Longitudinal multicohort analysis of tissue composition and continuous malignant states in recurrent glioblastoma
Year: 2026
Type: preprint
Abstract: Background Bulk expression changes in recurrent glioblastoma can arise from tissue composition or malignant-cell variation. We sought to distinguish these sources and evaluate the external validity of virtual perturbation. Methods This retrospective analysis integrated 84 paired patients from the Glioma Longitudinal Analysis Consortium (GLASS), 28 single-nucleus discovery pairs and 43 malignant-eligible pairs from an external single-nucleus cohort. Brain-reference deconvolution was benchmarked in 104 pseudobulks from 52 held-out donors. Patient-paired tests, frozen external hypotheses, baseline-adjusted continuous-state models and nine public KAT5-knockout cell models were evaluated separately. Results GLASS identified 1,432 differentially expressed genes. Calibrated estimates supported increased oligodendrocyte RNA weight (+ 8.25 percentage points; false discovery rate 0.00054) and decreased vascular weight (− 6.18 points; 0.0036), with concordant directions from a second method. Malignant RNA weights failed calibration. Expanded malignant programs showed no corrected average recurrence shift. Exploratory analysis identified 293 concordant gene-axis associations in the external cohort, reduced to 33 after baseline-state adjustment in both cohorts. None of 36 previously frozen gene-state hypotheses replicated. Geneformer discrimination declined from an internal area under the receiver operating characteristic curve of 0.696 to an external value of 0.486. Virtual KAT5 deletion did not outperform matched deletion controls against real knockout (P = 0.488). Conclusions Recurrent specimens show selected tissue-composition changes alongside heterogeneous malignant-state variation. The continuous-state associations remain exploratory, and the tested prediction and perturbation endpoints do not establish a therapeutic target.
Source: Research Square
Institution: Shanghai Ninth People's Hospital
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225414254