Genome-Wide Analysis of Glioblastoma Patients with Unexpectedly Long Survival

Abstract

Glioblastoma (GBM), representing WHO grade IV astrocytoma, is a relatively common primary brain tumor in adults with an exceptionally dismal prognosis. With an incidence rate of over 10 000 cases in the United States annually, the median survival rate ranges from 10–15 months in IDH1/2-wildtype tumors and 24–31 months in IDH1/2-mutant tumors, with further variation depending on factors such as age, MGMT methylation status, and treatment regimen. We present a cohort of 4 patients, aged 37–60 at initial diagnosis, with IDH1-mutant GBMs that were associated with unusually long survival intervals after the initial diagnosis, currently ranging from 90 to 154 months (all still alive). We applied genome-wide profiling with a methylation array (Illumina EPIC Array 850k) and a next-generation sequencing panel to screen for genetic and epigenetic alterations in these tumors. All 4 tumors demonstrated methylation patterns and genomic alterations consistent with GBM. Three out of four cases showed focal amplification of the CCND2 gene or gain of the region on 12p that included CCND2, suggesting that this may be a favorable prognostic factor in GBM. As this study has a limited sample size, further evaluation of patients with similar favorable outcome is warranted to validate these findings.

INTRODUCTION

Glioblastoma (GBM) is the most commonly encountered glioma, comprising 54% of cases, and is among the most common primary intracranial tumors (15.8%), second only to meningioma (1, 2). GBM has an annual incidence of 3.19/100 000 people in the United States, with more than 10 000 new cases diagnosed each year (2–4). Since 2016, this diagnostic “entity” has been split into 2 separate major categories based on the presence or absence of mutations in either IDH1 or IDH2. This diagnosis predicts a dismal clinical course, with a 5-year survival rate of ∼5% (4, 5), and a median overall survival of 10–15 and 24–31 months for IDH1/2-wildtype (also known as de novo or primary GBM) and IDH1/2-mutant (secondary GBM) groups, respectively (6–8). Rare cases of long-term survival in GBM patients have been reported in the literature (9, 10). More recent studies involving methylation patterns and whole genome analysis for copy number alterations (CNAs) have identified a significantly increased level of overall CNA and chromosomal instability in GBM cases compared to lower-grade gliomas (defined as WHO grades II and III in these studies), indicating a higher level of overall genomic instability (11), although other authors have suggested that the genomic instability observed in these high-grade tumors may actually be associated with longer survival in GBM patients, as these tumors may be more responsive to conventional therapies (12).

In this report, we identified 4 neurologically intact patients from The University of Texas Southwestern Medical Center with a diagnosis of IDH1-mutant GBM and postsurgical survival intervals ranging from 90 to 154 months (average survival: 119 ± 15 months thus far) and no documented tumor recurrence. We evaluated these tumors with a combination of immunohistochemistry, Sequenom mass spectrometry analysis, genome-wide methylation profiling, CNA analysis, and next-generation sequencing (NGS) to search for genetic or epigenetic signatures that could potentially predict and/or explain the unusually long survival in this cohort. We found high levels of overall genomic instability across all chromosomes in these IDH-mutant GBMs, as well as more specific alterations in specific genes and chromosomal regions classically associated with GBM. Interestingly, we also identified high-level amplifications of CCND2, a gene that encodes for cyclin D2, in 3 of 4 cases. This specific alteration is relatively rare among GBM cases with conventional clinical courses (13, 14).

MATERIALS AND METHODS

Case Selection and Clinical Review

A total of 4 patients with a diagnosis of “GBM” and a survival interval of ≥7.5 years were identified among 438 cases of GBM reviewed at The University of Texas Southwestern Medical Center between 2000 and 2011 (representing 0.9% of all cases). All 4 cases included here represent the initial resection in patients with recently identified brain masses on MRI. All available clinical history, imaging results, laboratory results, operative reports, subsequent follow-up encounters, pathologic findings, and treatment regimen were reviewed for these 4 patients (Table 1). The study was performed in accordance with a protocol approved by the Institutional Review Board of The University of Texas Southwestern Medical Center (IRB STU 022011-081).

