## Main

Epithelial tissues possess the remarkable ability to adapt their program of cell fate in response to increasing tissue demands, such as developmental cues and damage1,2,3,4,5,6,7. Yet, the mechanisms governing transitions between different modes of cell behaviour remain unclear.

To better understand epithelial cell transitions, we focused on one of the first, and most important, transitions that a specified tissue must face: the establishment of homeostasis. In most tissues, development continues after birth, with a rapid but physiological expansion that encompasses the postnatal growth of the animal and culminates in the establishment of homeostasis8,9. At this point, proliferation and differentiation must become balanced to ensure tissue maintenance10. However, adult homeostatic cells retain a degree of plasticity, being able to reactivate developmental features if needed11. Indeed, differentiating or lineage-committed cells in a range of tissues, including the intestine, lung and skin, have been shown to possess the ability to transition back to a stem cell-like state in response to tissue damage6,12,13.

The mouse oesophagus represents an ideal model with which to study changes in epithelial cell dynamics at the whole-tissue level due to its simple architecture10. Although the processes governing adult tissue maintenance have been studied widely, here, we aim to shed light on the unknown mechanisms that initially shape homeostasis following postnatal development. During embryonic development, the oesophagus originates as a single layer of cuboidal cells that matures into a squamous, multi-layered tissue14. The resulting adult oesophagus is lined by a squamous keratinized epithelium, formed by layers of keratinocytes (Fig. 1a). Proliferation is confined to progenitor cells within the basal layer. Upon commitment to differentiation, basal cells exit the cell cycle and stratify into the suprabasal layers, being ultimately shed at the surface. Quantitative studies have proposed that the homeostatic oesophageal epithelium is maintained by a single, functionally equivalent progenitor population that balances the production of proliferating and differentiating cells through stochastic cell fate decisions15.

Growing evidence suggest that cell behaviour is not solely regulated by classically known biochemical cues. Instead, cells are able to alter their behaviour in response to a combination of molecular and mechanical stimuli16,17,18,19,20,21,22,23,24,25,26,27. This is exemplified during postnatal development, when tissues growing at their own unique pace generate significant levels of mechanical stress at the organ level24,28. In this context, physical forces influence cell adhesion, shape and niche availability and, in return, feedback on tissue growth, organization and maturation26,29.

Here, we show that after birth the developing oesophagus grows at a slower pace than the rest of the body, resulting in the build-up of a physiological tensile strain along its longitudinal axis. Tissue-stretching experiments demonstrate the existence of a defined strain threshold that, once reached, promotes the emergence of a bright Krüppel-like factor 4 (KLF4Bright) committed basal cell population, primed for differentiation, in a yes-associated protein (YAP)-dependent fashion. This mechanical transition signals the tilt of progenitor cell fate from a proliferative mode, during tissue expansion, towards the balanced commitment state characteristic of adult homeostasis15. Our results reveal that naturally occurring changes in tissue mechanics orchestrate the establishment of tissue homeostasis.

## Results

### Oesophageal transition towards homeostasis

To study changes associated with the establishment of adult homeostasis, we started by measuring tissue dimensions from postnatal day (P) 2 to P147 (Extended Data Fig. 1a–c). The oesophagus showed a biphasic mode of growth, encompassing the growth of the animal (Fig. 1b–d and Extended Data Fig. 1b–d). The initial phase of rapid oesophageal expansion slowed down by ~P28 and stabilized at P70, coinciding with the establishment of adult homeostasis15.

Next, we looked at cell proliferation using short-term in vivo 5-ethynyl-2′-deoxyuridine (EdU) incorporation assays. Quantification of EdU+ cells after a 24-h chase revealed a reduction in the proliferative activity of progenitors over time (Fig. 1e–g and Extended Data Fig. 1e). The tilt away from progenitor cell duplication was further revealed through an estimate of fate imbalance (Fig. 1h, Extended Data Fig. 1f, Methods and Supplementary Note 1). The degree of fate imbalance was determined as the ratio of the differential change in total basal cell number to the cell division rate. When the fate imbalance is large, progenitor cell divisions are biased towards a symmetrical duplicative fate outcome (model in Fig. 1h, inset), whereas when the fate imbalance is small, cell duplication is approximately balanced by cell differentiation. During the early postnatal period, we observed a 60% imbalance towards cell duplication, which drastically fell, reaching a near-balance homeostatic-like state between P28 and adult.

Next, we performed immunostaining for keratin 14 (KRT14) and Keratin 4 (KRT4), which mark the basal and suprabasal layers of the adult oesophageal epithelium, respectively. At early time points, both keratins showed an atypical expression pattern, acquiring their characteristic adult profile by P28 (Fig. 1i and Extended Data Fig. 1g,h). The ongoing maturation of the tissue was further illustrated by the progressive keratinization and cornification of the uppermost luminal layers, as well as increasing tissue thickness (Fig. 1i,j and Extended Data Fig. 1h–k).

