50). Then about two-thirds of the samples were randomly selected as the training subsets and the remaining proportion were set as the test subsets for each age group, respectively. Ultimately, the training subsets in the young and old groups were combined into the total training set, as well as the total test set.

The feature selection pipeline of the aging predictor was performed as follows:

(1) The weight of each feature was calculated in the training sets with the help of the ReliefF algorithm. For classification, the ReliefF used K nearest neighbors per class, ranking indices of columns ordered by attribute importance, with large positive weights assigned to important attributes. In this work, the value of K was set to 10.

(2) According to the law of decreasing importance of the characteristics, the test set features were also sorted based on the training data.

(3) The prediction result of the age group was calculated using the Naive Bayesian classifier, assigning each sample to the most probable class (using the maximum a posteriori decision rule) (Equation 1):

Where X was each sample, Y was the phenotype (the class of each class, i.e. the young or old age group in this work), and π (Y=k) was the prior probability of each class (set as 0.5 in this work).

(4) To pick out the optimal model predictor which contained the maximum average correct rate, we adopted the five-fold cross-validation for the total training set; then taking advantage of the selected model, we computed the accuracy in the total test set. Consequently, both the learning curve and the ROC curve were summarized.

Accelerated aging patterns in GBM

We downloaded cancer genome data from the TCGA and GEO (GSE15745) database, and standardized 11,378 cancer genes’ (the genes shared by both normal tissue and cancerous tissue were selected) expression levels. As a comparison with healthy tissue, the best model was selected as mentioned in the normal samples. Then in line with prediction outcomes, a posteriori probabilities (in the Bayesian classifier) were treated as aging scores. The median aging scores were defined as an overall grade for the certain age groups (≥50 years old, ≥60 years old, ≥70 years old and ≥80 years old) and the differences between GBM patients and the healthy group were revealed. The Kruskal Wallis test was used to test the difference between the healthy people and GBM patients, and significance was set at p-value < 0.05.

Identifying accelerated aging related somatic mutations

Since the mutation and the transcriptional profiles contain different genes, we extracted the identical genes shared by both mutation and the expression profiles, and then selected relevant mutation data. To pick out the SNP set with the best fitting result, the elastic net algorithm was used. The five-fold cross-validation was performed, which calculated a least-squares regression coefficients for a set of regularization coefficients (Equation 2):

Where B was the regression coefficient, N was the number of training data, and λ was the penalty term (determined by the 5-fold cross validation in this work).

In order to select the optimal model in the five-fold cross-validation progress, the minimum average of the mean squared error (MSE) was used as the standard. As a result, the elastic net method was used on the total data to determine the number of selected mutations, if the coefficient was not 0.

Constructing the aging acceleration background network

Both the Pearson correlation coefficient and the partial correlation coefficient (based on the aging score) between each pair of genes were calculated. Whenever correlation and partial correlation coefficients had opposite signs, the relevant two genes might be related to aging acceleration. Then the criterion of P < 0.05 and the False Discovery Rate (FDR) < 0.2 in either correlation or partial correlation coefficient was used to determine each pair of genes to form the edge of the background network. As a result, each module of “accelerated aging related mutation to differential expression” was identified by exploring the shortest path in the aging acceleration network using the Dijkstra algorithm.

Enrichment analysis

To find vital biological significance, enrichment analysis was put to use, using Gene Ontology terms and KEGG pathways, both of which stem from Gene Set Enrichment Analysis (GSEA) platform (http://software.broadinstitute.org/gsea/downloads.jsp). We adopted the hypergeometric test to figure out significantly enriched KEGG pathways or Gene Ontology (GO) Biological Process (BP) terms. The hypergeometric distribution formula for test of the degree of enrichment about the selected gene sets is shown below (Equation 3).

Where N was the total number of genes in the gene list, M was the number of known gene sets (i.e. KEGG pathway, GO terms), n was the number of identified genes and i was the number of shared genes between known gene sets and identified genes. If P < 0.05 and FDR < 0.2, the selected GO terms or KEGG pathways were considered as a significantly enriched module.

Survival analysis

Whether the module had a significant effect on the patient’s survival time or not still was worth of exploring. In order to study survival results, all patients were divided into a high expression group and low expression group based on the median of each gene expression. The Kaplan-Meier method was used to draw the survival curve and the log-rank test was used to check if there was a significant difference between the survival curves of the two groups, where P < 0.05 was used as the primary standard.

Results

The aging model revealed crucial GBM markers based on aging acceleration

The aging predictor was modeled based on transcriptional profiles using the Naive Bayesian classifier (Table 1). The five-fold cross-validation was applied to select the optimal model. As a result, it illustrated that the top 402 important markers were selected (accuracy =0.7167 in the test data). The learning curve and ROC curve (AUC=0.72098) are shown in Figure 2. The results proved that our model could predict the aging process with enough accuracy.