TABLE 1.

Clinical, Radiologic, Pathologic, Treatment, and Long-Term Outcome Data for Patients 1–4

Patient	Age at Diagnosis	Gender	Imaging	Diagnosis (2016 WHO)	Molecular Results at Time of Diagnosis	Postsurgical Treatment	Outcome
1	37	Male	2.3 × 1.9 × 1.3 cm right parietal lobe mass with nodular enhancement	GBM (WHO IV), IDH1 R132H mutant	TP53 mutation; ATRX wildtype	Temodar + radiation	Alive at 154 months
2	49	Male	7.5 × 5.4 × 4.5 cm partially cystic right frontal lobe mass with heterogeneously enhancing solid component	GBM (WHO IV), IDH1 R132H mutant	TP53 mutation; ATRX mutation	Temodar + radiation	Alive at134 months
3	60	Male	6.9 × 5.7 × 4.8 cm mass in the bilateral frontal lobes with ring-enhancement	GBM (WHO IV), IDH1 R132H mutant	TP53 mutation; ATRX mutation	Temodar + radiation	Alive at 99 months
4	42	Male	6.1 × 4.8 × 4.6 cm partially cystic right frontal lobe mass with heterogeneous enhancement	GBM (WHO IV), IDH1 R132H mutant	TP53 mutation; ATRX mutation	Temodar + radiation	Alive at 90 months

Immunohistochemistry

Four-µm thick sections of formalin-fixed paraffin-embedded (FFPE) tissue underwent heat-induced epitope retrieval using CC1 (Ventana, Tucson, AZ), followed by immunohistochemical staining with a monoclonal mouse antibody to Ki-67 (Dako, Carpinteria, CA), polyclonal rabbit antibody to ATRX (Sigma-Aldrich, St Louis, MO), monoclonal mouse antibody to p53 (Ventana), and monoclonal mouse antibody to IDH1 R132H (Dianova, Hamburg, Germany) on either a Ventana Benchmark XT or Ventana Benchmark Ultra automated stainer, using Ventana UltraView Universal DAB Detection kits.

IDH1 and IDH2 Sequence Analysis

Tumor DNA extracted from deparaffinized tissue sections (QIAamp DNA FFPE Tissue Kit, Qiagen) was PCR amplified using iPLEX reagents and primers specific for exon 4 of IDH1 and IDH2. Codon 132 of IDH1 and codons 140 and 172 of IDH2 were then analyzed for any mutations by matrix-assisted laser desorption/ionization time of flight (MALDI-TOF) mass spectrometry of primer extension products using a MassARRAY Analyzer 4 (Agena Biosciences, San Diego, CA). IDH1 and IDH2 mutation status was assessed by manual inspection of spectrograms.

DNA Extraction

DNA was extracted from FFPE tissue using the automated Maxwell system (Promega, Madison, WI). Areas with the highest available tumor content were chosen.

Genome-Wide Methylation and Copy Number Profiling

The Illumina EPIC Array 850 Bead-Chip (850k) array was used to determine the DNA methylation status of >850 000 CpG sites (Illumina, San Diego, CA) according to the manufacturer’s instructions. Methylation profiles were compared to a reference cohort of 2801 cases from 82 tumor entities previously profiled and analyzed at the German Cancer Research Center using a random forest algorithm and customized bioinformatics packages as described previously (15). In addition, the array data were used to calculate a low-resolution copy number profile, as previously described (11, 16–21). Copy number profiles for the tumors in this study were also compared to a reference cohort of GBM cases from the German Cancer Research Center (15). “Gain” or “amplification” was determined by log2 ≥ 0.3.

Copy Number Analysis of CCND2

To determine precise copy number values for key genes identified by copy number profiling, we analyzed the raw iDAT files using the conumee package (http://bioconductor.org/packages/conumee/) (22). Tumor purity was determined using InfiniumPurify (23) and tumor ploidy was determined using the ABSOLUTE algorithm (24, 25). Amplification was defined here as ≥ 4 gene copies/cell.

