PROTECT YOUR DNA WITH QUANTUM TECHNOLOGY
Orgo-Life the new way to the future Advertising by Adpathway-
Loading metrics
Open Access
Peer-reviewed
Research Article
- Emma Legait,
- Charlotte Rulquin,
- Lauranne Bouteille,
- Raphaël Clément,
- Cédric Maurange
x
- Published: November 19, 2026
- https://doi.org/10.1371/journal.pbio.3004086
Abstract
How cell-state transitions are regulated along a tumor lineage to determine its overall cellular composition remains unclear. Using a simple hierarchical tumor model in Drosophila, we show that differentiation of cancer stem cells (CSCs) into transit amplifying progenitors (TAPs) depends on the fraction of TAPs among their immediate neighbors. This relationship is captured by a local feedback transition rule that quantitatively predicts CSC transition probability. Implementing this rule in 3D growth simulations spontaneously recapitulates CSC clustering and the emergence of a stable CSC fraction within tumors. Mechanistically, the local feedback is mediated by EGFR signaling from newly produced TAPs, propagating differentiation across CSC clusters. Consistently, we show that clustering protects the CSC pool from depletion driven by surrounding TAPs, indicating that spatial architecture and lineage dynamics are tightly coupled. Together, these findings show that a local feedback governing fate transitions drives emergent spatial organization that stabilizes the stem cell fraction in hierarchical tumors.
Citation: Legait E, Rulquin C, Bouteille L, Clément R, Maurange C (2026) Local feedback governs cell fate transition to drive self-organized phenotypic equilibrium in hierarchical tumors. PLoS Biol 24(11): e3004086. https://doi.org/10.1371/journal.pbio.3004086
Academic Editor: François Schweisguth, Institut Pasteur, FRANCE
Received: April 29, 2026; Accepted: October 16, 2026; Published: November 19, 2026
Copyright: © 2026 Legait et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Codes and underlying data generated for this study are available at: https://data.mendeley.com/datasets/w9bwgvxfrj. Single-cell data used in this study were already published [9] and are available at NCBI Gene Expression Omnibus ID GSE114986.
Funding: The salaries of EL and CR were supported by the Turing Center for Living Systems (CENTURI - https://centuri-livingsystems.org/), funded by France 2030, the French Government programme managed by the French National Research Agency (ANR-16-CONV-0001 - https://anr.fr/en/). The salary of EL was also funded by Fondation pour la Recherche Médicale (FRM - https://www.frm.org/en). The salary of LB was funded by la Ligue Contre le Cancer (Ligue 2020 “Enfants, adolescents, et cancer”). CM was supported by funding from La Ligue Contre le Cancer (Equipe Labellisée EL2019 and EL2025 - https://www.ligue-cancer.net/), and a government grant managed by the Agence Nationale de la Recherche under the France 2030 program, with the reference number ANR-24-EXCI-0001, ANR-24-EXCI-0002, ANR-24-EXCI-0003, ANR-24-EXCI-0004, ANR-24-EXCI-0005. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Abbreviations: CSC, cancer stem cell; NPC, non-proliferative cell; NSC, neural stem cell; PBS, phosphate-buffered saline; ROIs, regions of interest; RT, room temperature; RTKs, receptor tyrosine kinases; TAP, transit-amplifying progenitor; UAS, upstream activation sequences; UTRs, untranslated regions; VNC, ventral nerve cord
Introduction
Hierarchical tumors are sustained by cancer stem cells (CSCs), which can either self-renew or commit to a transit-amplifying progenitor (TAP) state with limited proliferative potential [1–4]. CSCs are known to be resistant to treatment and responsible for relapse. It is therefore essential to understand how they are regulated. A central unresolved question is how tumors robustly establish and maintain stable proportions of CSCs and differentiated progeny despite continuous growth, a phenomenon often referred to as phenotypic equilibrium [5–7].
Depending on tumor type and stage, the lineage hierarchy may be either rigid or exhibit varying degrees of plasticity. In both strict and plastic hierarchies, this equilibrium can be recapitulated by models invoking stochastic transitions between cell states [6,8–10]. However, the mechanisms that govern state transitions along a hierarchical lineage and determine the CSC fraction remain poorly understood. In particular, it is unclear whether the CSC-to-TAP transition rate is primarily encoded by cell-intrinsic programs or instead emerges from self-organizing properties of collective cell behaviors driven by local cell-cell interactions [11,12].
Dissecting these principles in vivo is challenging: human tumors generally coexist with diverse stromal, immune, and vascular cells, which create complex chemical and mechanical environments. In contrast, in vitro systems can be excessively reductionist and lack the three-dimensional and physiological context required to fully capture local cell-cell interactions [6,13]. On the other hand, non-mammalian model organisms can be used to generate genetically tractable tumors with reduced complexity while preserving three-dimensional architecture. Such minimal in vivo models enable quantitative mapping of cell states and their interactions, and may help formalize rules governing lineage transitions and their effects on tumor organization and composition.
Drosophila neural stem cell (NSC) tumors provide such a framework. During normal development, asymmetrically dividing NSCs, also known as neuroblasts, progress through well-defined temporal identity windows progressively limiting their proliferative potential. Early perturbations of asymmetric division or differentiation trigger NSC amplification and sustained proliferation, producing tumors that persist into adulthood [14,15]. These tumors are driven by co-option of the temporal-patterning program and are strictly hierarchical: a subset of cells, behaving as CSCs, retains expression of early temporal genes that act as an oncofetal-like module to sustain self-renewal. However, CSCs can also transition to a more differentiated, TAP-like state defined by the expression of later temporal genes (Figs 1A and 1B) [9,16]. As such, this system offers a powerful in vivo model to study developmental tumors with low mutational burden and highly constrained hierarchy as observed in pediatric cancers [17–21]. Drosophila brain tumors lack infiltration by stromal or immune cells, providing a minimal framework to dissect mechanisms specific to the tumor lineage [22]. Notably, tumor composition rapidly converges to phenotypic equilibrium in which CSCs form a stable minority of cells [9]. Moreover, CSCs appear spatially clustered rather than randomly distributed. In this context, it remains unclear whether spatial organization and cell fate choices are mechanistically linked.
Fig 1. Spatial organization of CSCs within in vivo and in silico models of hierarchical tumor.
(A) Asymmetrically dividing Drosophila NSCs transit through three main temporal windows during development defined by Chinmo/Imp (red), Syp (dark blue) and E93 (light blue), respectively. NSCs ultimately differentiate during metamorphosis (pupa). Knockdown of pros (pros-RNAi) during the Chinmo/Imp temporal window prevents the differentiation of intermediate progenitors (cells with a purple nucleus), inducing NSC amplification and tumors that persist in adults. Tumors are composed of NSC-like cells recapitulating the different temporal identities observed during development. (B) Tumors exhibit a strict hierarchy with Chinmo+ Imp+ NSC-like cells behaving as CSCs (red cells), Syp+ NSC-like cells behaving as TAPs (dark blue cells) and Syp+ E93+ cells behaving as NPCs (light blue). is the transition rate of CSCs, and
is the cell cycle exit rate of TAPs. (C) A mCherry reporter transgene reflecting chinmo expression (mChchinmo) (see description in Materials and methods section) is used to label CSCs in tumors of 5-day-old adults. Dapi labels the nucleus of all tumor cells. TAPs + NPCs are labelled with an anti-Syp antibody. Note the clustered organization of CSCs. The right panels are a magnification of the white dotted square on the left. (D) 3D time-lapse and Z-stack of a tumor growth simulation. (E) Plotted proportions of CSCs and TAPs + NPCs over time and number of cells over time (inset) in a simulation with intrinsically regulated CSC-to-TAP transition. (F) Qualitative description of the segregation index µ. (G) CSC segregation index distribution in vivo, and in silico after simulations with intrinsically regulated CSC-to-TAP transition (around n = 50,000 cells for both conditions). Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
Here, we leverage 3D segmentation to map CSC-to-TAP transitions at single-cell resolution in entire tumors. We find that CSC differentiation is determined locally by the fraction of TAP neighbors, revealing a quantitative, feedback-based transition rule. Implementing this local rule in 3D simulations spontaneously recapitulates system-level features, such as CSC clustering and a stable CSC fraction. Mechanistically, we show that the transition rule is mediated by EGFR signaling originating from TAPs which propagates differentiation throughout CSC clusters. By contrast, maintenance of CSC identity is promoted by an EGFR inhibitor (Kekkon-1) expressed in CSCs, but also by the spatial segregation of CSCs that emerges from the transition rule and shields them from differentiation signals.
Our results reveal that a local feedback between tumor cells couples transition dynamics and spatial tumor organization, unveiling the self-organizing principles that allow phenotypic equilibrium in hierarchical tumors.
