Clinical and Epidemiological Study of Intracranial Tumors in Children and Identification of Diagnostic Biomarkers for the Most Common Tumor Subtype and Their Relationship with the Immune Microenvironment Through Bioinformatics Analysis

Brain tumors are the second most common pediatric malignancy and have poor prognosis. Understanding the pathogenesis of tumors at the molecular level is essential for clinical treatment. We conducted a retrospective study on the epidemiology of brain tumors in children based on clinical data obtained from a neurosurgical center. After identifying the most prevalent tumor subtype, we identified new potential diagnostic biomarkers through bioinformatics analysis of the public database. All children (0–15 years) with brain tumors diagnosed histopathologically between 2010 and 2020 at the Department of Neurosurgery, Xijing Hospital, were reviewed retrospectively for age distribution, sex predilection, native location, tumor location, symptoms, and histological grade, and identified the most common tumor subtypes. Two datasets (GSE44971 and GSE44684) were downloaded from the Gene Expression Omnibus database, whereas the GSE44971 dataset was used to screen the differentially expressed genes between normal and tumor samples. Gene ontology, disease ontology, and gene set enrichment analysis enrichment analyses were performed to investigate the underlying mechanisms of differentially expressed genes in the tumor. Combined with methylation data in the GSE44684 dataset, we further analyzed the correlation between methylation and gene expression levels. Two algorithms, LASSO and SVM-RFE, were used to select the hub genes of the tumor. The diagnostic value of the hub genes was assessed using the receiver operating characteristic (ROC) curve. Finally, we further evaluated the relationship between the hub gene and the tumor microenvironment and immune gene sets. Overall, 650 children from 18 provinces in China were included in this study. The male-to-female ratio was 1.41:1, and the number of patients reached a peak in the 10–15-year-old group (41.4%).The most common symptoms we encountered in our institute were headache and dizziness 250 (28.2%), and nausea and vomiting 228 (25.7%). The predominant location is supratentorial, with a supratentorial to infratentorial ratio of 1.74:1. Low-grade tumors (WHO I/II) constituted 60.9% of all cases and were predominant in every age group. According to basic classification, the most common tumor subtype is pilocytic astrocytoma (PA). A total of 3264 differentially expressed genes were identified in the GSE44971 dataset, which are mainly involved in the process of neural signal transduction, immunity, and some diseases. Correlation analysis indicated that the expression of 45 differentially expressed genes was negatively correlated with promoter DNA methylation. Next, we acquired five hub genes (NCKAP1L, GPR37L1, CSPG4, PPFIA4, and C8orf46) from the 45 differentially expressed genes by intersecting the LASSO and SVM-RFE models. The ROC analysis revealed that the five hub genes had good diagnostic value for patients with PA (AUC > 0.99). Furthermore, the expression of NCKAP1L was negatively correlated with immune, stromal, and estimated scores, and positively correlated with immune gene sets. This study, based on the data analysis of intracranial tumors in children in a single center over the past 10 years, reflected the clinical and epidemiological characteristics of intracranial tumors in children in Northwest China to a certain extent. PA is considered the most common subtype of intracranial tumors in children. Through bioinformatics analysis, we suggested that NCKAP1L, GPR37L1, CSPG4, PPFIA4, and C8orf46 are potential biomarkers for the diagnosis of PA.


Introduction
Among childhood cancers, brain tumors follow acute lymphoblastic leukemia as the second most common disease, and accounts for approximately 20% of childhood cancers (Kaatsch et al. 2001;Steliarova-Foucher et al. 2004;Suresh et al. 2017). Currently, the predominant treatment is surgical resection, supplemented by radiotherapy and chemotherapy (Solomon et al. 2016). Although the survival rate has improved, the side effects have also greatly reduced the quality of patients' lives, and new treatments are vital (Finlay et al. 1996;Matthias et al. 2008).
Recently, mounting evidence has shown that the TME is genetically stable and available to determine abnormal tissue function, while also playing a key role in the development of malignant tumors (Chen et al. 2015;Quail and Joyce 2013). Immunity plays an important role in TME (Grivennikov et al. 2010). Inspired by such evidence, we may be able to find a novel approach to the treatment of pediatric brain tumors.
Studies have shown that the incidence rate of intracranial tumors in the Asian population is lower than that in other ethnicities. In addition, the incidence of certain types of brain tumors varies between races (Dabiao et al. 2007;Kuratsu et al. 2001;Wong et al. 2005). Consequently, based on the latest WHO classification and diagnostic criteria for nervous system tumors (2016 Edition) (Villa et al. 2018), we analyzed the clinical data of intracranial tumors in children from the Department of Neurosurgery, Xijing Hospital, Air Force Military Medical University in the past decade and summarized the epidemiological characteristics of childhood brain tumors in the northwest of China. We identified the key genes of the most common subtypes of pediatric brain tumors and analyzed the correlation between these genes and immune cells. This study aims to provide a theoretical basis for further improving the early diagnosis and treatment outcomes of children with intracranial tumors in China.

