The quiescent cells that tumor organoid screens cannot see
An analysis of single-nucleus data from ten glioblastomas concludes that roughly nine in ten glioma stem cells are slow-cycling or quiescent before a single dose of therapy, a population confirmed in patient-derived organoids and known to drive recurrence after chemoradiation. The same study shows these cells share activation programs with healthy neural stem cells, that they resist entering deep quiescence, and that blocking TGF-beta receptor 1 wakes them up. For drug screening built on proliferation and viability readouts, the population that matters most is the one the assay is least able to measure.
Source: Shared molecular regulation of quiescence in neural and glioma stem cells reveals therapeutic vulnerabilities, bioRxiv preprint, version 2, 2026. Primary source. Read the full text including the version 2 corrections note, all six main figures, and methods.
What the work claims
This is a primary computational and experimental study from QIMR Berghofer that combines reanalysis of published single-nucleus glioblastoma data, new in vitro quiescence models in patient-derived glioma stem cells (GSCs), a small-molecule screen, and validation in patient-derived glioblastoma organoids. Its claims stack in order. First, quiescent or slow-cycling GSCs are not a rare therapy-induced remnant but the dominant state in untreated tumors: on average 92 percent of transcriptionally defined GSCs across ten patient tumors sat in a slow-cycling or quiescent state before any treatment.1 Second, the molecular program these cells use to reactivate is substantially conserved with adult mouse neural stem cells, meaning the decades of quiescence biology built in the normal brain should be transferable to the tumor. Third, GSCs are strikingly resistant to entering deep quiescence; their resting state is better described as shallow, slow-cycling dormancy. Fourth, TGF-beta receptor 1 inhibition forces these cells out of that state, which could make them vulnerable to standard-of-care treatment that preferentially kills dividing cells.
How the quiescence machinery compares
The stemness ranking that drives the first claim was done with CytoTRACE, a genome-wide measure of transcriptional diversity used as a marker of stemness potential, run per patient to avoid artifacts from mixing cell sizes. Reclustering only the GSC compartment separated actively cycling from quiescent cells, the latter validated by their reduced total mRNA, a known quiescence signature. In a patient-derived glioblastoma organoid grown from surgical tissue, 85 to 95 percent of cells expressing the stem cell markers HOPX, PTPRZ1, or SOX2 were likewise negative for the proliferation marker Ki67, independently supporting a predominantly resting stem compartment.1
For the conservation claim, the authors ordered mouse subventricular zone neural stem cells and human GSCs along pseudotime trajectories from deep quiescence to activation. The activation signatures overlapped about 2.1-fold beyond chance: 1,403 genes in mouse neural stem cells, 766 in human GSCs at strict significance. Shared genes were enriched for neuronal development programs; GSC-unique genes pointed instead to biosynthesis, chromosome segregation, and ERK signaling, consistent with a shallower arrest that keeps growth machinery idling rather than dismantled.
Functionally, the shallow-versus-deep distinction is the paper's most useful contribution. BMP4, the niche signal that maintains normal neural stem cell quiescence, reduced proliferation in GSC cultures only at concentrations roughly 160 times higher than those effective in mouse neural stem cells, with a 1.6-fold shallower dose-response slope. A low 200 nanomolar dose of the CDK4/6 inhibitor palbociclib induced a deeper arrest, about 60 percent reduction in proliferating cells, and both states were fully reversible within three days of washout. Quiescence conferred radiation protection, but in a patient-specific pattern: the most radiosensitive line was protected by both models, a second only by palbociclib, and a third by neither. In the small-molecule screen, the TGF-beta receptor 1 inhibitor LY-364947, a DYRK1A inhibitor, and a KAT6A inhibitor all blocked BMP4-driven quiescence entry; in the palbociclib model, TGF-beta receptor 1 and DYRK1A inhibition still worked but BMP pathway inhibition did not, arguing that the tumor produces its own TGF-beta ligands to maintain the resting state.
