The TCGA gene expressions along with the clinical datasets were downloaded from the TCGA hub (https://tcga.xenahubs.net). The CGGA datasets are available at the http://www.cgga.org.cn/index.jsp website. The gene expression series matrix of glioma tissues was downloaded from the GEO website (www.ncbi.nlm.nih.gov/geo), including the GSE4412, GSE83300, GSE16011, GSE43378, GSE83294, and GSE16011 datasets.

The genomic alterations of EGFR and CDKN2A across 32 types of tumor patients or in glioma patients were downloaded from cBioPortal (version 3.2.0) based on the TCGA and MSKCC datasets (http://www.cbioportal.org/index.do).

GBM gene expression profile was analyzed using the TCGA GBM RNA-seq datasets. The differentially expressed genes between GBM patients with EGFR or CDKN2A alteration and GBM patients without EGFR or CDKN2A alteration were determined using paired Student's t-test. Genes with P < 0.01 were chosen to be significantly different.

The GEO expression datasets were annotated with corresponding probe sets and processed using the “plyr” package (version 1.8.5) in R software (version 3.5.0, https://www.r-project.org/). The plyr package was used for splitting, applying, and combining data and could be downloaded from Bioconductor (https://cran.r-project.org/web/packages/plyr/index.html).

Heatmaps were created using the R software “pheatmap” package (version 1.0.12). The “pheatmap” package and the basic usage were downloaded from Bioconductor (https://cran.r-project.org/web/packages/pheatmap/). The clustering scale was determined by the “average” method. The clustering distance was determined by the “correlation” method.

Kaplan–Meier plots were created using the “survival” package (version 3.1-8) in R statistics software (https://cran.r-project.org/web/packages/survival/index.html). GBM or LGG patients were divided into two clusters based on the mean expression levels of SPATS2L or alterations of EGFR or CDKN2A. The Kaplan–Meier estimator was applied to determine the clinical outcomes of the different clusters of glioma patients. P-values were determined using the log-rank test.

KEGG signaling pathways and tissue-specific expression were performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) tool (version 6.8; https://david.ncifcrf.gov). DAVID is a functional annotation tool for a list of genes. Enrichment results with P < 0.05 were considered to be statistically significant.

The box plots were generated from GraphPad Prism software (version 5.0; GraphPad Software, Inc.). Statistical analysis was performed using two-tailed paired Student's t-test in R software. A P < 0.05 was chosen to be significantly different.

Using cBioPortal (35, 36), the biological relevance of EGFR alteration in tumor patients derived from the TCGA Pan-cancer datasets was determined. Across 32 different tumor types, the frequency of EGFR amplification was the highest in patients with GBM (Figure 1A). More than 40% GBM patients had EGFR amplification. Moreover, tumor patients with EGFR amplification demonstrated worse prognosis compared with patients without EGFR amplification (P < 0.0001) (Figure 1B), suggesting that EGFR amplification was a biomarker correlated with the clinical outcomes across different tumor types. EGFR signaling was also activated by a specific EGFR mutation. EGFR mutation also particularly occurred in patients with GBM. More than 25% GBM patients had EGFR mutations (Supplementary Figure 1A). Similarly, tumor patients with EGFR mutations demonstrated worse prognosis compared with patients without EGFR mutations (P < 0.0001) (Supplementary Figure 1B).

GBM patients also had the highest frequency of CDKN2A deletion. More than 50% GBM patients had CDKN2A deletion (Figure 1C), and CDKN2A deletion was a bad prognosis across different tumor types (Figure 1D). Those results showed that EGFR amplification, EGFR mutation, and CDKN2A deletion particularly happened in patients with GBM and those genomic alterations may determine the bad prognosis of GBM.

We then tested the prognostic effects of EGFR amplification, EGFR mutation, and CDKN2A deletion in patients with GBM. GBM patients with EGFR amplification had worst prognosis than patients without EGFR amplification (P < 0.0001) (Figure 2A). GBM patients with CDKN2A deletion also demonstrated worse prognosis compared with patients without CDKN2A deletion (P < 0.0001) (Figure 2B). However, there was no difference in clinical overall survival between GBM patients with or without EGFR mutation (P = 0.98) (Supplementary Figure 1C). Furthermore, the expression levels of EGFR and CDKN2A were not associated with the clinical outcomes of patients with GBM (Supplementary Figure 2).

Moreover, GBM patients without EGFR amplification and CDKN2A deletion had the best prognosis than GBM patients with EGFR amplification or CDKN2A deletion or with both alterations (Figure 2C). Also, Spearman correlation demonstrated negative correlations of EGFR and CDKN2A expression in the TCGA GBM datasets (Figure 2D). All those results emphasized the synergetic effects of EGFR amplification and CDKN2A deletion in determining the clinical outcomes of GBM patients.

