Cell Structure and Function
Online ISSN : 1347-3700
Print ISSN : 0386-7196
ISSN-L : 0386-7196
Tissue tension release induces Vinculin-dependent epithelial delamination in Drosophila pupal wings
Souta Hiro, Keisuke Ikawa
著者情報
ジャーナル オープンアクセス HTML
電子付録

2026 年 51 巻 2 号 p. 249-262

詳細
Abstract

Mechanical forces play critical roles in tissue morphogenesis and homeostasis, yet how tissue mechanics are translated into epithelial cell delamination remains poorly understood. Here, we use the Drosophila pupal wing as an in vivo model to examine how release of tissue tension influences cell death and delamination. Surgical severing of the wing hinge or loss of dumpy (dpy) function alters the force distribution, promotes cell crowding, and thereby significantly increases epithelial cell delamination compared to wild-type controls. An RNAi analysis targeting mechanosensitive components identified Vinculin, an adherens junction–associated force adaptor, as a key suppressor of mechanically induced cell delamination. Both tissue tension release and Vinculin depletion reduced junctional Vinculin levels and elevated Notch transcriptional activity, while canonical Notch transcriptional activation remained undetectable within delaminating cells. In contrast, knockdown of the Notch ligand Delta suppressed cell death and delamination. These findings indicate that release of tissue tension causes changes in forces acting on cells and cell crowding, thereby altering Vinculin-dependent regulation of Delta–Notch signaling and promoting cell delamination through a non-canonical Notch pathway.

Key words: cell delamination, mechanotransduction, Delta–Notch pathways

Graphical Abstract

Introduction

Epithelial tissues are continuously exposed to mechanical forces generated by morphogenetic movements, cell proliferation, and interactions with surrounding structures. To preserve tissue integrity and function, cells must sense, transmit, and adapt to changes in mechanical load in a coordinated manner (Irvine and Shraiman, 2017; Panciera et al., 2017). Disruption of tissue mechanics can lead to defective morphogenesis, tissue degeneration, and disease (Vogel and Sheetz, 2006). One fundamental mechanism by which epithelia maintain homeostasis is the removal of damaged, overcrowded, or mechanically compromised cells via epithelial cell delamination. Delamination is often accompanied by apoptotic cell death and allows removal of cells while preserving epithelial barrier function. Previous studies have shown that delamination can be triggered by cell crowding, differential contractility, oncogenic transformation, or externally imposed mechanical stress (Cachoux et al., 2023; Brás-Pereira and Moreno, 2018; Levayer et al., 2016). However, how perturbations in tissue-level mechanical state are translated into molecular signals that bias individual cells toward delamination remains poorly understood.

Mechanotransduction at adherens junctions (AJs) is central to epithelial force sensing. The cadherin–catenin complex mechanically couples neighboring cells and transmits tension to the actin cytoskeleton (Yao et al., 2014; Yonemura et al., 2010). Under tension, α-catenin (α-cat) undergoes a conformational change that exposes binding sites for actin-associated junctional proteins such as Ajuba, Afadin, and Vinculin. Among these, Vinculin functions as a force-sensitive adaptor within the cadherin–catenin complex, reinforcing junctional integrity under mechanical load by linking the cadherin complex to F-actin and thereby strengthening intercellular force transmission (Yonemura et al., 2010). This mechanism helps maintain cellular and tissue architecture under mechanical stress.

Notch signaling is a highly conserved pathway that regulates cell fate decisions during tissue development and homeostasis. Canonical Notch activation requires ligand-mediated mechanical pulling that exposes proteolytic cleavage sites within the receptor, leading to release and nuclear translocation of the Notch intracellular domain (NICD) and subsequent transcriptional activation (Langridge and Struhl, 2017; Sachan et al., 2024). Beyond this classical pathway, several non-canonical modes of Notch signaling have been described, including ligand-independent and transcription-independent mechanisms (Palmer and Deng, 2015). In particular, studies in vascular systems have shown that mechanical inputs such as shear stress can engage non-canonical Notch signaling to regulate AJ assembly and tissue architecture (Polacheck et al., 2017). These reports suggest that mechanical forces can intersect with both canonical and non-canonical Notch signaling. Nevertheless, whether Notch signaling contributes directly to mechanically induced epithelial cell delamination, and how it is integrated with junctional force-sensing mechanisms, remains largely unexplored.

The Drosophila pupal wing offers a powerful in vivo system for probing the relationship between tissue mechanics and cell dynamics. During development, contractile forces generated in the hinge region impose tension across the wing blade (Aigouy et al., 2010; Sugimura and Ishihara, 2013). Surgical severing of the hinge releases this tension (Aigouy et al., 2010; Ikawa and Sugimura, 2018; Sugimura and Ishihara, 2013), whereas mutations in dpy disrupt mechanical attachment between the wing and surrounding cuticle, similarly altering tissue force balance (Etournay et al., 2015; Ray et al., 2015). These perturbations provide controlled means to study cellular responses to changes in epithelial mechanical state.

In this study, we use mechanical perturbations in the Drosophila pupal wing to examine how release of tissue tension promotes epithelial cell death and delamination. Surgical severing of the wing hinge or loss of dumpy (dpy) function alters the force distribution within the tissue, promotes cell crowding, and thereby significantly increases epithelial cell delamination. Through targeted RNAi analysis of mechanosensitive junctional components, we identify Vinculin—an AJ–associated force adaptor—as a key suppressor of mechanically induced delamination. Both tissue tension release and Vinculin depletion reduced junctional Vinculin levels and elevated Notch transcriptional activity, yet canonical Notch activation remained undetectable within delaminating cells. In contrast, knockdown of the Notch ligand Delta strongly suppressed cell death and delamination, including in a Vinculin–Delta double knockdown background. Together, these results indicate that release of tissue tension causes changes in forces acting on cells and cell crowding, thereby engaging Vinculin-dependent regulation of Delta–Notch signaling, driving cell delamination through a non-canonical Notch pathway, and linking tissue tension dynamics to junctional mechanotransduction in epithelial homeostasis.

