The Somatic Genomic Landscape of Glioblastoma

Authors

Cameron W. Brennan

Roel G.W. Verhaak

Aaron McKenna

Benito Campos

Houtan Noushmehr

Sofie R. Salama

Siyuan Zheng

Lynda Chin

TCGA Research Network

Doi

10.1016/j.cell.2013.09.034

PMID: 24120142 · DOI: 10.1016/j.cell.2013.09.034 · Journal: Cell (2013)

TL;DR

TCGA’s 2013 update on glioblastoma assembled multi-platform genomic, transcriptomic, epigenomic and proteomic data for 543 primary GBM patients — including 291 paired whole-exome sequencing (WES), 42 deep whole-genome sequencing (WGS) pairs, 164 RNA-seq transcriptomes, expanded DNA methylation, and 214 reverse-phase protein array (RPPA) samples. MutSig and InVEx jointly identified 71 significantly mutated genes including the novel driver LZTR1, while WGS/RNA-seq cataloged complex EGFR rearrangements (EGFRvIII in 11% at high TAF; novel C-terminal deletions and Δ12–13 variants) and recurrent double-minute amplifications of MDM2/CDK4. TERT promoter mutations (C228T/C250T) were detected in 21/25 deep-coverage cases and correlated with elevated TERT mRNA; non-mutated TERT samples all carried ATRX mutations (ALT pathway). Survival advantage of the proneural transcriptomic subtype was attributable to the G-CIMP phenotype, and MGMT promoter methylation predicted treatment response specifically in the classical subtype. Proteomic data showed that downstream pathway activation is non-linear with respect to upstream mutations, challenging the strategy of pathway-level (versus target-level) therapeutic inhibition.

Cohort & data

  • Cohort: 543 primary glioblastoma patients (GBM) accrued from 17 TCGA contributing sites, diagnosed 1989–2011. Median age 59.6 yrs; M:F = 1.6 (333:209). Median OS 13.9 months; 2-year survival 22.5%; 5-year 5.3%. IDH1 mutation in 6% (28/423 with adequate coverage). Combined TMZ + radiation documented for 40% (217/543).
  • Dataset: gbm_tcga_pub2013 — TCGA GBM 2013 publication freeze (data frozen 2013-07-15).
  • Assays:
  • Significance algorithms: MutSig and InVEx; mutual-exclusivity testing with MEMo; structural variant callers BreakDancer and BamBam.