NGS Targeted Cancer Gene Panel

The clinically validated Ion Torrent Oncomine Focus Assay (Thermo Fisher Scientific, Inc., Waltham, MA) was used according to the manufacturer’s protocols to identify mutations, copy number variants, and RNA alterations across 52 genes relevant to solid tumors (Table 2). The result of a second NGS cancer gene panel consisting of 595 genes was available on one of the tumors (patient 2) (Tempus xT panel; Tempus, Inc., Chicago, IL).

TABLE 2.

Genes Included in the Targeted Sequencing Panel Performed on Patients 1–4

DNA Panel				RNA Panel
Hotspot Genes		Copy Number Variants		Fusion Drivers
AKT1	IDH2	ALK	FGFR3	ABL1	FGFR2
ALK	JAK1	AR	FGFR4	AKT3	FGFR3
AR	JAK2	BRAF	KIT	ALK	MET
BRAF	JAK3	CCND1	KRAS	AXL	NTRK1
CDK4	KIT	CDK4	MET	BRAF	NTRK2
CTNNB1	KRAS	CDK6	MYC	EGFR	NTRK3
DDR2	MAP2K1	EGFR	MYCN	ERBB2	PDGFRA
EGFR	MAP2K2	ERBB2	PDGFRA	ERG	PPARG
ERBB2	MET	FGFR1	PIK3CA	ETV1	RAF1
ERBB3	MTOR	FGFR2		ETV4	RET
ERBB4	NRAS			ETV5	ROS1
ESR1	PDGFRA			FGFR1
FGFR2	PIK3CA
FGFR3	RAF1
GNA11	RET
GNAQ	ROS1
HRAS	SMO
IDH1

TCGA Case Selection

We performed a search of 378 GBM cases in the The Cancer Genome Atlas (TCGA) database with survival data, mutation analysis, and copy number profiling analysis available using cBioPortal (26, 27) and identified 10 cases with CCND2 amplification (2.6%). Three of these cases had IDH1 R132H mutations and 7 were IDH1/2-wildtype. The mean age at diagnosis was 41 ± 9.2 years (range 31–49 years) and 65 ± 7.3 years (range 58–74 years), for IDH1-mutant and IDH1/2-wildtype cases, respectively.

We performed a second search of these cases to identify cases with comparably long survival. We identified 5 cases with recurrence-free survival ranging from 73 to 127 months and overall survival ranging from 82 to 127 months. The mean age at diagnosis was 38 ± 11.7 years (range 25–53 years). Unlike our cohort of cases, 4/5 of these cases were IDH1/2-wildtype and all 5 patients had recurrences and all eventually died due to their tumors (Supplemental Table 1).

RESULTS

Unexpectedly Long-Term Survival in a Small Cohort of GBM Patients

We identified 4 currently living, neurologically intact patients with the diagnosis of GBM (0.9% of total cases with a diagnosis of neuropathologically confirmed GBM at University of Texas Southwestern Medical Center) with recurrence-free survival after initial diagnosis and surgery ranging from 90 to 154 months (mean of 119 ± 15 months) (Table 1). The survival times for the patients in these cases are well beyond the typical 24- to 31-month survival documented in patients with IDH-mutant GBM (4, 6–8). These cases were initially diagnosed as GBM based on histologic features; all were infiltrating astrocytic tumors with palisading necrosis or microvascular proliferation. All 4 cases were negative for 1p/19q codeletion, and had IDH1 R132H mutations identified by both immunohistochemistry (Fig. 1) and Sequenom mass spectrometry assay (18). A high nuclear p53 immunolabeling index, suggestive of a TP53 mutation, was identified in tumor tissue from all 4 patients. Loss of nuclear ATRX staining in tumor cells, indicative of ATRX mutations, was identified in the tumors from patients 2, 3, and 4 by immunohistochemistry (Fig. 1; Table 1). TP53 and ATRX mutations were separately confirmed in patient 2 by an NGS panel.

FIGURE 1.