Table 1.

The prediction results

	Old samples	Young samples	Accuracy
The training set	120	269	0.7632 (cross-validation)
The test set	60	135	0.7167

Figure 2.

Results of the aging predictor. A. The learning curves by 5-fold cross-validation; B. The ROC curves.

Further, to distinguish between the chronological age and aging acceleration, the aging scores were calculated using posterior probability of the Naive Bayesian classifier (Table 2). From the chart, the result shown that the aging scores indicated a growing trend with increasing ages; apart from this, the aging scores of the cancer samples were significantly higher than the scores of normal samples within different age groups, respectively. All involved p-values were lower than 0.05 using the Kruskal-Wallis test (Figure 3A-D). The details displayed in Figure 3A-D indicate aging acceleration in GBM.

Table 2.

The median aging scores

In GBM	In normal brain	Age (years old)
7.4103e-13	1.0879e-26	≥50
5.1013e-12	2.1579e-24	≥60
5.804e-12	6.3590e-24	≥70
7.3753e-06	2.5671e-23	≥80

Figure 3.

Aging acceleration results of GBM. A. 180 normal samples, 204 GBM samples (≥50 years old, p=4.4657e-10; B. 132 normal samples, 134 GBM samples (≥60 years old), P=7.82927e-07; C. 108 normal samples, 58 GBM samples (≥70 years old), P=0.005; D. 88 normal samples, 18 GBM samples (≥80 years old), P=0.0208.

To explore crucial genomic dysfunctions associated with accelerated aging, the relationships between somatic mutations and the aging acceleration pattern were identified using the elastic net. As a result, 24 somatic mutations were identified as key markers related to aging acceleration, whose frequencies are also summarized in Table 3 and Figure 4A. For example, ABCB4 had the greatest absolute weight (0.2268) and TP53 had maximum frequency (0.2400). The functions of ABCB4 have been deeply investigated, and it plays an important role in the synthesis of phosphatidylcholine [5] as well as the restraining of anticancer drug resistance during cancer treatment [6]. It had a great association with immunity, affecting both cancer and aging. Besides, TP53 has been widely recognized as the crucial marker that inhibits cancer [7-9] by regulating the cell cycle, cell apoptosis, cell senescence, and the DNA damage repair [10]. In short, TP53 played an important role in on aging and cancer.

Table 3.

The mutations associated with aging

Gene symbol
ITGA7	ITGAE	SLC38A1	USP16	ABCB4	AUTS2
PABPC3	DNAH17	SIGLEC8	KRT15	DMBT1	CEP76
CETP	DGKD	IKZF3	YY2	ZNF235	FGD6
CD7	AHNAK	PTPRM	ZBTB40	YIPF6	NEBL

Figure 4.

The statistical results of mutation frequencies of 24 mutated genes and relative function modules (the yellow is the driver gene, the blue is differential genes and the red is survival genes). A. The histogram of 24 mutation frequencies. B. KEGG function module driven by ITGAE; C. GO terms function module driven by CEP76; D. GO terms function module driven by IKZF3; E. GO terms function module driven by FGD6; F. GO terms function module driven by CD7; G. GO terms function module driven by YIPF6.

Critical functions were explored by integrating accelerated aging-related mutations and differential expression genes

To investigate GBM mechanism based on aging acceleration, herein the aging acceleration background network was constructed (totally containing 1948259 edges). Then the modules of “accelerated aging related SNP to differential expression” were identified based the aging acceleration background network. Further, an enrichment analysis was performed to explore biologic functions of these modules. As a result, 24 some key modules were enriched in biologic significance (Table S3 and Figure S1). These modules are shown in Figure 4B and 4C, and the total gene lists were shown in Table S1.

For example, the most enriched KEGG pathway was interactions with endocytosis of cells (FDR=0.0612) driven by Clathrin. Clathrin was ubiquitously distributed [11], and CHC17 (representative of Clathrin) performs key functions in formation of the mitotic spindle that contribute to cancer cell proliferation and division [12].

Further, there were a series of “accelerated aging SNP to differential expression” enriched critical BP terms, indicating the dysfunction of the immune system and nervous system as well as the cellular homeostasis of GBM based on aging acceleration (Figure 5).

Figure 5.

Enriched GO term results in the immune system and the nervous system. The left-most genes are mutated genes (driver genes), the medial contents are specific enriched GO terms, and the right-most values are the significance of GO terms in the module.

(1) The impact of the immune system on glioblastoma.