These results suggest that, unlike the skin, the squamous epithelium of the oesophagus has not reached full maturation and keratinization by birth30. Instead, our data point towards P28 as the approximate age at which the features of the homeostatic oesophageal epithelium become established.

### KLF4 marks the emergence of an early committed population

To further understand the transition towards homeostasis, we analysed the expression pattern of key transcription factors associated with cell fate in oesophageal wholemounts by three-dimensional (3D) confocal imaging. These included SRY-box 2 (SOX2; which marks basal progenitor cells in the oesophageal epithelium31,32) and KLF4 (a well-characterised transcription factor marking suprabasal differentiated cells33,34,35,36,37). Similar to KRT14 and KRT4 (Fig. 1i), SOX2 and KLF4 expression at early time points was not restricted to their typical basal and suprabasal location in adults, showing an immature pattern of expression. By P28, SOX2 and KLF4 acquired their characteristic adult distribution, reflecting the maturation of the tissue (Fig. 2a).

Interestingly, we noticed a progressive increase in the percentage of basal cells expressing high levels of KLF4 (hereafter, KLF4 bright and dim cells will be referred to as KLF4+ and KLF4, respectively; Methods). KLF4+ cells were initially rare, representing ~2–3% of the basal population. This increased to ~9% by P28, reaching ~19% in the adult oesophageal epithelium (Fig. 2a–g and Extended Data Fig. 2a). This resulted in a salt and pepper pattern in the basal progenitor compartment, with SOX2+ cells interspersed with SOX2+/KLF4+ cells (Fig. 2a,b). The temporal changes in basal KLF4 expression, coinciding with the transition window at P28, identified KLF4 as a signature marking the transition towards adult homeostasis.

Since KLF4 is a transcription factor characteristic of differentiated cells, we reasoned that the previously unreported basal KLF4+ population identified here could represent a pool of early committed cells transitioning towards differentiation in the oesophageal epithelium (Fig. 2c,d). Consistent with this notion, previous work has shown how targeted deletion of Klf4 in the mouse oesophagus results in increased proliferation in the basal layer33. To explore this hypothesis, we initially investigated the proliferative activity of the KLF4+ basal subpopulation. Short-term EdU incorporation assays (Fig. 2e–h), as well as quantitative imaging analysis of wholemounts obtained from the cell cycle reporter mouse line Fucci2a (R26Fucci2aR; Extended Data Fig. 2b–e), revealed that the KLF4+ population identified in the basal layer of the oesophageal epithelium was in a nonproliferative resting state. Next, we evaluated the expression profile of KRT4 and KRT14 in the KLF4+/− basal populations. We noted that, although KLF4+ basal cells could be found expressing any potential combination of these keratins (Fig. 2i,j), the adult basal KLF4+ population was enriched in KRT4, revealing a bias towards differentiation (Fig. 2k). Taken together, the data support the committed state of the KLF4+ basal population. The emergence of this committed population by P28 inevitably slows down the rapid tissue expansion that takes place immediately after birth, favouring the balanced proliferation/differentiation typical of adult homeostasis.

Next, we explored a debated question in the field: whether basal progenitors initiate commitment towards differentiation from the moment of division, or whether the cell fate outcome is assigned later, following a refractory period in the basal compartment15,38,39,40,41,42. For this, we quantified the expression of KLF4 in post-mitotic EdU basal pairs, as marked by a 12-h EdU chase43 (Fig. 2l,m). Our results revealed KLF4 levels being reliably detected in EdU+ sister cells just hours after their physical post-mitotic separation, supporting the notion that fate can be assigned from the time of division (Fig. 2n,o and Supplementary Note 1).

Collectively, we conclude that the emergence of KLF4 in the basal cell compartment defines an early committed population, marking the transition towards homeostasis. These data provide further insight into the temporal resolution of cell fate commitment, revealing that commitment towards differentiation can be made from the division stage.

### Single-cell RNA sequencing (scRNA-seq) captures the onset of homeostasis

To explore the mechanisms regulating the emergence of the KLF4+ basal population and the transition to homeostasis, we made use of large-scale transcriptomics at single-cell resolution, focusing on three critical time points spanning the postnatal transition: P7, P28 and adult. Single-cell RNA-seq was performed on the viable epithelial cell fraction (EpCam+/Cd45) isolated from the oesophagus of C57BL/6 wild-type mice (Fig. 3a, Extended Data Fig. 3a and Supplementary Table 1). The uniform manifold approximation and projection (UMAP) distribution of the single-cell data revealed a noticeable separation of P7 cells, while P28 and adult cells clustered closely together (Fig. 3b). This confirmed that by P28, although not fully mature, the oesophageal epithelium resembles the adult homeostatic tissue at the transcriptional level. Hence, P28 and adult samples were grouped when exploring gene expression patterns, unless otherwise stated.