Key findings

  • Mutation rate. 21,540 somatic mutations across 291 exomes; median 2.2 coding mutations/Mb (IQR 1.8–2.3). 20,448 SNVs, 39 dinucleotide, 1,153 indels. Validation by orthogonal targeted re-sequencing in 259 pairs validated 98% of SNVs, 84% of insertions, 82% of deletions.
  • 71 significantly mutated genes. MutSig + InVEx jointly identified PTEN, TP53, EGFR, PIK3CA, PIK3R1, NF1, RB1, IDH1, PDGFRA, and a novel driver LZTR1 (mutated in 10 samples; hemizygous deletion at 22q in 5/6 with copy number data). SPTA1 (9%), ATRX (6%), GABRA6 (4%), KEL (5%) were among the additional 61 MutSig-only genes (q<0.1).
  • Hotspots. IDH1 R132H in 28 cases; R132G and R132C each in 1; no IDH2 mutations. BRAF V600E in 5/291 (1.7%). Histone H3.3 (H3-3A) mutations were absent in this primary-GBM cohort.
  • Chromatin modifier genes. 135/291 samples (46%) had ≥1 non-synonymous mutation across a curated 161-gene chromatin modification gene (CMG) set; MEMo found CMG mutations overall mutually exclusive (p=0.0008).
  • Copy-number landscape. Recurrent amplifications on chr7 (EGFR/MET/CDK6), chr12 (CDK4, MDM2) and chr4 (PDGFRA) at higher frequencies than the 2008 freeze; new precision on SOX2, MYCN, CCND1, CCNE2 gains. CDKN2A/CDKN2B homozygous deletion remains the dominant deletion. QKI defined as the sole gene in the minimal 6q26 deletion region (homozygous deletion in 9 cases; mutated in 5 additional cases without deletion). Other single-gene deletion targets: LRP1B, NPAS3, LSAMP, SMYD3.
  • Structural rearrangements (WGS, n=42). 238 high-confidence somatic rearrangements (49 interchromosomal, 125 intrachromosomal, 64 intragenic); median 2/sample, range 0–32. One sample exhibited chromothripsis spanning a 7.5 Mb chr1 region. Recurrent intragenic events: EGFR (n=12), CPM, PRIM2 (n=3 each), FAM65B, PPM1H, RBM25, HOMER2 (n=2 each).
  • RNA-seq fusions. 228 fusion transcripts in 106/164 samples (48 interchromosomal, 180 intrachromosomal). FGFR3–TACC3 inversion detected in 2 cases; focal FGFR3/TACC3 amplification in 14/537 (2.6%) copy-number profiles. 7p11 EGFR-locus fusions included EGFR–SEPT14 (n=6), SEC61G–EGFR (n=4), LANCL2–SEPT14 (n=1), COBL–SEPT14 (n=1).
  • EGFR alteration spectrum. 38.4% of cases harbor an EGFR genomic rearrangement or expressed point mutation at ≥10% TAF; 57% have some EGFR alteration (mutation, rearrangement, splicing, or focal amplification). EGFRvIII (exon 1–8 aberrant junction) ≥10% TAF in 11%, ≥1% TAF in 19%. C-terminal deletion variants in 3.7% at ≥10% TAF (plus 9% at ≥1%). Novel recurrent EGFR Δ12–13 in 28.7% and Δ14–15 in 3% — mostly at low TAF and likely splicing rather than DNA deletion (14/15 lacked concordant DNA breakpoints).
  • PDGFRA variants. No recurrent fusions, but intragenic Δ8,9 splice variant at low TAF in 29/163 RNA-seq samples; a novel PDGFRA Δ2–7 variant in 1 sample (TCGA-28-5216).
  • Pathway integration (251 GBM with WES+CNA). RTK alteration in 67.3% (EGFR 57.4%, PDGFRA 13.1%, MET 1.6%, FGFR2/3 3.2%). PI3K mutations in 25.1% (mutually exclusive with PTEN, p=0.0047); together 89.6% of GBM had ≥1 PI3K-pathway alteration. p53 pathway dysregulated in 85.3% (TP53 27.9%, MDM1/2/4 amplification 15.1%, CDKN2A deletion 57.8%); Rb pathway in 78.9%. NF1 deletion/mutation in 10%, never co-occurring with BRAF mutation.
  • TERT promoter. C228T in 15/25 deep-coverage WGS cases, C250T in 6/25 (84% combined). TERT promoter mutations correlated with elevated TERT mRNA. All 4 non-TERT-mutant cases harbored ATRX mutations (concurrent with IDH1 + TP53); ATRX-mutant tumors did not show elevated TERT RNA — consistent with the ALT pathway.
  • Subtypes & survival. Proneural-subtype survival advantage was confined to G-CIMP+ tumors; non-G-CIMP proneural and mesenchymal tumors had less favorable early outcomes (p=0.07). Methylation re-analysis of 396 cases produced 6 classes (M1–M6 + G-CIMP); M1 enriched for mesenchymal (60%), M3 for classical (58%), M6 for non-mutant-IDH1 proneural with PDGFRA amplification. Enrichments by subtype: ATRX mutations and MYC amplifications in G-CIMP+; CDK4 and SOX2 amplifications in proneural; broad chr19/chr20 amplifications in classical; NF1 inactivation in mesenchymal.
  • MGMT methylation. Methylated in 174/359 (48.5%); 79% in G-CIMP versus 46% in non-G-CIMP. MGMT status distinguished responders from non-responders only in the classical subtype (n=96; p=0.01), and not in proneural (n=66; p=0.57), mesenchymal (n=104; p=0.62), or neural (n=55; p=0.12).
  • Proteomics (RPPA). 127/171 antibodies correlated significantly with transcriptomic subtype (Kruskal-Wallis p<0.05). EGFR amplification/mutation associated with elevated total EGFR (p=3.74e-15) and phospho-EGFR (p=1.44e-12); classical tumors showed concomitantly reduced pro-apoptotic markers. Mesenchymal: elevated CD31, VEGFR-2, fibronectin, COX-2, phospho-Raf/MEK/ERK. Proneural: elevated PI3K/Akt-mTorc1 activation. Notably, RTK-amplified samples had lower downstream phospho-AKT/S6K/MAPK signaling, and PI3K-mutant samples had lower AKT activity than PI3K-WT — non-linear genotype-to-signaling relationships.
  • Long-term survivors (>3 yrs, n=39; 7.7%). No specific genomic alteration over-represented; CDK4/EGFR amplification and CDKN2A deletion were under-represented. 79% of long-term survivors diagnosed at <50 yrs. Only one-third of G-CIMP+ patients survived beyond three years.