H&E, IDH1 R132H, p53, ATRX, and Ki-67 immunohistochemical images demonstrating the defining histologic features of GBM: focal necrosis (patients 1 and 2) and microvascular proliferation (patients 3 and 4), IDH1 R132H immunopositivity in all patients, TP53 mutations in all patients, ATRX in 3/4 patients (patients 2–4), and varying Ki-67 proliferation indices. H&E images for patients 1 and 2 are taken at a total magnification of 40×, all other images are taken at a total magnification of 100×. Scale bars = 100 µm in all images.

Methylation Profiling

Genome-wide methylation profiles obtained using the Illumina EPIC 850k Array were compared online to a reference cohort of 2801 cases from 82 tumor entities using a DNA methylation-based classification of human brain tumors (http://www.molecularneuropathology.org) with automated generation of methylation classification as described by Capper et al (15). The overall methylation profiles for all 4 cases matched to either “high-grade IDH-mutant glioma” or “IDH-mutant GBM.” A fifth additional IDH-wildtype case was tested with methylation profiling; however, the brain tumor methylation classifier results most closely matched to anaplastic pleomorphic xanthoastrocytoma (PXA), and this case was excluded from this cohort due to the discrepancy between original histologic diagnosis and methylation profile (15). A review of the histopathology prompted by the methylation profile suggested that the tumor could plausibly be reclassified as an anaplastic PXA. In contrast to previous studies which showed high rates of MGMT promoter methylation among GBM patients with survival >4 years (5, 28), the MGMT gene promoter was methylated in one only case (patient 3) and unmethylated in one case (patient 1). MGMT gene promoter was partially methylated in the remaining 2 cases.

CNA Analysis

Chromosomal abnormalities consistent with previously described IDH-mutant GBM were identified in all of the tumors included in this study (Fig. 2, Table 3). The overall level of CNA across the genome of all patients was high in all cases included in this study (Fig. 2).

FIGURE 2.

Copy number analysis derived from the Illumina EPIC 850k methylation array data demonstrating widespread genomic alterations in GBM cases with long-term survival, significant amplifications of CCND2 in 3 of 4 cases, and characteristic GBM alterations, including large-scale chromosomal gains and losses, and focal amplifications of CDK4 and deletions of CDKN2A/B.

TABLE 3.

Summary of CNAs for Patients 1–4, All Genes and Chromosomal Regions Meet the Criteria of Log2 ≥ ± 0.3

Patient	Specific Gene Amplifications	Specific Gene Deletions	Whole/Arm Chromosome Gain	Whole/Arm Chromosome Loss
1	CCND2	PTEN	20	5
	CDK4	MGMT	Xq	10
	CDKN2A/B	RB1		12p
	MDM2	SMARCB1		13q
	MYBL1			18
	MYC			Xp
	NF1			Yp
2	CCND2	MDM2	12p	11p
3	CDK4	C19MC	3q (partial)	2q (partial)
		MDM2		6q
		MGMT		9p
		MYB		10q (partial)
		PPM1D		18q (partial)
		TERT		19
4	CCND2	CCND2A/B	12p	7p (partial)
		EGFR		9p
		MDM2		10q (partial)
		MGMT		11p
		PPM1D		12q (partial)
				20p

When compared with a reference cohort of GBMs (15), patients 1, 2, and 4 all had amplifications involving CCND2 above the 0.3 log2 copy number cutoff level. Based on analysis of copy number profiling, case 1 contained 12 copies of CCND2 per cell on average, case 2 contained 6 copies per cell, case 3 contained 3 copies per cell, and case 4 contained 6 copies per cell. Confirmation of the CCND2 gain by a second method was available in one case (patient 2), where a gain involving CCND2 (7 copies/cell) was confirmed by a 595-cancer gene NGS panel (xT Panel, Tempus, Inc.). Other focal CNAs identified included high-level amplifications of CDK4 (patients 1 and 3) and deletion of CDKN2A/B (patient 4) (Fig. 2, Table 3).