The combined dataset underwent unsupervised clustering and annotation using signatures of known lineage and cell cycle markers (Fig. 3c and Extended Data Fig. 3b–i). We identified three different populations for all three time points: (1) actively cycling basal progenitors, defined by increased expression of both basal and cell cycle markers; (2) resting/committed basal cells, with enriched basal signature expression and reduced levels of cycling markers; and (3) differentiated cells, marked by increased expression of genes associated with differentiation, as well as reduced levels of both basal and cell cycle marker genes (Fig. 3c, Extended Data Fig. 3j,k and Methods). Integration analysis of the unsupervised clusters confirmed the equivalence of the assigned populations between time points (Extended Data Fig. 3c and Methods). We concluded that, consistent with the current model15, the basal layer of the oesophageal epithelium is formed by a mixture of actively cycling and resting cells.

To shed light on the regulatory processes governing the epithelial transition towards homeostasis, we focused our analysis on the basal cell compartment, where progenitor cells reside. We identified four major patterns defining changes in gene expression over time (Extended Data Fig. 4a,b, Supplementary Table 2 and Methods). Gene Ontology enrichment analysis of the pattern genes revealed a particularly strong WNT signature at P7 (Wnt5b, Fzd2, Sox9, Sfrp2, Lgr5, Dkk3 and Wnt11) (Extended Data Fig. 4b,c). Other signalling genes, with notable differences at P7, included increased expression of insulin-like growth factor-related genes (Igfbp3, Igfbp4, Igf2 and Igf1r), as well as Pdgfa, Fgfr2 and Fgfr3 (Extended Data Fig. 4b,c and Supplementary Table 3), hinting at an active crosstalk between epithelial and mesenchymal cells at early postnatal stages44,45. In contrast, later time points were indicative of a maturation process, drifting towards bone morphogenetic protein (BMP)-driven regulation (Bmp3, Bmp4, Btg2, Hoxc8 and Hoxc4) (Extended Data Fig. 4b,d)46,47,48 and showing changes in genes related to protein biogenesis and metabolism (Supplementary Table 3)49,50. Consistent with a transition away from the developmental state of the early postnatal oesophagus51,52,53, P28 and adult cells showed an enrichment in genes involved in tissue maintenance and homeostasis, including Cd44, Ly6a and Mt2 (ref. 54). This differed from the P7 epithelium, which presented a marked expression of genes directly associated with epithelial development, wound healing and cancer (Krt8, Krt17, Krt19 and Lgr5)14,55,56 (Extended Data Fig. 3g). These transcriptional profiles reflect the fact that, from P28 onwards, basal cells favour maturation, leaving behind the proliferative state that governs tissue expansion shortly after birth51,52,53.

A subset of related Gene Ontology terms were found to change at both early and later time points. We interpreted these shared changes as an indication of an active switch, suggesting their relevance during this developmental transition. These included Gene Ontology terms associated with extracellular matrix and cytoskeleton organization, cell adhesion and response to mechanical stimulus (Extended Data Fig. 4b–d), all of which are critical for the mechanical stability and behaviour of epithelial cells57,58. Indeed, the transcriptional signature of basal cells showed a marked transition from active cytoskeletal remodelling at P7 (Tnc, Pdpn, Rhoc, Evl, Vcan and Mmp14) to an increased cell adhesion at later time points (Dcn, Lad1, Cfl1, Dstn, Dsc3 and Rhob). To explore whether these processes were related to the emergence of the KLF4+ committed population in the basal layer (Fig. 2), we started by assessing Klf4 expression across all clusters and time points. As predicted from tissue observations, Klf4 expression was higher in differentiated clusters, as well as in a subset of the resting basal population that became progressively established from P28 onward, defined by cluster 12 (C12) (Extended Data Fig. 4e). To investigate what makes C12 distinctive, we further compared the transcriptional profile of this Klf4high cluster with the remaining resting basal clusters (C2 and C6 for P7 and C1 and C7 for P28 and adult; Fig. 3d–g and Supplementary Table 4). We identified three patterns of differential gene expression in the resting basal compartment: (1) pattern 1 (genes elevated at P7); (2) pattern 2 (genes highly expressed in Klf4high C12); and (3) pattern 3 (genes progressively increasing over time) (Fig. 3f). This analysis showed that Klf4 and Klf4-associated genes, including the downstream targets Krt4, Krt13, Cdkn1a and Cebpb59,60,61, presented a higher expression level at later time points (P28 and adult; Fig. 3f,g). Additionally, we were able to confirm that the Klf4 signature seen in C12 is a feature of an emerging committed basal population, which becomes more marked as cells fulfil their differentiation program (C14, C8 and C9; Fig. 3g). Therefore, the Klf4 expression profile strengthens histological observations suggesting that KLF4+ basal cells represent a committed population established at the onset of homeostasis.

