Biomarker-based prognostic stratification of young adult glioblastoma

While the predominant elderly and the pediatric glioblastomas have been extensively investigated, young adult glioblastomas were understudied. In this study, we sought to stratify young adult glioblastomas by BRAF, H3F3A and IDH1 mutations and examine the clinical relevance of the biomarkers. In 107 glioblastomas aged from 17 to 35 years, mutually exclusive BRAF-V600E (15%), H3F3A-K27M (15.9%), H3F3A-G34R/V (2.8%) and IDH1-R132H (16.8%) mutations were identified in over half of the cases. EGFR amplification and TERTp mutation were only detected in 3.7% and 8.4% in young adult glioblastomas, respectively. BRAF-V600E identified a clinically favorable subset of glioblastomas with younger age, frequent CDKN2A homozygous deletion, and was more amendable to surgical resection. H3F3A-K27M mutated glioblastomas were tightly associated with midline locations and showed dismal prognosis. IDH1-R132H was associated with older age and favorable outcome. Interestingly, tumors with positive PDGFRA immunohistochemical expression exhibited poorer prognosis and identified an aggressive subset of tumors among K27M mutated glioblastomas. Combining BRAF, H3F3A and IDH1 mutations allowed stratification of young adult glioblastomas into four prognostic subgroups. In summary, our study demonstrates the clinical values of stratifying young adult glioblastomas with BRAF, H3F3A and IDH1 mutations, which has important implications in refining prognostic classification of glioblastomas.


INTRODUCTION
Glioblastoma is the commonest and most devastating primary brain cancer [ 1 ]. The disease has a universally fatal prognosis despite aggressive treatment in which over 85% of patients die within two years [ 2 ]. While the median age group is middle age to elderly, a smaller numbers of cases are found in young adults and children [ 2 , 3 ]. The predominant group of middle aged and elderly glioblastomas has been extensively investigated. Molecular classifi cation in glioblastomas based on gene expression profi les classifi ed the heterogeneous disease into proneural, neural, classical and mesenchymal subtypes, with each subtype carried distinct genomic aberrations [ 4 ]. Isocitrate dehydrogenase-1 (IDH1) mutation or platelet-derived growth factor receptor alpha ( PDGFRA ) amplifi cation, epidermal growth factor receptor ( EGFR ) amplifi cation and neurofi bromin 1 ( NF1 ) mutations were associated with proneural, classical and mesenchymal glioblastomas, respectively [ 4 ]. Genome-wide methylation study further identifi ed a subset of adult glioblastoma with glioma-CpG island methylation phenotype (G-CIMP) which was enriched in proneural subgroup, tightly associated with IDH1 mutation and exhibiting favorable prognosis [ 5 ]. For the younger group, only the children's glioblastomas has been extensively studied [ 6 -9 ]. Distinct genomic aberrations including PDGFRA alterations and hotspot mutations in histone 3.3 ( H3F3A ) at codons 27 (K27) and 34 (G34) as well as histone 3.1 ( HIST1H3B ) at codon 27 (K27) were frequently found in glioblastomas in children [ 6 -8 ]. Genomic study in combined series of pediatric and adult glioblastomas further identifi ed age-specifi c biological subgroups which can be defi ned by driver events including H3F3A -K27M, H3F3A-G34R/V and IDH1 mutations, strongly indicating that glioblastomas are different diseases in different age groups [ 10 , 11 ]. Recent study also reported the activating mutation BRAF -V600E identifi ed a distinct clinical subgroup of pediatric high grade gliomas [ 12 ]. While current literature has been only focused in glioblastomas of either children or older patients and in particular, vast majority of adult glioblastomas studied were above 35 years as a result of the skewed distribution towards older age (median age 64 years) [ 2 ], the young adult age group was known to have better prognosis [ 3 , 13 ] but was understudied in the literature. In this regards, we investigate a set of subgroup-defi ning molecular biomarkers in young adult glioblastomas aged from 17 to 35 years and evaluate the prognostic impact of the biomarkers. Our study reveals that BRAF , H3F3A and IDH1 mutations are associated with distinct clinical features and can stratify young adult glioblastomas into prognostic subgroups, which have important clinical implications in refi ning the prognostic classifi cation of glioblastomas in young adults.

