Molecular Correlates of Long Survival in IDH-Wildtype Glioblastoma Cohorts

IDH-wildtype glioblastoma is a relatively common malignant brain tumor in adults. These patients generally have dismal progno-ses, although outliers with long survival have been noted in the literature. Recently, it has been reported that many histologically lower-grade IDH-wildtype astrocytomas have a similar clinical outcome to grade IV tumors, suggesting they may represent early or under-sampled glioblastomas. cIMPACT-NOW 3 guidelines now recommend upgrading IDH-wildtype astrocytomas with certain molecular criteria ( EGFR amplifications, chromosome 7 gain/10 loss, and/or TERT promoter mutations), establishing the concept of a “molecular grade IV” astrocytoma. In this report, we apply these cIMPACT-NOW 3 criteria to 2 independent glioblastoma cohorts, totaling 393 public database and institutional glioblastoma cases: 89 cases without any of the cIMPACT-NOW 3 criteria (GBM-C0) and 304 cases with one or more criteria (GBM-C1-3). In the GBM-C0 groups, there was a trend toward longer recurrence-free survival (median 12–17 vs 6–10months), significantly longer overall survival (median 32–41 vs 15–18months), younger age at initial diagnosis, and lower overall mutation burden compared to the GBM-C1-3 cohorts. These data suggest that while histologic features may not be ideal indicators of patient survival in IDH-wildtype astrocytomas, these 3 molecular features may also be important prognostic factors in IDH-wildtype glioblastoma.


INTRODUCTION
Glioblastoma (GBM) is the third most common intracranial tumor after pituitary adenoma and meningioma (comprising 14.7% of all cases), and is the most common malignant central nervous system tumor with an annual incidence of 3.21/100 000 individuals and >11 000 new cases diagnosed each year in the United States (1,2).Identification of IDH1/2 mutations in a subset of both histologically low-grade gliomas (LGGs) and GBMs (3,4) has led to a change in the diagnosis and reporting of these tumors with an integrated histologic/ molecular diagnosis focused primarily around IDH1/2 status (5).IDH-wildtype GBMs comprise 90% of all GBM cases, tend to occur in older individuals (mean age at diagnosis of 62 years), and have median survival intervals of approximately 10-15 months (4,5).Even with recent advances in treatment, the overall expected 5-year survival rate for GBM is <5% (2,6), although rare cases with extremely long-term survival have been reported in the literature (7)(8)(9).
Longer survival among some patients diagnosed with IDH-wildtype GBM raises hope of finding additional prognostic molecular markers that may surpass traditional histologic features in predicting survival or serve as therapeutic targets.Investigation into large cohorts of IDH-wildtype GBMs demonstrated that EGFR alterations are one of the most common features in these patients, present in >50% of cases.Cohorts with long-term survival (>36 months) had less-frequent EGFR amplification (or pathogenic activating mutation), although many of these studies included both IDH-wildtype and IDH-mutant GBMs (10,11).Other studies of the effect of EGFR in GBM have yielded conflicting results, with patient age being a possible confounding factor (11)(12)(13)(14)(15)(16)(17). Combined gain of whole chromosome 7 and loss of whole chromosome 10 (7þ/10-) is another frequent alteration that is considered definitional for GBM and may confer poor clinical outcome within this astrocytoma subset as well as histologically lowergrade IDH-wildtype astrocytomas (18)(19)(20).TERT promoter (TERTp) mutation is similarly common in GBM and has a negative prognostic value in some glioma subsets; however, this alteration is less specific for GBM than the other 2 factors (18,(21)(22)(23)(24).In the IDH-wildtype category, histologically LGGs (defined here as WHO grades II and III) with high-level EGFR amplification, 7þ/10-, and/or TERTp mutation have been shown to have aggressive clinical outcomes indistinguishable from IDH-wildtype GBM, and thus are now considered to be "molecular grade IV" by cIMPACT-NOW update 3 criteria (18,(25)(26)(27)(28)(29)(30).
Additional research suggests that groups of longer surviving patients also tend to have a lower incidence of CDK4 amplification and homozygous CDKN2A deletion (10,11).Other reports have suggested that within the IDH-wildtype GBM groups, co-gain of chromosomes 19 and 20 (19þ/20þ) is associated with longer overall survival (31,32), and MGMT promoter methylation results in longer patient survival as it impairs the protective response to alkylating agents in tumor cells and thus confers a better response to temozolomide therapy (33)(34)(35).
In this report, we identified 299 public dataset IDHwildtype GBM cases, including 65 cases without cIMPACT-NOW 3 factors (GBM-C0) and 234 cases with 1-3 of these factors, as well as an additional 15 LGG cases without cIMPACT-NOW factors and 51 with at least one factor from The Cancer Genome Atlas (TCGA) online repository.We analyzed the GBM and LGG groups with respect to total copy number variation (CNV), somatic mutation burden, specific mutations, and specific gene amplifications and deletions.In addition, we analyzed an institutional cohort of 24 GBM-C0 cases and 70 GBM-C1-3 cases as an independent validation cohort, using similar methods.In all 3 cohorts, the groups without cIMPACT-NOW criteria had significantly longer overall survival and younger age at initial diagnosis than those with at least one of these factors.Our results raise the possibility that these 3 molecular features may be as important in determining prognostic categories in IDH-wildtype GBMs as they are in histologically lower-grade IDH-wildtype gliomas.