Results

Tension release induces epithelial cell death and delamination in the pupal wing

To investigate how tissue mechanics influence epithelial homeostasis, we utilized the Drosophila pupal wing, in which contraction of the hinge region normally generates tensile forces that stretch the wing blade (Aigouy et al., 2010; Sugimura and Ishihara, 2013). Surgical severing of the hinge released this tension, leading to rapid tissue contraction (within 15 min) accompanied by a reduction in individual cell area and a marked increase in epithelial cell density (Fig. 1a–c, Supplementary Fig. S1b). Under this condition, force inference revealed that the distribution of inferred tensions across all cell-cell junctions in hinge-cut wings was narrower than that in wild-type wings (Fig. 1f, g), and that tension anisotropy (RT ; see Methods for definition) was reduced (hinge cut: RT = 0.0444, N = 2; wild-type: RT = 0.0821, N = 1). Together, these findings suggest that hinge severing changes the mechanical state within the analyzed wing region. As in hinge-cut wings, dpy RNAi wings, which lose mechanical attachment to the surrounding cuticle, exhibited spontaneous tissue contraction and elevated cellular packing density (Fig. 1a–c, Supplementary Fig. S1f). We therefore used these two methods as complementary models to investigate how changes in tissue mechanics affect epithelial cell behavior.

Fig. 1

Tension release induces epithelial cell death and delamination in the Drosophila pupal wing

(a) DE-cadherin–GFP–labeled wing epithelium under control conditions, after hinge severing, after hinge severing in an H99/+ background, and in dpy RNAi wings. Magenta indicates delaminating cells. (b–c) Quantification of apical cell area and epithelial cell density following mechanical perturbation. Hinge severing and dpy RNAi induce significant cell area reduction and increased packing density. In all images, the left side corresponds to the proximal hinge side and the right side corresponds to the distal side of the pupal wing. The proximal–distal axis is indicated by the arrow. (d) Delamination rate under control, hinge-cut, and dpy RNAi conditions. (e) Matched-stage and matched-region-of-interest analysis comparing delamination frequency between control and dpy RNAi wings. (f) Representative tension maps of control and hinge-cut wing epithelia generated by Bayesian force inference analysis of DE-cadherin–GFP images. Individual cell–cell junctions are color-coded according to inferred edge tension. (g) Density histogram (left) and cumulative distribution (right) of inferred edge tension values in control (N = 1) and hinge-cut tissues (N = 2). Hinge cutting shifted the distribution of junctional tension, suggesting that release of tissue tension alters the mechanical state of cell–cell interfaces. (h) Live imaging of the GC3Ai caspase reporter immediately after hinge severing and 60 min later in a control RNAi + hinge cut sample, showing increased apoptotic activity within the hinge cut tissue. Orange dashed lines indicate the vein region. (i) Quantification of normalized GC3Ai intensity under control conditions, hinge cutting, and genetic perturbations affecting Delta and p53 signaling. Unless otherwise noted, all analyses were performed with N = 3 individuals; only the Delta[DN] condition in the GC3Ai background was analyzed with N = 2 individuals. Error bars indicate SEM.

Live imaging of pupal wings expressing the caspase reporter GC3Ai revealed a pronounced increase in apoptotic activity following hinge severing compared with wild-type control wings (Fig. 1h). GC3Ai reports caspase-associated apoptotic activity rather than delamination itself; however, previous studies in the Drosophila notum have shown that GC3Ai activation accompanies epithelial cell extrusion/delamination events (Cachoux et al., 2023). In line with this relationship, hinge-severed pupal wings showed increased GC3Ai signal together with elevated cell delamination. Notably, delamination events occurred predominantly within intervein regions, the epithelial areas located between the wing veins (Fig. 1h, 60 min, intervein region; Supplementary Fig. S1a, intervein), whereas vein territories exhibited relatively low levels of cell delamination (Fig. 1h, 60 min, vein region), indicating spatial heterogeneity in the response to the release of tissue tension. Quantitative analysis confirmed that delamination frequency was significantly elevated in hinge-cut and dpy RNAi wings (magenta-colored cells in Fig. 1a, d). When hinge cutting was performed in H99/+ wings, in which apoptosis is genetically reduced by deletion of the pro-apoptotic genes (reaper, hid, and grim), delamination was reduced, suggesting that apoptosis contributes to mechanically induced cell delamination (Fig. 1a, Hinge cut + H99/+). Because dpy RNAi wings exhibited a slightly altered developmental timing and a reduced analyzable tissue area, we additionally performed stage- and region-matched analyses to directly compare corresponding epithelial domains at the same developmental stage. Even under these controlled conditions, delamination frequency remained significantly higher in dpy RNAi wings than in controls (Fig. 1e), confirming that the enhanced cell delamination was not attributable to developmental stage differences or sampling bias. Importantly, hinge severing did not impair wing hair formation or folding of the severed wing blade tissue, and hinge-cut individuals emerged without lethality (Supplementary Fig. S1c–e). These observations indicate that the delamination observed in the severed wing is not simply a consequence of progressive tissue degeneration. Previous studies have demonstrated that crowded epithelial environments promote cell delamination through mechanical stimuli (Cachoux et al., 2023; Brás-Pereira and Moreno, 2018; Levayer et al., 2016). In line with this idea, our results suggest that hinge severing alters the mechanical state of the analyzed wing region, including changes in inferred tension anisotropy, cell area, and cell density, thereby promoting cell death–associated delamination.