Genes & alterations

  • EGFR — amplification, point mutation (extracellular and kinase domain), EGFRvIII (exons 2–7 deletion), C-terminal deletion variants, novel Δ12–13 (28.7%) and Δ14–15 splice variants, focal fusions (EGFR–SEPT14, SEC61G–EGFR, etc.). 57% of GBM altered overall; 38.4% with rearrangement or expressed mutation at ≥10% TAF.
  • PDGFRA — focal amplification in 13.1%; intragenic Δ8,9 splice variant; novel Δ2–7; concurrent with EGFR alteration in 42.4% of PDGFRA-altered cases.
  • LZTR1 — novel significantly mutated GBM gene, putative transcriptional regulator at 22q; hemizygous deletion in 5/6 samples with CNA data.
  • QKI — sole gene in 6q26 minimal deletion region; homozygous deletion (n=9) and intragenic mutation (n=5); supports tumor-suppressor role.
  • TERT — promoter C228T (15/25) and C250T (6/25) hotspots correlate with elevated TERT mRNA.
  • ATRX — mutations in 6% overall; enriched in G-CIMP+ and in TERT-promoter-wild-type tumors; consistent with ALT pathway.
  • IDH1 — R132H/G/C mutations in 6% of cases; concurrent with TP53 in 12/13 and ATRX in many; all cases G-CIMP+ (plus 7 G-CIMP+ cases without IDH1 mutation).
  • TP53 — mutation/deletion in 27.9% of GBM; mutually exclusive with MDM amplification (p=0.0003) and CDKN2A deletion (p=1.99e-7).
  • NF1 — deletion/mutation in 10%; enriched in mesenchymal subtype; never co-occurring with BRAF mutation; mutation/deletion associated with elevated p-ERK/p-MEK.
  • PTEN — mutation/deletion mutually exclusive with PI3K mutations (p=0.0047); combined alterations in 59.4% of GBM.
  • PIK3CA / PIK3R1 — PI3K-family mutations in 25.1%; 18.3% in p110α/p85α subunits.
  • BRAF — V600E in 5/291 (1.7%); never co-occurring with NF1 alteration.
  • CDKN2A — homozygous deletion in 57.8% of pathway-analysis cohort.
  • CDK4 / CDK6 / MDM2 / MDM4 — recurrent focal amplifications; chr12 double-minute amplification of MDM2/CDK4 was a striking structural finding.
  • RB1 — direct mutation/deletion in 7.6%; Rb-pathway dysregulation overall in 78.9%.
  • FGFR3 / TACC3 — recurrent in-frame inversion fusion in 2 cases, with focal co-amplification in both; 2.6% of CNA profiles show focal FGFR3/TACC3 amplification.
  • SOX2, MYC, MYCN, CCND1, CCND2, CCNE2 — recurrent focal amplifications; subtype-correlated.
  • LRP1B, NPAS3, LSAMP, SMYD3 — single-gene focal deletion targets.
  • SPTA1, GABRA6, KEL — additional MutSig SMGs.
  • MGMT — promoter DNA methylation in 48.5% of cohort; predictive of TMZ response only in classical subtype.

Clinical implications

  • MGMT methylation as a subtype-specific predictive biomarker. Methylated MGMT distinguishes responders from non-responders to temozolomide + radiation in the classical GBM subtype (p=0.01) but not in proneural, mesenchymal, or neural subtypes. This refines the canonical Hegi finding and argues for stratifying biomarker interpretation by transcriptomic subtype.
  • G-CIMP drives proneural survival advantage. Survival benefit historically attributed to “proneural” GBM is in fact attributable to G-CIMP+ epigenotype; non-G-CIMP proneural tumors behave more like other subtypes. Implication: subtype-based prognostication should use methylation, not just transcriptome.
  • EGFR therapeutic targeting is complicated by intratumoral heterogeneity. Diverse co-existing EGFR alteration forms (EGFRvIII, C-terminal deletions, Δ12–13, point mutations) in the same tumor may yield differential responses to any single targeted inhibitor; strategies must address heterogeneity.
  • Pathway-level inhibition ≠ target-level inhibition. RPPA data show that downstream pathway activation is non-linear with respect to genomic alterations (e.g. PI3K-mutant samples show lower p-AKT than PI3K-wild-type). Inhibiting a pathway component is therefore not a guaranteed substitute for inhibiting the mutated target itself.
  • Actionable hotspots present in a minority. BRAF V600E (1.7%) in GBM provides a rationale for vemurafenib-class therapy in selected patients; FGFR3–TACC3 fusion (also rare) is a candidate for FGFR inhibitor trials.
  • TERT/ATRX dichotomy as a binary telomere-maintenance axis. Nearly every primary GBM activates telomere maintenance — TERT promoter mutation (telomerase reactivation) or ATRX loss (ALT) — supporting telomere maintenance as a putatively obligate step in GBM pathogenesis.
  • Treatment data in the cohort. 40% of all patients (217/543) and 50.2% of those diagnosed ≥2002 received concurrent temozolomide + adjuvant radiation. Bevacizumab is noted in the paper’s clinical-context discussion of GBM but is not the focus of biomarker analysis here.