Case Selection
We performed a search of histologically confirmed GBM cases across multiple publicly available datasets available on the cBioPortal interface (www.cbioportal.org)(36,37), TCGA database (https://portal.gdc.cancer.gov/),and other previously published publicly available databases to create a public database cohort (Cohort 1; Fig. 1) (10,21,(38)(39)(40)(41) (http://creativecommons.org/licenses/by/4.0/;http://creativecommons.org/publicdomain/zero/1.0/).The available histologic and molecular features were manually examined in each case; all cases with 1p/19q co-deletion, IDH1/2 mutations, or incomplete mutational analysis precluding IDH1/2 status evaluation were excluded.All cases were then screened for the availability of data regarding EGFR alterations, chromosome 7/10 status, TERTp status, and TERT mRNA expression levels.Notably, a number of the cases in the original TCGA datasets had only TERT mRNA expression levels without TERTp mutation status.In cases where both DNA and RNA sequencing were performed, the 2 measures were highly correlated; in the remainder of cases, TERT mRNA expression level was used to estimate the promoter mutation status, as previously described (10,24).We identified a total of 65 IDH-wildtype GBMs without cIMPACT-NOW update 3 factors and an additional 234 cases with 1-3 cIMPACT-NOW 3 factors for a control group.
In addition, we searched institutional cases (2006-2017) that had sufficient molecular analysis performed for clinical purposes at the time of pathologic diagnosis on the initial resection specimen, including targeted molecular profiling and/ or copy number profiling (https://www.pennmedicine.org/departments-and-centers/center-for-personalized-diagnostics/ gene-panels; https://www.foundationmedicine.com/genomictesting/foundation-one-cdx), as previously described (9,17,42,43).In total, we identified 94 total IDH-wildtype GBM cases with sufficient molecular data to test our central hypothesis (24 cases without cIMPACT-NOW factors, 70 cases with 1-3 factors) (Cohort 2; Fig. 2).All molecular data are derived from tissue from the initial resection specimen.No significant differences between tumor size, extent of resection, or postsurgical treatment were found between groups in the institutional cohort.All ethical standards were followed and this retrospective study was performed with Institutional Review Board approval.

Genetic and Epigenetic Analyses
The gene expression (Illumina HiSeq, RNASeq) and DNA methylation data (Illumina Human methylation 450) (Illumina, San Diego, CA) were downloaded for the selected TCGA GBM and LGG cases and analyzed with TCGAbiolinks (https://bioconductor.org/packages/release/bioc/html/TCGAbiolinks.html),Qiagen's IPA tool (www.qiagen.com/ingenuity) (Qiagen, Hilden, Germany), and R 3.4.1 (http:// www.R-project.org/)(44,45).The Affymetrix SNP 6.0 microarray data normalized to germline for copy number analysis for the same TCGA cases were downloaded from Broad GDAC Firehose (Broad Institute, Cambridge, MA).The fraction of the genome with copy number alterations was calculated from the above data as the fraction of the genome with  log2 of copy number >0.3 following the procedure used in cBioPortal (37).
The GISTIC 2.0 algorithm was used to identify individual regions of the genome that are significantly amplified or deleted (46).Each region with significant alteration was screened for tumor suppressor genes, oncogenes, and other genes associated with glioma and malignancy (46,47).GIS-TIC 2.0 analysis was run in GenePattern (https://www.genepattern.org/)(48).

Mutation Analysis
The mutation load is the number of nonsynonymous mutations seen in a sample.Differential analysis and visualization of mutations were done using Maftools (49).TERTp mutation was obtained from DNA sequencing data and TERT mRNA expression data were correlated to TERTp status in the subset of cases where both measures were available, as previously described (10,24,50).Variant annotation was performed using COSMIC (51), dbSNP (52), ClinVar (53), CanProVar 2.0 (54), The 1000 Genomes Project (55), and FATHMM-MKL (56).

