Computational triage of therapeutic and germline vulnerabilities in IDH-wildtype glioblastoma: a focal-adhesion dependency in CDKN2A/B-deleted tumors

Glioblastoma (GBM) remains almost uniformly lethal, and the CDKN2A/B-homozygous-deleted, IDH-wildtype subtype carries a particularly poor prognosis with no subtype-specific therapy. Using publicly available functional-genomic, transcriptomic, pharmacologic, clinical, and population-genetic datasets, we performed an integrated triage for actionable vulnerabilities in this subtype. We report three findings of differing strength. (1) A focal-adhesion/FAK dependency in CDKN2A/B-null GBM is supported by five complementary lines of evidence: it is selectively essential by CRISPR knockout (ITGAV q=1.8e-4, TLN1, VCL, FERMT2, ITGB5, PTK2/FAK, ILK and the Arp2/3 machinery); it reproduces under orthogonal RNAi silencing (TLN1 p=3.7e-6, FAK p=1.3e-3, ITGAV p=5e-3), ruling out a copy-number cutting artifact; a potent FAK inhibitor (PF-562271, GDSC dose-response) selectively sensitizes CDKN2A/B-null lines (p=2e-3), closing the gap left by the inert first-generation probe (GSK2256098) in PRISM; ITGAV mRNA is upregulated in CDKN2A/B-null TCGA-GBM tumors (q=0.014); and FAK is both more abundant and more activated (autophosphorylation-Y397 p=0.012; FAK protein p=3e-4; FERMT2 protein p=9e-5) in CDKN2A/B-null CPTAC-GBM tumors. The signal is not a mesenchymal-subtype artifact (null tumors are not mesenchymal-enriched, p=0.40). One major caveat constrains attribution: CDKN2A/B is co-deleted with the adjacent 9p21 gene MTAP in ~92% of null lines, and the two deletions are statistically inseparable in DepMap (our positive control, the canonical MTAP-loss dependency PRMT5, is itself misattributed to CDKN2A/B), so the dependency is most defensibly described as 9p21-codeletion-associated, though it does not vary with MTAP status among null lines, and CDKN2A/B is the more mechanistically plausible driver. It remains a genetic/protein-level dependency awaiting wet-lab pharmacologic confirmation. (2) A large, molecularly-definable CDK4/6-inhibitor-eligible subset (IDH-wildtype, RB1-intact, CDK4/6-axis-activated; ~65% of the full GBM cohort, ~70% of IDH-wildtype GBM) has never been prospectively enrolled with resistance stratification; eligibility conferred no survival advantage in TCGA or MSK-IMPACT cohorts, and a large fraction of eligible tumors carry PTEN/PI3K-axis alterations (PTEN mutation 34%, PTEN deletion 12%, PIK3R1/PIK3CA ~10% each) that plausibly explain prior null trials. (3) A Finnish-founder MUTYH pathogenic allele (p.Gly155Asp, rs587781864) shows a suggestive monoallelic association with GBM in FinnGen R13 (OR=24.9, ~4 case carriers) that is internally validated by expected colorectal signals but does not survive multiple-testing correction. Together these define one strong genetic dependency requiring better pharmacologic tools, one trial-design hypothesis with an explicit resistance map, and one replication-grade germline lead.