GBM Trial Failure Autopsy: Full Exploitation Decode

Framework: CrisPRO BrM Therapy Fit Layer | ZEB1→ITGAV SL Anchor Date: 2026-05-07

Preamble: The Decoding Logic

Every GBM trial failure carries a failure signature — a specific combination of target validity, delivery architecture, patient selection, and resistance mechanism. The question for each trial is not "did it fail?" but "which failure mode applies, and does that mode exist in the BrM context?"

The BrM context differs from primary GBM in four structurally important ways:

1. BBB disruption is more complete and uniform at metastatic lesions
2. The target cell population (extravasating, ZEB1-high, EMT-active NSCLC cells) is biologically distinct from GBM stem cells
3. The SL exploit (ZEB1→ITGAV) is a NSCLC-intrinsic dependency, not a GBM-derived hypothesis
4. The drug (cilengitide) was never tested in BrM — GBM failure is a context transfer problem, not a target invalidation

Trial Class 1: Cilengitide / Integrin αVβ3 Inhibition

Trials

Failure Signature Decode

Failure Mode 1 — BBB Paradox (PRIMARY): Cilengitide requires an intact or partially intact BBB to concentrate in tumor tissue. In GBM, the BBB is disrupted heterogeneously — the necrotic core is accessible but the invasive rim (where αV-dependent cells reside) retains partial BBB integrity. NABTC 03-02 directly confirmed this: drug reached resectable tumor but distribution was uneven [8]. The drug was fighting its own delivery architecture.

Failure Mode 2 — Unselected Patients (CRITICAL): No ZEB1-high or αV-dependency biomarker was used for enrollment. The CENTRIC biomarker analysis confirmed that αVβ3/αVβ5 IHC expression did not predict benefit [9]. This is not evidence that the target is wrong — it is evidence that the wrong patients were enrolled. The SL signal (ZEB1→ITGAV, delta=−0.72, FDR=0.001) provides exactly the stratification logic that was missing.

Failure Mode 3 — Schedule Subtherapeutic: Twice-weekly IV dosing produced trough plasma levels below αVβ3 IC₅₀ in many patients. Preclinical data showed schedule-dependent radiosensitization [16] — continuous infusion was more effective but never tested in Phase III.

Failure Mode 4 — Anti-Angiogenic Ceiling: In GBM, anti-angiogenic monotherapy consistently fails OS endpoints (see bevacizumab below). Cilengitide's anti-angiogenic component (endothelial αVβ3 blockade) may have been the dominant mechanism tested, not the anti-invasive/anoikis component that the ZEB1 SL logic predicts.

BrM Exploitation Signal

All four failure modes are partially or fully mitigated in BrM:

• BBB paradox → BrM disrupts BBB more completely and uniformly [55]
• Unselected patients → ZEB1-high NSCLC stratification is now available via DepMap SL signal
• Schedule → BrM trials can test continuous or more frequent dosing
• Anti-angiogenic ceiling → The BrM hypothesis is anti-invasive/anoikis, not anti-angiogenic

Exploitation verdict: STRONG. The GBM failure is a context-and-selection artifact. The target is valid. The BrM hypothesis is mechanistically distinct from what was tested in CENTRIC.

Trial Class 2: Bevacizumab / Anti-VEGF

Trials

Failure Signature Decode

Failure Mode 1 — Vascular Normalization Paradox: Bevacizumab reduces tumor vascularity and edema (explaining the PFS signal and steroid-sparing effect) but simultaneously reduces drug delivery to the tumor. Anti-VEGF therapy normalizes the BBB, which paradoxically traps subsequent drugs outside the tumor Neuro-oncology +1.

Failure Mode 2 — Invasion Escape: When VEGF-driven angiogenesis is blocked, GBM cells switch to a co-option invasion strategy — they migrate along existing blood vessels rather than inducing new ones. This is a well-characterized resistance mechanism that anti-VEGF cannot address Expert Review of Neurotherapeutics.

Failure Mode 3 — No OS Benefit Despite PFS Benefit: The PFS benefit is real but reflects tumor shrinkage and edema reduction, not tumor cell killing. Patients live longer without progression but die at the same time — the tumor adapts and recurs more aggressively Neuro-Oncology Advances +1.

