Glioblastoma is not one rogue gene. It is a wiring diagram with no single load-bearing node — which is exactly why every drug aimed at one node has, so far, lost.
Three control circuits fail in almost every glioblastoma: the growth-signal input is stuck on, the p53 damage-control system is disabled, and the RB cell-cycle gate is forced open. Add three modules — telomerase switched back on by a TERT promoter mutation, DNA repair capacity that decides whether chemotherapy works, and a motility programme that drives infiltration — and you have the disease.inference
The circuits are not broken one at a time. In one large cohort, 89.6% of tumours carried an alteration somewhere in the growth axis and 85.3% in the p53 pathway.fact Each circuit has several interchangeable failure modes, so blocking any single node leaves the others intact. That structural redundancy — not a missing drug — is the central problem.inference
Since 2021, glioblastoma has been a molecular diagnosis. Get this wrong and every statistic downstream is wrong too.
Under the fifth edition of the WHO Classification of Tumours of the Central Nervous System, glioblastoma is by definition IDH-wildtype and H3-wildtype. An adult diffuse astrocytic glioma qualifies as glioblastoma, IDH-wildtype, CNS WHO grade 4 if it meets that condition and shows at least one of five features — two histological, three molecular.fact
Toggle the switches to see how the 2021 criteria resolve. Note that any one of the three molecular findings is sufficient on its own — the histology is not required.
Teaching model of the published criteria only. Real diagnosis is an integrated neuropathological assessment; this widget cannot diagnose anything and must not be used for clinical decisions.
A tumour that looks grade 2 or 3 down the microscope can be grade 4 glioblastoma on molecular grounds alone. Before 2021 that same tumour was called a diffuse or anaplastic astrocytoma.fact
"Primary GBM" and "secondary GBM" are obsolete terms. What used to be called secondary GBM — grade 4 disease arising from an IDH-mutant lower-grade astrocytoma — is now astrocytoma, IDH-mutant, CNS WHO grade 4. Different entity, different natural history, not glioblastoma.fact Any pre-2021 cohort labelled simply "GBM" almost certainly contains some of these, which inflates the apparent frequency of IDH1, TP53 and ATRX alterations.inference
For scale: glioblastoma accounted for 13.7% of all primary brain and other CNS tumours in the United States, and 52.2% of the malignant ones, for diagnosis years 2018–2022.fact Five-year relative survival is about 7%.fact
Three symbols cause most of the trouble, because the gene and its protein look nothing alike.
gene / locus / DNA italic, uppercase EGFR TP53 MTOR CDKN2A transcript / mRNA same symbol as gene EGFR mRNA protein roman, own convention EGFR p53 mTOR p16INK4a
| Gene | Protein | Why it trips people up |
|---|---|---|
TP53 | p53 | Named for its apparent mass, not for its gene |
CDKN2A | p16INK4a and p14ARF | One locus, two proteins, two different circuits |
MTOR | mTOR | The lower-case m is a historical relic |
CDKN2A is the one to remember. The locus uses two alternative reading frames and makes two unrelated proteins: p16INK4a, which blocks CDK4/6 in the RB circuit, and p14ARF, which stabilises p53 by inhibiting MDM2. A homozygous deletion therefore knocks out two of the three core circuits in one event.fact
The framing comes from the first large integrated genomic analysis of glioblastoma, which found three core pathways altered in the great majority of tumours.fact Every gene on this page belongs to one of them, or to one of the modules in section 07.
Scroll the diagram sideways →
Read this as a wiring diagram, not as a sequence of events in any one tumour. Different tumours break different nodes to reach the same three outcomes.inference
The most frequently altered circuit in the disease. In a cohort of 251 tumours with both exome sequencing and copy-number data, at least one receptor tyrosine kinase was altered in 67.3%, and 89.6% carried an alteration somewhere in the RTK–PI3K–PTEN axis.fact
A receptor kinase in the cell membrane. Ligand binds, two receptors pair up, their tails phosphorylate each other, and the phosphorylated tails become docking sites for PI3K and others. The single most frequently altered gene in glioblastoma — usually amplified to very high copy number and often rearranged.
