Molecular Profiling Reveals Biologically Discrete Subsets and Pathways of Progression in Diffuse Glioma

Authors

Michele Ceccarelli

Floris P. Barthel

Tathiane M. Malta

Thais S. Sabedot

Sofie R. Salama

Bradley A. Murray

Olena Morozova

Yulia Newton

Amie Radenbaugh

Stefano M. Pagnotta

Houtan Noushmehr

Antonio Iavarone

Roel G.W. Verhaak

Doi

10.1016/j.cell.2015.12.028

PMID: 26824661 · DOI: 10.1016/j.cell.2015.12.028 · Journal: Cell (2016)

TL;DR

Ceccarelli et al. analyze 1,122 newly diagnosed adult diffuse gliomas (WHO grades II–IV) from TCGA — combining exome, whole-genome, RNA-seq, DNA copy-number, DNA methylation, and RPPA data — to redefine glioma classification on molecular grounds. They identify 75 significantly mutated genes (45 of them novel for glioma), nominate the cohesin complex (e.g., NIPBL, STAG2) as a recurrently disrupted pathway, and show that ATRX (not TERT promoter) mutations drive telomere lengthening, consistent with an alternative-lengthening-of-telomeres (ALT) mechanism. Pan-glioma DNA methylation clustering yields six subgroups (LGm1–6) that recapitulate IDH/1p-19q-based classification and reveal two clinically important new entities: a G-CIMP-low subset of IDH-mutant non-codel glioma with cell-cycle gene activation and poor outcome, and a “PA-like” IDH-wildtype subset of histologically diffuse glioma whose methylation, MAPK-pathway alteration spectrum, and silent copy-number landscape resemble pilocytic astrocytoma and confer favorable survival. Methylation subtype is an independent prognostic predictor on top of age and grade.

Cohort & data

  • 1,122 patients with newly diagnosed adult diffuse glioma from TCGA (516 LGG + 606 GBM); 590 grade IV (56%), 241 grade III (23%), 216 grade II (21%); histology: 590 GBM (56%), 174 oligodendroglioma (17%), 169 astrocytoma (16%), 114 oligoastrocytoma (11%). Clinical data available for 1,046/1,122 (93%) cases.
  • Cancer types: DIFG (umbrella), GBM, ASTR, ODG; PAST used as molecular comparator.
  • Datasets: primary cohort = lgggbm_tcga_pub (the merged TCGA pan-glioma study); supersedes prior independent analyses gbm_tcga and lgg_tcga. Validation cohort: 324 adult and pediatric gliomas from Lambert 2013, Mur 2013, Sturm 2012, and Turcan 2012; 221 predicted IDH-wildtype glioma (incl. 61 grade I PA) for IDH-wt validation.
  • Assay coverage: gene expression n=1,045; DNA copy number n=1,084; DNA methylation n=932 (hm27-methylation-array for 287 GBM + hm450-methylation-array for 516 LGG and 129 GBM); exome sequencing n=820 (513 LGG + 307 GBM); RPPA n=473; WGS available for telomere-length and TERT-promoter analyses. Mutation calling used mutect, indelocator, varscan, and radia (≥2-caller consensus); SMGs by mutsig (MutSigCV); copy number by gistic (GISTIC2); fusions by prada and defuse; integrative pathway inference by paradigm; co-clustering by tumor-map; stromal/immune scoring by estimate.

