A five‐microRNA signature for individualized prognosis evaluation and radiotherapy guidance in patients with diffuse lower‐grade glioma

Abstract Despite the prognostic value of IDH and other gene mutations found in diffuse glioma, markers that judge individual prognosis of patients with diffuse lower‐grade glioma (LGG) are still lacking. This study aims to develop an expression‐based microRNA signature to provide survival and radiotherapeutic response prediction for LGG patients. MicroRNA expression profiles and relevant clinical information of LGG patients were downloaded from The Cancer Genome Atlas (TCGA; the training group) and the Chinese Glioma Genome Atlas (CGGA; the test group). Cox regression analysis, random survival forests‐variable hunting (RSFVH) screening and receiver operating characteristic (ROC) were used to identify the prognostic microRNA signature. ROC and TimeROC curves were plotted to compare the predictive ability of IDH mutation and the signature. Stratification analysis was conducted in patients with radiotherapy information. Gene ontology (GO) analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were performed to explore the biological function of the signature. We identified a five‐microRNA signature that can classify patients into low‐risk or high‐risk group with significantly different survival in the training and test datasets (P < 0.001). The five‐microRNA signature was proved to be superior to IDH mutation in survival prediction (AUCtraining = 0.688 vs 0.607). Stratification analysis found the signature could further divide patients after radiotherapy into two risk groups. GO and KEGG analyses revealed that microRNAs from the prognostic signature were mainly enriched in cancer‐associated pathways. The newly discovered five‐microRNA signature could predict survival and radiotherapeutic response of LGG patients based on individual microRNA expression.


| INTRODUC TI ON
Glioma is a common type of intracranial malignant tumour, characterized by diffuse infiltration, no clear boundary, infinite proliferation and high invasion. According to histomorphological classification, World Health Organization (WHO) classifies gliomas into grades I to IV for diagnosis, guidance treatment and prognosis. Grade I gliomas have the lowest invasiveness and the best prognosis. Generally, patients can be cured by surgical removal of tumour tissue, and the 5-year survival rate reaches 94%. 1 On the contrary, grade IV tumours, in which glioblastomas are the majority, are the most aggressive and malignant. Even after maximal resection plus post-operative chemoradiotherapy, the median survival of patients with glioblastoma is as short as about one year. 2 Gliomas of grades II and III are intermediate in terms of malignant degree and prognosis and are combined into lower-grade gliomas because of their similarity in high invasiveness and the tendency to recur or develop advanced lesions such as secondary glioblastoma. Especially, the grade II/III astrocytic tumours and the grade II/III oligodendrogliomas (oligoastrocytomas are sorted into either astrocytoma or oligodendroglioma) are incorporated into diffuse lower-grade gliomas (LGG) after the 2016 CNS WHO Classification. 3 Some studies have found little prognostic difference between grade II diffuse astrocytomas and grade III anaplastic astrocytomas. 4,5 However, not all studies have confirmed the prognostic consistency of grade II and grade III gliomas. 6 Variabilities responding to treatment and differences in the timing of disease progression or recurrence are also frequently seen in LGG. Therefore, for patients with LGG, prognostic markers that can predict individual clinical outcome and treatment response are essential.
With the introduction of IDH mutation and 1p/19g codeletion into the classification of diffuse gliomas, it is increasingly recognized that molecular features play a significant role in the evaluation of prognosis and treatment response. MicroRNAs are a group of short-chain non-coding RNA molecules that are involved in the regulation multiple genes expression and activation of signalling pathways. Abnormal microRNA expression occurs in a variety of tumours and is closely related to tumour invasion, metastasis and more malignant lesions. Thus, microRNA could be used to predict treatment response and prognosis of many types of cancer and become a molecular marker. In addition, the stability and detectability of microRNAs in a variety of body fluids (serum/plasma/ cerebrospinal fluid) are more favourable for their use as prognostic markers. Therefore, microRNA signature containing multiple microRNAs has been demonstrated its promising prognostic role in various tumours including breast cancer, 7 hepatocellular carcinoma 8 and glioblastoma. 9 However, there is a lack of microRNA signature that can predict individual outcomes of patients with LGG.
In the present study, LGG patients with microRNA expression data were collected from the Cancer Genome Atlas (TCGA) database and the Chinese Glioma Genome Atlas (CGGA). We aim to use these expression data and clinical data to identify a prognostic microRNA signature and validate its predictive ability for survival and treatment response.