An in-frame deletion that removes part of the ligand-binding domain. The receptor can no longer bind ligand and signals constitutively — permanently on, with no off switch. Found in roughly half of EGFR-amplified tumours. Its junction sequence exists nowhere in normal tissue, making it a genuine tumour-specific neoantigen.
A second receptor kinase, marking a partly distinct biology: it associates with the oligodendrocyte-progenitor-like cell state where EGFR associates with the astrocyte-like state. About 42% of PDGFRA-altered tumours also carry an EGFR alteration — usually in different cells of the same tumour.
The principal brake on the PI3K arm. PI3K makes the membrane lipid PIP3, which recruits AKT; PTEN reverses that reaction. Lose PTEN and the signal persists with no receptor input at all. Alterations are largely mutually exclusive with activating PIK3CA or PIK3R1 mutations — 59.4% of tumours had one or the other. A single-gene frequency for PTEN alone was not confirmed in this pass, so none is quoted.
The catalytic and regulatory subunits of PI3K itself. These are activating mutations — they make the kinase work when it should not. Counting all PI3K-family genes, mutations were present in 25.1% of tumours.
A GTPase-activating protein that switches RAS off. Lose it and RAS signalling stays on. NF1 loss is the alteration most associated with the mesenchymal-like cell state, which carries the strongest hypoxia-associated and immunosuppressive programmes.
The integrator, working in two complexes — mTORC1 and mTORC2 — that ask whether there are growth signals, nutrients and energy, and whether the cell may build. Note the frequency: MTOR is rarely mutated or amplified in glioblastoma. The pathway is switched on from upstream, not by changing this gene.
The EGFR-to-mTOR connection is not one obligatory line. EGFR has been shown to signal toward mTOR through routes that do not require AKT, and oncogenic EGFR activates an mTORC2–NF-κB axis linked to chemotherapy resistance.fact Drawing EGFR → PI3K → AKT → mTOR as a single arrow is a teaching simplification.inference
The p53 pathway was dysregulated in 85.3% of tumours in the reference cohort.fact Only about a third of that comes from TP53 itself — the rest arrives by other routes to the same end.
Normally held at low abundance because MDM2 continuously tags it for destruction. On DNA damage the tagging is interrupted, p53 accumulates, and it induces p21 to arrest the cycle, induces repair genes, and — if damage is severe — triggers apoptosis. Most mutations are missense changes in the DNA-binding domain: the protein is still made, often more abundant than wild-type, and functionally dead.
The negative regulators. Amplifying them achieves what mutating TP53 achieves: p53 protein present, p53 output absent. The 15.1% figure is reported by the source as MDM1/2/4 amplification. Largely mutually exclusive with TP53 mutation — the genomic signature of two solutions to one problem.
p14ARF inhibits MDM2 and so stabilises p53. Deleting the CDKN2A locus removes this stabiliser — and simultaneously removes p16INK4a from the RB circuit. One deletion, two circuits.
This is why measuring p53 abundance is a poor proxy for p53 function. Strong p53 immunostaining usually indicates a stabilised mutant protein, not an active pathway.fact
TP53 mutation is far more characteristic of IDH-mutant astrocytoma than of IDH-wildtype glioblastoma.fact Frequencies quoted from pre-2021 "GBM" cohorts are inflated for exactly the reason set out in section 01.
The G1/S transition is the point of no return for cell division. RB1 holds it shut by sequestering E2F transcription factors. CDK4 and CDK6, partnered with D-type cyclins, phosphorylate RB1 and release E2F. p16INK4a inhibits CDK4/6 and keeps the gate closed.
Three ways to force it open, roughly interchangeable:
CDKN2A/B — remove the inhibitor. Present in 57.8% of tumours, one of the most common single events in the disease.factCDK4 or CDK6 — overwhelm the inhibitor.RB1 — remove the gate itself. RB1 is among the significantly mutated genes in glioblastoma.factAll three are mutually exclusive with one another, which confirms that they are alternative solutions to the same problem rather than additive hits.fact
One diagnostic asymmetry worth knowing: in IDH-mutant astrocytoma, homozygous CDKN2A/B deletion is itself a grading criterion that assigns CNS WHO grade 4.fact In IDH-wildtype glioblastoma the same deletion is common but is not part of the diagnostic definition.