Cohort characteristics
Clinical and molecular data of the cohort was summarized in Table 1 and Figure 1 . The age of the young adult glioblastoma cohort ranged from 17 years to 35 years. The mean and median ages were 25 years, respectively. The male-to-female ratio was 1:1.61. Eighty-eight tumors (82.2%) were located in cerebral hemisphere. There were 51 tumors (47.7%) involving frontal lobe, 18 tumors (16.8%) involving parietal lobe, 30 tumors (28%) involving temporal lobe and seven tumors (6.5%) involving occipital lobe, with 18 cases (16.8%) affected two cerebral lobes and fi ve frontal tumors also affected corpus callosum (four cases) and lateral ventricle (one case). Twenty-four tumors (22.4%) affected midline structures including one case (0.9%) in basal ganglia, seven cases (6.5%) in thalamus, six cases (5.6%) in ventricular system, fi ve cases (4.7%) in corpus callosum, one case (0.9%) in cerebellum and four cases (3.7%) in spinal cord. Treatment data in operation and chemo-radiotherapy was available in 87 patients (81.3%) and 80 patients (74.8%), respectively. Fifty-eight of 87 patients (67%) received total resection. Sixty-one of 80 patients (76.3%) received radiotherapy and 64 of 80 patients (80%) received chemotherapy, with 52 of 80 patients (65%) received concomitant chemo-radiotherapy. Survival data was available in 94 patients (87.9%). The median overall survival and median follow-up were 14 Figure  S1). Notably, with the exception of three BRAF mutated glioblastomas concurrently harboring TERT p mutation, BRAF , H3F3A , IDH1 , TERT p mutations and EGFR amplifi cation were mutually exclusive ( Figure 1 ). BRAF mutated glioblastomas show frequent CDKN2A homozygous deletion and younger patient age CDKN2A homozygous deletion was detected in 35.6% (31/87) of cases with analyzable data (Table 1 and Supplementary Figure S1). Data was non-analyzable in 20 cases due to either weak hybridization signals or strong background fl uorescence. Correlating with other molecular biomarkers examined, CDKN2A homozygous deletion was associated with BRAF mutation. Eighty percent (12/15) of BRAF mutated glioblastomas concurrently harbored CDKN2A homozygous deletion, compared with 26.4% (19/72) of BRAF wild-type glioblastomas ( p = 0.0002) (Figure 2a ). Comparing the age between BRAF mutated and BRAF wild-type glioblastomas, patients with mutated BRAF were younger than those with wild-type. The mean ages of BRAF mutated glioblastomas and BRAF wild-type glioblastomas were 22.3 years and 25.5 years, respectively ( p = 0.013) (Figure 2b ). Further comparison of patient age between the mutually exclusive molecular subgroups also revealed that BRAF mutated glioblastomas were younger

Tumor location differ between H3F3A -K27M mutated glioblastomas and other glioblastoma subgroups
Distribution of tumor location according to molecular biomarkers was shown in Figure 2c . H3F3A -K27 mutation was found in 58.3% (14/24) of glioblastomas affecting midline structures, including six thalamic tumors, four ventricular tumors and four cervical / thoracic spinal cord tumors. In contrast, only 3.6% (3/83) of hemispheric glioblastomas harbored the mutation ( p < 0.00001) (Figure 2d ). These three K27M mutated hemispheric glioblastomas were all located in temporal / frontal lobe. Young adult glioblastomas with BRAF , IDH1 , H3F3A -G34R/V mutations and EGFR amplifi cation predominantly developed in hemispheric locations without affecting midline structure, with the exception of one BRAF mutated glioblastoma in corpus callosum and three IDH1 mutated glioblastomas in basal ganglia (one case) and corpus callosum (two cases).