Epidemiological Analysis
Xijing Hospital is the largest hospital in northwest China, as well as the most comprehensive in terms of specializations. From January 2010 to October 2020, 650 children (0-15 years old) with intracranial tumors 18 came from 18 provinces, including Shaanxi, Gansu, Ningxia, and Shanxi, were treated in the Department of Neurosurgery, Xijing Hospital, Air Force Military Medical University. Only patients with a proven histopathological diagnosis were included in the study. All patients were confirmed to have intracranial tumors. We collected the patients' detailed information from the medical records, including age, sex, symptoms, tumor location, pathological examination results, followup adjuvant treatment, and other information. According to the most recent WHO classification and diagnostic criteria for tumors of the central nervous system (CNS) (2016 Edition), all 650 cases were classified and graded. Supratentorial tumors include tumors originating from the cerebral hemisphere, third ventricle, lateral ventricle, sellar region, and pineal region, as well as tumors originating from the cerebellum, brainstem, and fourth ventricle. The tumor grade was divided into low-grade and high-grade tumors: WHO grades I and II were classified as low-grade tumors, and grades III and IV were classified as high-grade tumors. The data were processed and scored using descriptive statistics, and categorical data were described in terms of frequencies and percentages.

Data Collection
Two datasets, the gene expression profiling dataset (GSE44971) and the gene methylation dataset (GSE44684), were retrieved from the publicly available GEO database (https:// www. ncbi. nlm. nih. gov/ geo/). The GSE44971 dataset, which was generated on the GPL11154 platform (Affymetrix Human Genome U133 Plus 2.0 Array), included 35 PA tumor samples and 9 normal cerebellum samples after removing 14 non-cerebellar samples. The GSE44684 dataset contained 39 PA tumor samples and 6 normal cerebellum samples (removed the non-cerebellar samples), which were acquired from the GPL13534 platform (Illumina HumanMethylation450 Bead-Chip). To find out whether the hub genes identified in our research occur in adult glioblastoma, we analyzed GSE90886 (9 GBM samples, 9 control samples) and GSE151352 (12 control samples, 12 GBM samples) in the GEO database.

Analysis of Differentially Expressed Genes
To appraise the differentially expressed genes (DEGs) between normal cerebellum and PA tumor samples, we applied the "limma" package for processing the GSE44971 dataset (Ritchie et al. 2015). Genes with |log 2 FC(fold change)|≥ 0.5, and P value < 0.05, were considered to be DEGs, which were presented by heatmap and volcano plot.

Functional Enrichment of DEGs
We used the R package "clusterProfiler" to implement the following enrichment analyses: gene ontology (GO) (Anonymous 2006), disease ontology (DO) (Osborne et al. 2009), and gene set enrichment analysis (GSEA) (Subramanian et al. 2005). The GO term enrichment analysis included three categories: biological process (BP), molecular function (MF), and cellular component (CC). For all analyses, statistically significant thresholds were set at p < 0.05.

Methylation Data Processing
The R package, "ChAMP" was used to identify the differential methylation sites between normal and PA samples. |∆β|> 2 and P value less than 0.05 were set as the cutoff criteria. The distribution of the differential methylation sites is shown in the volcano map. For the correlation analysis, we extracted 28 paired PA samples that contained both DNA methylation and gene expression data from the GSE44971 and GSE44684 datasets.