Another feature defining resting basal cells at later time points was the differential expression of genes associated with a response to mechanical stimuli and YAP mechanosensing (Fig. 3f)62,63,64,65. Interestingly, these genes were enriched at later time points (P28 and adult) and more particularly in the cluster containing the Klf4high population (C12; Fig. 3f,g). More specifically, P28 and adult samples were marked by an increase in Cyr61 expression (a direct downstream target of YAP in mouse keratinocytes), as well as other genes modulated by YAP, including (among others) Thbs1, Nrg1, Nrg2, Klf6, Hbegf, Ptgs2, Igfbp2 and Igfbp7 (Fig. 3f,g)66,67,68,69,70,71. Cav1 (a key cellular mechanoregulator also known to positively control YAP in response to substrate stiffness) showed increased expression at the P28 and adult stages66,72,73. Additionally, the expression of Klf2, a shear stress-responsive transcription factor known to be critical for heart valve development, was also significantly increased at later time points74,75 (Fig. 3f,g). Accordingly, Amotl1, Wtip and Ajuba, which have been proposed to act as negative regulators of the mechanosensing pathway Hippo76,77, showed increased expression at P7 compared with later time points (Fig. 3f).

Given the concurrence of genes associated with KLF4 and tissue mechanics, we then focused on the dynamic pattern of gene expression to unveil any clues suggesting potential co-regulation. For this, we performed pseudotime analysis, inferring the differentiation path within the resting basal population at early (P7; Extended Data Fig. 4f) and later time points (P28 and adult; Fig. 3h; Methods). Next, we examined gene expression along the pseudotime trajectory. At later time points, YAP-associated genes, including upstream regulators (Cav1) and direct targets (Cyr61, Ctgf and Thbs1), presented a distinctive pattern of expression that peaked in C12, decreasing thereafter as differentiation became established (C14, C8 and C9; Fig. 3g,h). Other mechanics-related genes, such as Dcn and Klf2, presented a similar trend, with Dcn showing a much wider peak of expression that spanned C7 and C12. Klf4 expression was also higher in C12, but as anticipated for a differentiation marker, its expression increased progressively throughout the differentiation trajectory (C14, C8 and C9; Fig. 3g,h). Accordingly, KLF4 downstream targets (Krt4, Krt13, Cdkn1a and Cebpb) showed a subsequent increase in the late phases of the pseudotime trajectory, representing the most differentiated stages (Fig. 3g,h). These results point to a link between a response to mechanical stimuli and the expression of KLF4 in the basal layer of the oesophageal epithelium at late time points; a similar association could not be made at P7 (Extended Data Fig. 4f).

Our scRNA-seq analysis highlights a clear shift in cell state through postnatal development, from being highly morphogenic at P7 to a more mature state at P28 and adult. Moreover, the uncovered transcriptional signature suggests a potential link between changes in tissue mechanics and a transition towards commitment in basal progenitor cells.

### Differential growth leads to build-up in mechanical strain

The biomechanical scRNA-seq signatures, together with considerable changes in basal cell morphology and density (Fig. 4a–c and Extended Data Fig. 5a,b), prompted us to look deeper into the architecture of the basal oesophageal epithelium. We developed a deep-learning image segmentation analysis pipeline (Extended Data Fig. 5c–e) to accurately quantify postnatal changes in cell shape, orientation and spatial organization. Our analysis shows that the anisotropic pattern and long-range orientational alignment observed in basal cells at early time points becomes less apparent from P28 onwards (Fig. 4d–f and Extended Data Fig. 5f–i), reflecting temporal changes in the physical properties of the tissue (Supplementary Note 1).

So far, our data hint to a possible link between changes in tissue mechanics and the transition towards homeostasis. We speculated that a potential source of mechanical stress may originate from growth anisotropy between the oesophageal epithelium, neighbouring tissues and/or the whole body of the animal. To explore this scenario, we measured the length of the oesophagus postmortem (in situ), within the animal and after tissue dissection (ex vivo) at different postnatal time points (Fig. 5a,b and Extended Data Fig. 6a). The increase in length relative to P7 (in situ and ex vivo) was compared with that of the body length (Fig. 5c). Interestingly, we observed a discrepancy in oesophageal length before and after dissection, with the ex vivo oesophagus becoming significantly shorter from P28 to adult. We established that the differential growth of the developing oesophagus, under normal physiological conditions, results in the build-up of a progressive strain, estimated to be between 20 and 40% from P28 to adult (Fig. 5d, Extended Data Fig. 6b,c and Supplementary Note 1). This growth pattern was not observed in the stomach, which showed a continuous increase in size and proliferative activity during the final postnatal stages (Extended Data Fig. 6d–g). Interestingly, this mechanical transition coincided with an active remodelling of the extracellular matrix, as evidenced by temporal changes in second harmonic generation (SHG; Extended Data Fig. 6h,i and Supplementary Note 1). We therefore conclude that the oesophageal epithelium undergoes a mechanically mediated transition, from being densely packed at earlier time points (when rapid proliferation promotes cell crowding; Fig. 4c) to becoming more loosely packed and stretched upon homeostasis (Fig. 5c,d).