Prognostication of young adult glioblastoma by molecular biomarkers
We investigated the survival data of the young adult glioblastoma cohort according to the clinical parameters and molecular biomarkers by univariate analysis as shown in Table 2 . Patients receiving total resection ( p = 0.002) and chemo-radiotherapy ( p < 0.0001) were associated with better prognosis and midline tumor location was associated with poor outcome ( p < 0.0001) (Supplementary Figure S2). Among the molecular biomarkers evaluated, BRAF , H3F3A-K27M, IDH1 and PDGFRA demonstrated prognostic relevance in young adult glioblastomas. BRAF mutated glioblastomas exhibited better prognosis than those with wild-type BRAF ( p = 0.032). The median overall survival was 43.2 months for BRAF mutated glioblastomas and 13.6 months for BRAF wild-type glioblastomas. IDH1 mutated tumors, as expected, also exhibited favorable prognosis comparing to IDH1 wild-type tumors, with median overall survival of 24.2 months in the mutant group and 13.5 months in the wild-type group ( p = 0.034). H3F3A-K27M mutation was associated with poor prognosis across the cohort. The median overall survival was 6 months in K27M mutated tumors, compared to 17.6 months in the wild-type counterparts ( p < 0.0001). Tumors with positive PDGFRA expression showed shorter survival (8.6 months) than those with negative expression (17.4 months) ( p = 0.03). Co-evaluation of BRAF , IDH1 and H3F3A-K27M status stratifi ed young adult glioblastomas into four prognostic groups across the cohort ( p < 0.00001). BRAF mutated tumors ( p = 0.038) and IDH1 mutated tumors ( p = 0.028) had better prognosis than BRAF/IDH1/ K27 wild-type tumors, which in turn showed better survival than K27M mutated tumors ( p = 0.002) (Figure 3a to 3e ). Subset analysis demonstrated that positive PDGFRA expression was associated with poor outcome within the K27M mutated glioblastoma subgroup ( p = 0.03) (Supplementary Figure S2). Multivariate analysis was performed by including molecular biomarkers showing prognostic    BRAF , IDH1 , H3F3A -K27M