Key findings

  • 75 significantly mutated genes (SMGs), of which 45 had not been previously associated with glioma; mutation frequencies for novel SMGs ranged from 0.5% to 2.6% (PMID:26824661, Table S2A).
  • Recurrent driver counts (across the integrated cohort): IDH1 n=457, TP53 n=328, EGFR n=314, ATRX n=220, PTEN n=168, CIC n=80, FUBP1 n=45.
  • Newly nominated glioma drivers: SETD2 (n=24), ARID2 (n=20), DNMT3A (n=11), and the KRAS / NRAS oncogenes (n=25 and n=5 respectively).
  • Pathway convergence (mutation + CNA + fusion across n=793 mutation and n=649 fusion profiles): Ras-Raf-MEK-ERK altered in 73% (578 cases; 92% of IDH-wildtype, 327/357), and chromatin-modification genes altered in 54% (423 tumors; 87% IDH-mutant non-codel, n=230).
  • Cohesin pathway is disrupted in 16% of LGG/GBM, including NIPBL (cohesin loader) by mutation/CNA and STAG2 mutations. Authors note cohesin-mutant tumors may be sensitive to DNA-damage agents and PARP inhibitors based on prior literature.
  • TERT promoter (TERTp) mutation status (n=287 targeted + n=42 inferred from WGS): 85% of diffuse gliomas carry either TERTp mutation (157, 48%) or ATRX mutation (120, 37%); the two are nearly mutually exclusive. TERT mRNA is significantly upregulated in TERTp-mutant samples (p<0.0001); TERT expression by RNA-seq is a 91% sensitive / 95% specific surrogate for TERTp mutation.
  • TERTp mutation tracks with chr7 gain / chr10 loss in IDH-wildtype glioma: 52/53 chr7-gained and 134/147 chr10-lost cases harbor TERTp mutations, vs only ~45% of IDH-wildtype samples lacking these chr7/10 events. Authors infer TERTp mutation may precede the chr7/10 events implicated in glioma initiation.
  • ATRX, not TERTp, drives telomere length. In 141 matched tumor-normal pairs analyzed for telomere length, ATRX-mutant gliomas had significantly longer telomeres than TERTp-mutant samples (t-test p<0.0001), supporting an ALT mechanism. Among TERTp-mutant gliomas, IDH co-mutation status did not affect telomere length.
  • Novel promoter-region somatic events. Beyond TERT (n=37), promoter mutations in TRIM28 (n=8) correlated with TRIM28 upregulation, and CACNG6 (n=7) promoter mutations with CACNG6 downregulation. TRIM28 promoter activation is mechanistically linked (via prior work by Pineda et al. 2015) to ubiquitin-mediated AMPK degradation and mTOR activation, hypersensitizing cells to AMPK agonists such as metformin.
  • Pan-glioma DNA methylation clustering of 932 samples on a merged 27K + 450K probe set (1,300 tumor-specific CpG probes) yields six clusters LGm1–6. LGm1/2/3 (IDH-mutant, 449/450 = 99%, enriched for LGG 421/454 = 93%) show genome-wide hypermethylation; LGm4/5/6 (IDH-wildtype, 429/430 = 99%, GBM-enriched 383/478 = 80%) do not. RNA-seq clustering of 667 samples yields four expression clusters LGr1–4, with LGr4 exclusively IDH-wildtype (376/387 = 97%).
  • Three IDH-mutant subtypes: Codel (1p/19q codeleted), G-CIMP-high, and G-CIMP-low — a newly defined non-codel subgroup with relatively reduced genome-wide methylation. G-CIMP-low contains both GBM (13/25) and LGG (12/25) and is associated with significantly worse survival than G-CIMP-high or Codel groups (Figure 3C, Figure S3D); G-CIMP-high outcomes match Codel (the historically best-prognosis group).
  • Cell-cycle activation defines G-CIMP-low. 15/18 G-CIMP-low cases carry cell-cycle pathway abnormalities (e.g., CDK4, CDKN2A) vs 36/241 G-CIMP-high and 2/172 Codels. Differential methylation between G-CIMP-low and G-CIMP-high is enriched in intergenic “open sea” regions (69% vs expected; 2.5-fold enrichment, χ² p<2.2×10⁻¹⁶) and depleted at CpG islands (3.4-fold). De novo motif analysis identifies a TGTT motif associated with SOX2 / OLIG2 families; SOX2 and 17/20 SOX-family members are upregulated in G-CIMP-low.
  • G-CIMP-high → G-CIMP-low progression at recurrence. In 10 IDH-mutant non-codel primary–recurrent pairs, 4/10 showed full demethylation pattern at recurrence and 6/10 showed partial demethylation, consistent with progression from G-CIMP-high toward G-CIMP-low.
  • Three IDH-wildtype methylation subtypes: Classic-like (enriched for classical expression subtype), Mesenchymal-like (enriched for mesenchymal subtype), and a third LGG-enriched cluster designated PA-like (LGG)/LGm6-GBM (GBM by grade). PA-like LGG patients had markedly longer survival (log-rank p=3.6×10⁻⁵) and were younger (mean 37.6 vs 50.8 yr, t-test p=0.002). Validation: 61 grade I PAs all clustered with the PA-like methylation group.
  • MAPK-pathway alteration enrichment in PA-like. Frequency of mutations/fusions/amplifications in 8 PA-associated genes (BRAF, NF1, NTRK1, NTRK2, FGFR1, FGFR2) rises from 11% (Classic-like) and 13% (Mesenchymal-like) to 32% (LGm6-GBM) and 52% (PA-like LGG) (Fisher exact p<0.0001). Conversely, only 8% of PA-like LGGs upregulate TERT vs 92% Classic-like and 84% Mesenchymal-like (FET p<0.0001). PA-like tumors have low rates of EGFR, CDKN2A/CDKN2B, and PTEN alterations and largely euploid CN profiles.
  • Methylation subtype is an independent prognostic predictor. Optimal multivariate model = age + grade + epigenetic subtype (LRT p<0.0001, C-index 0.836); validation cohort n=183 → C-index 0.746. After accounting for age and grade, neither TERT expression nor TERTp mutation contributed independent prognostic value (LRT p=0.82 and p=0.85), in contrast to Eckel-Passow et al. 2015.