The organoid validation delivered the translational headline and its warning in the same experiment. In early patient-derived glioblastoma organoids, LY-364947 at 30 micromolar increased the fraction of proliferating SOX2-positive cells 1.85-fold, confirming that waking quiescent GSCs works in a three-dimensional tumor model. But the BMP receptor inhibitor that had worked in 2D did nothing in organoids, and the DYRK1A inhibitor leucettine L41, which had looked equally promising in 2D, was likewise inactive in 3D. Two-dimensional results did not survive the transition to organoids; the TGF-beta result did.
Where a skeptic should push
The 92 percent figure deserves scrutiny before it travels. It rests on transcriptional clustering of single-nucleus data from ten tumors, where quiescence was defined by the absence of cell-cycle gene expression. Quiescence-by-transcript is vulnerable to both directions of error: cells too sparse in mRNA to classify, and the well-known problem that Ki67 loss is not proof of a defined quiescent state, a limitation the authors themselves acknowledge when they validate their models by single-cell sequencing. The organoid corroboration, 85 to 95 percent Ki67-negative stem-marker-positive cells, comes from a single resection explant. A dominant resting compartment is plausible and consistent with prior lineage-tracing literature, but the precise proportion should be read as a strong estimate, not a measured constant across glioblastoma.
The image-integrity history of this preprint must be part of any reading. Version 1 contained a drug-treated organoid image in Figure 6D from which debris had been digitally removed; version 2 restores the unmodified image, re-reports the Figure 6E organoid data after re-imaging and blinded re-analysis with slightly changed effect sizes, and expands the methods to document exclusion, normalization, and blinding procedures. The stated conclusions are unchanged, and the corrected effect, a 1.85-fold increase in proliferating GSC fraction, survived re-analysis. But a preprint that has already crossed an image-manipulation line once warrants maximal skepticism toward every figure until peer review, and the honest framing is that the quantitative details of the organoid experiment rest on the corrected version alone.
Third, the therapeutic logic cuts both ways and the paper is candid about it. Forcing quiescent stem cells to proliferate so that chemoradiation can kill them is a coherent strategy, but reactivation is also the definition of relapse if the timing fails: a patient whose residual GSCs are pushed into cycle without concurrent effective therapy has been handed a regrowing tumor. The radiation-protection data already show the strategy is patient-line-specific, with one of three lines protected by neither quiescence model. And everything mechanistic here was anchored in mouse neural stem cell biology plus a handful of patient-derived lines, so the generality of the TGF-beta dependency across the glioblastoma population remains an open, testable question.
What this means for tumor organoid drug screens
The structural problem this paper exposes is that the standard tumor organoid screen and the biology of recurrence are misaligned by design. Most patient-derived organoid drug screening measures viability or proliferation after short compound exposure, which selects for the cycling fraction and is structurally blind to the resting stem cells that survive therapy and reseed the tumor. This study quantifies that blind spot: in its data the cells driving recurrence are not a residual minority but the overwhelming pre-treatment majority. A screen that reads out ATP after 72 hours can rank compounds by how well they kill the 8 percent of cells that are cycling, while the 92 percent that matter sit outside the measurement entirely. The non-obvious implication is that adding a quiescence axis to screening, for example parallel assays run with BMP4 or low-dose CDK4/6 inhibition, and scoring reactivation rather than just killing, converts a blind screen into one that can actually rank recurrence-relevant activity.
The 2D-to-3D result is the second lesson, and it lands on a workflow many programs still use: validating hits in 2D patient-derived cell cultures before committing to organoid work. In this study the DYRK1A inhibitor leucettine L41 looked clean in 2D and did nothing in glioblastoma organoids, while the BMP pathway inhibitor told the opposite story, active in 2D, inert in 3D. Only the TGF-beta receptor 1 inhibition replicated across both. If that ratio, one of three pathways surviving dimensional translation, is even roughly representative, then 2D-first screening pipelines are generating hit lists whose majority may be artifacts of the flat substrate, in either direction. Three-dimensional validation is not the expensive confirmation step; it is the experiment.
