Background
Glioblastoma multiforme (GBM) is regarded as the most common malignant brain tumor in adults, accounting for 47.1% of all malignant brain tumors [
1], and the median survival time of untreated patients with GBM is only 3 months [
2]. For malignant brain tumors, according to the Central Brain Tumor Registry of the United States (CBTRUS), the incidence rate of GBM in the United States is extremely high (3.20/100,000 population) and increases with age [
1]. Maximal surgical resection, is considered as the first-line treatment for GBM patients relieving clinical symptoms, extending survival time and providing tissue to pathological diagnosis [
3]. A large-scale randomized phase III trial, initiated by the European Organization for Research and Treatment of Cancer and National Cancer Institute of Canada Clinical Trials Group, found that the 2-year survival rate of GBM patients was improved to 26.5% by radiotherapy plus temozolomide from 10.4% by radiotherapy alone [
4]. Since then, the standard therapeutic strategy for glioblastoma patients has become the multimodal treatment with radiotherapy and chemotherapy after surgery. Therefore, prediction of response to chemotherapy drugs or radiation and prediction of prognosis are crucial for post-surgical GBM patients.
In 1993, the Radiation Therapy Oncology Group-Recursive Partitioning Analysis (RTOG-RPA) classification system was developed for high-grade glioma patients with similar survival times [
5] and validated its prognostic significance in GBM patients [
6‐
8]. However, all the stratification variables of RTOG-RPA risk classification are clinical factors including age, tumor size and location, treatment, karnofsky performance score (KPS), cytologic, histologic composition and so on. Due to the intra- and inter-individual heterogeneity, the RTOG-RPA classification could not satisfactorily predict the survival and tumor response to therapy of each individual [
9]. Therefore, molecular markers are becoming more useful in the field of prognosis prediction [
10]. Currently, GBM related researches from genomics, epigenomics and transcriptomics level have led to unprecedented discoveries of potential prognostic and predictive indicators [
11]. Genomic analysis suggests survival-related genomic abnormalities in GBM patients, such as epidermal growth factor receptor (EGFR) amplification [
12,
13] and isocitrate dehydrogenase 1/2 (IDH1/2) mutations [
14,
15], have prognostic value. Some studies show that high expression of EGFR indicated poor prognosis [
16], and other research find the IDH mutations are associated with improved survival [
17]. From the epigenetic level,
O-6-methylguanine-DNA methyltransferase (MGMT) promoter methylation has been demonstrated that it is associated with improved progression-free and overall survival in GBM patients treated with alkylating agents [
18]. However, genomic prognostic classification of GBM is not yet clinically feasible, and the mechanism of how these multiple genomic alterations affect clinical prognosis is not clear [
19].
As far as the transcriptomics level is concerned, studies mostly focus on mRNA or protein coding gene (PCG) and long noncoding gene (lncRNA) because of their role as gene expression regulators, tumor suppressors and oncogenes. Using PCGs or lncRNAs, numerous studies have constructed transcriptome prognosis models for GBM survival prediction. Zhu et al. screened out an effective prediction system composed of 63 signature genes for glioblastoma prognosis [
20]. Marko et al. identified a 43-gene “fingerprint” from a population of 1478 differential expressed genes (
P < 0.01) that distinguished GBM survival phenotypes [
21]. Anindya Dutta et al. in a global analysis identified 584 lncRNAs correlated with a poor prognosis and 282 lncRNAs associated with better survival outcomes in GBM patients [
22]. Above researches verify PCGs and lncRNAs can be prognostic biomarkers of GBM. However, these studies found too many prognostic genes to provide a clinically feasible transcriptome signature with a small number of genes to predict the survival of GBM patients. Therefore, we focus our attention on find out a molecular signature which contains few prognostic genes and could more accurately predict the outcomes of postoperative GBM patients and guide the tailored therapy.
In the present study, we sought to explore the role of multi-transcriptome signature in the prognosis of GBM patients after surgery. We analyzed 233 postoperative GBM patients with the expression profiles of mRNAs and lncRNAs and screened out genes significantly associated with survival. Through further bioinformatics analysis, we aimed at constructing a prognostic transcriptome signature to divide patients into different risk groups, thereby assessing the survival and treatment response for GBM patients after surgical resection.
Discussion
Glioblastoma multiforme (GBM) is a heterogeneous disease characterized by poor prognosis. In order to extend the survival time of patients with GBM, in recent year, adjuvant and concomitant temozolomide with radiation are widely used. Despite advances in treatment such as radiation and chemotherapy, the prognosis and therapy response for post-surgical GBM patients with similar clinical risk factors varied tremendously. Considering the molecular heterogeneity of GBM, in this study, we identified a prognostic molecular indicator comprising five long non-coding RNAs and six protein coding genes, and confirmed the survival prediction power of the PCG-lncRNA signature in postoperative GBM patients.
Molecular markers are of great significance to disease diagnosis, treatment decision and prognosis assessment. With regard to the prognostic molecular characterization of GBM, the 2016 World Health Organization (WHO), for the first time, used the isocitrate dehydrogenase (IDH) gene mutation status as the classification molecular parameter to separate the GBM into three groups: GBM IDH-wild type, GBM IDH-mutant, and GBM NOS [
37], with different prognosis [
38]. In the past decade, GBM prognostic studies focused on mRNA or PCG as a result of the development of sequence technology and The Cancer Genome Atlas (TCGA) database. Chen et al. selected a gene expression signature score (GGESS) by incorporating ten glycolytic genes significantly correlated with patient survival and verified that the PCG signature could independently predict prognosis and response to chemotherapy of GBM patients [
39]. According to Chinese Glioma Genome Atlas (CGGA) RNA sequencing database and TCGA DNA methylation, another study established a gene signature comprising eight differentially expressed genes affected by DNA methylation and validated its prognostic value for GBM patients [
40]. A minimal multigene signature that correlated with patient survival and effectively separated the proneural and mesenchymal glioblastoma subtypes was developed from two patient-derived novel primary cell culture models (MTA10 and KW10) [
41].