Genes & alterations

  • IDH1 / IDH2 — defining hotspot mutation across LGm1–3 / LGr1–3 subtypes (449/450 = 99% IDH-mut macro-cluster); primary axis of methylome and transcriptome separation. n=457 IDH1 mutations recovered.
  • ATRX — n=220 mutations, nearly mutually exclusive with TERTp mutations; mutations associated with significantly longer telomeres (ALT phenotype).
  • TERT — promoter mutations in 48% of diffuse gliomas, drive TERT mRNA upregulation; co-occur with chr7 gain / chr10 loss in IDH-wildtype glioma but did NOT independently predict survival in this cohort once age + grade were modeled.
  • TP53 (n=328), EGFR (n=314), PTEN (n=168), CIC (n=80), FUBP1 (n=45) — recover all known glioma drivers.
  • SETD2 (n=24), ARID2 (n=20), DNMT3A (n=11) — chromatin-modification genes newly nominated as glioma drivers.
  • KRAS (n=25) and NRAS (n=5) — Ras-pathway oncogenes confirmed as recurrently mutated in human glioma; previously known mainly from engineered mouse models.
  • STAG2 / NIPBL — cohesin complex genes mutated/altered in 16% of LGG/GBM; nominated as therapeutic vulnerability (PARP / DNA-damage agents).
  • CDK4, CDKN2A, CDKN2B — cell-cycle alterations strongly enriched in G-CIMP-low (15/18) vs G-CIMP-high (36/241) and Codel (2/172).
  • BRAF, NF1, NTRK1, NTRK2, FGFR1, FGFR2 — MAPK-pathway / pilocytic-astrocytoma-associated alterations enriched in PA-like LGG (52%) and LGm6-GBM (32%).
  • SOX2, OLIG2 — neurodevelopmental TFs whose binding motif (TGTT) is enriched at intergenic CpG sites that lose methylation in G-CIMP-low; SOX2 and most SOX-family members are upregulated in G-CIMP-low.
  • TRIM28 — promoter mutation (n=8) correlates with TRIM28 upregulation; mechanistic link to AMPK degradation and mTOR activation noted as candidate therapeutic axis (metformin sensitization).
  • CACNG6 — promoter mutation (n=7) correlates with CACNG6 downregulation; functional consequence in glioma not established.
  • DAXX, H3-3A (HUGO-current symbol for H3F3A) — only 2 samples mutated in this WGS dataset, despite frequent involvement in pediatric gliomas; supports an ALT mechanism distinct from telomerase but not driven by DAXX/H3.3 in adult diffuse glioma.

Clinical implications

  • Methylation subtype is independently prognostic (C-index 0.836 with age + grade + epigenetic subtype) and outperforms histology- or TERT-based classifiers in this dataset; supports moving glioma classification toward an integrated molecular framework.
  • G-CIMP-low identifies high-risk IDH-mutant non-codel glioma (poor survival comparable to IDH-wildtype) that would be missed by IDH-only stratification; cell-cycle gene activation (CDK4 amplification, CDKN2A loss) provides a candidate biomarker and potential CDK4/6-inhibitor rationale (paper does not directly test this).
  • PA-like IDH-wildtype LGG has favorable prognosis despite histologic appearance of diffuse glioma; the authors argue these patients may be spared intensive treatments. MAPK-pathway alterations (BRAF, NF1, NTRK1/2, FGFR1/2) suggest candidate targeted therapies for this subset.
  • ALT-positive (ATRX-mutant) gliomas are nominated as candidates for ATR-inhibitor sensitivity (citing Flynn et al. 2015).
  • Cohesin-mutant gliomas (NIPBL/STAG2 ~16%) are nominated as candidates for PARP inhibitors and DNA-damage agents (citing Bailey et al. 2014).
  • TRIM28 promoter-mutant gliomas are nominated as a setting where metformin / AMPK agonists may have heightened activity (mechanism per Pineda et al. 2015).

