Glioblastoma (GBM) is an incurable and highly heterogeneous brain tumor,originating from human neural stem/progenitor cells(hNSCs/hNPCs) years ahead of diagnosis.Despite extensive efforts to characterize hNSCs and end-stage GBM at bulk and single-cell levels,the de novo gliomagenic path from hNSCs is largely unknown due to technical difficulties in early-stage sampling and preclinical modeling.Here,we established two highly penetrant hNSC-derived malignant glioma models,which resemble the histopathology and transcriptional heterogeneity of human GBM.Integrating time-series analyses of whole-exome sequencing,bulk and single-cell RNA-seq,we reconstructed gliomagenic trajectories,and identified a persistent NSC-like population at all stages of tumorigenesis.Through trajectory analyses and lineage tracing,we showed that tumor progression is primarily driven by multi-step transcriptional reprogramming and fate-switches in the NSC-tike cells,which sequentially generate malignant heterogeneity and induce tumor phenotype transitions.We further uncovered stage-specific oncogenic cascades,and among the candidate genes we functionally validated C1QL1 as a new glioma-promoting factor.Importantly,the neurogenic-to-gliogenic switch in NSC-like cells marks an early stage characterized by a burst of oncogenic alterations,during which transient AP-1 inhibition is sufficient to inhibit gliomagenesis.Together,our results reveal previously undercharacterized molecular dynamics and fate choices driving de novo gliomagenesis from hNSCs,and provide a blueprint for potential early-stage treatment/diagnosis for GBM.