In the process of GO term analysis, the immunomodulatory-related terms were concentrated in the 12th and the 18th module. The results are displayed in Figure 5A and Table 4. Moreover, the gamma interferon regulates apoptosis and tumor suppressor pathway through immunological genes [13]. Thus in the reactive conditions, myeloid leukocytosis mediates immune response to regulate the diseases [14]. Leukocyte degranulation effect plays a role in cancer. Besides, the mast cell activation is effective to maintain survival [15]. Neutrophils (primarily white blood cells) are associated with immune function. Once neutrophil degranulation occurs, its phagocytic function is lost, influencing the elimination of tumor factors [16]. Most members of the white blood cell family are involved in tumor suppression. Therefore, it was found that the enrichment results of GO terms revealed critical mechanisms of GBM.

Table 4.

Enriched GO terms related to immunity

BP term	FDR
INTERFERON GAMMA MEDIATED SIGNALING PATHWAY	0.059
MYELOID CELL ACTIVATION INVOLVED IN IMMUNE RESPONSE	0.069
MAST CELL ACTIVATION	0.097
LEUKOCYTE DEGRANULATION	0.100
REGULATION OF MAST CELL ACTIVATION INVOLVED IN IMMUNE RESPONSE	0.129
MYELOID LEUKOCYTE ACTIVATION	0.173
REGULATION OF MAST CELL ACTIVATION	0.173
MYELOID LEUKOCYTE MEDIATED IMMUNITY	0.173
REGULATION OF LEUKOCYTE DEGRANULATION	0.173
PRODUCTION OF MOLECULAR MEDIATOR INVOLVED IN INFLAMMATORY RESPONSE	0.176
NEUTROPHIL ACTIVATION INVOLVED IN IMMUNE RESPONSE	0.176
CELL ACTIVATION INVOLVED IN IMMUNE RESPONSE	0.197

(2) The nervous system-related reactions.

A total of 17 terms were enriched in the nerve cells and neuromodulation processes. Details are shown in Table 5 and in Figure 5B. The neurological biological processes were enriched in the 12th, 18th and 19th module.

Table 5.

Enriched GO terms related to nervous system

Biologic process	FDR
ASTROCYTE DIFFERENTIATION	0.050
GLIAL CELL DIFFERENTIATION	0.075
GLIOGENESIS	0.083
HEAD DEVELOPMENT	0.083
NEGATIVE REGULATION OF NEURON APOPTOTIC PROCESS	0.127
CENTRAL NERVOUS SYSTEM DEVELOPMENT	0.127
SYNAPTIC SIGNALING	0.131
REGULATION OF GOLGI ORGANIZATION	0.135
FOREBRAIN DEVELOPMENT	0.148
REGULATION OF NEURON DEATH	0.169
REGULATION OF CALCIUM ION IMPORT	0.173
NEURON PROJECTION EXTENSION	0.176
SUBSTRATE INDEPENDENT TELENCEPHALIC TANGENTIAL MIGRATION	0.176
TELENCEPHALON DEVELOPMENT	0.176
NEURAL CREST CELL DIFFERENTIATION	0.187
NEGATIVE REGULATION OF NEURON DEATH	0.194
FOREBRAIN CELL MIGRATION	0.197

The top enriched modules indicated the dysfunction of glial cell production and differentiation. Malignancy of glioma increased with the degree of differentiation of tumor cells increasing [17]. That is, the division and differentiation of various nerve cells were mostly annotated in terms of significant enrichment and affect the formation and spread of GBM [18]. Previous studies have shown that astrocytes and glial cells were prone to variability in an undifferentiated state [19]. Neuron projection extension was also enriched in the 12th module. The migration and connection of glioma cells cannot be separated from the extension of projections [20]. It could be inferred that the occurrence and deterioration of GBM were inseparable from the prolongation of neuronal processes. As well known, the movement activity of cells in the forebrain (including the telencephalon) plays an important role in maintaining the development of brain tissue, promoting the metastasis of brain tumors [21]. If the cell death were induced abnormally, the risk of cancer should increase [22]. The synapse is an important approach of neuron communication thus it has a significant impact on survival [23].

(3) Cells’ response and changes to various stimulation and substances.

There were 12 terms enriched in the variation in cells stimulated by changeable substances, including oxygenates, nitrogenous compounds, hormones and so on. Related chart content can be found in Table S4. Oxygenates may resist toxic substances (substances that are susceptible to cancer) of the invading cells [24] and partial nitrogen compounds inhibit certain processes in tumor cells [25].

(4) The mechanism of influence of the remaining terms.