LASSO and SVM-RFE Algorithms
LASSO is an algorithm that can obtain a more refined model by constructing a penalty function (Friedman et al. 2010). SVM-RFE is a feature selection algorithm based on a support vector machine that ranks the features based on the recursive feature deletion sequence (Huang et al. 2014). In this study, the SVM-RFE (e1071 packages) and LASSO regression (glmnet package) algorithms were used to select the characteristic genes from 45 candidate genes. Ultimately, we obtained the intersection of candidate genes screened by the LASSO and SVM-RFE algorithms for follow-up studies.

Immune Infiltration Analysis
We performed the ESTIMATE algorithm in the GSE44971 dataset using the "estimate" package in R to generate immune, stromal, and estimate score. PA samples were divided into high expression and low expression groups according to the median value of the expression of five hub genes. Next, we further compared the difference of immune, stromal, and estimate scores in the two groups using the Wilcoxon test, with the grouping was based again on the median value of the expression of five hub genes. The ssGSEA is a method used to calculate individual enrichment scores for each sample and gene set (Barbie et al. 2009). In this study, ssGSEA was used to quantify the enrichment levels of 28 immune-related gene sets in each sample of the GSE44971 dataset. The 28 immunerelated gene sets were extracted from the study by Bindea et al. (2013).

Statistical Analysis
All statistical analyses were performed using R software. The AUC of the ROC curve was used to determine the diagnostic value of the five hub genes. Unless otherwise stipulated, P < 0.05, was considered statistically significant.

Sex and Age Distribution
A total of 650 patients were included in our retrospective study, and we subdivided our patients into four age subgroups: less than 1, 1-5 years, 5-10 years, and 10-15 years. There was a slight overall male predominance, with a male to female ratio of 1.41:1, composed of 380 males (58.5%) and 270 females (41.5%). Although the sex ratio varied by age subgroup, males were consistently the dominant group. The number of patients increased steadily with age: 4.3% in the first year, 16.0% between 2 and 5 years, 38.3% between 5 and 10 years, and then reaching a peak between 10 and 15 years (41.4%). The most common age at presentation was between 10 and 15 years, with a mean of 8.8 years (Fig. 1).

Symptoms
As shown in Table 1, among the most common symptoms in our institute were headache and dizziness (250 cases, 28.2%), nausea and vomiting (228 cases, 25.7%), followed by visual disturbances (104 cases, 11.7%), and cerebellar signs (80 cases, 9.0%). Physical examination or routine examination after head trauma revealed 81 cases of intracranial tumors with no symptoms.

Grade
Low-grade tumors (WHO I/II) constituted 60.9% (396 cases) of all cases, while the rest were high-grade tumors  (WHO III/IV) (254 cases, 39.1%). The ratio of the two was 1.29:1. The statistics of tumor grade and children's age showed that the majority of high-grade and low-grade tumors were children aged 10-15 years, and the majority of tumors in all age groups were low grade (Fig. 3).

The Top Five Common Intracranial Tumors in Children
In our data, the top five tumors with the highest incidence rate constituted more than half of childhood brain tumors: astrocytomas (all subtypes) in 110 (16.9%) patients, medulloblastoma in 88 (13.5%), craniopharyngioma in 83(12.8%), ependymoma in 41 (6.3%), and germinoma in 38(5.8%), as shown in Table 3. In terms of tumor grade, astrocytomas and craniopharyngiomas were mostly low grade, medulloblastoma, and germ cell tumors were mainly high grade, and no significant difference in the grade of ependymoma. The top five common tumors were mostly low grade, with a ratio of 1.28:1. Sex distribution showed that except for craniopharyngioma, the majority of other types of tumors were male children, and the majority of craniopharyngiomas were female children. Among the top five common tumors, 204 of the case were male, and 154 were female, with a male to female ratio of 1.32:1. Regarding the location of tumor distribution, there was no significant difference in the location of astrocytomas, ependymoma was mostly subtentorial, and germinomas were mostly supratentorial. Among the top five common tumors, 190 were located in the supratentorial region, and 170 were subtentorial, with a ratio of 1.12:1.

The Most Common Subtype Intracranial Tumors in Children
In 110 patients with astrocytomas (all subtypes), there were 62 cases of PA, which is the most common, accounting for 61% (Fig. 4). There were 15 cases of diffuse astrocytoma (14%), 10 cases of oligoastrocytoma (10%), 7 cases of anaplastic astrocytoma (7%), 5 cases of pleomorphic xanthoastrocytoma (5%), and 3 cases of anaplastic pleomorphic xanthoastrocytoma (3%). PA is the most common intracranial tumor subtype in children; therefore, we used it as the follow-up research object.