To test whether the observed changes in the mechanical properties of the oesophagus stimulated a switch in tension and cytoskeleton remodelling of epithelial cells, we labelled the oesophageal epithelium for F-actin and phosphorylated myosin light chain II (pMLC2), a marker of cortical actomyosin contractility. Both presented a similar profile, increasing their signal as tissue matured (Fig. 5e–j and Supplementary Note 1). Next, we interrogated the cytoplasmic-to-nuclear translocation of the transcription factor YAP, an effector of the mechanosensing pathway Hippo78. YAP nuclear localization is associated with increased space availability, cell spreading and subsequent cytoskeletal reorganization. As predicted by the increase in tensile strain observed in the oesophageal epithelium, YAP was found to translocate from its cytoplasmic localization at P7 to the nucleus in adult animals, with an intermediate state at P28 reflecting a mechanosensing transition (Fig. 5k–m and Extended Data Fig. 6j–l). These data are indicative of basal progenitors sensing the physiological changes in strain and space availability experienced by the tissue.

We conclude that the differential growth of the oesophagus results in the build-up of a longitudinal tensile strain as the oesophageal epithelium transitions towards homeostasis. Changes in the cellular response to mechanical stimuli, which were initially evidenced at the transcriptional level, were further reinforced by the cytoplasmic-to-nuclear translocation of the Hippo effector YAP.

### YAP promotes basal commitment through KLF4

The experiments above suggest a temporal correlation between YAP and KLF4 expression patterns, with both transcription factors becoming localized to the nucleus of basal cells as the tissue transitions towards homeostasis (Fig. 2 and Fig. 5l,m). To explore the existence of a potential link between the two, we used three different approaches. First, we co-stained and correlated the levels of both proteins in the nucleus of adult basal cells (Fig. 6a–c). Second, we looked at KLF4 expression upon interfering with YAP signalling in adult mice. For this, we used an inducible mouse line (Krt5-rtTA; TEADi-GFP79), where genetic induction results in blockage of YAP- and tafazzin (TAZ)-induced TEAD (transcriptional enhancer factor) activity in the basal cell compartment, as detected by the TEADi-GFP (green fluorescent protein) reporter (Fig. 6d–f). Third, we investigated the impact of YAP overactivation in the early postnatal oesophageal epithelium (P6–P8), when basal cells are still negative for KLF4 (Fig. 2a,b,g), using the doxycycline (Dox)-inducible mouse strain Rosa26-rtTA; tetO-YAPS127A (rtTA/tetOYAP), which has a p.Ser127Ala substitution in YAP (Fig. 6g–k). Collectively, these experiments showed a correlation between YAP and KLF4 expression, and demonstrated that YAP nuclear activity promotes KLF4 expression in the basal cell compartment, balancing progenitor cell behaviour through commitment (Supplementary Note 1).

### Changes in tissue mechanics influence basal KLF4 expression

Having identified an increase in tensile strain throughout postnatal development, we next interrogated whether the mechanical stress imposed at the whole-organ level was sensed at the cellular level. To answer this question, we used a whole-organ in vitro approach utilising a 3D-printed stretching device80 (Fig. 7a,b and Extended Data Fig. 7a) that allows for controlled uniaxial tissue stretching (Methods). Adult oesophagi were stretched along their longitudinal axis at a static strain of 40%, representing the physiological levels measured in the adult oesophagus (Fig. 5d, Extended Data Fig. 6c and Methods). As a result, there was an increase in basal cell area (Extended Data Fig. 7b,c) that could be rescued by blebbistatin (BLEBB) treatment, which inhibits myosin II motor activity, disrupting tissue tension and cell junction stability (Extended Data Fig. 7b–d; 40% stretch). Accordingly, BLEBB led to a marked cellular disorganization. These results suggest that basal cells bear part of the mechanical load experienced upon whole-organ stretching (Supplementary Note 1).

