Lab-grown glioblastoma stem cells reveal how tumors evolve treatment resistance

Glioblastoma remains one of the most lethal cancers in adults, and its near-universal recurrence after surgery, radiotherapy and temozolomide chemotherapy has long been blamed on a stubborn subpopulation of tumor cells with stem-like properties. A research team now reports a way to watch that resistance take shape in the laboratory rather than infer it from scarce matched tissue. Writing in EMBO Molecular Medicine, the group describes “resistant GSC families”: within-patient collections of glioblastoma stem-like cells in which temozolomide-selected, radiation-selected and treatment-naive cultures are all derived in parallel from the same primary tumor. The design lets researchers separate the fingerprints of chemotherapy from those of radiotherapy, something primary-versus-recurrent comparisons routinely confound, and it reproduces established mechanisms of acquired drug resistance while exposing less familiar ones.

The protocol starts in the operating room. The team collected near-complete glioblastoma cell populations removed with a surgical aspirator, processed them into single-cell suspensions and cultured them in an enriched stem-cell medium containing EGF, FGF2, PDGFBB and HGF, each at 20 ng/mL, a cocktail chosen to mirror growth factors abundant in the brain and tumor microenvironment and to preserve the original heterogeneity of the stem-like compartment. After roughly 20 days of recovery, the cultures were split three ways. One flask received temozolomide at 50 µM every 3 days for three cycles, a concentration the authors note corresponds to drug levels measured in patients’ blood plasma, which runs 5- to 10-fold higher than concentrations inside the tumor. A second flask was irradiated at 2 Gy per day for 5 consecutive days, mimicking the first week of clinical treatment and calculated by radiobiological modeling to kill about 98% of the average glioblastoma cell population. The third flask was left untouched as the treatment-naive control. Cultures were considered established after passage 10, and any family included at least one therapy-selected member.

Applying the protocol to 25 grade 4 gliomas with wild-type IDH1/2 genes, the researchers successfully derived nine resistant GSC families, seven containing all three member types and two containing only the control plus one selected culture. Two further tumors yielded only control cells, and 14 produced no stem-like culture at all, giving a derivation efficiency of 44% among the 11 cases where at least the control line took. Genetic profiling of the original tumors offered a clue to who succeeded: tumors yielding families under therapeutic pressure carried significantly more pathogenic alterations in AKT pathway genes, including PTEN biallelic inactivation and activating PIK3CA mutations, and showed a strong association with TP53 mutations, though the authors caution these associations require validation in larger cohorts. TERT promoter mutations were frequent at 88%, and tumor cell content estimated from their variant allele frequency ranged from 0.24% to 64.88%, with no link between higher tumor content and successful family derivation.

The chemotherapy arm of the study turned on MGMT, the O6-methylguanine-DNA methyltransferase enzyme that repairs temozolomide-induced DNA damage and stands as the major determinant of primary drug resistance. Tumors with unmethylated MGMT promoters, considered MGMT-negative, yielded temozolomide-selected cultures in 6 out of 6 cases, while only 2 of 5 MGMT-methylated tumors did so, a trend toward significance. As expected, control cells expressing MGMT resisted the selection schedule, with survival above 50% of vehicle-treated controls, whereas MGMT-negative controls were sensitive. Crucially, the MGMT status of the original tumor was preserved across all family members over multiple passages, suggesting the promoter methylation state is an epigenetic trait stably inherited through cell generations, consistent with the frequent concordance reported between matched primary and recurrent glioblastomas.

The most striking chemotherapy result came from the two MGMT-negative families whose selected cells developed true secondary resistance. In the GBM17 and GBM151 families, the temozolomide-selected cells remained MGMT-negative yet became markedly more resistant than their matched sensitive controls, in both survival curves and long-term clonogenic assays. Sequencing revealed private mutations undetectable in the original tumors or controls: biallelic inactivation of MSH6 in GBM17, and MSH6 biallelic plus MSH2 monoallelic inactivation in GBM151. These genes encode core components of the DNA mismatch repair machinery, whose loss prevents the futile repair cycles that normally trigger cell death after temozolomide exposure. A fluorescence-based host cell reactivation assay confirmed the functional consequence, showing impaired mismatch repair activity in both resistant lines. Both MSH6 variants had previously been reported in association with temozolomide treatment, and the predominantly C-to-T and G-to-A mutational spectrum supports a model of drug-induced mutagenesis followed by selection of repair-deficient clones, mirroring findings in patient tissues. Notably, the alterations did not produce microsatellite instability at establishment or later passages.

A third MGMT-negative family, GBM149, charted a different route. Its temozolomide-selected cells survived two rounds of selection yet remained fully drug-sensitive, recovering and regrowing after treatment without acquiring any mismatch repair mutation or reactivating MGMT. The team classifies this behavior as compatible with a drug-tolerant persister phenotype, a resilient state better characterized in cancers driven by targeted therapies than in glioblastoma. A TP53 mutation appeared in the selected cells but not in controls, and its spectrum was poorly compatible with temozolomide mutagenesis, leaving open whether a pre-existing subclone was favored during selection. Even MGMT-expressing, primarily resistant cells were not spared genomic consequences: cytogenetic analysis revealed marked karyotypic changes, including ploidy shifts and chromosomal aberrations, in selected cells relative to controls, with no recurrent alteration across families, indicating that even cells proficient at repairing temozolomide’s signature damage accumulate chromosomal-level remodeling under drug pressure.