The Molecular Mechanisms of DEGs
To dissect the potential biological functions of 3264 DEGs in PA, we performed GO, DO, and GSEA analyses. The top 10 GO terms (BP, CC, and MF) are depicted in Fig. 6A, from which we found that DEGs mainly participated in the process of neural signal transduction. Of note, DEGs were also significantly correlated with several immune-related biological processes, such as regulation of leukocyte proliferation, regulation of lymphocyte proliferation, humoral immune response, and macrophage chemotaxis (Supplementary Table 3). DO enrichment analysis suggested that DEGs were highly associated with "lung disease", "periodontal disease", "skin disease", and "obesity" (Fig. 6B and Supplementary Table 4). The results of GSEA showed that DEGs were only markedly enriched in 5 KEGG signaling pathways, including "neuroactive ligand receptor interaction", "olfactory transduction", "basal cell carcinoma", "hedgehog signaling pathway", and "taste transduction" (Fig. 6C and Supplementary Table 5).

DNA Methylation Analysis in the GSE44684
We obtained a total of 36,787 differentially methylated sites between normal cerebellum and PA tumor samples in the GSE44684 dataset, of which 14,924 were hypermethylated sites (∆β > 2) and 21,863 were hypomethylated sites (∆β < 2) ( Fig. 7A and Supplementary Table 6). The results of all differentially methylated sites from each autosomal chromosome are shown in the outer circle of the circos plot. The two innermost circles in the circos plot displayed differential hypermethylation and hypomethylation frequencies in a 10 Mb sliding window across the genome (Fig. 7B). The pyramid plot showed the frequencies of  (Fig. 7C). Subsequently, we further examined the correlations between promoter DNA methylation and corresponding gene expression, and a total of 45 DEGs were identified for further analysis (Supplementary Fig. 1 and Table 7).

Identification of Characteristic Genes in PA
To screen for biomarkers of PA from the 45 DEGs, we adopted the LASSO and SVM-RFE algorithms. Eight characteristic genes (NCKAP1L, GPR37L1, CSPG4, PPFIA4, CD86, CD1D, SLC15A3, and C8orf46) were identified using the LASSO algorithm (Fig. 8A, B). A total of 20 characteristic genes were selected using the SVM-RFE algorithm, as shown in Fig. 8C and Supplementary Table 8. After that, using the Venn diagram (Fig. 8D), we obtained five hub genes by overlapping the characteristic genes screened by the two algorithms, namely Nck-associated protein 1-like (NCKAP1L), G protein-coupled receptor 37-like 1 (GPR37L1), chondroitin sulfate proteoglycan 4 (CSPG4), chondroitin sulfate proteoglycan-4 (PPFIA4), and chromosome 8 open reading frame 46 (C8orf46). Except for PPFIA4, the other four hub genes were highly expressed in PA. Then, through analyzing GSE90886 and GSE151352, we found that these five genes did not show significant differences, which would suggest that CSPG4, PPFIA4, NCKAP1L, GPR37L1, Fig. 8 Identification of characteristic genes. A, B Eight characteristic genes were discerned using the LASSO algorithm. C Twenty characteristic genes were selected with the SVM-RFE algorithm. D The Venn diagram of the characteristic genes identified by LASSO algorithm and SVM-RFE algorithm. We had obtained five hub genes and C8orf46 have a relatively important role in PA, while the role in GBM was not as significant ( Supplementary Fig. 2). Besides, we distinguished PA patients in the GSE44971 dataset into pediatric and adult groups according to the 18-year age cutoff to perform a difference-in-difference analysis, and found that the five hub genes did not show significant differences ( Supplementary Fig. 3).

Diagnostic Capacity of Hub Genes
We evaluated the diagnostic value of five hub genes in the GSE44971 dataset using ROC analysis. The results indicated that these genes had a powerful ability to discriminate PA from normal samples. The AUC of GPR37L1 was 0.994, and the AUC of the other genes was 1.000. Specific information is displayed in Fig. 9A-E.