Tension-dependent Vinculin localization regulates cell delamination and is associated with increased Notch activity

To identify molecular components mediating mechanically induced delamination, we performed a targeted RNAi analysis focusing on mechanosensitive components. As an initial candidate-based survey, we examined genes related to mechanosensation, cell adhesion, extracellular matrix organization, and stress-response signaling, including brv1 RNAi, piezo heterozygous mutants, trpl RNAi, laminin RNAi, and a dominant-negative form of bsk (bsk DN). Among the candidates tested, knockdown of Vinculin (vinc) markedly enhanced cell delamination under hinge-cut conditions compared with control RNAi (Fig. 2a, b, hinge cut + control RNAi vs. hinge cut + vinc RNAi). Notably, unsevered vinc RNAi wings showed only a low delamination frequency, indicating that Vinculin depletion alone is not sufficient to induce overt delamination (Fig. 2a, b, vinc RNAi vs. hinge cut + control RNAi). These results suggest that loss of Vinculin increases epithelial susceptibility to cell delamination in response to changes in mechanical states. To further examine the temporal progression of delamination after hinge severing, we quantified cumulative delamination over time using 10-min bins. This time-course analysis showed that delamination progressively accumulated after hinge severing and that vinc RNAi increased the cumulative delamination frequency compared with control RNAi (Fig. 2c). Thus, Vinculin depletion not only increased the endpoint delamination rate but also enhanced the temporal accumulation of delamination events following tension release. Given Vinculin’s known role as a tension-sensitive junctional adaptor, we next examined its localization following hinge severing. Using a Vinculin–GFP knock-in line, we observed strong enrichment of Vinculin at AJs in control wings, where tensile stress emerges via hinge contraction. In contrast, hinge cutting led to a pronounced reduction in junctional Vinculin signal, consistent with decreased junctional tension at cell–cell contacts (Fig. 1f, g, Fig. 2d, e).

Fig. 2

Vinculin regulates mechanically induced epithelial cell delamination

(a) DE-cadherin–GFP images showing increased delaminating cells following vinc RNAi under uncut or hinge-cut conditions. In all images, the left side corresponds to the proximal hinge side and the right side corresponds to the distal side of the pupal wing. The proximal–distal axis is indicated by the arrow. (b) Quantification of delamination rates in vinc RNAi, control RNAi + hinge cut, vinc RNAi + hinge cut, delta RNAi + hinge cut, and vinc–delta RNAi + hinge cut. For vinc RNAi, N = 1 and T = 61. (c) Cumulative delamination after hinge severing in control and vinc RNAi wings. Thin gray lines indicate individual samples/replicates, and thick black lines indicate the mean. Solid lines show control RNAi, and dashed lines show vinc RNAi. Error bars indicate SEM across three biological replicates. (d) Vinculin–GFP localization at AJs in control and hinge-cut wings. (e) Quantification of junctional Vinculin intensity demonstrating rapid loss of Vinculin from cell–cell junctions after tension release. Unless otherwise noted, all analyses were performed with N = 3 individuals.

Previous studies have suggested functional coupling between junctional mechanics and Notch signaling. In the Drosophila midgut, Vinculin has been implicated in regulating Notch-dependent epithelial turnover (Bohere et al., 2022). Notably, intestinal epithelia represent tissues in which mechanical compression or cell crowding actively promote cell extrusion from flies to mammals (Eisenhoffer et al., 2012; Brás-Pereira and Moreno, 2018), highlighting a conserved link between tissue mechanics and Notch-mediated epithelial remodeling. To assess whether tissue tension release similarly influences Notch signaling in the pupal wing, we monitored Notch activity using the NRE-EGFP, a Notch-responsive element reporter. Under normal developmental conditions, NRE-EGFP signal is known to be enriched in intervein cells adjacent to vein regions, reflecting spatially patterned Notch activation during wing morphogenesis (Herszterg et al., 2025). Consistent with previous reports, control wings exhibited the same spatial pattern, with low reporter activity in most of the intervein epithelium and relative enrichment near vein-adjacent intervein regions (Fig. 3a Control). Strikingly, release of tissue tension by hinge cutting led to a pronounced increase in NRE-EGFP fluorescence throughout the Notch-activated intervein region, indicating elevated Notch transcriptional activity in response to tissue tension release (Fig. 3a Control and Hinge cut, b). A similar enhancement of NRE signal was observed under conditions of junctional destabilization, including partial loss of α-cat function (Fig. 3a, α-cat [3]/+, b), suggesting that disruption of mechanical coupling at AJs broadly promotes Notch pathway activation. Importantly, depletion of Vinculin alone was sufficient to elevate NRE-EGFP activity, even without tissue tension release, suggesting that Vinculin normally suppresses Notch signaling under tensile conditions (Fig. 3a, vinc RNAi, b). Together, these findings indicate that release of tissue tension and weakening of junctional force transmission converge on enhanced Notch pathway activation, linking mechanical state to signaling modulation in the wing epithelium.

Fig. 3

Mechanical destabilization enhances Notch signaling but delamination occurs independently of canonical Notch activation in dying cells

(a) NRE-EGFP Notch reporter signal in control (N = 3), hinge-cut (N = 4), vinc RNAi (N = 4), and α-cat[3] heterozygous (N = 3) wings. Orange dashed lines indicate the vein region. For NRE-EGFP analysis, samples were obtained from MS1096-Gal4, rn-Gal4, or Gal4-independent backgrounds, depending on the genotype. (b) Quantification of NRE intensity showing elevated Notch activity following tension release and junctional destabilization. (c) GC3Ai activity under control, delta RNAi, and dominant-negative Delta conditions following hinge severing. The control image is the same as that shown in Fig. 1(h). (d) DE-cadherin–GFP images showing suppression of delamination by delta RNAi and vinc-delta RNAi. In all images, the left side corresponds to the proximal hinge side and the right side corresponds to the distal side of the pupal wing. The proximal–distal axis is indicated by the arrow. (e) Low-magnification images showing NRE-EGFP and myosin-II signals around extruding cells. (f) Time-lapse imaging of extruding cells labeled with myosin II and NRE-EGFP. Merged images, unmerged NRE-EGFP images, and unmerged myosin-II images are shown. Dashed outlines indicate the extruding cells. (g) Quantification of cell area dynamics and Notch reporter intensity in NRE-positive and NRE-negative cells. Each trace corresponds to one cell; 11 cells were sampled from 2 wings (2 individuals). (h) Schematic model illustrating tension release–dependent Vinculin reduction, Notch activation at the tissue level, and subsequent cell delamination.