Genetically, EGFR alteration and CDKN2A deletion were also connected. Among the 378 GBM patients in the TCGA datasets, 53% of the patients had EGFR alteration and 58% of the patients had CDKN2A deletion. Interestingly, 138 (36%) GBM patients had both EGFR and CDKN2A alterations. Statistical analysis showed that the coexistence of EGFR alteration and CDKN2A deletion was significant (P < 0.0001) (Figure 2E).

Next, we further tested the correlations of the expression levels of EGFR and CDKN2A. As expected, compared with GBM patients without EGFR amplification, EGFR was highly expressed in GBM patients with EGFR amplification (Figure 3A). On the contrary, CKDN2A was highly expressed in GBM patients without EGFR amplification (Figure 3A). Also, with the deletion of CDKN2A in GBM patients, CDKN2A was downregulated, while EGFR was overexpressed (Figure 3B).

Next, all the genes regulated by EGFR amplification or CDKN2A deletion were identified. We found that 3,745 genes were differentially expressed in GBM patients with or without EGFR amplification, and 2,417 genes were differentially expressed in GBM patients with or without CDKN2A deletion. Moreover, 864 genes were commonly regulated by EGFR amplification and CDKN2A deletion (Figure 3C). Those genes classified GBM patients into two distinct subgroups, and each subgroup demonstrated different expression profiling and different genomic alterations (Figure 3D).

We further determined the transcriptional relevance of EGFR amplification and CDKN2A deletion commonly regulated genes. We found that those genes were enriched in brain tissues (Figure 4A). Moreover, cell cycle, ERBB signaling pathway, and MAPK signaling pathway were associated with EGFR amplification and CDKN2A deletion (Figure 4B). The enriched EGFR amplification and CDKN2A deletion regulated genes in the cell cycle, ERBB, and MAPK signaling pathways were illustrated in the heatmaps (Figure 4C).

We further illustrated that CDK6 from the cell cycle, EGF from the ERBB signaling pathway, and PDGFA from the MAPK signaling pathway were all upregulated in GBM patients with EGFR amplification (Figure 4D). Also, CDK6, EGF, and PDGFA were all overexpressed in GBM patients with CDKN2A deletion (Figure 4E).

Next, we determined the prognostic relevance of EGFR amplification and CDKN2A deletion on commonly regulated genes. Among the 864 genes commonly regulated by EGFR amplification and CDKN2A deletion, 62 genes were significantly associated with the clinical outcomes of GBM derived from the TCGA GBM datasets (Figure 5A). Also, using the CGGA datasets, we found that 41 genes out of the 864 genes demonstrated prognostic effects in GBM (Figure 5A). Interestingly, the two genes SPATS2L and Kinase suppressor of ras 2 (KSR2) were both significantly correlated with the prognosis of GBM in the TCGA and CGGA datasets (Figure 5A).

SPATS2L was upregulated in GBM patients with EGFR amplification, compared with GBM patients without EGFR amplification (Figure 5B). Also, compared with GBM patients without CDKN2A deletion, the expression levels of SPATS2L were higher in patients with CDNK2A deletion (Figure 5B). Correspondingly, the higher expression levels of SPATS2L were associated with worse prognosis in patients with GBM in both TCGA and CGGA datasets (Figure 5C).

The prognostic effects of SPATS2L were further validated using the GEO datasets. In the GSE4412 dataset, we found that the higher expression levels of SPATS2L were associated with worse prognosis in patients with GBM (P = 0.01) (Figure 5D). However, GBM patients with higher expression levels of SPATS2L had no significant worse clinical outcomes in the GSE83300 dataset (P = 0.087) (Figure 5D). Those results highlighted that SPATS2L was an important prognostic marker in patients with GBM.

KSR2 was specifically expressed in brain tissues (33). We found that KSR2 was downregulated in GBM patients with EGFR amplification or GBM patients with CDNK2A deletion (Supplementary Figure 3A). However, we found the contradictive prognostic effects of KSR2 in the TCGA and CGGA datasets. The higher expression levels of KSR2 were associated with worse prognosis in patients with GBM in the TCGA datasets, while KSR2 was a good prognostic marker in the CGGA datasets (Supplementary Figure 3B). So, KSR2 was not further studied.

GBM is divided into four subtypes by the TCGA GBM research group: classical (Clas), neural (Neu), proneural (Pro), and mesenchymal (Mes) (37). The present study then assessed the expression levels of the SPATS2L in patients with different subtypes of GBM. Compared with the Clas, Neu, and Pro subtypes, the expression levels of SPATS2L were higher in patients with the Mes subtype of GBM in the TCGA datasets (Figure 6A). GBM was also classified into three subtypes—proneural (PN), mesenchymal (Mes), and proliferative (Pro)—based on their molecular signatures (38). This higher expression of the SPATS2L in the Mes subtype of GBM patients was also observed in the GSE13041 dataset, compared with the PN and Pro subtypes (Figure 6A). Also, compared with primary GBM tissues, SPATS2L was highly expressed in recurrent GBM tissues in the CGGA datasets (Figure 6B).