Correlation of the Five Hub Genes with Immune Infiltration
Previous results have shown that DEGs were markedly enriched in BP terms related to the immune system.
Studies have shown that the TME plays a significant role in suppressing or enhancing the immune response. Consequently, we investigated the association between the expression levels of five hub genes and the TME. As shown in Fig. 10A-C and Supplementary Fig. 4, the immune, stromal, and estimate scores in the NCKAP1L high expression group were higher than those in the NCK-AP1L low expression group. Unfortunately, there was no significant correlation between the other four hub genes (GPR37L1, CSPG4, PPFIA4, and C8orf46) and TME. Subsequently, ssGSEA analysis revealed a difference in the abundance of 28 immune-related gene sets in normal and PA tumor samples. Compared with the normal group, the PA tumor group had a higher proportion of immune gene sets, except activated CD4 T cells and effector memory CD4 T cells (Fig. 10D). We further evaluated the correlation between the expression of NCKAP1L and immune gene sets. Interestingly, NCKAP1L was found to be positively correlated with all immune gene sets, especially in macrophages (Fig. 10E). Fig. 9 Diagnostic capacity of hub genes. A-E ROC analysis to evaluate the diagnostic value of 5 hub genes. The AUC of GPR37L1 was 0.994, and the AUC of other genes was 1.000