Delta–Notch signaling is required for Vinculin-dependent cell delamination, but canonical Notch activation is not detected within delaminating cells

To determine whether Notch signaling contributes to mechanically induced delamination, we genetically disrupted the Notch ligand Delta. RNAi-mediated knockdown of Delta reduced caspase activation (Fig. 1i, Fig. 3c). Consistent with this, expression of a dominant-negative Delta construct also reduced caspase activation (Fig. 1i, Fig. 3c). Although both perturbations reduced GC3Ai activation, the inhibitory effect of delta RNAi was weaker than that of the dominant-negative Delta construct. This difference may reflect incomplete depletion of Delta by RNAi. Delta knockdown also significantly reduced cell delamination following hinge severing when directly compared with hinge-severed control RNAi wings (Fig. 2b, hinge cut + control RNAi vs. hinge cut + delta RNAi; Fig. 3d). Furthermore, simultaneous knockdown of Vinculin and Delta suppressed the enhanced delamination phenotype observed in hinge-severed, Vinculin-depleted wings (Fig. 2b, hinge cut + vinc RNAi vs. hinge cut + vinc, delta RNAi; Fig. 3d). Combined knockdown of Vinculin and Delta also reduced the delamination rate relative to hinge-severed control RNAi wings, although delamination was not completely abolished (Fig. 2b, hinge cut + control RNAi vs. hinge cut + vinc, delta RNAi). Together, these results demonstrate that Delta–Notch signaling functionally promotes epithelial cell delamination following tissue tension release and weakened localization of Vinculin at AJs. Notably, however, unsevered vinc RNAi wings did not exhibit overt cell delamination (Fig. 2a, b, vinc RNAi), indicating that increased Notch activity caused by Vinculin depletion alone is not sufficient to initiate cell delamination. This suggests that additional mechanisms, beyond the Vinculin–Notch axis, likely contribute to delamination onset. Consistent with this idea, expression of dominant-negative p53 reduced hinge cut–induced GC3Ai activation (Fig. 1i, Supplementary Fig. S1h), suggesting that p53-dependent apoptotic signaling also contributes to the caspase response after tension release.

Given the involvement of Delta–Notch signaling, we next examined whether canonical Notch activation occurred in delaminating cells themselves. Surprisingly, although NRE-EGFP activity increased following tissue tension release and Vinculin knockdown, delaminating cells did not exhibit elevated reporter signal. Live imaging revealed that cells undergoing delamination showed relatively low or variable NRE-EGFP fluorescence compared with the surrounding epithelium (Fig. 3e–g). Because Delta has been reported to undergo reverse signaling upon interaction with Notch (Vázquez-Ulloa et al., 2022), we considered the possibility that this signaling contributes to delamination. To test this possibility, we directly examined Delta localization following hinge cutting. However, Delta signal was inherently weak within the intervein region analyzed and did not exhibit appreciable enrichment upon tissue tension release (Supplementary Fig. S1g). Thus, although Delta function is genetically required for delamination, our data do not support the idea that enhanced Delta-mediated signaling in Delta-expressing cells (Vázquez-Ulloa et al., 2022) accounts for the observed Notch response in the intervein epithelium. Together, these observations suggest that cell delamination is regulated by Delta-Notch signaling distinct from conventional canonical Notch signaling or Delta-mediated signaling.

Exploratory morphodynamic trajectory analysis suggests distinct trajectories of surviving and delaminating cells

Canonical Notch transcriptional readouts did not uniquely predict which cells would be eliminated. In addition, although subtle differences in apical area distinguished delaminating cells, this morphological feature was not a deterministic predictor of delamination (Fig. 4a). These observations suggested that delamination is not specified by a single molecular threshold or simple morphological cutoff, but may instead depend on the combined influence of multiple shape features and signaling inputs on cellular stability. To examine this possibility, we performed a morphodynamic analysis using time-series measurements of cell shape. We defined a PCA space using the wild-type control condition and constructed a reference landscape based on the density distribution of control cells in this space. A basin was then defined as the most probable baseline state. Frame-wise cell shape features were aggregated into temporal bins, converted into multiparametric morphodynamic descriptors, and projected onto this PCA space. End-aligned trajectory analysis revealed that, although surviving and delaminating cells started from similar regions of morphodynamic space, delaminating cells later exhibited a marked change in trajectory, whereas surviving cells remained relatively stable (Fig. 4b). Consistent with this, delaminating cells moved farther away from the baseline basin toward the trajectory endpoint (Fig. 4c, d). Thus, our exploratory energy landscape analysis revealed a trend toward separation between surviving and delaminating trajectories in morphodynamic space, suggesting the presence of a boundary-like structure between the two states.