Statistical Analysis
Differences in patient age, total mutation burden, and CNV were calculated using ANOVA.Significance of survival curves was calculated using the log-rank test (Mantel-Cox test).All univariate and multivariate regression analyses and other statistical calculations were performed with MedCalc and GraphPad Prism version 8 (GraphPad, La Jolla, CA).
The GBM-C0 group in Cohort 1 had a significantly younger age at initial diagnosis (55.2 6 2.2 years) compared to tumors in the GBM-C1 (61.The Cohort 2 GBM-C0 cases also had a significantly younger age at initial diagnosis (52.2 6 2.2 years) than the corresponding GBM-C1-3 (61.9 6 1.4 years; p ¼ 0.0006).These differences were significant after correcting for multiple comparisons, and with both univariate and multivariate analyses (Tables 1  and 2).There was a significantly lower level of overall mutation burden in the Cohort 1 GBM-C0 group compared to tumors with at least one cIMPACT-NOW 3 factor (p < 0.0001) (Table 1).Unlike IDH-mutant astrocytoma and oligodendroglioma cohorts (57)(58)(59), the overall CNV levels were not significantly different between the GBM-C0 and the GBM C1-3 groups (p ¼ 0.5920) (Table 1).
In Cohort 1, there was a lower percentage of cases with PTEN alterations in the GBM-C0 group compared to the GBM-C1-3 group, however, this difference was not found in Cohort 2, and there was a higher percentage of GBM-C0 cases with homozygous loss of CDKN2A in Cohort 2 but not in Cohort 1.After correcting for multiple comparisons, only the RFS and OS, patient age, mutation burden, and frequency of PTEN alterations were significantly different between the GBM-C0 and GBM-C1 cohorts (Table 1).Notably, there was no difference in frequency of 19þ/20þ or MGMT promoter methylation between the GBM-C0 and GBM-C1-3 groups after correcting for multiple comparisons (Table 1).
Additionally, we pooled all LGG and GBM cases without cIMPACT-NOW 3 factors into a single group (all C0 cases) and all LGG and GBM cases with at least one cIMPACT-NOW 3 factor into a single group (all C1-3 cases).There were significant differences between these groups in terms of RFS (median survival 17.0 vs 7.0 months; p ¼ 0.0004) (Fig. 5C), OS (median survival of 43.7 vs 15.0 months; p < 0.0001) (Fig. 5D), and age at initial diagnosis (52.3 6 1.7 vs 61.6 6 0.7 years; p < 0.0001).

Analysis of MGMT Promoter Methylation Status As an Additional Prognostic Factor in Gliomas With cIMPACT-NOW 3 Criteria
No significant difference in frequency of MGMT methylation was found between the GBM-C0, -C1, -C2, or -C3 groups, or between the GBM-C0 and pooled GBM-C1-3 groups in univariate or multivariate analysis (Tables 1 and 2).In addition, no significant difference was found between tumors with and without MGMT promoter methylation within the GBM-C0 cohorts.Within the GBM-C1-3 groups, no significant difference was found in terms of RFS between cases with methylated and unmethylated MGMT in Cohort 1 (p ¼ 0.6077) (Fig. 6A) or Cohort 2 (p ¼ 0.2932) (Fig. 6C).In terms of OS, there was a significant difference in cases with methylated versus unmethylated MGMT in Cohort 1 (p ¼ 0.0492) (Fig. 6B) and Cohort 2 (p ¼ 0.0006) (Fig. 6D).