Oncomine Targeted Cancer Gene Panel

Additional DNA and RNA sequencing was performed by a targeted gene panel. All 4 cases had well preserved DNA for analysis; however, only case 3 had preserved RNA. RNA analysis showed no detectable fusions in known oncogene drivers. DNA analysis showed a high-level amplification of CDK4 in patient 1 (19 copies/cell; 95% CI: 15.0–23.6) and patient 3 (21 copies/cell; 95% CI: 18.1–25.2), supporting the initial copy number profile results. High-level amplifications in CDK4 in these 2 tumors with long recurrence-free survival is a surprising result, as CDK4 gene amplification has been associated with aggressive behavior and poor prognosis in lower-grade astrocytomas (29–31).

TCGA Analysis

CCND2 amplification provided no survival advantage in GBM cases in the TCGA data set with IDH1/2-wildtype status (n = 7). The median recurrence-free survival was 4 months (range 3–11 months) and median overall survival was 7 months (range 3–16 months) in this patient group with a single patient still alive and free of tumor recurrence at the date of last follow-up (9 months after initial diagnosis). The small sample size and follow-up interval prevented any conclusions to be made from the 3 patients with both IDH1 R132H mutations and CCND2 amplification in the TCGA dataset. One patient had GBM recurrence at 8 months and died at 19 months, and the other 2 were still alive and disease-free at the date of last follow-up, although this was limited in these 2 cases to 7 and 31 months.

None of the 5 GBM TCGA cases with comparatively long survival (recurrence-free survival ranging from 73 to 127 months and overall survival ranging from 82 to 127 months) had CCND2 amplification and surprisingly only a single case had an IDH1 mutation, despite the relative survival advantage usually associated with mutations in IDH1/2. All 5 of these cases had eventual tumor recurrence and subsequently died of their disease, unlike the 4 patients in our institutional cohort. Other molecular anomalies associated with GBM that were found in this subset were TP53 mutation (3/5 cases), CDKN2A mutation (1/5 cases), BRCA2 mutation (1/5 cases), CDKN2A deletion (2/5 cases), CDK4 amplification (1/5 cases), MYCN amplification (1/5 cases), PDGFRA amplification (1/5 cases), KIT amplification (1/5 cases), KDR amplification (1/5 cases), and FGFR2 amplification (1/5 cases).

These results suggest that CCND2 amplification is a rare event in GBM (2.6% in the TCGA dataset) and does not seem to confer a survival advantage in IDH1/2-wildtype tumors. The limitations in patient follow-up and low case numbers prevent any conclusions from being made about the prognostic implications of CCND2 amplification in IDH1/2-mutant GBMs from this data.

DISCUSSION

Although there are rare reports of GBM patients surviving unexpectedly long periods of time (9, 10), patients with these tumors rarely survive past 5 years (4, 5), and most studies define “long-term survival” for GBM patients as survival of 3 years or more (12, 32–35). In our study, we evaluated the genetic and epigenetic features of a group of GBM patients with unusually long progression- and recurrence-free survival. First, we correlated the histologic diagnoses with unbiased molecular classification using the methylation-based classifier (15), as inclusion of PXA or other histologically unusual tumors in cohorts of GBM long-term survivors can hamper identification of biomarkers relevant to GBM. A fifth patient with a tumor originally diagnosed as IDH1/2-wildtype GBM but reclassified as anaplastic PXA following the methylation array result and subsequent re-examination of the relevant histology was excluded from this series on this basis. This series, comprising 4 IDH1 R132H-mutant GBMs, had recurrence-free survival of 90–154 months as of this writing; all 4 patients are still alive and neurologically intact (Table 1). Analysis of methylation patterns in these patients matched the histologic diagnoses in these 4 patients. Analysis of overall copy number variation across the genome revealed large-scale amplification and deletion, gains and losses of chromosomes or chromosomal regions, and focal amplification and deletion of specific genes (Table 3) characteristic of high-grade glioma cases (11–13, 18, 36–39). In addition, we found high-level CCND2 amplification in 3 of 4 patients (6–12 copies of CCND2 per tumor cell) (Fig. 2).