Next, we investigated the relevance of mechanical stress in driving the transition to homeostasis. Oesophageal samples from P7, P28 and adult animals were stretched to adult physiological levels (40%). The expression of KLF4 was analysed as a proxy to identify changes in cell commitment in the basal layer. Reassuringly, the unstretched in vitro controls exhibited a similar KLF4 expression profile to that observed in vivo (Fig. 7c (top panels) versus Fig. 2b). Moreover, KLF4 expression remained unchanged in adult stretched samples, confirming that the strain was within the physiological range (Fig. 7c,d). P7 and P28 samples, instead, showed a significant increase in KLF4 basal levels that reached control adult values upon stretching (Fig. 7c,d). These data suggest that physiological levels of tensile stress promote KLF4 expression in the basal compartment. Further experiments exposing P7 samples to a range of tensile strains revealed that changes in basal KLF4 expression respond to a strain threshold between 20 and 40% (Fig. 7e), matching the levels found at the transition between P28 and adult (Fig. 5d, Extended Data Fig. 6c and Methods). Accordingly, the proliferative activity of the oesophagus at P7 was reduced at 40% strain (Extended Data Fig. 7e–h). Contrary to physiological levels of strain, stretching by 100% promoted an increase in proliferative activity, in line with previous studies on overstretched tissues81. These results support the notion of a biomechanical switch operating during the homeostatic transition of the oesophageal epithelium that dictates a shift in cell behaviour.

Remarkably, KLF4 levels in adult unstretched controls closely resembled those of in vivo counterparts (Fig. 7c). This suggests that, while an increase in strain is sufficient to activate KLF4 expression in basal cells, it does not seem necessary to maintain it in adult cells, where suprabasal keratinization provides the means to retain tissue architecture to a certain extent (Extended Data Figs. 1j,k and 7i–k)82,83.

Supporting the idea that basal KLF4 expression responds to changes in tissue mechanics, in vitro treatment with BLEBB to lower the tissue tension (Extended Data Fig. 7l), by disrupting actomyosin contractility, led to a reduction in KLF4 expression in the adult oesophageal epithelium (Fig. 7f–h), although no impact on cell proliferation was observed (Extended Data Fig. 7l,m). Further mechanistic insight was achieved by treating stretched P7 oesophagi with verteporfin, an established pharmacological inhibitor of YAP84,85. Verteporfin rescued the strain-induced KLF4 response (Fig. 7i–k), demonstrating that the mechanical regulation of KLF4 expression is mediated by YAP.

Taken together, our results reveal that the longitudinal tensile strain, generated by the anisotropic growth of the oesophagus during postnatal development, must reach a critical threshold in order to affect cell behaviour. Once this threshold is reached, basal KLF4 expression increases, modulated via YAP. This results in progenitor cells tilting their balance towards commitment, which ensures the establishment of tissue homeostasis (Fig. 7l).

## Discussion

Here, we report that the physiological strain emerging during normal postnatal development of the oesophagus controls its maturation, temporally guided by a mechanical threshold. In particular, we show that the differential growth of the oesophagus relative to the entire body generates a level of mechanical stretch that influences epithelial cell fate and promotes the transition towards homeostasis. Mechanistically, our data indicate that physiological stretching of the oesophagus results in nuclear re-localization of the YAP mechanosensor, which in turn promotes the emergence of a KLF4+ committed population in the basal progenitor compartment. The onset of basal cell commitment towards differentiation balances proliferation and ultimately defines the transition to adult homeostasis.

Tissue morphogenesis has long been known to be orchestrated by finely balanced gradients of morphogens. More recent studies in a growing number of developmental systems are starting to uncover the impacts that diverse types of mechanical forces have on major organs in the body24,29,81,86,87, from the shear stress dictating the development of the heart and lungs in mouse and zebrafish embryos18,74,75 to the mechanical stress guiding the patterning of the intestines and skin follicles in the developing chick and mouse19,20,21,88. However, there is surprisingly limited knowledge on the mechanisms regulating how developmental processes come to an end, safeguarding adult tissue homeostasis from abnormal growth and disease. Our findings in the postnatal mouse oesophagus reveal that this transition is linked to the mechanical changes experienced by the tissue. The rapid expansion of the oesophagus after birth is progressively restricted as the oesophageal epithelium matures and keratinizes. When that occurs, the oesophagus no longer grows at the same rate as the rest of the body, resulting in progressive stretching of the oesophageal epithelium. This is marked by a transcriptional switch, showing an upregulation in the expression of genes known to regulate the response to mechanical stimuli (Klf2, Cav1 and Dcn)72,74,75, including YAP-associated genes (Cyr61, Thbs1, Igfbp2 and Klf6, among others)66,67,69,71. Interestingly, this transcriptional signature is a feature of a particular committed cell population that emerges at P28 in the basal progenitor compartment, and is marked by high levels of KLF4 expression. The presence of a differentiating population in the basal layer promotes a more balanced progenitor cell behaviour, signalling the transition towards adult homeostasis15. Ex vivo stretching experiments reveal that the build-up in tissue strain experienced by the oesophagus during its postnatal development is responsible for the emergence of this KLF4+ basal committed population. Remarkably, we demonstrate the existence of a defined strain threshold that triggers the tilt in basal cell fate towards commitment, offering a means by which the temporal regulation of this process may be achieved. Genetic mouse models provide further mechanistic insights, revealing that KLF4 expression is regulated in a YAP-dependent manner. Our results establish a direct link between differential growth, tissue stretching and the acquisition of a committed basal population that defines the transition towards homeostasis. This supports the idea that changes in tissue mechanics can regulate epithelial behaviour in response to specific tissue needs.