The opportunity and the threat are the same mechanism viewed from two timescales. If TGF-beta receptor 1 inhibition wakes quiescent GSCs, combining it with the Stupp chemoradiation regimen is a rational, testable sensitization strategy, made more actionable by the number of TGF-beta inhibitors already in clinical trials for solid cancers. The threat is that forced exit from dormancy, in the wrong sequence or the wrong patient, accelerates exactly the recurrence it aims to prevent, and the patient-line specificity of the radiation-protection data is a standing warning that one line's organoid result, the single most common evidence level in this field, cannot carry a treatment strategy. What is genuinely new here is not another resistance mechanism but a quantified argument about what screening architectures can and cannot see, and a concrete, mechanistically grounded way to build the assay that sees it.
The bottom line
Established: slow-cycling or quiescent GSCs dominate untreated glioblastoma in transcriptional data and in a patient-derived organoid; their reactivation program is partially conserved with adult neural stem cells; their BMP4-induced arrest is far shallower than normal stem cell quiescence; and TGF-beta receptor 1 inhibition reverses that arrest in 2D cultures and in glioblastoma organoids. Hypothesis: that waking quiescent GSCs with TGF-beta inhibition, timed with standard-of-care therapy, improves glioblastoma control in patients, and that the quiescence axis generalizes across tumors beyond the lines tested. What would confirm the strategy is patient-derived organoid panels, plural donors, showing TGF-beta inhibition sensitizes quiescent GSCs to chemoradiation rather than merely proliferating them; what would break it is patient-line heterogeneity of the kind already visible here, where one of three lines was protected by neither quiescence model, because that is the heterogeneity a single-line result can never see.
Frequently asked questions
Why do quiescent glioma stem cells matter if they are not dividing?
Because they are the cells that survive chemotherapy and radiation, which preferentially kill dividing cells, and then reactivate to regrow the tumor after treatment stops. A therapy that clears the cycling bulk but spares this compartment treats the measurement, not the disease.
How many glioblastoma stem cells are quiescent before treatment?
In this study's reanalysis of single-nucleus data from ten untreated tumors, an average of 92 percent of transcriptionally defined glioma stem cells were slow-cycling or quiescent, corroborated by 85 to 95 percent Ki67-negative stem-marker-positive cells in one patient-derived organoid. Treat the exact figure as an estimate tied to the classification method.
What does shallow quiescence mean?
Quiescence exists on a spectrum. Deeply quiescent cells need strong or prolonged signals to re-enter the cell cycle; shallow or slow-cycling cells stay alert and reactivate fast. Glioma stem cells resisted BMP4-driven deep quiescence at concentrations roughly 160 times higher than normal neural stem cells required, suggesting their resting state is shallow by design.
How does TGF-beta inhibition help?
The tumor appears to produce its own TGF-beta ligands to keep stem cells resting. Blocking the TGF-beta receptor 1 with LY-364947 drove quiescent glioma stem cells back into cycle in 2D cultures and raised the proliferating stem fraction 1.85-fold in glioblastoma organoids, a 30 micromolar exposure over three days, potentially sensitizing them to therapies that kill dividing cells.
Why did some 2D results fail in organoids?
Dimensionality changes biology. A DYRK1A inhibitor that reversed quiescence in 2D culture had no effect in organoids, and a BMP pathway inhibitor showed the reverse pattern. Only TGF-beta receptor 1 inhibition worked in both, illustrating why 2D hit lists need 3D confirmation before any therapeutic claim.
What happened with the preprint's corrected figures?
Version 1 of this preprint contained a digitally edited organoid image in Figure 6D, with debris removed from the drug-treated panel. Version 2 restores the unmodified image, re-reports the organoid data after re-imaging and blinded re-analysis with slightly changed effect sizes, and expands methods documentation. The core conclusions, including the 1.85-fold effect, are unchanged, but the episode is a reason for maximal figure-level skepticism until peer review.
References
- Choudhury C, Singleton M, Brauer S, Friess D, et al. Shared molecular regulation of quiescence in neural and glioma stem cells reveals therapeutic vulnerabilities. bioRxiv. 2026 (version 2). doi:10.1101/2025.01.22.634421. Accessed 2026-09-05.