Results
A uniform transition rate recapitulates phenotypic equilibrium but not spatial organization
We used a well-characterized Drosophila model of NSC-derived tumors with a developmental origin [9,15,23]. In this model, the knockdown of the transcription factor Prospero (Pros) by RNAi in six larval NSCs (poxn>pros-RNAi) blocks differentiation of NSC progeny generated by asymmetric division, leading to NSC amplification and sustained tumor growth in adults, until death around 10–12 days after eclosion (Figs 1A and S1A–S1C). These tumors follow a rigid hierarchy (Fig 1B) [9]. CSCs express high levels of the transcription factor Chinmo and the RNA-binding protein Imp, the ortholog of mammalian IGF2 BP. Both genes are normally expressed during early larval stages where they promote self-renewal [15,24]. TAPs express the RNA-binding protein Syncrip (Syp), a late larval temporal gene known to post-transcriptionally repress Imp and Chinmo [9,25,26]. Upon Syp expression, TAPs rapidly transition to a non-proliferative cell (NPC) state that expresses high levels of the transcription factor Eip93f (E93), normally expressed in late larval and pupal NSCs (Fig 1C) [9]. Thus, the cell state hierarchy in the tumor mirrors the temporal gene expression program operating in NSCs during development. However, whereas the mechanisms regulating temporal transitions in developing NSCs are relatively well understood and involve both cell-intrinsic and hormonal cues [9,15,16,27], the mechanisms controlling cell-state transitions along the tumor hierarchy remain unknown.
In adults, tumors reproducibly stabilize at ~15% to 20% CSCs, and previous clonal analyses yielded precise measurements of the tumor-average CSC-to-TAP transition rate and TAP cell-cycle exit rate [9]. Because these tumors, although being surrounded by glia, are largely devoid of non-tumor infiltrates, we reasoned that the CSC-to-TAP transition rate is primarily governed by tumor cell-derived mechanisms. These could operate either through a cell-intrinsic mechanism, resulting in a uniform transition rate across CSCs, or through an extrinsic mechanism, whereby the transition rate depends on the local cellular environment.
To test the cell-intrinsic scenario, we designed 3D particle-based simulations of tumor growth. Cells are represented as spheres in a densely packed tissue-like environment. They can divide, differentiate, and exit the cell cycle with prescribed rates. We first used a uniform, cell-intrinsic transition rate for all CSCs, using parameters constrained by our previous clonal measurements (see Materials and methods) [9]. CSCs divided every 1.6 days, TAPs every 1.3 days, CSC-to-TAP transitions occurred at rate , and TAPs exited the cell-cycle at rate
. Starting from a small founder population arranged as in early larval tumorigenesis, the system converged to stable cell-type proportions closely matching those observed in vivo, with ~20% CSCs and ~80% non-CSCs (TAPs + NPCs, hereafter simply called TAPs) (Figs 1D and 1E).
However, CSCs appeared randomly distributed in silico (Fig 1D; S1 and S2 Movies), contrasting with the spatial segregation observed in vivo (Fig 1C). To quantify this difference, we defined for each CSC a normalized segregation index, , where
indicates exclusive contact with CSCs,
exclusive contact with TAPs, and
the neighborhood expected from a random mixture matching the tumor-scale proportions (Fig 1F). Automated 3D segmentation of full tumors in 5-day-old adults revealed strong segregation in vivo, whereas simulations under the intrinsic model yielded segregation values close to random (Fig 1G).
The simulations demonstrate that a uniform, cell-intrinsic transition rate can recapitulate stable heterogeneity, but fails to account for spatial segregation between cell states. This suggests that CSC fate is not hard-wired, but might instead depend on local cellular context. We therefore set out to assess whether the CSC-to-TAP transition is spatially biased in vivo.
In vivo cell state maps reveal a local feedback transition rule
To that end, we first sought to identify and locate cells undergoing the CSC-to-TAP transition. During normal development, NSCs progress from an early Chinmo+ Imp+ state to a later Syp+ state during larval stages [9,27]. We found that Chinmo is downregulated slightly earlier than Imp (Fig 2A). We therefore reasoned that transitioning cells in tumors should have lost Chinmo while retaining low Imp. Consistently, we detected a tumor cell population that was Chinmo− and Implow and Syplow, thus an intermediate state between CSCs and TAPs (Fig 2B). We refer to these cells as transitioning TAPs (tTAPs). Taking advantage of this observation, we used 3D segmentation on confocal stacks coupled to fluorescence quantification to systematically locate CSCs, tTAPs and TAPs in 5-day-old adult tumors. We used Dapi to segment nuclei, and an mCherry-based chinmo reporter (mChchinmo) [9,23] and Imp immunostaining to determine cell state. Based on these markers, we could categorize all cells in three possible states: (1) Imp+ Chinmo+ cells as CSCs, (2) Imp+ Chinmo− cells as tTAPs, and (3) Imp− Chinmo− cells as TAPs (Figs 2C and S1C–S1D; S3 Movie).
Fig 2. Cell state maps and transition rule.
(A) Developmental time course of Chinmo (anti-Chinmo antibody) and Imp expression pattern in NSCs (labelled with Miranda (Mira)) during larval stages revealing a transient Chinmo− Imp+ state in early L3 (eL3). (B) A subset of tumor cells (examples delineated by the white dotted lines) exhibits an absence of Chinmo, but low expression of both Imp and Syp suggesting that they are undergoing the CSC-to-TAP transition. (C) Segmentation and categorization of cell states: CSC, tTAP, TAP based on mChchinmo, Imp and Dapi stainings. (D) Definition of the TAP score s. (E) 2D sections of the 3D cartography of cell states and TAP scores. (F) TAP score distribution in CSCs and tTAPs. tTAPs have on average more TAP neighbors than CSCs. (G) Probability for a CSC to be undergoing transition as a function of its TAP score. This defines a sigmoid-like transition rule (data pooled from n > 200k cells from 10 different tumors). Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
To assess whether transition was spatially biased, we quantified the local 3D neighborhood of each cell in our tumors using a normalized TAP score, , that measures the fraction of neighbors in the TAP state (Fig 2D). Combining the 3D cell state maps and TAP score maps (Fig 2E) revealed a clear spatial bias: tTAPs were enriched in TAP-rich neighborhoods and had higher TAP scores than CSCs, placing them preferentially at the periphery of CSC clusters (Fig 2F). Thus, CSC-to-TAP transitions do not occur uniformly throughout the tumor, but are favored in CSCs adjacent to differentiated progeny. Pooling data from hundreds of thousands of cells, we quantified the probability
for a CSC to be undergoing transition as a function of the fraction of neighboring TAPs. We found that CSC differentiation follows a simple quantitative rule determined by local neighborhood composition (Fig 2G). CSCs surrounded exclusively by other CSCs (
) displayed an almost null transition probability. CSCs located at the periphery of clusters, in contact with both CSCs and TAPs (
), exhibited intermediate transition probabilities. In contrast, isolated CSCs surrounded entirely by TAPs (
) differentiated with near-certainty.
Together, these measurements reveal a local feedback transition rule. The CSC-to-TAP transition probability is not uniform, but locally determined by the fraction of TAPs in their immediate neighborhood. Interestingly, the sigmoidal relationship mirrors classical ligand-receptor dose-response curves governed by Hill dynamics [28], suggesting a possible receptor-mediated local control of CSC fate. Thus, the transition rate is not intrinsic but instead emerges as a predictable function of adjacent cell states.
The local feedback transition rule drives emergent stable heterogeneity and segregation
To investigate the tumor-level implications of the local feedback transition rule, we implemented it directly into our simulation framework. Experimentally, static images provided the probability of observing a CSC in the—transient—Chinmo⁻ Imp⁺ transition state at a given TAP score
. This occupancy reflects the rate at which CSCs enter the transition state,
, together with the average time
they spend in it. We used Little’s law to relate the experimentally observed occupancy of the transient state
to its entry rate:
. Because
is not known a priori, we explored a range of transition rate amplitudes corresponding to different values of
(Fig 3A). Strikingly, simulations implementing the local transition rule spontaneously generated stable coexistence between CSCs and TAPs. Lower transition rates preserved a persistent CSC population, whereas higher transition rates led to progressive CSC depletion (Fig 3B). These differences also affected tumor growth dynamics (Fig 3C), consistent with the dominant contribution of CSCs to tumor expansion [9,15,23]. Spatial segregation between CSCs and TAPs also emerged spontaneously, as a direct consequence of the local transition rule (Fig 3C). The experimentally observed CSC fraction (15%–20%) was reproduced for
(Fig 3D; S4 Movie). This prediction provides a stringent quantitative test of the model. Indeed, the tumor-average transition rate implied by the rule can be calculated as:
Fig 3. The local feedback transition rule drives emergent stable heterogeneity and segregation.