Failure Mode 4 — Proneural Subtype Signal (Missed): Retrospective AVAglio analysis showed proneural GBM subtype may derive OS benefit Journal of clinical oncology : official journal of the American Society of Clinical Oncology. This subgroup was not prospectively selected — a recurring theme across GBM trials.

BrM Exploitation Signal

Bevacizumab is already in brm.json as a DELIVERY_INTERCEPTION_HYPOTHESIS. The GBM failure data actually supports this classification:

• The PFS benefit (without OS benefit) in GBM is consistent with a delivery-interception mechanism — bevacizumab changes the tumor microenvironment and drug access, not tumor cell biology
• In BrM, the hypothesis is different: bevacizumab may reduce BBB permeability changes that facilitate metastatic seeding, or improve co-delivery of other agents
• The MMP9 biomarker signal from AVAglio ancillary data Acta Neuropathologica Communications is directly relevant — MMP9 is in the BrM universe (receipt-backed, score_basis=HYBRID)

Exploitation verdict: MODERATE. Bevacizumab's GBM failure confirms it is not a tumor-cell-killing agent. The BrM DELIVERY_INTERCEPTION_HYPOTHESIS is the correct classification. The MMP9 biomarker connection (MMP9 in BrM universe) is an underexploited provenance link — worth annotating in brm.json.

Trial Class 3: EGFR Inhibitors (Erlotinib, Gefitinib, Dacomitinib, Depatux-m)

Trials

Failure Signature Decode

Failure Mode 1 — BBB Exclusion (DOMINANT for small molecules): Erlotinib and gefitinib are P-glycoprotein substrates. The intact BBB actively pumps them out of the CNS. Brain tumor tissue concentrations are 10–100× below plasma levels Molecular Cancer Therapeutics +1. This is a pharmacokinetic failure, not a target failure.

Failure Mode 2 — PTEN Loss Decouples EGFR from PI3K: In GBM, ~40% of tumors have PTEN loss. When PTEN is absent, PI3K/AKT/mTOR is constitutively active regardless of EGFR inhibition — the downstream pathway runs without the upstream signal Cancer research. EGFR inhibition becomes irrelevant. This is the most important mechanistic failure mode.

Failure Mode 3 — EGFRvIII Upregulation Under Pressure: Erlotinib resistance in EGFR-amplified GBM is associated with upregulation of EGFRvIII and PI3Kp110δ Neuro-oncology. The drug selects for the most treatment-resistant variant.

Failure Mode 4 — Depatux-m (ADC) — Different Failure: Depatux-m is an antibody-drug conjugate targeting activated EGFR (overexpressed WT + EGFRvIII). It failed despite CNS penetration being better than small molecules. The failure here is likely tumor heterogeneity — EGFRvIII is expressed in only a subset of cells, and non-EGFRvIII cells survive and repopulate Neuro-Oncology.

BrM Exploitation Signal

EGFR inhibitors are already in brm.json as STANDARD_OF_CARE_ANCHOR (osimertinib). The GBM failure data is directly instructive:

• Osimertinib has superior CNS penetration vs. erlotinib/gefitinib — it is not a P-gp substrate Oncotarget. The BBB failure mode does not apply.
• PTEN loss in NSCLC BrM is a known resistance mechanism to EGFR TKIs — this is the key vulnerability to annotate. PTEN loss → PI3K constitutive activation → EGFR TKI resistance → need for combination with PI3K/mTOR inhibitor.
• The ZEB1→ITGAV SL signal is independent of EGFR status — it operates in the EMT compartment, not the EGFR-driven proliferative compartment. This means cilengitide + osimertinib is a mechanistically non-overlapping combination hypothesis.

Exploitation verdict: HIGH STRATEGIC VALUE. The EGFR TKI GBM failures define exactly why osimertinib (CNS-penetrant, non-P-gp substrate) is the right anchor. The PTEN loss resistance mechanism in NSCLC BrM is the key gap that the ZEB1→ITGAV SL exploit could address in a combination setting.

Trial Class 4: EGFRvIII Vaccine — Rindopepimut (ACT IV)

Trials

Failure Signature Decode

Failure Mode 1 — Immunoediting / Antigen Loss (DEFINITIVE): This is the most instructive failure in the entire GBM trial history. Rindopepimut generated robust EGFRvIII-specific immune responses — the drug worked immunologically. But the tumor responded by downregulating or losing EGFRvIII expression entirely. At recurrence, EGFRvIII-negative clones dominated The Lancet. Oncology +1. The vaccine killed EGFRvIII+ cells and selected for EGFRvIII− cells. This is textbook immunoediting.