Discussion
Intracranial tumors in children are not the comparable with those of adults, and also have obvious differences in terms of topographical distribution, pathological types, treatment, prognosis, and outcome. Consistent with the vast majority of other studies in the Asian region (Ahmed et al. 2007;Cho et al. 2002;Makino et al. 2010;Wong et al. 2005), our study also revealed a predominance of boys with intracranial tumors, with a sex ratio of 1.41:1, which was slightly higher than that in Western countries (Bauchet et al. 2009;El-Gaidi 2011;Hjalmars et al. 1999). Tumors at different life stages reflect different dynamics and biological behaviors (Rickert and Paulus 2001). In our study, the mean age of the affected children was 8.8, which is similar to the results of studies in Iran and Pakistan (Ahmed et al. 2007;Mehrazin and Yavari 2007). Our study suggested that the number of patients increased steadily with age and peaked at the age of 10-15 years. This phenomenon was also reported in a study conducted at the Beijing Neurosurgical Institute, Beijing, China (Zhou et al. 2008), and the findings were corroborated by studies in England, Sweden, and India (Ehrstedt et al. 2016;Govindan et al. 2018;Stiller et al. 2019). Conversely, some studies have suggested that the highest incidence occurs in the 5-9-year age group (Cho et al. 2002;Makino et al. 2010;Riaz et al. 2019). We speculated that this may be caused by the shyness of children in western China to express their symptoms, and when they grow up, they are more able to complain about their headache, visual disturbances, or endocrinopathies such as increased thirst or urination. Regardless of tumor location, intracranial hypertension (53.9%), and ocular symptoms (11.7%) were the main clinical manifestations in our cohort of patients. However, our study also highlighted several other symptoms that indicate the possibility of developing tumors, such as weight loss, behavioral changes, school difficulties, developmental delay, head tilt, macrocephaly, diabetes insipidus, abnormal menstruation, and growth arrest. Furthermore, we were also concerned about the fact that 81 patients (9.1%) were asymptomatic before diagnosis. Hence, improving the comprehensive examination of children and raising awareness of the various complex and atypical symptoms of brain tumors in children will help in diagnosing the tumor and in alleviating the suffering of the child expeditiously.
In this study, we followed the WHO Classification of Central Nervous System Tumors 2016 for the diagnosis of different tumors. As for the most common intracranial tumors in children, most studies have reported that astrocytoma followed by medulloblastoma are the two most common intracranial tumors (Aghadiuno et al. 1985;El-Gaidi 2011;Katchy et al. 2013;King et al. 2015;Lannering et al. 2009;Ogun et al. 2016;QT et al. 2020;Rickert and Paulus 2001;Shanmugavadivel et al. 2020;Udaka and Packer 2018). However, some studies have reported medulloblastomas to be the most common pediatric intracranial tumors ahead of astrocytomas (Almutrafi et al. 2020;Govindan et al. 2018;Kadri et al. 2005;Nasir et al. 2010;Shah et al. 2015). According to the present study, astrocytomas with a frequency of 16.9% were the most common brain tumors, with the average age of onset of this tumor being 8.5 years. Among the 110 patients, there was a slight male predominance, with a ratio of 1.2:1, and there was no specific location preference. Obviously, most astrocytomas belong to low grade tumors, which is consistent with previous research (El-Gaidi 2011; Riaz et al. 2019;Shah et al. 2015). Medulloblastoma, an embryonal tumor of the posterior fossa, comprises up to 20% of all pediatric brain tumors (Ostrom et al. 2019;Shtil 2016). In our study, medulloblastoma was the second most common brain tumor, accounting for 13.5% of all brain tumors with a male predominance (ratio of 1.8). It has two incidence peaks in patients aged 4-5 years (18%) and in patients aged 7-8 years (23.8%) with a mean age of 7.8 years, which is slightly different from previous studies (Udaka and Packer 2018) but consistent with the study in China (Zhou et al. 2008). Craniopharyngioma is a benign epithelial tumor with slow growth that accounts for approximately 5 to 10% of pediatric brain tumors (Garre and Cama 2007;Muller et al. 2019). In this study, craniopharyngioma, the third most common, had a mean age of onset of 8.7 years and no discrepancy in incidence between sexes. Moreover, the critical location of the lesion and the difficulty in treating postoperative complications make the prognosis of patients generally poor (Muller et al. 2006). In our follow-up results of the 83 pediatric craniopharyngioma patients in our study, 10 patients died and 18 relapsed. Ependymomas are the fourth most common type, accounting for 6.3% of all pediatric brain tumors with a mean age of 6.9 years, consistent with some studies in Asia and Europe (Ehrstedt et al. 2016;El-Gaidi 2011;Makino et al. 2010;Zhou et al. 2008). Our study showed that there was a subtentorial predominance with a ratio of 1.56:1 and a slightly male predominance, with a ratio of 1.16:1. Germinoma, which accounts for 5.8% of all brain tumors in children, is the fifth most common, consistent with the consensus (Fetcko and Dey 2018).
Consistent with previous studies (Riaz et al. 2019), pilocytic astrocytomas were the most common childhood brain tumor subtypes in the present study. Studies have shown that pilocytic astrocytoma is associated with high levels of myeloid and lymphocyte infiltration and activation marker expression (Forsyth et al. 1993;Huang et al. 2005;Rosemberg and Fujiwara 2005). Therefore, it is inferred that immune cells are related to the occurrence and development of pilocytic astrocytomas. Based on the incidence rate of this type of tumor and the current treatment methods that are still not perfect, we decided to carry out bioinformatics analysis to provide theoretical guidance for further research.
In the present study, by analyzing the GEO44971 dataset, we identified 3264 DEGs in the tumor tissue. We then performed GO, DO, and GSEA analyses to explore the biological functions of the DEGs. The results showed that DEGs participated in the functions of neurotransmitter transfer and transmembrane transport, and are highly associated with lung disease, periodontal disease, skin disease, and obesity, and markedly enriched in neuroactive ligand receptor interaction, olfactory transduction, basal cell carcinoma, hedgehog signaling pathway, and taste transduction KEGG pathways. Notably, immune-related functional items were found in the GO-BP pathway. These results suggest that DEGs were highly correlated with the immune system, which confirms our hypothesis.
DNA methylation causes transcriptional silencing to regulate gene expression. Therefore, we carried out methylation analysis on the GSE44864 dataset to determine whether methylation affects gene expression in pilocytic astrocytoma, and further precise exploration of differential genes. The results showed that methylation had the greatest effect on the expression of chromosome 17. The results of methylation site analysis were correlated with the gene expression of corresponding samples, and 45 genes with negative regulation of promoter methylation level and gene expression level were screened out. Therefore, the differential expression of some differential genes may be due to differential methylation of their own genes.
To screen the hub genes among these 45 differentially expressed genes, we used LASSO regression and the SVM-RFE algorithm. At the intersection of the two algorithms, we obtained five hub genes: NCKAP1L, GPR37L1, CSPG4, PPFIA4, and C8orf46, and their efficiency was also confirmed by the ROC curve.
In recent years, an increasing number of studies have focused on TME, in which immune cells and stromal cells play an important role in the diagnosis and prognosis of tumors (Hanahan and Coussens 2012). Therefore, we first performed ssGSEA analysis on the GSE44971 dataset. We found that compared with the normal group, the PA tumor group had a higher proportion of immune gene sets, except for activated CD4 T cells and effector memory CD4 T cells. This result suggests that we should perform further immunological analysis of the five hub genes. Then, using the ESTIMATE algorithm, we found that the high and low expression groups of NCKAP1L gene had significant differences in the immune, stromal, and estimate scores. Consequently, we further evaluated the correlation between the expression of NCKAP1L and immune gene sets, and found that this gene was positively correlated with all immune gene sets, especially in macrophages.
Our study is the first to report the relationship between NCKAP1L and PA. NCKAP1L, or Nck-associated protein 1-like, is a key component of the actin cytoskeleton machinery. The protein encoded by this protein is hematopoietic protein-1 (HEM-1), a hematopoietic lineage-restricted member of the Nap1l subunit of the WAVE (WASP-family verprolin-homologous protein) complex (Weiner et al. 2006). The role of this gene in tumors has not been extensively studied. Recently, it has been found that this gene plays an important role in the activation, migration, and cell contact formation of lymphoid and myeloid cells, including the formation of immune synapses in effector cells (Park et al. 2008). Subsequently, a few studies have revealed that it is associated with immunodeficiency, lymphoproliferation, and high inflammation, and may be a biomarker of some tumors (Castro et al. 2020;Hu et al. 2020;Wang et al. 2020). Similarly, our study also found that as a hub gene of PA, NCKAP1L plays an important role in the immune system. It is closely related to B cells and macrophages and may affect the TME in this way, thereby affecting the development of PA. Therefore, we provide a new basis and direction for the pathogenesis of PA and the study of the TME.
G protein-coupled receptor 37-like 1 (GPR37L1) and chondroitin sulfate proteoglycan 4 (CSPG4) have been shown to be expressed in astrocytomas and gliomas. GPR37L1 interacts with patched 1 (Ptch1) in the periciliary membranes of astrocytes (La Sala et al. 2020), while CSPG4, a type I transmembrane protein, plays a role in the origin, progression, and angiogenesis of glioma (Mellai et al. 2020). Both these genes play a role in TME and may be promising as biomarkers for detecting PA progression and patient survival (Mellai et al. 2020). PPFIA4, also known as LIP.1 or Liprin alpha1, encodes liprin, which is related to neural signal transmission. Previous studies have shown that it is correlated with pancreatic cancer and small cell lung cancer (BA et al. 2009). Our study found that it may be one of the characteristic biomarkers of PA, and further research is needed to explore its role in tumor development and its relationship with TME. Chromosome 8 open reading frame 46 (C8orf46), a human protein-coding gene, has been named Vexin. Previous studies have suggested that Vexin is involved in embryonic neurogenesis and is related to cancer progression, but its function has not yet been elucidated clearly (Koshimizu et al. 2020;Moore et al. 2018). Our study is the first to report that this gene is associated with PA, which provides a new direction for future research.