(A) Fit of the experimental transition rule for implementation in the numerical model (top) and resulting transition rates assuming different values of the transition time (bottom). (B) Final proportion of CSCs in simulations (n = 10 per condition) using the transition rates shown in (A). The proportion of CSCs decreases when the average transition rate increases. Color codes as in (A). (C) Total tumor cell number over time (top) and segregation index distribution comparison (bottom). Color codes as in (A). No segregation index is plotted for
as this condition has no sustainable CSC pool. (D) For a transition rate corresponding to
, the system reaches stable heterogeneity and cell type segregation with
15% CSCs. Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
where is the probability distribution for a CSC to have score
. Using the experimentally measured distributions
and
(Fig 2F and 2G) and
yields
. Remarkably, this value closely matches the average transition rate independently measured by clonal analysis [9],
. We therefore used
in following simulations.
Together, these new simulations show that the local feedback transition rule identified in vivo is sufficient to spontaneously generate tumor-scale stable heterogeneity and spatial segregation. They further predict that tumor composition should be highly sensitive to parameters affecting the transition rate or the spatial organization of cell types within the tumor.
EGFR signaling mediates the local feedback controlling CSC transitions
Because CSC-to-TAP transitions occur at CSC-cluster boundaries and follow a Hill-like rule, we hypothesized that TAPs might provide a short-range differentiation cue. From single-cell data obtained from tumors in adults [9], we prioritized Notch, Wnt, and EGFR as candidate local pathways (S2A Fig), but found that only perturbations of the EGFR pathway (Figs 4A, 4B and S2B) significantly altered the CSC fraction. Depletion of Spitz (the most highly expressed EGFR ligand in tumors (S2C Fig)) strongly expanded the CSC pool, and EGFR or Erk knockdown phenocopied this effect, whereas constitutively active EGFR reduced CSC abundance (Figs 4B and S2D). Importantly, single-cell RNA-seq revealed enrichment of EGFR expression in CSCs, consistent with the idea that they represent the target population of this signaling pathway (S2C Fig). ERK phosphorylation (P-Erk), used as a readout of EGFR pathway activity [29], was enriched in the cytoplasm of a subset of cells with tTAP characteristics. Indeed, P-Erk+ cells expressed low levels of Imp and were located at the boundary of CSC clusters (Figs 4C and S2E), with a TAP score similar to tTAPs (Fig 4D). Moreover, P-Erk decreased upon spitz knockdown in the tumor (Figs 4E and S2F). Finally, other receptor tyrosine kinases (RTKs) potentially able to promote Erk phosphorylation such as breathless (btl), heartless (htl), PDGF/VEGF receptor (Pvr), anaplastic lymphoma kinase (Alk), torso (tor) and sevenless (sev) were either not detected or at much lower levels by single-cell RNA-seq (S2C Fig), further supporting the idea that EGFR signaling is the main driver of Erk phosphorylation in these tumors.
Fig 4. TAP-mediated EGFR signaling promotes CSC differentiation.
(A) Schematic of the EGFR signaling pathway. (B) CSC proportions in control, rho-RNAi, spitz-RNAi, EGFR-RNAi, erk-RNAi, and constitutively active Egfr (Egfr-CA) tumor conditions. (C) Confocal section of a control tumor showing P-Erk staining. Nuclei are labeled with Dapi. Arrows indicate examples of cytoplasmic P-Erk signal in transitioning cells (tTAPs), identified by absence of Chinmo reporter (mChchinmo) and low Imp::GFP expression. (D) Distribution of TAP scores for CSCs, tTAPs, P-Erk+ tTAPs, Rho+ tTAPs and TAPs. (E) Fraction of P-Erk+ cells in control and spitz-RNAi tumors (left), and fraction of rho-lacZ+ cells in control tumors (right). (F) Confocal section of a Syp-RNAi tumor, composed exclusively of CSCs (mChchinmo+), showing strongly reduced P-Erk signal. (G) Confocal section of a control tumor showing a P-Erk+ tTAP with activity of the rho-lacZ reporter gene. (H) Single-cell RNA-seq feature plots showing expression of Imp, Eip93F, and kek1, with corresponding distributions in CSC and TAP populations. (I) Fraction of P-Erk+ cells in control, kek1-RNAi, and kek1 overexpression (OE) tumors. (J) CSC proportions in control tumors, and tumors mis-expressing either of two independent kek1-RNAi transgenes or a kek1 OE transgene. (K) Tumor volumes in 9-day-old adults in control and kek1-RNAi conditions. (L) Model of a TAP-mediated feedback, involving EGFR signaling, driving the CSC-to-TAP transition along the tumor lineage. Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
To identify the source of EGFR ligand, we forced tumors to be only composed of CSCs, lacking TAPs, by knocking down Syp (Syp-RNAi). In such tumors, P-Erk became barely detectable, indicating that the paracrine input is mainly supplied by committed tumor cells (Figs 4F and S2G). Importantly, in control tumors, reporter activity detected expression of rhomboid (rho), a constitutively active protease required for Spitz secretion [30], in a small fraction of P-Erk+ cells (Figs 4E, 4G and S2H) exhibiting intermediate TAP scores, between TAPs and tTAPs (Fig 4D). Thus, these results support a model in which tTAP-derived Spitz activates EGFR–MAPK signaling to drive CSC-to-TAP transition. The scarcity of Rho+ tTAPs suggests transient local Spitz production after CSC-to-TAP commitment. This phenomenon is reminiscent of relay mechanisms propagating EGFR-mediated differentiation fronts in developmental systems [31–34].
We next asked how CSCs buffer this differentiation pressure to maintain a stable fraction. Simulations indicated strong sensitivity to the effective transition rate , prompting us to search for CSC-enriched modulators of EGFR signaling. Among potential candidates, single-cell RNA-seq profiles identified the canonical EGFR inhibitors Sprouty (Sty) and Kekkon-1 (kek1) [35–38] as being significantly expressed in tumors (Figs 4H and S2C). We decided to focus on the LRIG1 ortholog Kek1, its mRNA being strongly enriched in CSCs. Although endogenous Kek1 protein was low (S3A Fig), forced expression revealed preferential accumulation in CSCs, suggesting cell-state-dependent stabilization (S3A and S3B Figs). Functionally, kek1 knockdown broadly increased P-Erk, while kek1 overexpression nearly abolished ERK activation (Figs 4I and S3C) consistent with a negative-feedback role on EGFR signaling. Importantly, kek1 depletion largely reduced CSC abundance, whereas overexpression expanded it (Figs 4J and S3D). Moreover, promoting commitment through kek1 depletion reduced tumor size in adults, indicating that shifting fate balance toward differentiation depletes the CSC pool and ultimately limits growth (Fig 4K).
Notably, ERK phosphorylation was not detected in normal larval NSCs at the time of the Imp-to-Syp transition (S4A Fig), and knockdown of spitz or kek1 in normal NSCs did not block or delay this transition as the number of Syp+ progeny remained unchanged in all conditions (S4B and S4C Figs). Thus, EGFR-mediated induction of the Imp-to-Syp transition appears to be a tumor-acquired feature. In addition, ERK phosphorylation was not detected in larval tumors suggesting that EGFR signaling is initiated during later stages in more mature tumors (S4D Fig).
Together, these results support a self-organizing feedback mechanism arising in late tumors, in which newly generated TAPs provide short-range Spitz to promote and propagate CSC-to-TAP transitions, while CSC-intrinsic Kek1 dampens EGFR responsiveness (Fig 4L). The balance between paracrine activation (Spitz/Rho/EGFR) and intrinsic inhibition (Kek1) thus tunes the local transition rate and thereby sets the steady-state CSC fraction, and ultimately the tumor growth rate.
Spatial self-organization regulates CSC differentiation dynamics
The local feedback transition rule predicts that CSC differentiation is strongly influenced by spatial organization within the tumor. Because the local transition rate increases with the fraction of neighboring TAPs, the spatial distribution of cell types should determine the tumor-average transition dynamics. In particular, increased clustering should shield CSCs from TAP-derived differentiation signals, thereby reducing the average transition rate and stabilizing an increased CSC fraction. To test this in silico, we selectively increased homotypic CSC adhesion with respect to homotypic TAP adhesion
and heterotypic
adhesion, with
and
(see Materials and methods). As anticipated, this led to a strong increase in the steady-state CSC fraction (Fig 5A) and in overall segregation and tumor growth (Fig 5B and S5 Movie).
Fig 5. Emergent CSC clusters promote CSC maintenance.