Fig. 4

Energy landscape analysis based on multiple cell shape features suggests a boundary between surviving and delaminating cells

(a) Apical area aligned by trajectory end (bin median; bin size = 10 frames). Cells were classified as “dead” when tracking terminated before the predefined endpoint and “alive” otherwise. Dead-cell trajectories were aligned so that the final observed bin corresponded to time 0; for visualization, apical area was set to zero immediately after t = 0. Alive-cell trajectories were compared using repeatedly sampled random windows matched to dead-cell trajectory lengths. Thin lines indicate individual trajectories or sampled windows; thick lines indicate group medians with quantile bands. Only subtle differences in apical area were observed between surviving and delaminating cells. (b) End-aligned median trajectories projected onto the baseline morphodynamic landscape. A two-dimensional PCA space was fitted using the wild-type control dataset (N = 1, 162 cells), and baseline state density was estimated by Gaussian kernel density estimation. The corresponding reference energy landscape was defined as Uctrl = −log(Pbaseline + ε) − Umin, and the basin center was defined as the global minimum of this landscape. Surviving and delaminating cell states from the hinge-cut condition (N = 3 wings, comprising 578 surviving cells and 31 delaminating cells) were projected onto this fixed baseline space, and end-aligned median trajectories were plotted. Representative individual-cell trajectories are also shown for reference. Delaminating trajectories showed a tendency to diverge from surviving trajectories in morphodynamic space. Start and end points of the median trajectories are indicated. (c) Distance to the baseline basin center (dist_basin) aligned by trajectory end. The basin center was defined as the global minimum of the baseline energy landscape. Delaminating cells showed a tendency to move farther from the basin center toward the trajectory endpoint, whereas surviving cells remained closer to the baseline basin. (d) Change in distance to the basin center from trajectory start to end. Negative values indicate movement toward the basin, whereas positive values indicate movement away from the baseline basin. Delaminating cells exhibited a larger increase in distance from the basin between the start and end of the trajectory than surviving cells. For the end-aligned comparisons in panels a–d, 300 length-matched random windows were sampled from surviving-cell trajectories.

Discussion

In this study, we demonstrate that tissue tension release in the Drosophila pupal wing induces epithelial cell death and delamination through Vinculin and Delta–Notch signaling. By integrating mechanical manipulation, genetic perturbation, live imaging, and quantitative analysis, we reveal how junctional mechanotransduction and signaling pathways mediate mechanically induced cell delamination. We propose a conceptual model in which release of tissue tension changes tissue mechanics, leading to reduced Vinculin levels, increased Notch activity, and eventual cell delamination (Fig. 3h). These findings may provide insight into the mechanisms by which crowding- and compression-related forces promote epithelial cell delamination.

A central outcome of our study is the identification of Vinculin as a key mechanical stabilizer that suppresses delamination under conditions of reduced tension anisotropy. Loss of Vinculin enhanced cell delamination under conditions of tissue tension release, indicating that Vinculin modulates epithelial susceptibility to changes in tissue tension. Consistent with its established tension-dependent recruitment to AJs, junctional Vinculin levels decreased following tissue tension release. These observations suggest that Vinculin contributes to maintaining intercellular force coupling under tensile conditions in the wild-type wing and that weakening of this junctional reinforcement sensitizes the epithelium to mechanically induced cell delamination. Interestingly, several features of our system parallel recent findings showing that epithelial apoptosis patterns emerge through global and local regulation of apical area and Ajuba localization (Cachoux et al., 2023). In both contexts, cells with modestly reduced apical area exhibit increased delamination propensity (Fig. 4a), while clear deterministic molecular markers are lacking at the single-cell level. These similarities suggest that force-sensitive AJ components, potentially including Vinculin and Ajuba, act as a broader mechanosensing module that biases delamination probability by modulating tissue mechanical stability rather than triggering binary cell death-versus-survival switches.

In this study, we were unable to define the precise forces that drive cell delamination. Our results suggest that hinge severing alters the distribution of tension at cell–cell junctions, and that this change is important for the localization of Vinculin. However, the relationship between such changes in junctional tension and the crowded tissue state remains unclear. Clarifying how tissue-level changes in tension reshape the distribution of forces at cell–cell interfaces and how this relates to crowding will be important for understanding the force-sensing mechanisms that trigger cell delamination.

Because Vinculin depletion alone activated Notch but did not induce delamination in wings without hinge severing, pathways other than the Vinculin–Notch axis may also respond to mechanical changes to regulate cell delamination. In mammalian systems, p53, one of the key factors involved in the stress response, has been implicated as a key mediator of mechanically induced cell competition and elimination (Wagstaff et al., 2016). Consistent with this idea, we found that expression of dominant-negative p53 suppressed hinge cut–induced caspase activation in the pupal wing (Fig. 1i, Supplementary Fig. S1h). Together with the delamination phenotype, the suppression of hinge-cut-induced GC3Ai activation by p53DN raises the possibility that p53-dependent apoptotic signaling contributes to mechanically induced cell death and delamination after tension release, in addition to the Vinculin–Delta/Notch-related mechanism described above.

Delta knockdown significantly suppressed mechanically induced delamination and rescued the Vinculin RNAi phenotype, indicating that Delta–Notch signaling is required for this process. However, delaminating cells in the intervein region did not show elevated NRE-EGFP activity. In other words, cells lacking detectable canonical Notch reporter activation preferentially underwent delamination. These observations suggest that the mechanism linking Delta–Notch signaling to delamination may involve a non-canonical mode of Notch signaling. One possibility is that tissue tension release alters the output of Delta–Notch signaling. Canonical Notch activation requires sufficient mechanical tension generated by ligand pulling, and disruption of this force-dependent process—such as when tension is reduced or when receptor endocytosis precedes productive cleavage—can abort transcriptional signaling (Langridge and Struhl, 2017). This raises the possibility that such abortive Notch signaling may generate a distinct signaling state that contributes to cell delamination. Relatedly, studies in vascular systems have shown that mechanical inputs such as shear stress can engage non-canonical Notch signaling to regulate AJ assembly and tissue architecture (Polacheck et al., 2017). In addition, endosome-dependent activation mediated by the E3 ubiquitin ligase Deltex has been shown to generate signaling-competent Notch fragments independently of classical ligand-induced cleavage (Wilkin et al., 2008), raising the possibility that altered trafficking or endosomal processing may also contribute to a distinct Delta–Notch output in this context. These possibilities are not distinguished by our current data and are presented as testable mechanisms for future work.