Several in vitro and in vivo studies have shown that activation of typical mechanosensors such as YAP/TAZ and Piezo1 triggers an epithelial proliferative response66,89,90,91,92,93,94. Additional work in the mouse intestinal epithelium indicated that YAP can also regulate cell differentiation, specifically through its interaction with KLF4 (ref. 95). Our observations suggest a model in which changes in cell behaviour are regulated by integrating the physiological levels of mechanical stress happening at the organ level with those sensed by local neighbouring cells in a YAP-dependent fashion. This provides a simple solution to orchestrate the maintenance of a whole tissue at the cellular level (model in Fig. 7l).

Our study offers a new perspective in understanding the mechanisms governing epithelial cell transitions. These observations are likely to have wider implications and open exciting new questions as to whether cell behaviour is regulated similarly in other mechanically active tissues96,97,98,99,100. Future studies should elucidate whether regulatory processes controlled through differential growth are a broader phenomenon operating in other tissue models, as well as their translational relevance in regeneration and cancer6,87,101.

## Methods

### Mouse strains and allele induction

All animal experiments were approved by the local ethical review committees. Work was approved by the University of Cambridge and conducted according to Home Office project licence PPL70/8866 at the Wellcome–Medical Research Council (MRC) Cambridge Stem Cell Institute, Cambridge University. Additional mouse work was conducted according to approved protocols LCMB-034 and 19-329 from the National Institutes of Health Intramural Animal Care and Use Committee of the National Cancer Institute, or under permit number 2017-15-0201-01381, reviewed and approved by the National Animal Ethics Committee in Denmark.

Unless otherwise specified, wild-type C57BL/6J mice (Charles River; strain code 632) were used. We also used the following genetically engineered mouse lines: Fucci2a (R26Fucci2aR; kindly provided by I. J. Jackson102); tetO-YAPS127A (kindly provided by J. Larsson103); Rosa26-rtTAM2 (stock number 006965; The Jackson Laboratory); Krt5-rtTA;TEADi (kindly provided by R. Iglesias-Bartolome79); and Rosa26-mT/mG (stock number 007676; The Jackson Laboratory104). Animals cohorts were used at the exact postnatal days indicated, with the adult cohort constituting mice between 10 and 14 weeks of age.

All strains were maintained in a C57BL/6 background. All experiments comprised male and female mice, with no gender-specific differences observed. Only male mice were used in the scRNA-seq experiments, to avoid confounding effects due to oestrous cycles. All animals were housed between 19 and 23 °C, 45 and 65% humidity and a 12 h day/12 h night light cycle.

### Tissue dimensions

Oesophageal dimensions were measured from P2 to adult. Wild-type mice were culled and the oesophagi were excised, taking as reference two anatomical landmarks: the cricoid cartilage and the opening of the stomach fundus. Oesophageal tubes were longitudinally opened and flattened, to measure the length and width under a dissecting microscope, and the tissue area was calculated. The average tissue area was calculated for n = 3–22 per time point.

For in situ versus ex vivo comparison, we performed an immediate in situ postmortem tissue measurement, followed by a subsequent measurement after dissection (ex vivo) (Fig. 5a). Wild-type mice were culled at the indicated time points and organs surrounding the oesophagus and stomach were removed for a clear view. In situ images were taken, from which the length (oesophagus) and perimeter outline (stomach) were measured. This was repeated immediately after tissue dissection. Measurements were expressed as a percentage relative to their value at P7.

### Histology

Haematoxylin and eosin staining was performed by the Histology Core Service at the Cambridge Stem Cell Institute and imaged using an EVOS XL Core Imaging System.

### Tissue preparation and immunofluorescence

For in situ fixation of the oesophagus (Extended Data Fig. 5a,b), wild-type mice were culled and the organs surrounding the oesophagus and stomach were removed. The thoracic cavity was filled with 4% paraformaldehyde (Alfa Aesar; 043368) in phosphate-buffered saline (PBS) for 10 min for fixation. The oesophagus was then excised and processed for immunostaining, as indicated below.