(A) Final proportion of CSCs in simulations (n = 10 per condition) with differential adhesion (). The proportion of CSCs increases when
increases. (B) Total tumor cell number over time (top) and segregation index distribution comparison (bottom). Color codes as in (A). (C) Schematic of the Toll-1Ex condition. (D) Confocal section of the Toll-1Ex tumor condition with tumor cells labelled with Mira, and CSCs labelled with the mChchinmo reporter transgene. Toll-1Ex expression is reflected by Venus detection. Lower panels are enlargements of the white dotted box. The fusion protein Toll-1Ex::Venus is enriched at the interface between adjacent CSCs demonstrating homotypic interactions. (E) CSC clusters (detected with the mChchinmo reporter) in control tumors and tumors mis-expressing Toll-1Ex. (F) Quantification of CSC segregation in Control and Toll-1Ex conditions (around n = 60,000 cells for both conditions). (G) Quantification of the CSC proportion in Control and Toll-1Ex conditions. (H) Quantification of the proportion of P-Erk+ cells in Control and Toll-1Ex condition. Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
To test this prediction in vivo, we engineered a Drosophila line in which CSCs specifically overexpress a truncated form of the adhesion protein Toll-1 (lacking the intracellular signaling domain) fused with the Venus GFP variant (Toll-1Ex::Venus) (Fig 5C) (see description in Materials and methods). Toll-1 has previously been shown to sharpen compartment boundaries in Drosophila developing tissues by increasing cell–cell adhesion through homotypic interactions [39]. Since Toll-1 is not expressed in tumors (not detected in single-cell transcriptomic data produced from former studies [9]), this strategy enables us, in principle, to selectively increase adhesive interactions among CSCs without inducing ectopic Toll-1 signaling (Fig 5D). Consistently, we observed an expression pattern of the transgene restricted to CSCs, and enrichment of Toll-1Ex::Venus at the interface between adjacent CSCs, compared to the CSC/TAP interface. This confirms that homotypic interactions stabilize Toll-1Ex at the membrane, as previously described [39].
As anticipated, we found that CSC segregation was increased in these tumors (Figs 5E and 5F), consistent with increased adhesion. Strikingly, this was accompanied by a significant increase of the CSC fraction (Fig 5G), associated with a reduced proportion of P-Erk+ tTAPs compared to control tumors (Fig 5H). These findings validate the model predictions: modifying tumor spatial organization—here by specifically increasing inter-CSC adhesion—changes the average neighborhood composition and therefore alters the differentiation dynamics by shifting the local transition landscape, leading to an expanded pool of CSCs.
Together, these results demonstrate that spatial segregation directly promotes CSC maintenance.
Discussion
Tumors develop within complex ecosystems shaped by immune, stromal, and vascular signals, making it difficult to extract general principles governing tumor organization. Using a reductionist in vivo model, we identify a lineage-intrinsic mechanism by which local interactions between CSCs and their progeny determine CSC fate, drive spatial self-organization, and stabilize tumor heterogeneity. Rather than being encoded at the single-cell level, the maintenance of a stable CSC fraction emerges from a local feedback that couples cell fate decisions to neighborhood composition. The strictly hierarchical nature of our tumor model shows that plasticity of cell-state transitions along the lineage is not required to reach phenotypic equilibrium.
This equilibrium is governed by a balance between short-ranged differentiation signals (EGF) produced by newly differentiated cells and CSC-intrinsic inhibitory signals (Kek1) that sets the differentiation rate. In addition, this local feedback is sufficient to generate emergent spatial organization which acts as another regulatory layer that controls differentiation dynamics by governing neighborhood composition. Simulations were essential to reach these conclusions. First, they showed that a uniform transition rate can match cell proportions without matching spatial organization, providing a drive for experiments seeking to establish the influence of surrounding cells on transition. Second, that local differentiation signals from CSC progeny are sufficient to generate both stable proportions and segregation. Finally, that modifying the average neighborhood of CSCs by increasing CSC homotypic adhesion should increase the CSC fraction—a prediction validated experimentally using Toll-1Ex. Because the model is inferred from static images, it would benefit from dynamic measurements obtained by live imaging. However, long-term ex vivo imaging of Drosophila tumors remains technically challenging, limiting our ability to directly resolve transient pathway activation and differentiation dynamics. Such data would be particularly valuable for incorporating dynamics of the differentiation signal, notably rho expression and EGFR pathway activation, which our model captures only qualitatively by assuming that all TAPs contribute equally.
Coupling between local cell-cell interactions and global cellular composition parallels mechanisms governing normal stem cell compartments, indicating that tumors exploit conserved principles of tissue self-organization to preserve stemness under dynamic conditions [32,40,41]. This study highlights the importance of understanding tumors as highly self-organized systems to be able to control or eliminate their CSC fraction. Interestingly, recent live imaging of a 2D in vitro breast cancer cell-line model implicated cell–cell communication in self-organization and phenotypic equilibrium [13]. Remarkably, even this rudimentary in vitro context gave rise to CSC segregation, illustrating the robustness and conservation of this emergent spatial property of hierarchical tumors across evolution and diverse microenvironmental contexts.
Our study also sheds light on how developmental programs are co-opted during tumor initiation. During Drosophila development, the Imp-to-Syp temporal transition is precisely scheduled in larval NSCs by the coordinated action of an intrinsic timing mechanism and hormonal cues [20,27,42]. However, these mechanisms do not appear to operate in tumors. Instead, we show that the EGFR pathway is specifically activated in tumors after larval stages to promote the Imp-to-Syp transition. How such signaling modules arise during tumorigenesis and rewire developmental trajectories to regulate phenotypic heterogeneity and proliferation remains a central question in pediatric cancer biology.
Our finding that EGFR signaling promotes CSC differentiation in this tumor model challenges the prevailing paradigm in the cancer field that EGFR is uniformly associated with proliferation, survival, and tumor maintenance. However, pro-differentiation roles of EGFR signaling are well-established during development or after seizure in the mammalian brain too [31,43,44]. This duality underscores the context-dependent nature of EGFR signaling, where outcomes are shaped by cellular/developmental context, signal intensity, and cross-regulation with other pathways, all of which are subject in tumors to ongoing evolution driven by accumulating mutations and therapeutic pressure. Our observations prompt a deeper investigation into the structural and regulatory roles of EGFR signaling in tumor architecture and cell state dynamics at the different stages of tumor progression.
Interestingly, it has recently been shown in Drosophila cell culture, that EGFR/ERK signaling can directly phosphorylate the oncofetal gene Imp to disassemble the cytoplasmic condensates that they form with their target mRNAs [45]. This could lead to inhibition of Imp activity and subsequent chinmo silencing, providing a potential post-translational mechanism by which transient EGFR/ERK signaling triggers the CSC-to-TAP transition. It remains unclear whether EGFR signaling is the only differentiation cue in our Drosophila tumor model. The perdurance of some P-Erk levels upon EGFR knockdown associated with an incomplete blockage of the CSC-to-TAP transition suggests either partial inactivation or that other RTK-mediated signals may act redundantly.
Our work establishes a formal framework in an experimentally tractable hierarchical tumor, and it suggests testable principles—local progeny-to-CSC feedback, CSC buffering of differentiation signals, spatial shielding—that can be examined in other systems. Capturing the full complexity of mechanisms that regulate phenotypic equilibrium will also require future models to account for additional micro-environmental or mechanical inputs, which are likely to modulate spatial organization in more complex 3D tumors. Linking single-cell state transitions to emergent tumor-scale spatial organization provides a means to understand how tumor composition and growth are regulated, and may reveal opportunities to control them by steering the feedback that sets the balance between self-renewal and differentiation.
Materials and methods
Key resources (Table 1).
Fly culture
Drosophila lines were kept at 18°C on standard medium (8% cornmeal, 8% yeast, 1% agar).
Fly lines
The GAL4 line used is:
poxn-GAL4 (active in six homologous thoracic NSCs of the ventral nerve cord (VNC) (Bloomington Stock Centre #66685)) [53].
The UAS lines used are:
UAS-mCherrychinmoUTRs [46]: The mCherry coding sequence is flanked by the 5′ and 3′ UTRs of chinmo mRNA reflecting the post-transcriptional regulation of endogenous chinmo mRNA. Consistently, we observe that mCherry always overlaps with endogenous Chinmo as detected by immunostaining [9,23].
We used this approach to generate UAS-Toll-1Ex::VenuschinmoUTRS transgenic line allowing specific expression of Toll-1Ex::Venus in CSCs.
Other stocks: UAS-prosRNAi1 (Bloomington Stock Centre #26745), UAS-dicer2 (Bloomington Stock Centre #24650 and #24651) was used in combination with GAL4 lines in order to improve RNAi efficiency, UAS-Toll1Ex::VenusChinmoUTRS (Homemade with plasmid from Daiki Umetsu [39], UAS-EGFR-CA (Bloomington Drosophila Stock Centre #9533), UAS-EGFRRNAi (Vienna Drosophila Resource Center #43267), UAS-spitzRNAi (Vienna Drosophila Resource Center #3922), UAS-rhoRNAi (Vienna Drosophila Resource Center #51953), UAS-erkRNAi (Vienna Drosophila Resource Center #43123), UAS-kek1RNAi (Bloomington Drosophila Stock Centre #57000), UAS-kek1RNAi (Vienna Drosophila Resource Center #101166), UAS-kek1 (from Christian Ghiglione [35]), UAS-NotchRNAi (Vienna Drosophila Resource Center #27228), UAS-DeltaRNAi (Vienna Drosophila Resource Center #3720), UAS-Fz2RNAi (Vienna Drosophila Resource Center #108998), UAS-Wnt5RNAi (Vienna Drosophila Resource Center #101621).