DISCUSSION
While the predominant middle-aged and elderly glioblastomas as well as the pediatric glioblastomas have been extensively investigated, there is lack of focusing study in the young adult age group. In this study, we showed that prognostication and possibly classifi cation of young adult glioblastoma can be biomarker-based and demonstrated that a signifi cant portion of young adult glioblastomas could be genetically defi ned by mutually exclusive BRAF-V600E mutation (15%), H3F3A -K27M mutation (15.9%), H3F3A-G34R/V mutation (2.8%) and IDH1 -R132H mutation (16.8%). BRAF-V600E mutation was frequently identifi ed in pediatric low-grade gliomas including pleomorphic xanthoastrocytoma, ganglioglioma and extra-cerebellar pilocytic astrocytoma [ 14 , 15 ]. In contrast to these pediatric low-grade glioma subtypes, mutation frequency was much lower in adult diffuse gliomas [ 15 -19 ]. Knobbe and co-workers investigated 94 glioblastomas and identifi ed three cases (3.2%) harboring BRAF mutation [ 16 ]. In the study by Schindler and colleagues examining 1,320 nervous system tumors, BRAF mutation was detected in less than 2% of adult glioblastomas investigated [ 15 ]. Dahiya and colleagues evaluated the BRAF status in 39 adult glioblastomas and identifi ed 7.7% (three cases) possessing the mutation [ 19 ]. The mutation was reported in up to 54% of epithelioid glioblastoma, an uncommon histologic variant of glioblastoma in young adults [ 17 ]. In our current study focusing on young adult patients aged from 17 to 35 years, 15% of patients had BRAF mutation but no BRAF mutation was identifi ed in seven cases of secondary young adult glioblastomas. And patients had BRAF mutation were signifi cantly younger than those without the mutation. Mutated cases had mean age of 22.3 years and median age of 21 years, compared to mean age of 25.5 years and median age of 26 years in wild-type cases ( p = 0.013). Additionally, BRAF mutation identifi ed a subset of patients with favorable prognosis in our cohort. In univariate analysis, the median overall survival of patients with BRAF mutated glioblastomas was 43.2 months and those with wild-type BRAF was 13.6 months ( p = 0.032). Notably, among the 16 BRAF mutated glioblastomas, 11 cases had operation record and total resection could be achieved in all cases ( p = 0.013), suggesting BRAF mutated glioblastomas were more amendable to surgical resection. Seven cases received concomitant chemotherapy and radiotherapy, out of nine cases with adjuvant treatment data available. Since majority of the BRAF mutated tumors received total resection and concomitant chemo-radiation which could account for the favorable outcome of this subset of patient in our cohort, we tried to analyze the prognostic value of BRAF mutation in patients receiving total resection and concomitant chemoradiation ( n = 37). Seven BRAF mutated tumors showed strong trend of better prognosis than 30 BRAF wild-type tumors (Supplementary Figure S2). The potential prognostic value of BRAF -V600E mutation was also recently reported in pediatric high-grade gliomas [ 9 , 12 ]. Mistry and colleagues described a series of pediatric secondary high-grade gliomas harboring BRAF mutation and CDKN2A deletion, showing longer latency to transformation from low-grade lesion and better clinical outcome [ 12 ]. By conducting genome-wide DNA methylation profi ling in 202 pediatric glioblastomas, Korshunov and colleagues identifi ed an epigenomic subset of glioblastoma showing methylation pattern similar to pleomorphic xanthoastrocytoma, enriched for BRAF mutation and CDKN2A homozygous deletion, and showed favorable prognosis [ 9 ]. Although those studies were on pediatric glioblastomas, our study provided complementary results to theirs and supported the clinical signifi cance of BRAF -V600E testing in glioblastoma of young person [ 9 , 12 , 17 , 19 ]. Apart from the potential prognostic value, BRAF -V600E mutation also served as a novel therapeutic target in this subset of glioblastomas. Robinson and colleagues recently reported a case of BRAF mutated pediatric glioblastoma treated by BRAF inhibitor vemurafenib and showed complete response [ 20 ]. Given the potential clinical utility in prognostication and treatment selection, BRAF mutational testing, either by direct sequencing or immunohistochemistry [ 21 ], should be conducted in glioblastomas of young patients.