Since copy number profiling has become more commonplace as a research tool, studies have shown that CNA is reflective of and can be predictive of glioma grade and prognosis (11, 18, 40). It has been demonstrated that there is a significantly higher level of CNA in IDH1/2-mutant GBM compared with IDH-mutant lower-grade astrocytomas, but this difference is not present when comparing wildtype astrocytomas in these WHO grade categories. These differences in CNA patterns among IDH-mutant and wildtype classes are reflective of the prognostic differences seen in large epidemiological studies of these tumors (11). Recently, we have shown that an increased level of CNA is seen in lower-grade IDH1/2-mutant gliomas before transformation into GBM in rapidly progressing and rapidly fatal cases and we hypothesized that these early CNA patterns could be predictive of this significantly worse outcome in this subset (18, 40). There is still debate over the clinical impact of high levels of chromosomal instability. Increased heterogeneity of tumor clones could potentially result in treatment resistant tumor cell clones, although there is an upper limit to large-scale DNA disruptions that can occur without compromising tumor cell viability, and further genome instability could make tumors more responsive to conventional therapy (12, 41).

The significance of amplification involving CCND2 in patients with extremely long survival currently remains unclear. CCND2 encodes cyclin D2, a member of a family of proteins integral to cell cycle progression. Cyclin D2 has been shown to interact with CDK2, CDK4, and CDK6 with similar effects to cyclin D1: phosphorylation of RB and release of transcription factors, moving the cell cycle from G1- to S-phase, allowing for cell replication (42, 43). CCND2 amplification is relatively uncommon in gliomas in general, occurring in 3%–11% of GBMs (13, 14) and 23%–24% of all WHO grades II and III IDH-mutant astrocytomas, but is much more common in gemistocytic subtypes of IDH-mutant astrocytoma (87%–89%) (14). Our finding that CCND2 gene amplification may be favorable feature is unexpected in light of the fact that most previous in vitro and in vivo studies have focused on cyclin D2 as a mediator of cell proliferation and tumor growth (44–46).

Other studies have found that pep5, a small peptide derived from cyclin D2, when added to breast cancer cell lines with low intrinsic cyclin D2 levels, can interact with cytoskeleton and proteosome proteins, causing apoptosis in cells arrested in G1/S or in S-phase (47). It was also demonstrated that pep5 can reduce tumor volume by ∼50% when added to a rat GBM model (48). Additionally, the loss of cyclin D2 mRNA and protein expression due to promoter hypermethylation was shown to be an early event in the evolution of primary human ductal breast carcinoma (49), raising the possibility that cyclin D2 may act as a tumor suppressor rather than tumor promoter in some cancers. Sahm et al (14) further hypothesized that increased levels of cyclin D2 may halt the S/G2 transition in GBM cells.

The methylation and CNA analysis used in this study is currently not used for diagnostic purposes; however, recent studies have demonstrated that the analysis used here to identify these alterations can be performed relatively inexpensively and rapidly enough to be of clinical utility (50). As a result, information regarding overall methylome and methylation status of key genes, CNAs, and specific mutations may soon be available at the time of initial tumor diagnosis by neuropathologists. Studies identifying genetic and epigenetic prognostic factors could therefore become increasingly important for real time diagnosis and guidance of personalized treatment options for patients with a wide range of cancers.

This study presents a rare subset of GBM patients with excellent clinical outcomes despite the dismal prognosis classically associated with this malignancy. We used methylation and copy number profiling to evaluate the genetic and epigenetic alterations associated with this tumor group and demonstrate that although there is no difference in genome-wide CNA levels between these cases and IDH-mutant GBMs with conventional clinical outcomes (15), 3/4 of these cases show focal amplification of CCND2. Although found in a small number of cases, this genetic alteration appears to be relatively uniform among these 4 patients, suggesting that these findings may be generalizable to a larger population of GBM patients with extremely favorable clinical outcomes, although evaluation of the TCGA GBM data sets indicates that this survival benefit may only include cases with mutations in IDH1 or IDH2. Further studies are needed to confirm our results in a larger series of GBM patients with unexpectedly long survival.

This study was supported by the Friedberg Charitable Foundation grant to Matija Snuderl.

The authors have no duality or conflicts of interest to declare.

Supplementary Data can be found at academic.oup.com/jnen.