Imp::GFP also known as G00080 is a protein trap line [47].
Other lines: rho-lacZ (also known as rhoBB02 Bloomington Stock Centre #9231).
Fly crosses are set up and raised at 29°C.
Generation of the UAS-Toll-1Ex::VenuschinmoUTRs plasmid and transgenic flies
The Toll-1Ex::Venus coding sequence was cloned between the 5′ and 3′ untranslated transcribed regions (UTRs) of chinmo under the regulation of upstream activation sequences (UAS) using the In-Fusion Snap Assembly Master Mix kit (Takara). The entry vector used was pUAS-mcherrychinmoUTRs [46] digested with PmeI to remove mCherry and to allow insertion of the Toll-1Ex::Venus sequence. The Toll-1Ex::Venus coding sequence was obtained from pUAS-Toll1Ex::Venus (a gift from Daiki Umetsu [39]). The primers used were: pUAS-ChinReg_Toll-1ExV_iF GGTGCGGCCGCGTTTAAACATGAGTCGACTAAAGGCCGCTTCC; pUAS-ChinReg_Toll-1ExV_iR TCGGAAGTGGAGTTTAAACTTACTTGTACAGCTCGTCCATGCC. Bold and underlines indicate sequences used for PCR amplification and overlapping sequences for In-Fusion, respectively.
Plasmid was sent to BestGene (https://www.thebestgene.com) for transgenesis via targeted insertion in attP-3B using Bloomington Stock #9748.
Immunostaining
Larval and adult (5 days old) VNCs containing tumors were dissected in phosphate-buffered saline (PBS) and fixed for 30 min at room temperature (RT) in 4% paraformaldehyde/PBS. VNCs were rinsed with PBT (PBS containing 0.5% Triton X-100) and incubated with primary antibodies for 5 days at 4°C. Secondary antibodies were incubated for 3 days at 4°C. VNCs were mounted in Vectashield (Clinisciences, France) without Dapi for image acquisition. The following primary antibodies were used: chicken anti-GFP (1:1,000, Aves #GFP-1020), rabbit anti-RFP (1:500, Rockland #600-401-379), rat anti-RFP (1:500, Chromotek #5F8), rat anti-PH3 (1:500, Abcam #AB10543), rabbit anti-Imp (1:500) [48], rabbit anti-Syp (1:500, gift from C. Desplan), rabbit anti-Miranda (1:500) [49], mouse anti-Miranda (1/50, gift from A. Gould), rabbit P-ERK (1:500, Cell signaling #4370), mouse anti-β-galactosidase (1:500, Promega #Z3781), rat anti-Elav (1:50, DSHB #7E8A10), guinea pig anti-Chinmo (1:250, gift from J. Enriquez).
Production and validation of anti-Kek1 antibody
Anti-Kekkon1 polyclonal antibody was generated in rat by GenScript (https://www.genscript.com/). The amino acid sequence used as epitope for immunization is:
MLVLTLLPGMILGTRYNQLHLYANGGASSSGPGGYRPAPSSQNEVYSIADSQPMTEDGYMPPSQHFPPTHSDLDPPAQQQSTCQTVCACKWKGGKQTVECIDRHLIQIPEHIDPNTQVLDMSGNKLQTLSNEQFIRANLLNLQKLYLRNCKIGEIERETFKGLTNLVELDLSHNLLVTVPSLALGHIPSLRELTLASNHIHKIESQAFGNTPSLHKLDLSHCDIQTISAQAFGGLQGLTLLRLNGNKLSELLPKTIETLSRLHGIELHDNPWLCDCRLRDTKLWLMKRNIPYPVAPVCSGGPERIIDRSFADLHVDEFACRPEMLPISHYVEAAMGENASITCRARAVPAANINWYWNGRLLANNSAFTAYQRIHMLEQVEGGFEKRSKLVLTNAQETDSSEFYCVAENRAGMAEANFTLHVSMRAAGMASLGSGQ.
The antibody was validated in wing discs compared to its known expression pattern, as well as with gain and loss-of-function conditions (S5 Fig).
Image analysis
For reproducibility and unless stated otherwise, all image acquisition and analysis were performed in 5-day-old adults (about 13 days after tumor induction) when tumors have reached a phenotypic equilibrium [9].
Image acquisition.
Confocal image stacks were acquired using a Zeiss LSM880 microscope equipped with a 40× objective. Whole tumors were imaged in 3D using z-stacks with 1 μm intervals between optical sections.
Volume quantification.
Tumor volumes and CSC proportions were quantified using FIJI (ImageJ). Binary masks of tumor tissue and CSC clusters were generated through intensity thresholding. Volumes were subsequently measured using the MorphoLibJ plugin based on voxel counting.
Tumor segmentation.
Cell segmentation was performed using Cellpose-SAM [50], employing the pre-trained models for nuclear (Dapi) and cytoplasmic (Imp and P-Erk) labeling. This allowed the segmentation of individual cells within the densely packed tumor tissue and the classification of cells based on marker expression. To prevent erroneous merging of adjacent cells during downstream quantification, a 2-pixel erosion was applied to all segmented masks.
ROI extraction.
Quantifications post-segmentation were made on FIJI (ImageJ) to manually correct the segmentation [52]. Regions of interest (ROIs) corresponding to individual cells were extracted using the “Analyze Particles” function, providing information such as the XYZ coordinates and mean fluorescence intensity for each marker.
Data processing and classification.
Data processing and cell classification were performed using Python.
Segregation Index (
).
In all quantifications, Chinmo+ cells are identified based on intensity of the mCherry::chinmoUTRs reporter, and classified as CSCs. All Chinmo− cells are classified as TAPs, thus here also including NPCs. For each cell, the segregation index (μ) is calculated based on the identity of the neighboring cells. We use a distance-based criterion for neighborhood: cells in a 7 µm radius around the target cell are considered as neighbors. μ is normalized between −1 for perfect mixing and +1 for perfect segregation. μ = 0 corresponds to a balance of CSC and TAPs that matches the tumor-level proportions. Thus, for a cell with the identity (either CSC or TAP), we have:
If qx ≥ Qx
If qx < Qx
where is the proportion of neighbors with the
identity, and
is the tumor-level proportion of cells with the
identity.
TAP score (s).
In all quantifications, CSCs are identified using the mCherry and Imp intensities obtained after immunostaining. Chinmo⁺ Imp⁺ cells were classified as CSCs, Chinmo⁻ Imp⁺ cells as tTAPs, and Chinmo⁻ Imp⁻ cells as TAPs. For each cell, the TAP score is calculated based on the identity of its neighboring cells in 3D. Again, cells in a 7 µm radius around the target cell are considered as neighbors.
For each cell, is the number of neighbors, and the contribution
of neighbor
depends on its identity. For CSC neighbors,
, for TAPs,
and for tTAPs,
. Consequently, a cell only surrounded by CSCs has a score of 0, whereas a cell only surrounded by TAPs has a score of 1.
Statistical analysis
For each volume or transition rule experiment, at least 4 tumors were quantified. To test the significance of volume or proportion differences between conditions, we performed a Mann–Whitney test. To test the significance of TAP distribution differences, we performed a Student’s test. The statistical analysis was performed using the Scipy Python library. Results are represented as boxplots or violin plots using the seaborn or matplotlib Python libraries. Sample sizes are reported in the figures. *, **, *** correspond to p<0.05, p<0.01 and p<0.001 respectively.
Numerical model
Rationale of the model.
We developed an agent-based Monte-Carlo stochastic numerical model, which simulates 3D tumor development driven by two distinct cell types: Chinmo+ cells (C)—the CSCs—and Syncrip+ cells (S),—the TAPs. Starting from an initial configuration at time t = 0 with a tunable number of cells (typically 30 C cells surrounded by 10 S cells—resembling what is observed at tumor initiation during early larval stages), the simulation progresses up to a final time fixed by the user. Throughout the simulation, cells can undergo division, change identity (C → S), enter a non-proliferative state, and move to account for steric repulsion when new cells are generated. The rates governing stochastic division and transition events can be modulated by the user and depend on the scenario tested.
Cell–cell physical interactions are governed by a Lennard-Jones potential to account for steric constraints and dense cell packing. The Lennard-Jones potential sets an effective, preferred distance between cells, typically corresponding to their own size. The interaction strength, tuned by the potential depth, can either be the same for all cells () or heterogeneous, enabling modeling of differential affinity/adhesion between cell types.
Model parameters and consistency with [9].