Limitations & open questions

  • Conflict with Eckel-Passow et al. 2015 on TERT prognostic value. This study found no independent prognostic effect of TERT expression or TERTp mutation after adjusting for age + grade (LRT p=0.82 / 0.85), in contrast to Eckel-Passow’s TERT/IDH/1p19q triad. The discrepancy is acknowledged but not resolved; reflects different modeling and cohorts.
  • Validation cohort for IDH-wildtype subtypes mixed adult and pediatric samples and used distinct methylation platforms; the LGm6-GBM vs PA-like LGG distinction was made on grade rather than methylation signature, leaving the boundary somewhat operational.
  • Primary–recurrent G-CIMP-high → G-CIMP-low progression is supported by only 10 paired cases; whether this is the dominant route to malignant progression in IDH-mutant glioma requires larger longitudinal series.
  • Therapeutic claims are inferential. Sensitivity of cohesin-mutant glioma to PARP inhibitors, of ALT-positive glioma to ATR inhibitors, and of TRIM28-promoter-mutant glioma to metformin are extrapolated from prior literature, not tested in this paper.
  • Clinical confounders such as extent of resection and performance status were not available for the full cohort; future updates of TCGA clinical annotation may refine the prognostic model weights.
  • Drug treatments (e.g., temozolomide, radiotherapy) are not catalogued per patient in this analysis, so survival models cannot adjust for therapy.
  • The 6-cluster pan-glioma methylation classification is presented as a basis; authors acknowledge larger and more refined cohorts are likely to subdivide further.

Citations from this paper used in the wiki

  • “We defined the complete set of genes associated with 1,122 diffuse grade II-III-IV gliomas from The Cancer Genome Atlas…” (Summary, p. 2 of manuscript) — anchors the cohort size and scope.
  • “Whole genome sequencing data analysis determined that ATRX but not TERT promoter mutations are associated with increased telomere length.” (Summary) — anchors the central telomere-mechanism claim.
  • “16% of the LGG/GBM showed mutations and/or CNAs in multiple genes involved in the cohesin complex, thus nominating this process as a prominent pathway involved in gliomagenesis.” (Results, “Identification of novel glioma-associated genomic alterations”).
  • “Glioma samples harboring ATRX mutations showed significantly longer telomeres compared to TERTp mutant samples (t-test p-value < 0.0001; Figure 1C).” (Results, “Telomere length is positively correlated with ATRX but not TERT promoter mutations”).
  • “We concluded that IDH-mutant glioma is composed of three coherent subgroups: 1. The Codel group … 2. The G-CIMP-low group … 3. The G-CIMP-high group…” (Results, IDH-mutant subtypes section).
  • “15 of 18 G-CIMP-low cases carried abnormalities in cell cycle pathway genes such as CDK4 and CDKN2A, relative to 36/241 and 2/172 for G-CIMP-high and Codels, respectively (Figure 3D).” (Results).
  • “The frequency of mutations, fusions and amplifications in eight PA-associated genes (BRAF, NF1, NTRK1, NTRK2, FGFR1, and FGFR2) rated from 11% (n=12/113) of Classic-like, 13% (n=21/158) of Mesenchymal-like IDH-wildtype tumors to 32% (n=7/22) of LGm6-GBM and 52% (n=13/25) of PA-like LGG (Fisher Exact Test (FET) p-value < 0.0001; Figure 4C).” (Results, PA-like section).
  • “the optimal survival prediction model includes age, grade and epigenetic subtype (LRT p-value< 0.0001, C-Index 0.836; Table 2).” (Results, prognostic-model section).
  • “In contrast to previous reports (Eckel-Passow et al., 2015), we failed to observe a statistically significant and independent survival association with TERT expression (LRT p-value = 0.82 …) or TERTp mutations, after accounting for age and grade (LRT p-value = 0.85, not shown).” (Results) — anchors the conflict-with-prior-work note.

This page was processed by crosslinker on 2026-05-14.