GBM is grade IV glioma. We found that, compared with grade II–III glioma (LGG), the expression levels of SPATS2L were higher in GBM patients in the GSE4412, GSE16011, GSE43378, GSE83294, TCGA, and CGGA datasets (Figure 6C). Those results suggested the potential prognostic significance of SPATS2L in patients with LGG.

The EGFR and CDKN2A alteration frequency was relatively lower in patients with LGG (Figure 1A). Among the 507 LGG patients found in the TCGA datasets, 11% patients had EGFR alteration and 11% patients had CDKN2A deletion. Interestingly, 29 LGG patients had both EGFR and CDKN2A alterations. Statistical analysis showed that the coexistence of EGFR alteration and CDKN2A deletion was significant in patients with LGG (P < 0.0001) (Figure 7A).

Next, we tested the prognostic effects of EGFR amplification and CDKN2A deletion in LGG patients. We found that LGG patients with EGFR amplification demonstrated worse prognosis compared with patients without EGFR amplification (P < 0.0001) (Figure 7B). Similarly, LGG patients with CDKN2A deletion also had worse prognosis compared with patients without CDKN2A deletion (P < 0.0001) (Figure 7B). Moreover, LGG patients without EGFR amplification and CDKN2A deletion had the best prognosis than LGG patients with EGFR amplification or CDKN2A deletion or with both alterations (Figure 7B). All those results emphasized the associations of EGFR amplification and CDKN2A deletion in determining the clinical outcomes of LGG patients.

However, unlike GBM patients, the higher expression levels of EGFR were associated with worse prognosis in patients with LGG in the TCGA LGG datasets (Figure 7C). On the contrary, the lower expression levels of CDKN2A were associated with worse prognosis in patients with LGG (Figure 7C). Spearman correlation demonstrated high negative correlations between EGFR and CDKN2A expression in the TCGA LGG datasets (Figure 7D).

Moreover, in LGG patients with EGFR amplification, EGFR was highly expressed. However, CKDN2A was downregulated in LGG patients with EGFR amplification (Figure 8A). Also, compared with LGG patients without CDKN2A deletion, CDKN2A was downregulated, and EGFR was overexpressed in LGG patients with CDKN2A deletion (Figure 8B).

Since EGFR and CDKN2A alterations and expression levels were all associated with the clinical outcomes of patients with LGG, we then tested the prognostic effects of EGFR and CDKN2A alterations on the regulated gene SPATS2L. SPATS2L was upregulated in LGG patients with EGFR amplification, compared with LGG patients without EGFR amplification (Figure 8A). Also, compared with LGG patients without CDKN2A deletion, the expression levels of SPATS2L were relatively higher in patients with CDKN2A deletion (Figure 8B). Moreover, the higher expression levels of SPATS2L were associated with worse prognosis in patients with LGG in both TCGA and CGGA datasets (Figure 8C).

We also assessed the expression levels of SPATS2L in patients with different subtypes of LGG. LGG was divided into astrocytoma, oligoastrocytoma, and oligodendroglioma subtypes (3). Compared with the oligoastrocytoma and oligodendroglioma subtypes, the expression levels of SPATS2L were higher in patients with the astrocytoma subtype of LGG in the TCGA datasets (Figure 8D). The high expression levels of SPATS2L in the astrocytoma subtype of LGG patients were also observed in the GSE16011 dataset, compared with the oligodendroglioma subtype (Figure 8D). The present study also assessed the expression levels of SPATS2L in different stages of patients with LGG. Compared with stage II LGG, SPATS2L was highly expressed in patients with stage III LGG in the TCGA datasets (Figure 8E). However, there were no significantly different expression levels of SPATS2L in different stages of LGG patients in the CGGA datasets (Figure 8E).

Finally, we validated the genetic connections and the prognostic significance of EGFR and CDKN2A alterations using the MSKCC glioma datasets (39). Similarly, in the MSKCC datasets, EGFR alteration was coexistent with CDKN2A deletion. Among the 1,004 glioma patients in the MSKCC datasets, 27% patients had EGFR alteration and 33% patients had CDKN2A deletion. One hundred and sixty-one glioma patients had both EGFR and CDKN2A alterations (Figure 9A).

Moreover, in the MSKCC datasets, EGFR amplification was also associated with bad prognosis of glioma. Glioma patients without EGFR amplification had longer overall survival (Figure 9B). Similarly, glioma patients without CDKN2A deletion also had better prognosis, compared with glioma patients with CDKN2A deletion (Figure 9B). Furthermore, glioma patients without EGFR amplification and CDKN2A deletion had the best prognosis than glioma patients with EGFR amplification or CDKN2A deletion or with both alterations in the MSKCC datasets (Figure 9C). Those results were consistent with previous results derived from the TCGA datasets.