The model presented here requires setting a few key parameters. Importantly:
Note that ,
,
and the average transition rate
(used for the intrinsic scenario, see below) are directly extracted from our previous study using clonal analysis [9], in which we quantified transition and cell-cycle exit probabilities per division cycle. Here we will detail the consistency between the rates used here and the parameters inferred in [9].
In the clone model, we obtained the division times and
. We also inferred the average probabilities for C cells to undergo differentiation during a division cycle. The probability for a C cell to self-renew upon division was found to be p(C
CC)=0.64, the probability to give rise to two S cells p(C
SS)=0.21 and the probability to give rise to one cell of each type p(C
CS)=0.15. In this paper, we chose to reformulate these probabilities in terms of a single average transition rate
per unit time. This was done for multiple reasons. First, it makes the model more parsimonious, second, we have no a priori information on a link between transition and division, and third, experimentally we only have access to the probability
for single cells to be undergoing transition, which is directly related to the transition rate. Equating the expected number of new S cells per C cell division in both formulations, we obtain that:
We use a similar approach to obtain the rate of cell cycle exit of S cells, . Equating the rate of creation of non-proliferative cells in both formulations, we find that
, such that:
where is the probability for a new S cell to be non-proliferative in the clone model, C is the number of C cells, and S the number of S cells. When the system reaches stable proportions, C*/S* is constant and C and S populations have the same growth rate, such that
and
, which leads to:
Combining the above equations, we find:
Simulation workflow.
Once these parameters are defined, the simulation proceeds through the following steps, repeated at each time step (we set the time step to 1 hour in the simulations):
- 1. Cell division and cell cycle exit
C and S cells undergo probabilistic division, and S cells undergo cell-cycle exit, based on their respective rates (see above). For each cell, the probability that division or cell cycle exit occur during one time step of duration is:
where k corresponds to the rate of the considered event (,
,
). When a division event occurs, the two resulting daughter cells positions are initialized at a distance
, where
is the cell radius, centered around the position of the mother cell and at a random angle. This is followed by the energy minimization step.
- 2. Movement through energy minimization
New cells generated in the bulk push on the surrounding cells. In the simulations, the steric interactions between cells are mimicked by the Lennard-Jones potential, and we use global energy minimization to calculate cells’ displacement at each time step. Note that cells have no other active movements, consistent with the observation that clones remain compact over time [9]. The energy is calculated by summing the Lennard-Jones potentials
of all possible N interactions between cell pairs, such that:
With the interaction strength,
the prescribed cell radius,
the distance between the 2 cells of cell pair
and
the maximum distance considered (since the potential vanishes at long distance, we only consider cells within a certain range to compute the total energy).
Energy minimization is performed using a Monte Carlo–type relaxation scheme. Each sweep consists of single-cell displacement attempts, where at each attempt a cell is selected at random, ensuring statistically uniform sampling of all cells. The proposed displacement combines a deterministic force-driven term (local energy gradient) and a Gaussian random term. A displacement is always accepted if it lowers the energy, and accepted with probability
if it increases it (Metropolis criterion). The process is iterated over successive sweeps until convergence is reached. Convergence is reached when
consecutive displacement attempts produce sufficiently small energy variations, ensuring that the criterion scales with system size. Variations are considered small enough if
, i.e., the energy change per cell is below a threshold independent of system size.
- 3. Cell rearrangements via swapping for differential adhesion
Some of our simulations were designed to specifically test the effect of increased segregation, without changing the transition rules (Figs 5A and 5B). As discussed earlier, this is done by increasing the interaction strength among CSCs. A common strategy to implement differential affinity or adhesion in agent-based models is to allow neighboring cells with different identities to swap positions if it lowers the energy minimum. Note that in real cells, sorting also requires motility. Thus, we implemented test-swapping at a rate
. For each cell, the probability that it is chosen for test-swapping during one time step of duration
is thus given by:
To determine with which neighbor swapping occurs, the post-swap energy is computed for all possible swaps. The neighbor that lowers the overall energy the most is chosen for swapping. If none of the tested swaps lower the energy, no swapping occurs. Of note, this ensures that no swapping happens if .
- 4. C→S transition
Cell-intrinsic scenario (Figs 1D and 1E): In the cell-intrinsic transition scenario, all CSCs have the same transition rate such that for all CSCs:
and the probability for a CSC to initiate transition at each simulation time step of duration T is:
Cell-extrinsic scenario (Figs 3A–D): Our TAP score analysis revealed that the probability to transit is a quantifiable function of neighborhood identity. Hence, we devised simulations where the probability to transit is determined locally for each cell. Specifically, we measured experimentally the proportion of CSCs currently undergoing transition, . This observed proportion depends on the transition rate
but also on the time
spent by CSCs in the transition state, such that the transition rate we input in simulations is
(Little’s law of occupancy). For
, we directly use a sigmoid fit of the experimental curve. We conducted sensitivity analyses for
, and show the simulations outcome for the resulting
. The probability for a transition to occur in a given CSC with TAP score
at each simulation time step is:
Cells that have just undergone transition are considered tTAPs. They are endowed with the same division and cell cycle exit rates as TAPs. We assume that they retain their tTAP identity during a time or until they undergo division.
Supporting information
S1 Fig. Experimental system and segmentation pipeline.
(A) Asymmetrically dividing Drosophila NSCs transit through three main temporal windows during development defined by Chinmo/Imp (red), Syp (dark blue) and E93 (light blue), respectively. NSCs ultimately differentiate during metamorphosis (pupa). During development, the transition from a Chinmo/Imp+ state (in red) to a Syp+ state (dark blue) is scheduled in asymmetrically dividing NSCs by an intrinsic timing mechanism involving the rapid sequential expression of the temporal transcription factors Hunchback→Kruppel→Pou-domain 1 → Castor→Grainyhead→Seven-up [27,42]. Knockdown of pros during the Chinmo/Imp temporal window prevents the differentiation of intermediate progenitors, inducing NSC amplification and tumors that persist in adults. Tumors are composed of NSC-like cells recapitulating the Chinmo/Imp and Syp/E93 temporal identities observed during development. (B) The poxn>prosRNAi system allows the generation of tumors from six NSCs of origin located in the VNC. These tumors persist in adults sustained by the Chinmo/Imp + NSC subpopulation. (C) Tumors persisting in adults can be immunostained and confocal stacks with a 40 × objective allow 3D reconstitution with cellular resolution. Dapi shows all the tumor nuclei and mChchinmo show the CSCs. (D) Segmentations of Dapi and Imp stainings allow separating TAPs from the other cell types (tTAPs and CSCs). mChchinmo intensity thresholding allows to separate tTAPs from CSCs. This pipeline allows the complete categorization of all tumor cells in 3D. Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
https://doi.org/10.1371/journal.pbio.3004086.s001
(TIFF)
S2 Fig. Expression and role of signaling pathways in the CSC-to-TAP transition.
(A) Single-cell RNA-seq feature plots from poxn>pros-RNAi tumors in adults depicting expression of various genes including members of signaling pathways. (B) CSC proportions in tumors upon inhibition of components of the Notch and Wnt signaling pathways. (C) Feature plots and violin plots for mRNA levels of various components of the EGFR pathways and RTKs obtained from single-cell RNA-seq data. Note that EGFR is enriched in CSCs, while the EGFR ligand Gurken, and the RTKs heartless (htl) and torso (tor) are not detected in the tumors. (D) Mira labels NSC-like tumor cells. mChchinmo labels CSCs in control, UAS-spitz-RNAi, UAS-erk-RNAi and UAS-Egfr-CA conditions. White lines delineate tumors. (E) A tumor labelled with Dapi, anti-PH3 for mitotic cells and anti-P-Erk. Note that a nuclear P-Erk signal is detected in all mitotic cells. In contrast, cytoplasmic P-Erk (arrow) is restricted to a subset of tumor cells (usually localized in the vicinity of CSC clusters—see Fig 4C). (F,G) Dapi labels nuclei, mChchinmo labels CSCs. Absence of cytoplasmic P-Erk shows that EGFR pathway is not activated upon knockdown of spitz or Syp in tumors. P-Erk antibody constitutively labels mitotic nuclei (see in E) (40× objective). (H) Confocal section of control tumor with rho-lacZ expressed in a tumor cell (arrow) adjacent to a CSC cluster labelled with mChchinmo. Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
https://doi.org/10.1371/journal.pbio.3004086.s002
(TIFF)
S3 Fig. Kek1 in CSCs dampens EGFR-mediated differentiation.
(A) Dapi labels all tumor nuclei. Top panels—The anti-Kek1 antibody does not detect Kek1 in tumor cells. Bottom panels - The anti-Kek1 antibody detects enriched expression of Kek1 in tumor cells upon over-expression. (B) Quantification of Kek1 signal in CSCs versus TAPs upon kek1 over-expression. (C) P-Erk expression in control, kek1-RNAi and kek1-OE tumors (D) mChchinmo expression in control, kek1-RNAi and kek1-OE tumors. Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
https://doi.org/10.1371/journal.pbio.3004086.s003
(TIFF)
S4 Fig. EGFR pathway and Erk activation in NSCs during development and in larval tumors.