Failure Mode 2 — Single Antigen Targeting: Targeting one tumor-specific antigen in a heterogeneous tumor is inherently fragile. Any pre-existing EGFRvIII− clone survives and repopulates. The Phase II trials worked because they enrolled patients with minimal residual disease (MRD) after resection — low tumor burden, less heterogeneity. ACT IV enrolled a broader population Neuro-oncology +1.

Failure Mode 3 — Immunosuppressive TME: Even when EGFRvIII-specific T cells were generated, the GBM TME (TGF-β, IDO, regulatory T cells, M2 macrophages) suppressed their effector function Journal of Experimental & Clinical Cancer Research : CR +1.

BrM Exploitation Signal

This failure is the most directly translatable to the CrisPRO framework:

The rindopepimut failure is the canonical proof-of-concept for why SL-based targeting is superior to single-antigen targeting:

• SL exploits target a dependency (ITGAV essentiality in ZEB1-high cells), not an antigen (EGFRvIII expression). Dependencies cannot be simply downregulated without fitness cost.
• ZEB1-high cells that lose ITGAV expression lose their EMT-driven adhesion and survival advantage — they cannot escape the way EGFRvIII− cells escaped rindopepimut.
• The immunoediting failure of ACT IV is the strongest possible argument for the SL approach: you cannot immunoedit away a synthetic lethal dependency.

Exploitation verdict: MAXIMUM STRATEGIC VALUE. ACT IV is the negative control that proves the SL hypothesis. Use this explicitly in CrisPRO positioning: "Unlike antigen-targeted approaches (rindopepimut/ACT IV), SL-based targeting exploits fitness dependencies that cannot be immunoedited away."

Trial Class 5: Checkpoint Inhibitors (Nivolumab, Pembrolizumab, Ipilimumab)

Trials

Failure Signature Decode

Failure Mode 1 — Immunosuppressive TME (DOMINANT): GBM is one of the most immunosuppressive tumors known. The TME is dominated by M2-polarized tumor-associated macrophages (TAMs), regulatory T cells, TGF-β, IL-10, and IDO Journal of Experimental & Clinical Cancer Research : CR +1. Pembrolizumab window-of-opportunity trial directly showed that even after drug delivery, the TME remained macrophage-dominated with minimal T cell infiltration Neuro-oncology.

Failure Mode 2 — Low Tumor Mutational Burden (TMB): GBM has low TMB (except hypermutated TMZ-treated recurrences). Low TMB = few neoantigens = few T cell targets. Checkpoint inhibitors work by unleashing pre-existing T cell responses — if there are no T cells to unleash, the drug has nothing to work with Arquivos de Neuro-Psiquiatria +1.

Failure Mode 3 — PTEN Loss → Immune Exclusion: PTEN loss in GBM (40% of cases) directly suppresses T cell trafficking into tumors and reduces T cell-mediated killing Cancer discovery. PTEN loss creates a cell-intrinsic immune exclusion mechanism that checkpoint inhibitors cannot overcome.

Failure Mode 4 — Blood-Brain Barrier Limits T Cell Trafficking: Even systemically activated T cells face the BBB as a trafficking barrier. Intracerebral administration of checkpoint inhibitors showed better local immune activation Journal for Immunotherapy of Cancer but is not scalable.

Failure Mode 5 — ZEB1-Driven Immune Evasion: ZEB1 directly induces PD-L1 and M2 macrophage polarization, creating an immunosuppressive envelope around invading cancer cells [26]. This means ZEB1-high cells — the exact population the SL exploit targets — are also the most checkpoint-inhibitor-resistant cells.