In the present study, the energy landscape analysis was based only on morphodynamic features derived from cell shape. Although exploratory, this analysis revealed a trend toward separation between surviving and delaminating trajectories in state space. Future integration of additional time-resolved information, including gene expression, protein localization, and mechanical forces, may help determine more clearly whether a biologically meaningful boundary exists between these cellular states. If such a boundary-like structure proves to be biologically meaningful, it may reflect a probabilistic mechanism of cell fate selection (Xu and Youk, 2024; A conceptual model is shown in Supplementary Fig. S1i). Integrating mechanical forces, molecular signaling, and probabilistic cell-state dynamics may therefore provide a framework for understanding how specific cells are selected for elimination within mechanically active epithelia.

These findings reconcile the lack of clear deterministic markers of cell delamination with the reproducible increase in delamination observed following mechanical perturbation. Rather than acting as simple binary cell death-versus-survival switches, mechanochemical cues appear to influence the propensity for epithelial elimination by shaping the mechanical and signaling context of the tissue. In this perspective, mechanical forces function not only as structural constraints but also as modulators of cellular behavior that affect when and where delamination occurs. By linking junctional mechanotransduction with Delta–Notch signaling during mechanically induced delamination, our work contributes to understanding of how epithelial tissues integrate physical cues to preserve homeostasis under changing tissue mechanics.

Materials and Methods

Drosophila genetics

The following fly lines were used in this study. Reporter and marker lines included DE-cadherin–GFP (Huang et al., 2009), sqh-mKate2 (Pinheiro et al., 2017), UAS-GC3Ai (Bloomington Drosophila Stock Center, BDSC #84346), NRE-EGFP (BDSC #30727), Delta-GFP (BDSC #59819), and Vinculin-GFP (Pinheiro et al., 2017). Gal4 driver lines used were MS1096-Gal4 (BDSC #8860) and rn-Gal4 (BDSC #7405). The yw strain (BDSC #1495) was used as the genetic background where indicated. The deficiency line H99/TM6B (BDSC #106395) was used to modulate apoptosis. RNA interference and dominant-negative constructs were expressed using the UAS/Gal4 system. The following UAS lines were used: UAS-delta RNAi (BDSC #36784), UAS-delta[DN] (BDSC #26698), UAS-dpy RNAi (Vienna Drosophila Resource Center, VDRC #44029), UAS-vinculin RNAi (National Institute of Genetics, NIG #3299R-2). To perturb p53 function, UAS-p53[DN] (BDSC #8421) was used. Junctional components were analyzed using the α-cat[3] mutant line (BDSC #67418). LexA-based experiments employed UAS-LexA RNAi (BDSC #67946) as a control RNAi line where indicated. Balancer chromosomes CyO, FM7C, and TM6B were used to maintain stocks and identify genotypes. Unless otherwise noted, crosses, white pupae collection, and imaging were all performed at 25°C. For the dpy-RNAi experiment, crosses were maintained at 18°C to keep Gal4-UAS-driven RNAi expression relatively weak during development. Animals were shifted to 25°C from the white pupal stage to increase Gal4-UAS activity during the pupal stage analyzed. This temperature-shift condition was used to match the timing of dpy-RNAi activity with the live-imaging window. Fly genotypes are summarized in Table 1.

Table 1

Drosophila genotypes and experimental conditions used in this study

Genotype Condition Figure
1 MS1096/+; DE-cadherin-GFP/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 1a–g
2 MS1096/+; DE-cadherin-GFP/+; H99/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 1a, d
3 MS1096/+; DE-cadherin-GFP/+; UAS-dpy RNAi/+ Crosses at 18°C; white pupae collection and imaging at 25°C Fig. 1a–e
4 MS1096/+; UAS-GC3Ai/sqh-mKate2 Crosses, white pupae collection, and imaging were all performed at 25°C. Suppl. Fig. S1c
5 MS1096/+; UAS-GC3Ai/+; UAS-LexA RNAi/+ (control RNAi) Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 1h, i, Fig. 3c, Suppl. Fig. S1h
6 MS1096/+; DE-cadherin-GFP/+; UAS-LexA RNAi/+ (control RNAi) Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 2b, Fig. 4a–d
7 MS1096/+; DE-cadherin-GFP/UAS-vinculin RNAi Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 2a, b
8 vinculin::GFP/FM7C Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 2d, e
9 MS1096/+; NRE-EGFP/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 3a, b
10 Sp/NRE-EGFP; rn-Gal4/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 3a, b
11 MS1096/+; NRE-EGFP/UAS-vinculin RNAi Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 3a, b
12 UAS-vinculin RNAi/NRE-EGFP; rn-Gal4/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 3a, b
13 NRE-EGFP/+; α-cat[3]/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 3a, b
14 MS1096/+; DE-cadherin-GFP/UAS-delta RNAi Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 2b, Fig. 3d
15 MS1096/+; UAS-GC3Ai/UAS-delta RNAi Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 1i, Fig. 3c
16 MS1096/+; UAS-GC3Ai/+; UAS-delta[DN]/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 1i, Fig. 3c
17 MS1096/+; DE-cadherin-GFP, UAS-vinculin RNAi/UAS-delta RNAi Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 2b, Fig. 3d
18 sqh-mKate2/+; delta::GFP/+ Crosses, white pupae collection, and imaging were all performed at 25°C. Suppl. Fig. S1g
19 MS1096/+; UAS-GC3Ai/UAS-p53[DN] Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 1i, Suppl. Fig. S1h
20 MS1096/+; UAS-mCD8-GFP/+ Crosses at 18°C; white pupae collection and imaging at 25°C Suppl. Fig. S1b
21 MS1096/+; sqh-mKate2/+; UAS-dpy RNAi/+ Crosses at 18°C; white pupae collection and imaging at 25°C Suppl. Fig. S1f
22 DE-cadherin-GFP Crosses, white pupae collection, and imaging were all performed at 25°C. Suppl. Fig. S1d, e
23 sqh-mKate2/NRE-EGFP Crosses, white pupae collection, and imaging were all performed at 25°C. Fig. 3e–g