(A) Around late L2/early L3, no P-Erk signal is observed in normal NSCs (Mira+) preceding the Imp-to-Syp transition. Elav labels neurons. (B) Knockdown of kek1 or spi does not prevent the Imp-to-Syp transition in the poxn lineage (labelled in magenta with mCherry (mCh)) during larval development. Syp is expressed in all conditions in the NSC (arrow) and late-born larval neurons. (C) The number of Syp+ cells in each poxn lineage is not affected by the knockdown of kek1 or spitz. (D) P-Erk is not detected in late L3 larval tumors (delineated with white dotted line). Underlying data can be found at https://data.mendeley.com/datasets/w9bwgvxfrj in the Data Sheets file.
https://doi.org/10.1371/journal.pbio.3004086.s004
(TIFF)
S5 Fig. Characterization of the anti-Kek1 antibody and kek1-RNAi line.
(A) Anti-Kek1 reveals the wild-type expression pattern of Kek1 in late L3 wing imaginal discs, as previously reported [36]. (B) Top panels—Kek1 conditional knockdown via RNAi in the posterior compartment using en-GAL4 leads to efficient silencing as detected by immunostaining. Bottom panels Over-expression of kek1 in the posterior compartment is strongly detected by the anti-Kek1 antibody. These experiments validate both the specificity of the newly generated anti-Kek1 antibody, and the efficiency of transgenic lines for knockdown and mis-expression of kek1.
https://doi.org/10.1371/journal.pbio.3004086.s005
(TIFF)
S1 Movie. Timelapse of a simulated tumor grown in 3D with cell-intrinsically regulated CSC-to-TAP transitions (refers to Fig 1D).
3D simulation depicting the growth of a tumor composed of CSCs (red balls), TAPs (dark blue balls) and NPCs (light blue balls). The division rates of CSCs () and TAPs (
) are fixed as well as the CSC-to-TAP transition rate (
), and the rate of cell-cycle exit (TAP-to NPC) (
). All rates were previously calculated from clonal analysis in Genovese and colleagues 2019 [9] (see Materials and methods).
https://doi.org/10.1371/journal.pbio.3004086.s006
(AVI)
S3 Movie. z-stack of a full 5-day-old adult tumor analyzed with the segmentation and analysis pipeline.
Tumors were immunostained to label the Chinmo reporter (red), Imp (yellow) and nuclei are labelled with Dapi. Spacing between confocal sections is 4 µm. Top right panel: Dapi. Top left panel: Imp (yellow) and Chinmo (Magenta). Bottom left panel: segmented cells are colored based on their cell state (CSC (red), tTAP (yellow) or TAP+ NPC (dark blue). Bottom right panel: segmented cells are colored based on their TAP score.
https://doi.org/10.1371/journal.pbio.3004086.s008
(AVI)
S4 Movie. z-stack through a 3D simulated tumor grown with the transition rule without differential CSC/TAP adhesion.
. Z-stack shows tumor at the end of the simulation. The CSC-to TAP transition depends on cellular neighborhood as defined by the transition rule extracted from in vivo analysis. CSCs (red circles) show some segregation from TAPs (dark blue) and NPCs (light blue). tTAPs are in yellow.
https://doi.org/10.1371/journal.pbio.3004086.s009
(AVI)
S5 Movie. z-stack through a 3D simulated tumor grown with the transition rule with differential CSC/TAP adhesion.
. Z-stack shows tumor at the end of the simulation. The CSC-to TAP transition depends on cellular neighborhood as defined by the transition rule extracted from in vivo analysis. CSCs (red circles) show some segregation from TAPs (dark blue) and NPCs (light blue). tTAPs are in yellow. When a strong differential adhesion between CSCs and TAPs is implemented in the simulation (
), CSCs are highly segregated, leading to limited contacts with TAPs, reduced transition probability and over time an increased CSC proportion.
https://doi.org/10.1371/journal.pbio.3004086.s010
(AVI)
Acknowledgments
We warmly thank many colleagues from the Drosophila community for flies, plasmids and antibodies. We are grateful to Julien Leclercq for help on single-cell analysis. Stocks obtained from the Bloomington Drosophila Stock Center (NIH P40OD018537), Vienna Drosophila Resource Center (VDRC) and the Drosophila Genomics Resource Center were used in this study. We also thank Developmental Studies Hybridoma Bank (DSHB). We are grateful to the imaging facility at IBDM, member of the National Infrastructure France-BioImaging (https://ror.org/01y7vt929) supported by the French National Research Agency (ANR-24-INSB-0005 FBI BIOGEN). We also thank the Centre South-ROCK (INCa-Cancer_18695 - https://www.cancer.fr/) for meetings and discussions.
Declaration of generative AI and AI-assisted technologies in the manuscript preparation process: During the preparation of this work the authors used Lechat Mistral and ChatGPT in order to improve language and readability. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.
References
- 1. Batlle E, Clevers H. Cancer stem cells revisited. Nat Med. 2017;23(10):1124–34. pmid:28985214
- 2. Chu X, Tian W, Ning J, Xiao G, Zhou Y, Wang Z, et al. Cancer stem cells: advances in knowledge and implications for cancer therapy. Signal Transduct Target Ther. 2024;9(1):170. pmid:38965243
- 3. Gimple RC, Bhargava S, Dixit D, Rich JN. Glioblastoma stem cells: lessons from the tumor hierarchy in a lethal cancer. Genes Dev. 2019;33(11–12):591–609. pmid:31160393
- 4. Vegliante R, Pastushenko I, Blanpain C. Deciphering functional tumor states at single-cell resolution. EMBO J. 2022;41(2):e109221. pmid:34918370
- 5. Bonnet D, Dick JE. Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell. Nat Med. 1997;3(7):730–7. pmid:9212098
- 6. Gupta PB, Fillmore CM, Jiang G, Shapira SD, Tao K, Kuperwasser C, et al. Stochastic state transitions give rise to phenotypic equilibrium in populations of cancer cells. Cell. 2011;146(4):633–44. pmid:21854987
- 7. Iliopoulos D, Hirsch HA, Wang G, Struhl K. Inducible formation of breast cancer stem cells and their dynamic equilibrium with non-stem cancer cells via IL6 secretion. Proc Natl Acad Sci U S A. 2011;108(4):1397–402. pmid:21220315
- 8. Driessens G, Beck B, Caauwe A, Simons BD, Blanpain C. Defining the mode of tumour growth by clonal analysis. Nature. 2012;488(7412):527–30. pmid:22854777
- 9. Genovese S, Clément R, Gaultier C, Besse F, Narbonne-Reveau K, Daian F, et al. Coopted temporal patterning governs cellular hierarchy, heterogeneity and metabolism in Drosophila neuroblast tumors. Elife. 2019;8:e50375. pmid:31566561
- 10. Niu Y, Wang Y, Zhou D. The phenotypic equilibrium of cancer cells: From average-level stability to path-wise convergence. J Theor Biol. 2015;386:7–17. pmid:26365152
- 11. Ramos R, Swedlund B, Ganesan AK, Morsut L, Maini PK, Monuki ES, et al. Parsing patterns: Emerging roles of tissue self-organization in health and disease. Cell. 2024;187(13):3165–86. pmid:38906093
- 12. Adler FR, Anderson ARA, Bhushan A, Bogdan P, Bravo-Cordero JJ, Brock A, et al. Modeling collective cell behavior in cancer: Perspectives from an interdisciplinary conversation. Cell Syst. 2023;14(4):252–7. pmid:37080161
- 13. Brulé M, Horochowska A, Fontaine E, Torero-Ibad R, Woesteland F, Denoulet M, et al. Spatial self-organization of cancer stem cell niches revealed by live single-cell imaging. Stem Cell Res Ther. 2025;16(1):580. pmid:41146294
- 14. Caussinus E, Gonzalez C. Induction of tumor growth by altered stem-cell asymmetric division in Drosophila melanogaster. Nat Genet. 2005;37(10):1125–9. pmid:16142234
- 15. Narbonne-Reveau K, Lanet E, Dillard C, Foppolo S, Chen C-H, Parrinello H, et al. Neural stem cell-encoded temporal patterning delineates an early window of malignant susceptibility in Drosophila. Elife. 2016;5: e13463. pmid:27296804
- 16. Leclercq J, Maurange C. From the Making of a Neural Lineage to the Making of a Tumor: Lessons from the “Simple” Drosophila Brain. Adv Exp Med Biol. 2025;1482:181–208. pmid:40745142
- 17. Ghasemi DR, Okonechnikov K, Rademacher A, Tirier S, Maass KK, Schumacher H, et al. Compartments in medulloblastoma with extensive nodularity are connected through differentiation along the granular precursor lineage. Nat Commun. 2024;15(1):269. pmid:38191550
- 18. Hovestadt V, Smith KS, Bihannic L, Filbin MG, Shaw ML, Baumgartner A, et al. Resolving medulloblastoma cellular architecture by single-cell genomics. Nature. 2019;572(7767):74–9. pmid:31341285
- 19. Jessa S, Blanchet-Cohen A, Krug B, Vladoiu M, Coutelier M, Faury D, et al. Stalled developmental programs at the root of pediatric brain tumors. Nat Genet. 2019;51(12):1702–13. pmid:31768071
- 20. Maurange C. Temporal patterning in neural progenitors: from Drosophila development to childhood cancers. Dis Model Mech. 2020;13(7):dmm044883. pmid:32816915
- 21. Werbowetski-Ogilvie TE. From sorting to sequencing in the molecular era: the evolution of the cancer stem cell model in medulloblastoma. FEBS J. 2022;289(7):1765–78. pmid:33714236
- 22. Voutyraki C, Choromidis A, Meligkounaki A, Vlachopoulos NA, Theodorou V, Grammenoudi S, et al. Growth deregulation and interaction with host hemocytes contribute to tumor progression in a Drosophila brain tumor model. Proc Natl Acad Sci U S A. 2023;120(33):e2221601120. pmid:37549261
- 23. Gaultier C, Foppolo S, Maurange C. Regulation of developmental hierarchy in Drosophila neural stem cell tumors by COMPASS and Polycomb complexes. Sci Adv. 2022;8(19):eabi4529. pmid:35544555
- 24. Degrauwe N, Suvà M-L, Janiszewska M, Riggi N, Stamenkovic I. IMPs: an RNA-binding protein family that provides a link between stem cell maintenance in normal development and cancer. Genes Dev. 2016;30(22):2459–74. pmid:27940961
- 25. Lee JY, Huang N, Samuels TJ, Davis I. Imp/IGF2BP and Syp/SYNCRIP temporal RNA interactomes uncover combinatorial networks of regulators of Drosophila brain development. Sci Adv. 2025;11(6):eadr6682. pmid:39919181
- 26. Yang C-P, Samuels TJ, Huang Y, Yang L, Ish-Horowicz D, Davis I, et al. Imp and Syp RNA-binding proteins govern decommissioning of Drosophila neural stem cells. Development. 2017;144(19):3454–64. pmid:28851709
- 27. Syed MH, Mark B, Doe CQ. Steroid hormone induction of temporal gene expression in Drosophila brain neuroblasts generates neuronal and glial diversity. Elife. 2017;6:e26287.