BrM Exploitation Signal

This is the most complex failure class for BrM translation:

• In NSCLC BrM, the TME is different from primary GBM — it retains more NSCLC-like immune characteristics Targeted Oncology. Checkpoint inhibitors have demonstrated activity in NSCLC BrM (pembrolizumab, nivolumab active in CNS lesions in NSCLC).
• The ZEB1→PD-L1 connection [26] is a double-edged sword: ZEB1-high cells are checkpoint-resistant AND ITGAV-dependent. This means cilengitide + checkpoint inhibitor is a mechanistically motivated combination — cilengitide kills ZEB1-high cells via SL, removing the immunosuppressive envelope, potentially sensitizing residual tumor to checkpoint blockade.
• The PTEN loss → immune exclusion connection Cancer discovery links the EGFR TKI resistance mechanism (PTEN loss → PI3K constitutive activation) to checkpoint inhibitor resistance. This creates a three-way vulnerability: PTEN-loss tumors are EGFR-TKI resistant AND checkpoint-resistant AND potentially more ITGAV-dependent (PI3K/AKT promotes integrin-mediated survival signaling).

Exploitation verdict: HIGH STRATEGIC VALUE. The checkpoint inhibitor failures in GBM define the immunosuppressive context that ZEB1-high cells create. The SL exploit (cilengitide targeting ITGAV in ZEB1-high cells) is mechanistically upstream of checkpoint resistance. This is a combination hypothesis worth encoding.

Trial Class 6: mTOR/PI3K Inhibitors

Trials

Failure Signature Decode

Failure Mode 1 — mTORC1 Inhibition → AKT Feedback Activation: Rapamycin/rapalogs inhibit mTORC1 but release negative feedback on IRS-1, leading to AKT hyperactivation via mTORC2. The drug suppresses one arm of the pathway and activates another PLoS Medicine +1. This is the canonical rapalog resistance mechanism.

Failure Mode 2 — STAT3 Feedback: In PTEN-deficient cells, PI3K/mTOR inhibition triggers STAT3 activation as a compensatory survival signal Oncogenesis. STAT3 is a known GBM survival pathway — blocking PI3K/mTOR simply routes survival signaling through STAT3.

Failure Mode 3 — Pathway Redundancy: GBM has multiple parallel survival pathways (EGFR, PDGFR, MET, FGFR) that can all activate PI3K. Blocking PI3K at one node is insufficient when upstream inputs are redundant Medical Oncology +1.

Failure Mode 4 — BBB Penetration: Many PI3K/mTOR inhibitors have poor CNS penetration, limiting tumor exposure Disease Markers.

BrM Exploitation Signal

The mTOR/PI3K failure class is the most important for understanding the PTEN-loss resistance node:

• PTEN loss → constitutive PI3K/AKT → integrin-mediated survival signaling upregulation. This creates a direct mechanistic link: PTEN-loss NSCLC BrM cells may be MORE dependent on ITGAV-mediated survival (as a compensatory adhesion-survival pathway) precisely because their PI3K pathway is constitutively active.
• The rapamycin Phase I data PLoS Medicine showed antitumor activity specifically in PTEN-deficient GBM — confirming that PTEN status is a real biomarker for PI3K pathway dependency. In NSCLC BrM, PTEN loss is a known osimertinib resistance mechanism.
• Three-way convergence: PTEN-loss NSCLC BrM cells are (1) osimertinib-resistant, (2) checkpoint-resistant Cancer discovery, and (3) potentially more ITGAV-dependent. This is the highest-priority combination hypothesis in the framework.

Exploitation verdict: HIGH STRATEGIC VALUE. PTEN loss is the convergence node linking EGFR TKI resistance, checkpoint resistance, and potential ITGAV dependency. Encode PTEN-loss status as a biomarker enrichment hypothesis for the cilengitide entry.

Master Exploitation Matrix

Three Actionable Outputs for brm.json / CrisPRO

1. Cilengitide entry — ADD_AS_HYPOTHESIS (approved in dossier) Add pten_loss_enrichment_hypothesis: true and combination_hypothesis: ["osimertinib", "pembrolizumab"] to the entry. The GBM failure decode provides the mechanistic rationale for both.

2. Bevacizumab entry — annotate MMP9 biomarker link The AVAglio ancillary study Acta Neuropathologica Communications identified plasmatic MMP9 (from tumor-infiltrating neutrophils) as a predictive biomarker for bevacizumab efficacy. MMP9 is in the BrM universe (receipt-backed). This is a provenance link that should be encoded.

3. CrisPRO positioning statement — ACT IV as negative control The rindopepimut/ACT IV failure is the canonical proof that antigen-targeted approaches fail via immunoediting. SL-based targeting (ZEB1→ITGAV) exploits a fitness dependency that cannot be immunoedited. This is the single most powerful differentiation argument for the CrisPRO platform