Histone H3 mutation was exclusively observed in H3F3A in 18.7% of young adult glioblastomas. HIST1H3B mutation was not found in any of the 107 young adult samples examined, suggesting the mutation was specifi c to diffuse intrinsic pontine gliomas (DIPG) but not for non-brainstem high grade gliomas in both pediatric and adult patients [ 7 , 22 ]. K27M and G34R/V mutated tumors showed distinct tumor locations and clinical outcome in our cohort as in pediatric glioblastomas [ 9 , 10 ]. In our cohort, vast majority of K27M mutated tumors were located in midline structures including thalamus, ventricular system and cervical / thoracic spinal cord. These rare midline glioblastomas, despite of their different anatomical locations, shared the same driver mutation and suggested a closely-related origin [ 10 , 23 ]. Notably, K27M mutated glioblastomas showed aggressive clinical course in our cohort with median overall survival of 6 months (range 0.1 months to 19.8 months). Within this aggressive subset of glioblastoma, tumors with positive PDGFRA immunohistochemical expression exhibited a signifi cantly shorter survival (median 2.5 months) than tumors with negative expression (median 11.7 months) ( p = 0.03). Interestingly, Puget and colleagues previously reported a distinct transcriptional subgroup of DIPG characterized by oligodendroglial differentiation, driven by PDGFRA upregulation and exhibited signifi cantly worse outcome [ 23 ]. The association of PDGFRA expression with poor prognosis in K27M mutated glioblastomas in our study was in line with previous observation. The prognostic value of PDGFRA expression / alterations in the K27M mutated tumors warranted further evaluation in a larger cohort.
IDH1 mutation was present in 16.8% of young adult glioblastomas of which the mutation frequency was comparatively higher than the predominant group of adult / elderly glioblastoma as well as pediatric glioblastomas [ 8 , 9 , 24 -27 ]. Previous study by Pollack and colleagues reported that IDH1 mutation was common in adolescent malignant gliomas in which 16.3% of high grade gliomas between 3 to 21 years harbored the mutation and was signifi cantly associated with age greater than 14 years [ 28 ]. As expected, IDH1 mutated glioblastomas showed favorable prognosis in our cohort which was independent of age, tumor location, operation and adjuvant treatment among the young adult patients. Favorable prognostic value of IDH1 mutation was also recently demonstrated in pediatric glioblastomas [ 9 ]. Notably, although IDH1 mutation was identifi ed in nearly 90% of secondary elderly glioblastomas [ 27 ], only 43% (3/7) of IDH1 mutation was detected in our young adult glioblastomas, suggesting that IDH1 mutation wasn't a main contributing marker for young secondary glioblastomas. Collectively, our study provided complementary evidence to previous studies that IDH1 mutated glioblastoma was a glioblastoma subgroup with favorable prognosis prevalent in adolescents and young adults [ 11 ].
Hotspot mutations in TERT promoter region were identifi ed in up to 80% of adult / elderly glioblastomas [ 29 -31 ] but rare in pediatric glioblastomas [ 32 ]. The prevalence of TERT p mutation in the young adult age group was not precisely reported. In our study, only 8.7% of young adult glioblastomas harbored TERT p mutation, suggesting that mutation induced telomerase activation might not be a major mechanism in telomere deregulation in that age group. It remained to be investigated if promoter methylation of TERT causing telomerase upregulation or the alternative lengthening of telomeres (ALT) represented the major mechanism of telomere maintenance in young adult glioblastomas [ 6 , 33 ]. TERT p mutations were associated with poor prognosis in glioblastomas [ 30 , 31 , 34 , 35 ] as well as lower-grade gliomas with wild-type IDH [ 34 , 36 , 37 ]. Although not reached statistical signifi cance, the median overall survival of patients with TERT p mutated glioblastomas was only 5.1 months, compared to 15.6 months in those with wild-type TERT p. Notably, three TERT p mutations were overlapped with BRAF -V600E mutation in the cohort and overall survival was 5.1 months in one patient with survival data available. BRAF -V600E and TERT p mutations were recently found cooperatively identifying the most aggressive subset of papillary thyroid cancer with high recurrence rate [ 38 ]. TERT p mutations, either C228T or C250T, generated a consensus binding site (5′-TTCC-3′) for E-twenty-six (ETS) transcription factors and upregulated TERT expression [ 34 , 39 , 40 ]. On the other hand, activation of mitogen-activated protein kinase pathway was also shown to cause upregulation of the ETS system [ 41 -43 ]. The synergistic effect of BRAF -V600E and TERT p mutations in promoting tumorigenesis was therefore biologically explainable [ 38 ]. Although only 2.9% (3/104) of patients harbored concurrent BRAF -V600E and TERT p mutation, this subgroup accounted for 18.8% (3/16) of BRAF mutated glioblastomas of young adults. Further study should be conducted to evaluate the clinical value of concurrent BRAF and TERT p mutations in young adult glioblastomas.
In summary, our study demonstrates recurrent BRAF , IDH1 and H3F3A mutations in young adult glioblastomas with clinical impacts. BRAF mutation and IDH1 mutation identify glioblastomas with less aggressive clinical course and H3F3A -K27M mutation defi nes glioblastomas with dismal prognosis. The biomarkerbased stratifi cation has clinical implications and refi nes the prognostic classifi cation of young adult glioblastomas.