- 28. Gesztelyi R, Zsuga J, Kemeny-Beke A, Varga B, Juhasz B, Tosaki A. The Hill equation and the origin of quantitative pharmacology. Arch Hist Exact Sci. 2012;66(4):427–38.
- 29. Corson LB, Yamanaka Y, Lai K-MV, Rossant J. Spatial and temporal patterns of ERK signaling during mouse embryogenesis. Development. 2003;130(19):4527–37. pmid:12925581
- 30. Bier E, Jan LY, Jan YN. rhomboid, a gene required for dorsoventral axis establishment and peripheral nervous system development in Drosophila melanogaster. Genes Dev. 1990;4(2):190–203. pmid:2110920
- 31. Aguirre A, Rubio ME, Gallo V. Notch and EGFR pathway interaction regulates neural stem cell number and self-renewal. Nature. 2010;467(7313):323–7. pmid:20844536
- 32. Jörg DJ, Caygill EE, Hakes AE, Contreras EG, Brand AH, Simons BD. eLife. 2019;8:e40919.
- 33. Ogura Y, Wen F-L, Sami MM, Shibata T, Hayashi S. A Switch-like Activation Relay of EGFR-ERK Signaling Regulates a Wave of Cellular Contractility for Epithelial Invagination. Dev Cell. 2018;46(2):162-172.e5. pmid:29983336
- 34. Yasugi T, Sugie A, Umetsu D, Tabata T. Coordinated sequential action of EGFR and Notch signaling pathways regulates proneural wave progression in the Drosophila optic lobe. Development. 2010;137(19):3193–203. pmid:20724446
- 35. Ghiglione C, Carraway KL 3rd, Amundadottir LT, Boswell RE, Perrimon N, Duffy JB. The transmembrane molecule kekkon 1 acts in a feedback loop to negatively regulate the activity of the Drosophila EGF receptor during oogenesis. Cell. 1999;96(6):847–56. pmid:10102272
- 36. Ghiglione C, Amundadottir L, Andresdottir M, Bilder D, Diamonti JA, Noselli S, et al. Mechanism of inhibition of the Drosophila and mammalian EGF receptors by the transmembrane protein Kekkon 1. Development. 2003;130(18):4483–93. pmid:12900463
- 37. MacLaren CM, Evans TA, Alvarado D, Duffy JB. Comparative analysis of the Kekkon molecules, related members of the LIG superfamily. Dev Genes Evol. 2004;214(7):360–6. pmid:15179511
- 38. Avraham R, Yarden Y. Feedback regulation of EGFR signalling: decision making by early and delayed loops. Nat Rev Mol Cell Biol. 2011;12(2):104–17. pmid:21252999
- 39. Iijima N, Sato K, Kuranaga E, Umetsu D. Differential cell adhesion implemented by Drosophila Toll corrects local distortions of the anterior-posterior compartment boundary. Nat Commun. 2020;11(1):6320. pmid:33303753
- 40. Lander AD, Gokoffski KK, Wan FYM, Nie Q, Calof AL. Cell lineages and the logic of proliferative control. PLoS Biol. 2009;7(1):e15. pmid:19166268
- 41. Dray N, Mancini L, Binshtok U, Cheysson F, Supatto W, Mahou P, et al. Dynamic spatiotemporal coordination of neural stem cell fate decisions occurs through local feedback in the adult vertebrate brain. Cell Stem Cell. 2021;28(8):1457-1472.e12. pmid:33823144
- 42. Maurange C, Cheng L, Gould AP. Temporal transcription factors and their targets schedule the end of neural proliferation in Drosophila. Cell. 2008;133(5):891–902. pmid:18510932
- 43. Pastor-Alonso O, Durá I, Bernardo-Castro S, Varea E, Muro-García T, Martín-Suárez S, et al. HB-EGF activates EGFR to induce reactive neural stem cells in the mouse hippocampus after seizures. Life Sci Alliance. 2024;7(9):e202201840. pmid:38977310
- 44. Burrows RC, Wancio D, Levitt P, Lillien L. Response diversity and the timing of progenitor cell maturation are regulated by developmental changes in EGFR expression in the cortex. Neuron. 1997;19(2):251–67. pmid:9292717
- 45. Graeve FD, Debreuve E, Pushpalatha KV, Zhang X, Rahmoun S, Kozlowski D, et al. An image-based RNAi screen identifies the EGFR signaling pathway as a regulator of Imp RNP granules. J Cell Sci. 2024;137(23):jcs262119. pmid:39479884
- 46. Dillard C, Narbonne-Reveau K, Foppolo S, Lanet E, Maurange C. Two distinct mechanisms silence chinmo in Drosophila neuroblasts and neuroepithelial cells to limit their self-renewal. Development. 2018;145(2):dev154534. pmid:29361557
- 47. Morin X, Daneman R, Zavortink M, Chia W. A protein trap strategy to detect GFP-tagged proteins expressed from their endogenous loci in Drosophila. Proc Natl Acad Sci. 2002;98(26):15050–5. 10.1073/pnas.261408198
- 48. Geng C, Macdonald PM. Imp associates with squid and Hrp48 and contributes to localized expression of gurken in the oocyte. Mol Cell Biol. 2006;26(24):9508–16. pmid:17030623
- 49. Narbonne-Reveau K, Erni A, Eichner N, Sankar S, Kapoor S, Meister G, et al. In vivo AGO-APP identifies a module of microRNAs cooperatively preserving neural progenitors. PLoS Genet. 2025;21(4):e1011680. pmid:40299997
- 50. Pachitariu M, Rariden M, Stringer C. Cellpose-SAM: superhuman generalization for cellular segmentation. bioRxiv. 2025;:2025.04.28.651001.
- 51.
Van Rossum G, Drake FL. Python 3 Reference Manual. Scotts Valley, CA: CreateSpace. 2009.
- 52. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, et al. Fiji: an open-source platform for biological-image analysis. Nat Methods. 2012;9(7):676–82. pmid:22743772
- 53. Boll W, Noll M. The Drosophila Pox neuro gene: control of male courtship behavior and fertility as revealed by a complete dissection of all enhancers. Development. 2002;129(24):5667–81. pmid:12421707






















English (US) ·
French (CA) ·