DISCUSSION
Since the introduction of the 2016 WHO Classification of Tumours of the Central Nervous System dividing GBM and other adult astrocytomas into IDH-wildtype and IDH-mutant groups (5), much work has been performed to better understand the underlying molecular drivers of these tumors, and to identify reliable prognostic markers and targetable genomic alterations (8,10,11,13,21,28,31,32,35,38,40,57,60,61).The recent cIMPACT-NOW 3 update defines the minimum molecular criteria required for upgrading an IDH-wildtype astrocytoma with WHO grade II or III histologic features to IDH-wildtype astrocytoma, molecular grade IV (26).Since these factors are now considered "definitional" of grade IV within the IDH-wildtype astrocytoma class and reliably convey a worse prognosis in histologically lower-grade tumors, there is the implication that these factors may form a "molecular baseline" for higher-grade biologic behavior, although other alterations including homozygous deletion of CDKN2A have been considered as well (10,11,26).
In this context, we applied the cIMPACT-NOW 3 paradigm to IDH-wildtype GBMs to determine if there was a correlation between these factors and clinical outcomes in histologically grade IV tumors.Approximately 3% of the total GBM cases in the TCGA and cBioPortal datasets lack all 3 of these cIMPACT-NOW 3 molecular GBM criteria (GBM-C0), although they are designated as WHO grade IV tumors on the basis of histologic features (microvascular proliferation and/or tumor necrosis).It should be noted, however, that a portion of the total IDH-wildtype GBM cases do not have TERTp mutation status available in the TCGA and cBioPortal datasets, so this "triple-negative" GBM-C0 subgroup may be somewhat more frequent.It is also important to note that there may be an inherent selection bias in the institutional cases that were sequenced for clinical purposes and in the TCGA cases in terms of the cases initially sent from various institutions, as well as bias in sample type and molecular analysis of these cases (including batch effects) (62)(63)(64)(65).
In this study, the GBM-C0 groups had significantly longer OS intervals compared to the GBM-C1-3 group and individual GBM-C1, -C2, and -C3 subgroups (Fig. 4).In the GBM-C1 group, there was no significant difference in RFS or OS with regard to which of the 3 criteria is present, so the presence of any of these factors appears to be sufficient to produce a worse clinical outcome.The GBM-C0 groups also presented at a younger age and had fewer overall somatic mutations (Table 1) (60).Additional analysis of LGGs revealed that the LGG-C0 group had statistically indistinguishable survival intervals with the GBM-C0 group.
Our results do not validate the previous observation that CDK4 amplification and homozygous CDKN2A deletion are less frequently found in GBM cases with more favorable outcomes (10).We did, however, identify a significantly better OS in the cases with one or more cIMPACT-NOW 3 factor and 19þ/20þ than the cases without chromosome 19/20 cogain in both cohorts (Fig. 7 and Table 2), indicating that this may be an important additional factor to include when evaluating the prognosis in specific subsets of GBM cases.As previously reported (33,34), GBM-C1-3 cases with MGMT methylation had significantly longer OS intervals than those with unmethylated MGMT in both Cohort 1 and Cohort 2, but no significant effect was seen in terms of RFS (Fig. 6 and Table 2).
The current report is the first to establish a statistically significant role of these combined cIMPACT-NOW update 3 factors in predicting clinical outcome in IDH-wildtype GBMs, suggesting that like histologically lower-grade astrocytomas, these molecular features may be more useful for prognostic stratification than classic histologic findings in certain subsets.While testing all IDH-wildtype GBMs for these factors may prove cost-prohibitive, our findings suggest that there may be a benefit to screening younger IDH-wildtype GBM patients for these cIMPACT-NOW 3 criteria to help refine prognosis in these cases.

FIGURE 1 .
FIGURE 1. Summary chart for Cohort 1 showing key molecular alterations in the 299 assessed public dataset IDH-wildtype glioblastomas.

FIGURE 2 .
FIGURE 2. Summary chart for Cohort 2 showing key molecular alterations in the 94 assessed institutional IDH-wildtype glioblastomas.

FIGURE 3 .
FIGURE 3. Summary chart showing key molecular alterations in the 66 assessed IDH-wildtype histologically lower-grade gliomas.

FIGURE 6 .
FIGURE 6. Kaplan-Meier survival curves demonstrating no significant difference in the Cohort 1 GBM-C1-3 cases with methylated MGMT compared to those with unmethylated MGMT in terms of recurrence-free survival (RFS) (p ¼ 0.6077) (A), however, a significant difference was observed in terms of overall survival (OS) (p ¼ 0.0492) (B).Kaplan-Meier survival curves demonstrating no significant difference in the Cohort 2 GBM-C1-3 cases with methylated MGMT compared to those with unmethylated MGMT in terms of RFS (p ¼ 0.2932) (C), however, a significant difference was observed in terms of OS (p ¼ 0.0006) (D).

FIGURE 7 .
FIGURE 7. Kaplan-Meier survival curves demonstrating no significant difference in the Cohort 1 GBM-C1-3 cases with co-gain of chromosomes 19þ/20þ compared to the GBM-C1-3 cases without 19þ/20þ in terms of recurrence-free survival (RFS) (p ¼ 0.4159) (A), however, there was a significantly longer overall survival (OS) in the Cohort 1 cases with 19þ/20þ (p ¼ 0.0013) (B).There was a trend toward longer survival in the Cohort 2 GBM-C1-3 cases with co-gain of chromosomes 19þ/ 20þ compared to the GBM-C1-3 cases without 19þ/20þ in terms of RFS, although this was not a significant effect (p ¼ 0.0961) (C), however, there was a significantly longer OS in the Cohort 2 cases with 19þ/20þ (p ¼ 0.0073) (D).

TABLE 1 .
Clinical and Molecular Variables in LGG Cohort and GBM Cohorts

TABLE 2 .
Hazard Ratios for Molecular Variables