| Developing prognostic microRNA signatures in the TCGA dataset
Through univariate COX proportional regression analysis, we first identified the microRNAs associated with patients' survival time from the training group. Then, we further reduced the number of the survival related microRNAs using the random survival forestsvariable hunting (RSFVH) algorithm. In RSFVH, an iteration procedure was performed to narrow down the microRNAs set in which the 1/4 least important microRNAs were discarded at each iteration step and five hundred trees were grown at each step until without changing the error rate too much and then 9 microRNAs were selected. 10,11 Finally, COX regression analysis of selected candidate mi-croRNAs was performed and prognostic models were constructed as follows:

| Statistical and bioinformatics analysis
The median risk score was used as the cut-off value to subdivide patients with LGG into two risk groups. Kaplan-Meier analysis was performed to assess the survival difference of two risk groups, and statistical significance was assessed using the two-sided logrank test. Multivariate Cox regression analysis was performed to explore the independence of the screened signature. ROC and TimeROC were used to compare predictive performance. All analyses were performed with R program (www.r-proje ct.org), including packages named pROC, TimeROC, randomForestSRC and survival downloaded from Bioconductor (http://www.bioco nduct or.org/). we found that the median age of the enrolled patients was 40 years (14-87 years) and male patients were more than female ones. The number of grade II cases was basically the same as that of grade III, accounting for roughly half. Every patient had survival information, all but 186 of whom died. In addition, we listed the histological type, IDH mutation and radiotherapy and chemotherapy information (Table 1).
Using microRNA profiling data and survival information from the training set or TCGA set, we performed univariate Cox regression analysis to explore the association of patients' overall survival (OS) with microRNA expression. A total of 400 microR-NAs significantly associated with OS (P < 0.05, Table S1) were discovered and displayed as grey or red dots in Figure 1A. Next, we further reduced the number of prognostic microRNAs available for risk model by random survival forests-variable hunting (RSFVH) analysis, which were analysed by gradually removing one-fourth of the least important genes at each step ( Figure 1B).

| The microRNA signature predicts survival of patients with diffuse lower-grade glioma in the training and the independent dataset
When the median risk score was used as the cut-off point, the five-microRNA signature divided the training dataset into either the high-risk (n = 262) or low-risk group (n = 263). Kaplan-Meier analysis identified the low-risk group had significantly longer survival time than those in the high-risk group (median survival: 11.59 years vs 4.08 years, P < 0.001; Figure 2A). Then, we treated the independent CGGA diffuse lower-grade dataset with the specific follow-up survival time and vital status (n = 99, Table 1) as the test dataset to validate the prognostic value of the five-microRNA signature. We calculated the microRNA signature-based risk scores of the patients and then separated patients into high-and low-risk score group with the median risk score. Kaplan-Meier verified that the five microRNA could distinguish the low-risk group from the high-risk group in the test dataset (log-rank test P < 0.001; Figure 2B).

| The five-microRNA signature is an independent predictive factor
Through chi-square test in the training or test dataset, we found microRNA signature was associated with clinical variables such as grade (Table 3).
In addition, there are numerous known prognostic factors for LGG patients.

| Comparison of IDH mutation and the five-microRNA signature
Considering the important role of IDH mutation in the prognosis prediction of LGG, we performed ROC analysis and compared the area under the ROC curve (AUC) of IDH mutation and the five-microRNA signature to analyse their predictive power. 17,18 After congregating the TCGA and CGGA sets of data, we found a total of 216 patients with IDH mutation information. ROC analysis for vital status showed the predictive performance of the microRNA signature was better than IDH mutation status (AUCIDH = 0.607; AUCsignature = 0.688, Figure 4A). TimeROC analysis found the AUC of the signature was 0.737/0.720/0.728 at 1/3/5 years ( Figure 4B), and the AUC of IDH was 0.601/0.646/0.673 at 1/3/5 years ( Figure 4C) indicating that signature was better than that of IDH mutation in survival prediction.

| Radiotherapy stratification analysis
From the clinical information in Table1, we can find that most patients from the TCGA or CGGA datasets have clear radiotherapy information. To determine whether the signature can guide radiation therapy in LGG patients, we performed stratified analyses. As in the entire radiotherapy group, the overall survival time of the high-risk (n = 216) patients was significantly shorter than that of low-risk (n = 158) group patients (median survival 4.24 years vs 12.09 years, log-rank test P < 0.001, Figure 5A). Patients undergoing non-radiotherapy (n = 183) can also be divided into two groups with different prognosis by the signature (5-year/10-year survival rate: 45.60%/36.48% vs 92.40%/64.52%, log-rank test P < 0.001, Figure 5B).