Recently, emerging evidence suggests that lncRNA play a vital role in cancer occurrence and development, such as regulating gene transcription [
42] and post transcriptional processing of mRNA [
43], participating in chromatin remodeling [
44]. Subsequently, a great deal of lncRNAs have been shown to be closely associated with the survival of patients in different cancer types, indicating its prognostic prediction role. For GBM patients, some researchers identified a six-lncRNA signature associated with the overall survival by analyzing lncRNA expression profiling in 213 GBM tumors from TCGA [
45]. An immune-related six-lncRNA signature was found by performing a genome-wide analysis of lncRNA expression profiles form 419 GBM patients and demonstrated its ability to stratify patients into high- and low-risk groups with significantly different survival [
46]. All these above mentioned studies highlighted that it is feasible to mine the reliable and readily available expression profiles from TCGA database in GBM prognostic PCG/lncRNA marker studies. Moreover, a recent work found the dysregulated lncRNAs and mRNAs associated with acquired TMZ resistance in glioblastoma cells in vitro and may provide novel targets for GBM chemotherapy [
47].
Therefore, in the present study, we combined the PCG expression profile with the tissue-specific lncRNA expression profile to explore a signature indicating the prognosis and therapy effectiveness of postoperative GBM patients. We obtained 233 postsurgical GBM patients with corresponding PCG, lncRNA expression profiles and clinical information as the study object. After summarized clinical characteristics, we found the median age of the postsurgical GBM patients was 60 and more common in men, almost consistent with most research reported [
48‐
50]. Clinical treatment information of these 233 GBM patients provided convenience for our research on treatment response. Subsequently, we used two powerful bioinformatics analysis methods for identification of prognostic genes. Firstly, univariable cox regression analysis was performed and identified 707 genes that was significantly associated with the overall survival of GBM patients in the training dataset. Secondly, the random survival forest method further minimized the prognostic genes to 6 PCGs and 6 lncRNAs. Then we screened out a PCG-lncRNA signature with biggest AUC from 4095 combinations including different number of PCGs and/or lncRNAs, comprising six PCGs (EIF2AK3, EPRS, GALE, GUCY2C, MTHFD2, RNF212) and five lncRNAs (LINC00618, LINC02015, AC068888.1, CERNA1, CTD-2140B24.6), which separated patients into low-risk or high-risk group with different survival in the training and test dataset. The biggest AUC value of the PCG-lncRNA signature suggests it was better than any PCG alone signature or lncRNA alone signature. Multivariable Cox regression analysis verified the independence of the selected PCG-lncRNA signature from clinical factors like sex, age, KPS in predicting survival in postoperative GBM patients. As we mentioned, radiotherapy plus concomitant and maintenance TMZ chemotherapy after operation is the standard treatment for GBM patients, which means most postoperative GBM patients experienced TMZ-chemoradiation. Notably, the stratification analysis found that the PCG-lncRNA signature could further classify the TMZ-chemoradiation patients into low-risk or high-risk group with different survival, indicating the PCG-lncRNA signature could be helpful in predicting GBM treatment outcome, especially in TMZ-chemoradiation treated patients. Previous studies reported that age, MGMT promoter and IDH1 mutation were one of the main prognostic factors for GBM [
45], so we compared the predictive ability of age, MGMT promoter and IDH1 mutation with that of the PCG-lncRNA signature, and the ROC analysis results confirmed the signature had a superior survival predictive power.
To further explore the characteristics of the prognostic PCGs and lncRNAs in the signature, we found EIF2AK3, EPRS, MTHFD2, RNF212, LINC02015, CTD-2140B24.6 were protected factors for GBM patients highly expressed these genes with a long survival time (univariable cox coefficient < 0), and the remaining genes (GALE, GUCY2C, LINC00618, AC068888.1, CERNA1) associated with short survival time were risk factors (univariable cox coefficient > 0) according to the univariable cox result in Table
2. Due to relevant functional research of the prognostic 11 genes are limited, we performed bioinformatics functional analyses including co-expression network analysis and pathway analysis. However, the biological roles of the selected genes in tumorigenesis are still not clear and should be investigated in further experimental studies.
There are some limitations in this work. Firstly, after rejecting missing data, only 6613 lncRNAs were included, which might neglect some potential lncRNAs. Secondly, only 233 patients were included in the analysis, thus the efficiency of the PCG-lncRNA signature should be confirmed in more GBM patients. Moreover, the molecular mechanisms how these prognostic genes or the PCG-lncRNA signature influence patients risk stratification and clinical treatment responses need to be explained.
Although the above shortcomings, this article still has advantages and novelty. Firstly, we used few genes which predict survival and construct a PCG-lncRNA signature with satisfactorily prognosis predictive power, giving the postoperative GBM patients and clinicians a potential signature to evaluate survival. Secondly, in the post-operative GBM patients, treated with radiotherapy or chemotherapy, we found the stratification power of the signature in TMZ-chemoradiation, which is helpful for clinical treatment guiding.
Authors’ contributions
The authors contributed in the following way: Gao WZ, Guo LM, Xu TQ: data collection, data analysis, interpretation and drafting; Jia F, Yin YH: study design, study supervision and final approval of the manuscript. All authors read and approved the final manuscript.