Abstract
BACKGROUND: The brain comprises ~2 % of body mass yet consumes ≈ 20 % of total body oxygen and ATP, creating an exceptionally competitive metabolic niche. To survive, glioblastoma (GBM) cells adopt normal brain-cell programs, spanning the proneural-to-mesenchymal phenotype axis. Multiple lines of evidence implicate karyotypic change as a key adaptation: (i) across organs, tissue pO₂ correlates with mean tumor ploidy (Spearman r = 0.65, P = 0.002); (ii) GBM cell lines reach ploidy 3.36 versus 2.23 in tumors (P <0.001); (iii) high-ploidy karyotypes shift to glycolysis at oxygen levels that still support oxidative phosphorylation in low-ploidy karyotypes (ANOVA: P = 1.9E-4); (iv) chronic nutrient deprivation induces recurrent chromosome mis-segregations altering chemotherapy sensitivity; (v) untreated lower-grade gliomas accrue karyotype changes during progression; and (vi) mesenchymal GBMs are enriched for specific karyotypes (e.g., 6q loss ~4.5×). We hypothesized that metabolic stress sculptskaryotype fitness landscapes (KFLs)—tensors linking chromosome mis-segregations to fitness. METHODS: snRNA-seq from 18 longitudinally sampled GBMs (103,855 nuclei) was processed with Numbat to infer cell-resolved karyotypes. The ALFA-K algorithm fit KFLs; uncertainty was estimated by Bayesian bootstrap resampling (1,000 replicates), and accuracy assessed via cross-validation. RESULTS: Robust KFLs (cross-validation> 0.4) emerged for 15 / 18 tumors, resolving 166,393 karyotype–fitness pairs (median =11,093). Each KFL yielded a beneficial karyotype proportion (BKP)—the fraction of mis-segregations predicted to increase fitness. Hierarchical clustering of BKP across autosomes stratified patients into two groups: low BKP (n = 10) and high BKP (n = 5), supported by multiscale bootstrap resampling (AU = 0.95). BKP-high tumors up-regulated Angiogenesis (GSVA: adjusted P = 1.8 × 10⁻⁴) and down-regulated TNFA/NF-κB, Glycolysis, and PI3K/AKT/mTOR hallmarks (P < 0.03 each). Temporal-lobe GBMs showed higher BKP than frontal/parietal tumors (χ² -test: P = 0.018), mirroring the temporal lobe’s lower aerobic-glycolysis index. CONCLUSIONS: GBMs in metabolically stringent niches exhibit KFL topologies shaped by negative selection against CIN, imposing karyotypes suited for neural infiltration and metabolic hijacking.