Shared molecular regulation of quiescence in neural and glioma stem cells reveals therapeutic vulnerabilities

Quiescent glioma cells evade chemoradiation and drive glioblastoma recurrence, but are hard to study due to their rarity and the difficulty of maintaining quiescence in vitro. Using single-cell RNA-sequencing and patient-derived glioma stem cell lines, we found that although glioma cells resist entering deep quiescence, they share conserved activation trajectories with neural stem cells. Exploiting this conservation to use a tractable neural progenitor line as a screening surrogate, we queried LINCS L1000 signatures across more than 3,000 compounds and validated mTOR, ATPase and histone lysine methyltransferase inhibitors as quiescence promoters in glioma stem cell lines. This record contains the single-cell RNA-sequencing data generated for this study, together with the analysis code. Five patient-derived glioblastoma stem cell lines (QBC382, QBC385, QBC391, QBC392, QBC395, passages 4-9) were cultured for three days in one of four conditions: control medium (bFGF + EGF), BMP4 16 ng/mL, BMP4 plus palbociclib 200 nM, or BMP4 plus leucettine L41 1 uM. Within each condition the five lines were pooled in equal numbers into one library, so every count matrix contains cells from all five donors, separated afterwards by genotype. Per-donor analysis therefore requires the demultiplexing tables provided here, and because scSplit numbers its clusters arbitrarily the cluster-to-line mapping differs between conditions. Included are cell-called count matrices (one barcodes/features/matrix triplet per condition), the raw scSplit output with the cluster-to-line mapping, and cell_metadata.csv giving the condition, donor line, QC metrics and genotype cluster assignment for every analysed cell. A README documents the full manifest and all columns. The record also contains the complete analysis code, in code/. This is the full pipeline behind every figure and supplementary table in the paper: quiescence-inference benchmarking, the glioma and neural stem cell trajectory analyses and their comparison, the treatment single-cell cohort, and the LINCS L1000 in silico screen. code/README.md is a rerun guide giving the software environment, the expected data layout, the run order with runtimes, per-script inputs and outputs, expected results, and the practical pitfalls encountered. The code is released under the MIT licence. Libraries were prepared with the DNBelab C Series High-throughput Single-cell RNA Library Preparation Set V2.0 (MGI) and sequenced on the MGI DNBSEQ platform; this is not 10x Genomics data. Reads were processed with dnbc4tools v2.1.1, SNPs called with FreeBayes v1.3, and donors separated with scSplit v1.0.9. Of the 23,408 cells called across the four libraries, 20,032 (85.6%) were retained for analysis after exclusion of cross-genotype multiplets and quality-control filtering. Raw FASTQ data are not deposited in any public or controlled-access repository: the participant consent under which the specimens were collected does not extend to deposition of individual-level genomic sequence data. Individual-level SNP genotypes used for demultiplexing are likewise not deposited, although the resulting per-cell donor assignments are released here. Requests for access to raw sequencing data will be considered by the sample custodian, subject to ethics approval and a Data Transfer Agreement. Third-party datasets reanalysed in this study (GSM5039270, GSE160930, GSE92742, and published data from Richards et al. and Couturier et al.) remain available from their original sources.