F I G U R E 2
The prognostic microRNA signature predicts overall survival of patients with diffuse lower-grade glioma. Kaplan-Meier plots indicated patients could be classified into high-and low-risk groups according to the signature in the training (A) and test (B) datasets, and P values were calculated by log-rank test Survival probability (%) 49

TA B L E 3 Association of the microRNA signature with clinical characteristics in
LGG patients

| Function analysis of the five microRNAs
To explore the role and function of microRNAs screened in this study, we first detected the target genes of the five microRNAs and then performed functional analysis of the target genes by mirPath (see method). Figure 6A showed the target genes of the five microRNAs were significantly enriched in multiple cancer-related pathways, among which the glioma pathway was located in the top 10 differential pathways. Figure 6B displayed the target genes of the five prognostic microRNAs involved in the glioma pathway, indicating the pathway or mechanism in which the prognostic microRNAs might play a key role.  According to the European Association for Neuro-Oncology (EANO) guideline, the treatment of patients with LGG is mainly conservative management and radiotherapy, and a few patients with high malignancy require combination chemotherapy. However, there is no consistent conclusion as to which patient can no longer continue to 'watch and wait' but need radiotherapy or which patient have good radiotherapeutic response and can benefit from radiation therapy. Whether patients currently receive radiotherapy depends mainly on some prognostic indicators, such as age, Karnofsky performance score. 25 This study found that the signature can classify patients after radiotherapy into high-and low-risk groups with significant differences in prognosis, making the signature can be used to guide radiotherapy.

| D ISCUSS I ON
Notably, the five microRNAs in the signature are all risk factors. As for the functions of microRNAs in LGG, our GO and KEGG analysis found that the five risky microRNAs enriched in glioma related pathways. In the Pubmed database, we searched related studies to explore the role of the five microRNAs in glioma. Since the expression of miR-10b was found to be up-regulated and be closely related to the poor prognosis of glioma, 26-28 researchers have discovered that miR-10b may be involved in tumorigenesis by activating the p53 pathway 29 and induced glioma cell invasion by miR-10b/HOXD10/MMP-14/uPAR signalling pathway. 30 MiR-10b down-regulation inhabited proliferation, migration and invasion of glioma cell through regulating TGF-β1 stimulation 31 or homeobox B3 (HOXB3) expression. 32 For miR-148a, Wang et al 33 the NF-κB/miR-148a/TGF-β pathway was a critical mechanism of glioblastoma aggressiveness. Moreover, consistent with our study, researchers identified that miR-148a was an oncogenic gene associated with glioma survival and affected EGFR activation or regulated glioma growth by mediating HIF1α and Notch signalling. 34,35 Encouragingly, circulating exosome miR-148a was found in serum and promoted tumour progression by targeting CADM1 to activate STAT3 pathway, suggesting its clinical value. 36 MiR-155 has also F I G U R E 5 Radiotherapy stratification analysis. The five-microRNA signature could further divide patients undergoing radiotherapy (A) or patients undergoing non-radiotherapy (B) into two groups with significantly different survival  as an oncogene 44 and a risk factor for glioma prognosis. 45 Other studies, contrary to our finding, regarded miR-15b as a tumour suppressor in the progression of glioma. [46][47][48][49] Therefore, the role of miR-15b in glioma tumorigenesis remains unclear. Considering that the five microRNAs found in this study are all glioma prognostic risk genes, we still need to figure out whether there is some connection between the five microRNAs and how they work together to promote glioma progression.
In summary, our study findings revealed an expression-based five-microRNA signature can predict individual survival and radiotherapeutic response of patients with LGG. This paper highlights the promising potential of the novel five-microRNA signature as a good prognostic biomarker for individual survival prediction and therapeutic decision making.

CO N FLI C T O F I NTE R E S T
The authors confirm that there are no conflicts of interest.

E TH I C A L A PPROVA L
This article does not contain any studies with human participants or animals performed by any of the authors.
Informed consent: Informed consent was obtained from all individual participants included in the study. For specific informed consent, please refer to the source of public data.

DATA AVA I L A B I L I T Y S TAT E M E N T
Publicly available datasets were analysed in this study. This data can