Imaging of the Drosophila pupal wing

Live imaging was performed on pupal wings of Drosophila melanogaster. For imaging, pupae were placed on double-sided tape, and the pupal case was carefully removed using needles and fine forceps to expose the wing. Zeiss Immersion Oil W 2010 was applied to the exposed wing, and the pupa was mounted on a glass-bottom dish (Matsunami D11130H) with the wing facing downward. For hinge cut experiments, the wing was surgically severed by gently rubbing the wing through the pupal case with forceps prior to removal of the pupal case, while avoiding damage to regions other than the hinge (Supplementary Fig. S1b). Only wings in which the hinge region was reproducibly severed and the wing blade remained intact and suitable for live imaging were included in the analysis. Wings with incomplete hinge severing or obvious damage outside the hinge region were excluded. Developmental stages were synchronized using the white pupa stage as 0 h after puparium formation (0 h APF), and experiments were performed at 22–25 h APF. Imaging and subsequent analyses focused on the C-region of the wing, as indicated in Supplementary Fig. S1a. Mounted samples were subjected to live imaging using a confocal microscope (LSM900) equipped with a C-Apochromat 63×/1.2 W Corr objective (441777-9970-000).

Time window for observation of cell death and delamination

For observation of cell death using GC3Ai, wing hinge cutting was performed at 23 h APF, and imaging was conducted from 24 h APF to 25 h APF. For analysis of cell delamination events shown in Fig. 1 and Fig. 3e–g, hinge cutting was performed at 23 h APF, followed by time-lapse imaging for 60 minutes starting at 24 h APF. In dpy RNAi experiments, imaging was performed for 60 minutes starting at 23 h APF, and the number of delaminating cells was quantified. For experiments shown in Fig. 2, 3d and 4, hinge cutting was performed at 22.5 h APF to analyze long-term cellular behaviors immediately after cutting, and imaging was conducted for 100 minutes starting at 23 h APF. For energy landscape analysis, imaging was performed at approximately twice the magnification used for DE-cadherin conditions. For time-course analysis of cell delamination, cells were classified as delaminated when their segmented apical area first reached zero during time-lapse tracking. Cumulative delamination frequency was calculated in 10-min bins as the number of cells that had delaminated by each time point divided by the total number of tracked cells in the analyzed region.

Cell segmentation and tracking

To quantitatively analyze cell delamination and accurately assess cell dynamics, cell outlines were segmented. Cell segmentation is the process of identifying and partitioning individual cells or cellular structures into defined regions through image analysis. In this study, segmentation was performed using Fiji and its plugin Tissue Analyzer (Aigouy et al., 2016). Images acquired using DE-cadherin-GFP were first cropped within the C-region by defining a region of interest (ROI) in Fiji. The ROI size was set to 300 × 120 pixels for hinge-cut samples and 200 × 120 pixels for dpy RNAi. Subsequently, epithelial cells labeled by DE-cadherin-GFP were segmented using Tissue Analyzer, followed by cell tracking within the same software. Cell tracking enabled measurement of cell area, cell number, and identification of cell death events.

Force inference of wing tissue

DE-cadherin-GFP images were acquired from the C region for analysis. The cropped images were then segmented and skeletonized to generate binary images of cell boundaries. Vertex positions were extracted from the skeletonized images using the Fiji/ImageJ plugin GetVertex, following the developer’s instructions provided online (https://github.com/Sugimuralab/GetVertexPlugin; Sugimura et al., 2026). This plugin detects vertices and junctions from binary skeletonized cell-boundary images and outputs a text file containing vertex, junction, and cell information. The resulting output files were subsequently used as input for Bayesian force inference. Junction tensions and cell pressures were then estimated using the BayesianForceInference GoogleColab pipeline (https://github.com/Sugimuralab/BayesianForceInference_GoogleColab). This notebook performs estimation of junction tensions and cell pressures based on force-balance relationships at cell vertices (Ishihara and Sugimura, 2012; Ishihara et al., 2013). Tension anisotropy (RT) was quantified from inferred junction tensions and edge orientations. For each edge i, tension Ti and orientation θi were used to calculate an anisotropy vector based on cos 2θi and sin 2θi, and RT was defined as the magnitude of this vector normalized by the mean tension. Higher RT values indicate stronger directional bias in inferred junctional tension (Sugimura and Ishihara, 2013).

Quantification of GC3Ai signal

GC3Ai fluorescence intensity was quantified from acquired images. Fluorescence intensities were measured at each time point (1-minute intervals) over a 60-minute period from 24 h APF to 25 h APF, and total GC3Ai intensity across the entire wing was quantified.

Quantification of NRE (Notch-Responsive Element)-EGFP signal

NRE-EGFP fluorescence was quantified specifically within the C-region. An ROI was manually defined around the C-region using Fiji (Supplementary Fig. S1a). After brightness adjustment, a Gaussian filter was applied to enhance signal detection, followed by thresholding and mask generation to identify pixels exceeding the defined intensity threshold. Fluorescence intensity was then measured in the original image only within the masked region inside the ROI.