The radiation arm told a different story. Six of ten control cultures were primarily radioresistant, with cell survival above 60% after a 2 Gy × 3 day challenge, and radiation-selected cultures derived from all six showed no further increase in resistance, maintaining the primary phenotype in survival and clonogenic assays. Of the four radiosensitive tumors, only one, GBM95, yielded a radiation-selected culture, and that line showed a significant though incomplete gain in radioresistance, qualifying as secondarily resistant. Genetic analysis found no new driver mutations or copy-number changes in any radiation-selected line compared with controls, consistent with evidence that clinical radiotherapy does not raise single-nucleotide variant burden. Instead, all radiation-selected lines accumulated chromosomal-level alterations, including ploidy variation and significantly more chromosomal aberrations than controls, though the analysis could not distinguish treatment-induced adaptation from selection of pre-existing repair-competent subclones.

Mechanistically, the secondarily resistant GBM95 line mounted a stronger DNA damage response than its matched control. Time-course experiments after a single 5 Gy dose showed sustained CHK2 phosphorylation, indicating heightened response activation, and faster resolution of γH2AX, the histone mark of DNA double-strand breaks, in both western blot and immunofluorescence, pointing to more efficient break repair. The line also induced p21, a p53-targeted cyclin-CDK inhibitor, more strongly and earlier, which translated into a more effective G2/M arrest: 6 hours after irradiation the selected cells had accumulated in G2 with a complete absence of M-phase cells, while controls were still progressing through G2/M and only matched that distribution at 24 hours. Across the radiation-selected panel, transcriptomic analysis reinforced the theme, with prominent enrichment of double-strand break repair, homologous recombination and checkpoint-control programs relative to controls.

Transcriptional profiling of all families showed that global subtype identity was largely stable under pressure. All Wang subtypes, classical, mesenchymal and proneural, were represented, most cultures comprised multiple cell states, and deconvolution estimates showed those proportions changed little after selection. Principal component analysis clustered cells primarily by patient identity and molecular subtype rather than treatment, meaning inter-patient heterogeneity dominated the transcriptomic landscape. Yet marker-level shifts did appear in secondarily resistant lines: temozolomide-resistant cells upregulated mesenchymal and neural-lineage markers while reducing the astrocytic marker GFAP, and the radioresistant GBM95 line moved the opposite way, shedding mesenchymal markers and gaining neuronal-lineage markers. Unbiased gene set enrichment analysis then uncovered what subtyping missed: both selected populations showed significant enrichment of DNA replication, cell-cycle progression, genome maintenance and mitotic chromosome segregation programs, with temozolomide-selected cells leaning toward spindle-kinetochore and centrosome organization and radiation-selected cells toward break repair and checkpoint control, while both showed reduced immune-inflammatory and microenvironment-interaction programs.

Finally, receptor tyrosine kinases emerged as a convergent adaptive layer. Flow cytometry of EGFR, FGFR1-4, PDGFRA-B and MET showed a trend toward increased surface expression in therapy-emergent cells, particularly radiation-selected ones, reaching statistical significance for MET, a receptor previously implicated in glioblastoma radioresistance. In secondarily resistant lines, EGFR, MET and PDGFRA were markedly more abundant at the cell surface or intracellularly, and in the GBM95 radiation-selected line the fraction of MET-positive cells jumped from 40.15% to 97.35%. Western blots confirmed the protein increases, which occurred without gene copy-number gains or elevated mRNA, pointing to post-transcriptional regulation. Functionally, the changes mattered: radiation-selected cells showed enhanced proliferative and viability responses to FGF2 and HGF, secondarily resistant cells responded more strongly to individual growth factors, and only the two families harboring RB1 mutations could proliferate without exogenous growth factors at all. The authors frame the platform as a controlled, within-patient system that captures durable genetic escape through mismatch repair loss, MGMT-independent drug tolerance, therapy-associated chromosomal change and convergent programs of genome maintenance and RTK remodeling, while acknowledging it lacks microenvironmental contributions from immune cells and normal brain tissue and complements rather than replaces primary-recurrent tissue profiling.

Subject of Research: An ex vivo platform modeling therapy-driven resistance evolution in glioblastoma stem-like cells from matched primary tumors.

Article Title: Ex vivo therapy-driven selection of glioblastoma stem-like cells models resistance evolution

Article References: Prelli, M., De Bacco, F., Casanova, E., Maniscalco, S., Biagioni, G., Reato, G., Mahmoudi, S., Calogero, R., Panero, M., Boasso, E., Casorzo, L., Crisafulli, G., Bartolini, A., Macagno, M., Nagel, Z. D., Bertero, L., Cassoni, P., Zeppa, P., Cofano, F., … Boccaccio, C. (2026). Ex vivo therapy-driven selection of glioblastoma stem-like cells models resistance evolution. EMBO Molecular Medicine. https://doi.org/10.1038/s44321-026-00547-w

Image Credits: AI Generated

DOI: 10.1038/s44321-026-00547-w

Keywords: glioblastoma, glioblastoma stem-like cells, temozolomide, ionizing radiation, MGMT, mismatch repair, drug tolerance, radioresistance, DNA damage response, receptor tyrosine kinases, resistance evolution, ex vivo model