TERT — the promoter mutation that switches telomerase back onOrdinary cells lose telomere sequence at each division and eventually stop. Glioblastoma usually solves this with a point mutation in the promoter of TERT, not in the coding sequence — the two recurrent hotspots are C→T changes 124 and 146 base pairs upstream of the start codon, each creating a new binding site for an ETS-family transcription factor.fact
Reported frequency in IDH-wildtype glioblastoma runs from roughly 62% to 80% depending on cohort and method.contested It is high enough to be one of the three molecular criteria sufficient for diagnosis. It is also a neat demonstration of why exome sequencing alone can miss a defining driver: promoters are non-coding, and standard exome capture does not cover them.inference
The alternative route — ALT, alternative lengthening of telomeres, associated with ATRX loss — belongs to IDH-mutant astrocytoma. TERT promoter mutation and ATRX loss are largely mutually exclusive.fact
MGMT — the only routinely actionable result on the reportMGMT encodes a repair protein that strips alkyl groups from the O6 position of guanine — precisely the lesion temozolomide creates to kill the cell. The logic is a double negative, so read it slowly:
promoter UNMETHYLATED → gene expressed → repair protein present → damage repaired → less benefit promoter METHYLATED → gene silenced → repair protein absent → damage persists → more benefit
Reported in roughly 30–45% of glioblastomas depending on cohort, assay and cut-off. In a national cohort of 20,734 IDH-wildtype patients, 30.9% were methylated, 43.7% unmethylated and 25.4% unknown — about 41% of those with a determined result.factcontested
Two things MGMT is not: it is not a driver — silencing it does not cause the tumour, it changes how the tumour answers chemotherapy; and it is not a diagnostic criterion — it does not appear in the WHO definition at all.fact
IDH1, IDH2, ATRX, H3-3A — the genes that define what glioblastoma is notHotspot IDH mutations (canonically IDH1 R132H) give the enzyme a new activity: instead of α-ketoglutarate it produces D-2-hydroxyglutarate, an oncometabolite that inhibits the dioxygenases which demethylate DNA and histones, producing a genome-wide hypermethylator phenotype. Under WHO CNS5 an IDH mutation moves the tumour out of the glioblastoma category entirely.fact
Loss of nuclear ATRX is a practical marker of IDH-mutant astrocytoma; glioblastomas typically retain it.fact And a H3-3A K27M-mutant midline tumour is diffuse midline glioma, H3 K27-altered — not glioblastoma, however it looks histologically.fact
The practical rule for reading a molecular report: check IDH first. It changes the name of the disease, the expected survival, and the relevance of everything else on the page.inference
WNT5A — the motility armGlioblastoma kills by infiltration, not by bulk. Cells migrate along white-matter tracts and blood vessels well beyond the enhancing rim on MRI, which is why complete surgical removal is not achievable. Infiltration is a motility programme, regulated separately from growth.
WNT5A ligand ↓ FZD receptors, with ROR1 / ROR2 co-receptors ↓ Ca²⁺ / PKC / JNK / Rho-family GTPase signalling ↓ actin cytoskeleton and cell polarity ↓ directed migration, invasion-associated behaviour
WNT5A/ROR signalling drives directional migration and invasion across several tumour types, and in glioma Wnt-5a signalling has been correlated with infiltrative activity through induction of cellular migration and MMP-2.fact
Three honest limits. WNT5A is not a core driver in the way EGFR or CDKN2A/B are — it is not recurrently amplified or mutated at high frequency, and its relevance is expression-level and microenvironmental.inference Its effects are strongly context-dependent: the same ligand can promote or restrain migration depending on which receptors a cell displays.contested And there is no approved WNT5A-directed therapy in glioblastoma.fact
Glioblastoma is a copy-number disease more than a point-mutation disease.inference Three structural phenomena carry most of the weight.