Patients and tissue samples
A total of 107 tissue samples (all formalin-fi xed paraffi n-embedded) were obtained from young adult patients (age from 17-35 years) of Department of Anatomical and Cellular Pathology, Prince of Wales Hospital (Hong Kong) and Department of Neurosurgery, Huashan hospital (Shanghai) with a histological diagnosis of "glioblastoma, WHO grade IV". Seven out of the 107 samples were secondary glioblastomas progressed from low grade gliomas, including oligoastrocytoma and astrocytoma. Other 102 cases were diagnosed as primary glioblastomas. Clinical and survival data of the patients were retrieved from the respective institutional medical record systems. This study was approved by the Ethics Committee of Shanghai Huashan Hospital and the New Territories East Cluster-Chinese University of Hong Kong Ethics Committee.

Molecular analysis
Mutational analysis was performed as described previously [ 36 , 44 ]. Tissues from representative tumor area with tumor content 70% were scrapped off from dewaxed sections and treated with proteinase K at a fi nal concentration of 2 mg/ml in 10 mM Tris-HCl buffer (pH 8.5) at 55°C for 2-18 hours and then at 98°C for 10 minutes. Crude cell lysate was centrifuged and supernatant was used for subsequent PCR analysis. The forward and reverse primers then were used to amplify gene BRAF , H3F3A , HIST1H3B , IDH1 and TERT . PCR was performed in 10ul reaction mixture for different thermal protocol (Supplementary Table S1, S2). Sequencing was performed using BigDye Terminator Cycle Sequencing kit v1.1 (Life Technologies). The products were resolved in Genetic Analyzer 3130 × l and analyzed by Sequencing Analysis software. Hotspots BRAF -V600E, www.impactjournals.com/oncotarget H3F3A -K27M, H3F3A -G34R/V, HIST1H3B -K27M, IDH1 -R132H and TERT p were detected (Supplementary Figure S1). All base changes were confi rmed by sequencing of a newly amplifi ed fragment.

Fluorescence in situ hybridization analysis for EGFR amplifi cation and CDKN2A deletion
Dual-probe fl uorescence in situ hybridization (FISH) assay was performed on paraffi n-embedded sections, with locus-specifi c probes for EGFR and CDKN2A paired with centromere probes for chromosome 7p12 and chromosome 9p21 (Supplementary Figure S1, Table S4). Deparaffi nization of the sections was carried out, followed by dehydration in 100% ethanol, retrieval by 1M sodium thiocyanate at 80°C for 10 minutes, and digestion in 0.04% pepsin at 37°C was applied on each section for 10 minutes. Simultaneous probe per specimen was denatured at 80°C for 10 minutes with subsequent overnight incubation at 37°C. The sections were washed next day in 1.5 M Urea/2X saline sodium citrate at 50°C for 10 minutes twice. After washing, sections were stained with Vectashield mounting medium containing 4′,6-diamidino-2-phenylindole (Vector Laboratories) and visualized under a fl uorescent microscope (Carl Zeiss Microscopy LLC, NY, USA). Hybridizing signals in at least 100 non-overlapping nuclei were counted. EGFR amplifi cation was considered when 50% of the tumor cells harbored more than fi ve signals per nuclei CEP7 or innumerable tight clusters of signals of the locus probe [ 45 ]. CDKN2A deletion was considered if both signals were lost (homozygous deletion) in at least 20% of tumor nuclei [ 46 ].

Immunohistochemistry of PDGFRA
FFPE tissue sections of 4 micron thickness were deparaffi nized in xylene and rehydrated in graded alcohols. For PDGFRA, antigen retrieval was carried out by treating the sections in 1M Citrate buffer (PH = 6.0) in a microwave oven. After antigen retrieval, all sections were processed by BondMax automade staining systems (Leica BondMax) using validated protocols. Tissue sections were incubated at 37°C for 30 mins with relevant antibodies of PDGFRA (Supplementary Table S3). Antigen detection was performed using Ultra View diamino benzidine chromogen step (BondMax). The presence of cytoplasmic and membrane staining indicated positivity for PDGFRA [ 47 ] (Supplementary Figure S1, Table S5).

Statistical analysis
Statistical analysis was performed using IBM SPSS Statistics 20 (IBM Corporation, NY, USA). Correlation between molecular markers and clinical parameters were examined by X 2 -test. Comparison between two groups was performed by Student's t -test or Mann-Whitney U -test. Comparison between three or more groups used one-way analysis of variance (ANOVA). Overall survival (OS) was defi ned as the duration between the diagnosis and death or last follow-up [ 25 ]. Survival curves were plotted by Kaplan-Meier method and analyzed by Log-rank test. Multivariate analysis for independent prognostic marker was performed by Cox-proportional hazards model. Tests with a p value below 0.05 were considered signifi cant.
Supplementary information is available at Oncotarget 's website.