Conclusion
The present analysis, based on a large series of pediatric patients, provides a reliable profile of the epidemiology of nervous system tumors in children and valuable information for worldwide epidemiological research. We found that PA was the most common tumor subtype in childhood brain tumors. Then, through bioinformatics analysis, we suggested that NCKAP1L, GPR37L1, CSPG4, PPFIA4, and C8orf46 are potential biomarkers for the diagnosis of PA. NCKAP1L plays an important role in the immune system and may affect the TME.
The findings of this study must be considered in light of some limitations. First, the data of this study were obtained from a single center of pediatric neurosurgery in Northwest China, which may cause possible bias in the population source. Second, five hub genes were not tested in human tumor tissues. Therefore, we should verify the results and further explore the value of key genes in the diagnosis and treatment of PA. We are looking forward to multi-center and multidisciplinary collaborative research to improve the prognosis of children with intracranial tumors in China and other countries by establishing reasonable diagnosis, treatment systems, and evaluation standards for intracranial tumors, and perfecting the epidemiological follow-up system.
Author Contribution GW had the idea for the article; GW and YJ designed the experiments; GW, YJ, and EK carried out the experiments. GW and HC analyzed the data and experimental results. The manuscript was written by GW and YY. YJ and JW critically revised the work. All authors have read and approved the final manuscript.