Limitations & open questions

  • Static single-time-point snapshot. TCGA samples are pre-treatment biopsies; the paper cannot map signaling/genomic states at recurrence or under therapy. The non-linear genotype-to-protein relationships argue against extrapolating static states to dynamic drug response.
  • Bulk-tissue analysis obscures intratumoral heterogeneity. Co-existing EGFR variants and patterns of intratumoral RTK heterogeneity (EGFR/PDGFRA/MET) cannot be resolved by whole-tumor profiling; single-cell or in-situ approaches are needed.
  • Underestimation of EGFR C-terminal deletions. Complete C-terminal loss yields aberrant junctions not mappable by transcriptome sequencing; reported prevalence is a lower bound.
  • Long-term-survivor signal is unexplained. Beyond age <50 and G-CIMP status (which itself only confers ~one-third 3-yr survival), no specific genomic alteration distinguishes the 7.7% long-term-survivor subset.
  • Seven G-CIMP+ cases lack IDH1/IDH2 mutation. Mechanism of hypermethylator phenotype in IDH-wild-type G-CIMP remains unresolved.
  • ceRNA hypothesis is correlative. Predicted miR/ceRNA networks for PDGFRA, EGFR, NF1, PTEN are statistically suggestive but untested functionally.
  • WGS subset is small (n=42). Structural variant frequency estimates and double-minute findings are derived from a fraction of the cohort; generalization should be cautious.
  • Selection bias toward primary GBM. TCGA preferentially accrued primary GBM, so IDH1 mutation prevalence (6%) is lower than other glioma series; conclusions about secondary GBM are limited.

Citations from this paper used in the wiki

  • “We describe the landscape of somatic genomic alterations based on multi-dimensional and comprehensive characterization of more than 500 glioblastoma tumors (GBMs).” (Abstract)
  • “Overall, of the 291 tumor exomes sequenced, 21,540 somatic mutations were identified, with a median rate of 2.2 coding mutations per megabase…” (Results, p.3)
  • “When both InVEx and MutSig algorithms were run on the same dataset, a total of 71 genes were identified as significantly mutated genes (SMG).” (Results, p.3)
  • “MutSig additionally identified 61 additional genes (71 overall) with mutation frequency above background with a q-value of < 0.1… LZTR1, a putative transcriptional regulator associated with the DiGeorge congenital developmental syndrome… has not previously been implicated in cancer. It is located at chromosome 22q, and in five of six samples with available copy number data it was simultaneously targeted by hemizygous deletion.” (Results, p.3)
  • “135 samples or 46% of the sample cohort harbored at least one non-synonymous mutation in this CMG gene set… CMG mutations were found to be mutually exclusive overall by MEMo analysis (p=0.0008).” (Results, p.4)
  • “QKI as the sole gene within the minimal common region and the target of homozygous deletion in 9 cases. The QKI gene was also mutated in 5 cases without evidence of deletion.” (Results, p.4)
  • “In 11% of tumors, the aberrant exon 1–8 junction characteristic of EGFRvIII was highly expressed (≥10% TAF), while 19% showed at least a low level expression (≥1%).” (Results, p.6)
  • “In total, 38.4% of cases harbored an EGFR genomic rearrangement or a point mutation expressed in at least 10% of transcripts… Overall, 57% of GBM showed evidence of mutation, rearrangement, altered splicing and/or focal amplification of EGFR.” (Results, p.6)
  • “TERT promoter mutations at these two hot spots were correlated with up-regulated TERT expression at the RNA level… the four GBMs with non-mutated TERT promoters all harbored ATRX mutations.” (Results, p.7)
  • “MGMT status distinguished responders from non-responders amongst samples classified as classical (n = 96; p = 0.01) but not among samples classified as proneural (n = 66; p = 0.57), mesenchymal (n = 104; p = 0.62) and neural (n = 55; p = 0.12).” (Results, p.9)
  • “samples with RTK amplification had lower levels of canonical RTK-target pathway activities as measured by phospho-AKT, phospho-S6 kinase and phospho-MAPK co-cluster levels… PI3K-mutant samples had lower AKT activity than samples lacking PI3K mutations.” (Results, p.11)
  • “this observation directly challenges the notion that therapeutic inhibition of downstream signaling components along a pathway would yield similar efficacy of targeting the mutated gene itself.” (Discussion, p.12)

This page was processed by crosslinker on 2026-05-09.