+7 / −10Gain of the whole of chromosome 7 together with loss of the whole of chromosome 10, in the same tumour. Chromosome 7 carries EGFR; chromosome 10 carries PTEN. One event raises the accelerator and lowers the brake at the same time. It is so characteristic that it is one of the three sufficient molecular criteria for the diagnosis.fact
Losing both copies of a locus. CDKN2A/B is the canonical example, and because that locus encodes both p16INK4a and p14ARF, a single deletion hits two circuits.fact
Amplified oncogenes in glioblastoma are frequently not on a chromosome at all, but on circular DNA elements with no centromere. Because they do not segregate evenly at mitosis, copy number can rise and fall very fast under selection, and ecDNA-borne oncogenes reach far higher copy numbers and transcript levels than chromosomally amplified ones.
In a pan-cancer analysis, ecDNA was detected in 49.1% of glioblastomas — among the highest of any tumour type — against 17.1% across all cancers.fact In IDH-wildtype glioblastoma, ecDNA is the dominant mechanism of focal oncogene amplification, and EGFR structural variants including vIII were found exclusively on ecDNA in at least one series.fact
ecDNA is probably the most important recent change to how this network should be pictured. The genotype is not a fixed list. It is a distribution the tumour can re-tune within weeks of drug pressure.inference
Bulk expression profiling originally produced four transcriptional subtypes; three survived scrutiny — proneural, classical, mesenchymal. The fourth, "neural", is now attributed largely to contamination by non-malignant oligodendrocytes and neurons rather than to a tumour programme.fact
Single-cell sequencing then reframed the picture. Rather than each tumour being a subtype, each tumour contains cells in four states that interconvert.fact
An astrocytic programme. The state most closely tied to the classical bulk subtype.
An oligodendrocyte-progenitor programme. Typically co-occurs with the NPC-like state.
A neural-progenitor programme. Together with OPC-like it accounts for the proneural bulk subtype.
Mesenchymal. Enriched in hypoxic regions, and the state carrying the strongest immunosuppressive programme.
Most tumours contain all four states; the mixture varies between tumours and even between regions of the same tumour.fact This is the bridge from genotype to behaviour: a mutation biases a cell state; the microenvironment and the treatment then shift the mixture.inference
PTEN/PIK3CA, across TP53/MDM2, across CDKN2A/CDK4/RB1 proves each circuit has several equivalent failure modes.factCDKN2A deletion alone is in 57.8%.factEGFR-amplified and PDGFRA-amplified cells coexist in one tumour. Clear one and you make room for the other.factTaken together: the network has no single load-bearing node.inference
Maximal safe resection, then radiotherapy with concurrent and adjuvant temozolomide — the regimen that produced a median overall survival of 14.6 months.fact Adding Tumor Treating Fields to maintenance temozolomide extended median overall survival from 16.0 to 20.9 months in the phase 3 EF-14 trial.fact MGMT promoter methylation status guides the expected benefit from the alkylating component.fact
| Target | Approach | Outcome |
|---|---|---|
| EGFRvIII | Rindopepimut peptide vaccine | Negative in phase 3 |
| EGFR | Depatuxizumab mafodotin, an antibody–drug conjugate | Negative |
| EGFRvIII | CAR-T, given systemically | Antigen loss, no durable response |
| EGFRvIII + EGFR | CARv3-TEAM-E, given intraventricularly | Rapid regression, largely transient |
| VEGF | Bevacizumab | Improves progression-free survival and oedema, not overall survival |
The EGFRvIII story is the cleanest illustration of section 10: an ideal-looking tumour-specific target, hit hard by three different modalities, defeated each time by heterogeneity and antigen loss.inference The intraventricular CAR-TEAM result is genuinely striking — rapid radiographic regression in the first reported patients — but responses were largely transient, and it remains an early-phase finding rather than a standard of care.fact
Active directions as of mid-2026 include B7-H3-directed antibody–drug conjugates and CAR-T, multi-antigen and pooled CAR designs, sonodynamic therapy, and biomarker-selected EGFR inhibition in MGMT-unmethylated, EGFRvIII-positive disease. These are trials, not treatments.fact
If a source claims a targeted therapy has changed survival in glioblastoma, check whether the claim rests on a randomised phase 3 overall-survival endpoint. Most do not.inference
This is the reasoning core of the page, and it applies to every gene above. A tumour or a cell line can be interrogated at eight levels. They are not interchangeable. Select a rung to see what it answers — and what it does not.