Energy landscape construction and trajectory analysis

Cell tracking and time-series feature extraction

Time-lapse images of DE-cadherin-GFP were acquired and a region of interest (ROI; 300 × 250 pixels) was cropped from the C region for analysis. Time-resolved cell morphodynamic features were extracted using the Fiji plugin Cellpose-TrackMate and the R package CellPhe (Wiggins et al., 2023; Stringer et al., 2021; Tinevez et al., 2017). Cellpose-TrackMate was used for automated cell segmentation and tracking, enabling the acquisition of cell shape dynamics and motility information over time. Based on these tracking data, CellPhe was used to quantify time-resolved morphological features, including cell area, shape descriptors, and dynamic motion-related parameters. The resulting time-series datasets were further processed using the tsfresh library to compute standardized time-series features.

Feature extraction and PCA embedding

For each tracked cell, frame-wise morphodynamic measurements were aggregated into fixed-width temporal bins (bin size = 10 frames). Bins were retained only when they contained at least 8 frames and when frames were contiguous within the bin. For each cell × bin segment, time-series features were computed using tsfresh with MinimalFCParameters (Christ et al., 2018), and non-finite values in the extracted feature matrices were imputed using tsfresh.utilities.dataframe_functions.impute, which replaces NaN, positive infinity, and negative infinity with the feature-wise median, maximum finite value, and minimum finite value, respectively. Feature matrices were z-scored using a StandardScaler. A two-dimensional PCA space was then fitted only on the unperturbed no-cut control dataset (“baseline”), and all experimental conditions were projected onto this fixed baseline PCA space using the same scaler and PCA transform, implemented in scikit-learn (Pedregosa et al., 2011).

Baseline probability density and energy landscape

To define a reference baseline state distribution, we estimated the probability density of baseline points in the PCA plane using scipy.stats.gaussian_kde with the default Scott’s rule for bandwidth selection. The resulting probability landscape was converted into an energy-like landscape using a negative log transformation, following a general landscape-based framework for state stability analysis (Jin et al., 2018; Wang et al., 2008). We defined the energy landscape as

  
U ctrl ( x , y ) = −log ( P baseline ( x , y ) + ε ) − U min

where Pbaseline is the KDE-estimated baseline probability density, ε = 1 × 10–12 is a small constant introduced to avoid numerical divergence, and Umin is the minimum value of the energy grid, such that the minimum baseline energy is set to zero. For each projected query point, Uctrl was computed directly from the baseline KDE, providing a quantitative measure of how “baseline-like” a given cell state is, with higher values indicating lower baseline probability.

Basin definition and distance-to-basin metric

The basin center was defined as the global minimum of Uctrl on the KDE grid. For each query point, we computed the Euclidean distance to this basin center in PCA coordinates (“dist_basin”), which captures how far the cell state lies from the most probable baseline region.

Trajectory alignment and end-aligned comparisons (dead vs alive)

Trajectory comparisons between surviving and delaminating cells were performed primarily using end-aligned analyses. Cells were classified as “dead” (i.e., delaminated) when their tracks terminated before the predefined endpoint (maximum FrameID<100), and as “alive” otherwise. Dead-cell trajectories were aligned so that the last observed bin corresponded to time 0. For alive cells, we repeatedly sampled random windows (n = 300) from alive-cell trajectories, with each sampled window matched in length to a dead-cell trajectory and aligned so that the end of the window corresponded to time 0. Group-level trends in dist_basin were summarized by median trajectories with quantile-based bands at each aligned time point, whereas PCA trajectories were summarized by median trajectories projected onto the baseline landscape. This procedure enabled direct comparison of trajectory evolution preceding the endpoint while reducing bias due to unequal track duration.

Statistics

Statistical analyses were performed in R using EZR (Saitama Medical Center, Jichi Medical University, Saitama, Japan), a graphical user interface for R (R Foundation for Statistical Computing, Vienna, Austria), together with custom-written scripts. More precisely, it is a modified version of R commander designed to add statistical functions frequently used in biostatistics (Kanda, 2013). Statistical analyses were performed as indicated for each figure and are summarized in Table S1. All tests were two-sided unless otherwise specified. For multiple comparisons, Dunnett’s test, two-tailed unpaired Student’s t-tests with Holm correction or Holm-adjusted Brunner–Munzel tests were applied as appropriate. Welch’s two sample t-test was used for pairwise comparisons when data satisfied parametric assumptions. Statistical significance was defined as p<0.05. Significance levels are indicated as follows: * p<0.05, ** p<0.01.

Author Declaration Statements

Funding

This study was financially supported by JSPS KAKENHI Grant (24K09469), Grant-in-Aid for Transformative Research Areas (A) (24H01400), the Inamori Foundation, and the Takeda Science Foundation to KI.

Conflict of Interest Statement

The authors declare no competing interests.

Data Availability Statement

The datasets generated during the current study are available from the corresponding author upon reasonable request.

Author Contribution Statement

KI and SH designed the research. SH and KI performed the experiments. KI and SH analyzed the data. KI drafted the manuscript. SH modified the manuscript. All authors approved the final manuscript.

Ethics Approval and Consent to Participate

Ethics approval is not required in this study. Consent to participate is not applicable because this study did not involve human participants.

Patient Consent for Publication

Patient consent for publication is not applicable because this study did not involve human participants.

Acknowledgments

The authors would like to thank Kaoru Sugimura for critical reading and suggestion, Yang Hong, Roger Kares, François Schweisguth, Yohanns Bellaïche, the Bloomington Stock Center, the Kyoto Stock Center, the Vienna Drosophila Resource Center (VDRC), NIG-Fly, the Drosophila Genomics Resource Center for reagents, Shizue Ohsawa for imaging equipment.

References
 
© 2026 The Author(s)

This is an open access article distributed under the terms of the Creative Commons BY (Attribution) License (https://creativecommons.org/licenses/by/4.0/legalcode), which permits the unrestricted distribution, reproduction and use of the article provided the original source and authors are credited.
https://creativecommons.org/licenses/by/4.0/legalcode
feedback
Top