The RHO proteins (Ras homolog gene family) act as molecular switches to control cellular processes, including cytoskeletal formation and cell migration [26]. Additionally, secreted substances affect the development of GBM to varying degrees. GBM migration is inhibited by both histone deacetylase SAHA and natural product andrographolide [27]. Furthermore, dexamethasone induces expression of neuronal and glial cell apoptosis markers [28]. During development, the diversity of forebrain complexity is also influenced by this signal path [29]. To sum up, all significantly enriched terms revealed crucial cancer mechanisms and these complex terms demonstrated the utility of integrated modules.

The “accelerated aging related mutation-differential expression” module revealed an important prognostic index

To investigate the relationship between the “accelerated aging related mutation-differential expression” modules and the clinical characteristics, a survival analysis was performed. As a result, 256 out of 11378 genes were identified to notable prognostic genes (Table S2). Some of the survival genes were integrated in modules (Figure 4B and 4C), and details are shown in Table 6. Interestingly, MYOZ3 with the minimum p-value (about 0.001) has relevance to part of cell proliferation genes, hence it might be involved in regulating the aging and cancer growth progress [30]. Significant survival patterns were uncovered by the survival curves, even indicating that key survival genes were integrated in enriched modules (Figure 6A-H).

Table 6.

The survival genes with modules

Gene symbol
ABLIM3	ADRB3	AFF3	ANGPTL2	BTN2A2	C14orf162
CACNA1G	CACNA2D2	CALCRL	CAMKV	CDK5RAP3	CDS1
CLIP3	CYB561	DCHS1	EPS8	GBP2	IQSEC2
KCNG1	MYLC2PL	NDUFA8	NHP2L1	NR1I2	PAK1IP1
PTPRT	RHBDL1	RPP25	SAT1	SLC22A1	SSBP3
TFF2	TOR3A	TUBD1	WASF1	WDR8

Figure 6.

Survival curves of genes that were integrated in function modules. A. The survival curves of CDS1. B. The survival curves of BTN2A2. C. The survival curves of CAMKV. D. The survival curves of GBP2. E. The survival curves of RPP25. F. The survival curves of SAT1. G. The survival curves of SSBP3. H. The survival curves of CACNA2D2.

Discussion

The genomic dysfunction in molecular networks is the chief culprit that should be responsible for development of GBM, where aging acceleration is one of the driving incentives. The purposes of this study were as follows: (1) discriminating different age groups based on the normal brain of healthy people; (2) verifying the accelerated aging pattern of GBM compared by the normal group; (3) integrating accelerated aging related SNPs and differential expression thus exploring potential mechanisms of GBM based on aging acceleration.

In the immunosenescence theory of aging, a decline in functions in the immune system promote further dysfunctions in the neuro-endocrine-immune axis as well as inflammation during the aging process [31]. In other words, accelerated aging is considered to lead to a disruption of the immune system, and dysregulation of the nervous system. As a result, cancer might be triggered. Our results further confirmed that TP53 and ABCB4 in mutated genes possessed the highest mutation rate and influence ratio, respectively. These results have been highly noted [5,7]. In a prior study, the advanced aging process induced DNA damage, cell life reduction and apoptosis [32]; then cellular effects are caused by these changes in cellular and molecular levels irreversibly. The enriched GO functions in GBM modules coordinated with aging acceleration indicated that aging and GBM formation were affected by not only the nervous system, but also the endocrine and immunologic function [33]. In other words, all members of the intricate multifunctional system interact with others and then promote GBM progression together. Interestingly, significant GO terms showed that the occurrence of GBM was inseparable from the endocrine regulation, the immunity system and the nervous system, thus indicating aging acceleration is one of the most important factors of GBM.

Integrating multi-omics profiles based on the proper network was informative to explore potential mechanisms of GBM, thus presenting a survival analysis [4,34]. In addition, the aging acceleration was vital to promote the progression of cancers (also including GBM) [35]. Our results also indicated that the accelerated aging may be a risk index of GBM. In other words, molecular network modules in GBM provided a foundation for clinical therapeutics, and relative molecular markers (i.e. the immune system and the nervous system) were crucial to study the relationship between accelerated aging and GBM. Moreover, drug molecular treatment and targeted gene therapy came to insight of cancers. For instance, a signaling pathway suggested that dexamethasone can repress the development of the tumor [29].

Briefly, the tumor markers did not drive their functions alone, but interacted with each other within proper modules, then promoted development of GBM coordinately. In addition, GBM goes through a complicated process. Therefore, multi-scale analyses were apparently essential for GBM by focusing on potential biological functions. Our work identified important mechanisms (i.e. the immune system, the nervous system and the endocrine system) regulating GBM during the abnormal aging process coordinately. In summary, there are a series of dysfunctions promoting developments of GBM, in which aging acceleration was identified as a core factor in our work.

Disclosure of conflict of interest

None.