| Do not write | Write instead |
|---|---|
| "The treatment activated mTOR" | MTOR transcript abundance was higher; pathway activity was not measured |
"TP53 increased, so repair improved" | TP53 transcript abundance was higher; p53 function was not assessed |
"WNT5A caused faster migration" | WNT5A abundance and closure rate moved together; no perturbation was performed |
| "p > 0.05, so there was no effect" | No difference was detected at the stated threshold |
These frequencies come predominantly from a 2013 exome-and-copy-number cohort of 251 tumours assembled before the 2021 reclassification. That cohort therefore includes IDH-mutant tumours that would today be excluded from a glioblastoma series. Frequencies for TP53 are inflated relative to a modern IDH-wildtype-only cohort; those for EGFR and TERT are, if anything, understated.inference Treat every number as an order of magnitude, not a constant.
| Event | Reported range | Note |
|---|---|---|
TERT promoter mutation | ~62–80% | IDH-wildtype glioblastoma contested |
MGMT promoter methylation | ~30–45% | Assay- and cut-off-dependent contested |
| EGFRvIII expression | ~25–30% | Roughly half of EGFR-amplified tumours |
| ecDNA present | ~49% | Pan-cancer analysis, glioblastoma subset |
Radiotherapy with concurrent and adjuvant temozolomide becomes the standard of care; median overall survival 14.6 months.
The first integrated genomic characterisation names three core pathways: RTK/RAS/PI3K, p53, RB.
Bulk expression profiling defines transcriptional subtypes — proneural, classical, mesenchymal, and a fourth later reattributed to normal tissue.
Exome and copy-number analysis of 251 tumours produces the frequency table still quoted today.
Single-cell sequencing replaces subtypes with four interconvertible cell states present within each tumour.
WHO CNS5 redefines glioblastoma as IDH-wildtype and adds three sufficient molecular criteria.
Intraventricular CAR-TEAM cells produce rapid but largely transient regression in the first reported patients.
Extrachromosomal DNA is confirmed as the dominant mode of focal oncogene amplification in IDH-wildtype glioblastoma.
DEFINITION — WHO CNS5, 2021
IDH-wildtype + H3-wildtype diffuse astrocytic glioma
+ any one of: microvascular proliferation | necrosis
TERT promoter mutation | EGFR amplification | +7/-10
THREE CIRCUITS
GROWTH EGFR, PDGFRA up · PTEN, NF1 lost · PI3K > AKT > mTOR
p53 TP53 mutated | MDM2/MDM4 amplified | p14ARF lost
RB CDKN2A/B lost | CDK4/6 amplified | RB1 lost
MODULES
TERT promoter mutation telomerase switched back on
MGMT promoter methylated predicts temozolomide benefit
WNT5A / noncanonical Wnt cytoskeleton, motility, infiltration
ecDNA oncogenes on circles, copy number unstable
ONE LOCUS, TWO CIRCUITS
CDKN2A > p16INK4a (RB circuit) + p14ARF (p53 circuit)
READING RULE
DNA != methylation != mRNA != protein != protein state
!= pathway output != phenotype
A measurement at one rung licenses a claim only at that rung.
Claims were split into atomic statements and checked against primary literature, official classification documents and population registries before being written. Where cohorts disagree, a range is given and the claim is marked contested. Where a statement is this page's synthesis rather than a source's, it is marked inference. Last evidence check: 16 August 2026.
No numeric gene identifiers are listed here, because they were not individually verified in this pass and an unverified identifier is worse than none. Symbols follow HGNC-approved nomenclature. No claim is made about any drug approval that was not confirmed against a source during this pass.