For oesophageal wholemounting, oesophagi were excised and opened longitudinally. Then, the muscle layer was removed and the tissue was flattened under a dissecting microscope. The epithelial–stromal composite was then fixed in 4% paraformaldehyde in PBS for 30 min. Samples were incubated for 30 min in permeabilization buffer (0.5% bovine serum albumin (VWR International; 126575-10), 0.25% fish skin gelatin (Sigma–Aldrich; G7765) and 1% Triton X-100 (Fisher Scientific; 10102913) in PBS) and blocked for 2 h in permeabilization buffer containing 10% donkey serum (PBDS; Scientific Laboratory Supplies; D9663). Primary antibodies (Supplementary Table 5) were incubated in PBDS for 3 d at 4 °C, followed by four washes over 24 h with 0.2% Tween-20 (Promega; H5151) in PBS. Secondary antibodies were incubated overnight at 4 °C in PBDS and washed four times over 4 h with 0.2% Tween-20 in PBS. Cell nuclei were stained with 1 µg ml−1 4′,6-diamidino-2-phenylindole (DAPI; Sigma–Aldrich; D9542) in PBS at 4 °C before mounting. Wholemounts were cleared in 1.52 RapiClear mounting media (SunJin Lab; RC152001). Organ culture samples were stained following the same procedure.

Wholemounts of the oesophageal epithelium were stained with specific antibodies. These were prepared by incubating oesophagi in 5 mM ethylenediaminetetraacetic acid (EDTA; Life Technologies; 15575020) in PBS for 2.5 h at 37 °C following muscle layer removal. The epithelium was peeled from the stroma and fixed as above. Staining was performed using low Triton permeabilization buffer (0.5% bovine serum albumin, 0.25% fish skin gelatine and 0.5% Triton X-100 in PBS) and the following altered timings: 10 min of permeabilization in low Triton permeabilization buffer; 1 h of blocking in PBDS; primary antibodies in PBDS overnight at 4 °C; 2 h of washes using 0.2% Tween-20 in PBS; 3 h of secondary antibodies in PBDS; washes as for the primary antibodies; and cell nuclei staining as above.

For the staining of 10 µm cryosections, fixed oesophagi were embedded in optimal cutting temperature compound (Thermo Scientific; 12678646) for cryosectioning onto glass slides. Sections were permeabilized in Triton permeabilization buffer for 10 min, blocked in PBDS for 30 min and incubated overnight with primary antibodies in Triton permeabilization buffer at 4 °C, followed by a 1-h incubation with secondary antibodies in Triton permeabilization buffer. Sections were washed four times for 5 min in 0.2% Tween-20 in PBS between incubations.

Where indicated, EdU incorporation was detected using Click-iT imaging kits according to the manufacturer’s instructions (Invitrogen).

### EdU tracing

For in vivo tracing, EdU (Life Technologies; A10044) was administered at 10 µl g−1 bodyweight and 0.1 mg ml−1 via subcutaneous injection. Tissues were collected 2 and 24 h after injection. In organ cultures, EdU was added to the culture media at a final concentration of 10 µM and incubated for 2 h at 37 °C and 5% CO2. EdU+ cells were quantified in wholemounts from three to nine confocal z-stacks per animal.

For tracking EdU+ post-mitotic cell pairs in adult mice, EdU was administered at 18:00 (subcutaneously), when the proportion of S-phase cells was low in the circadian cycle, making clonal density labelling feasible, and chased for 12 h43,105. The KLF4 expression profile was determined in 225 EdU pairs from three different animals.

### Basal cell density

The basal cell density in oesophageal wholemounts was calculated by quantifying the number of DAPI+ basal cells per field in three to six images per animal.

### Total cell production and imbalance

First, we estimated the total number of basal cells in the whole oesophagus over time, by multiplying the basal cell density by the total tissue area at each time point. The total basal cell production rate, $$\dot B$$, was then estimated as the average change in basal cell number, B, divided by the time step.

The degree of fate imbalance (that is, the degree to which basal cell fate is biased towards duplication over differentiation and cell loss) was estimated as $$\left( {1/\lambda } \right)\dot B/B$$, where λ is the cell division rate. The division rate was determined from the results of a short-term (2 h) EdU incorporation assay (Fig. 2e) and normalized by the reported division rate of basal cells in the homeostatic adult tissue15.

### Tissue thickness

Tissue thickness was measured using side views of rendered confocal oesophageal wholemount images. Eighteen individual measurements were made for each of three animals.

### Imaging, analysis and segmentation

Confocal images were acquired using either an inverted Leica SP5 with standard laser configuration, an SP8 X with a white light laser or an upright SP8 X with a standard laser configuration with LAS X 3.5.5.19976 Leica software. Typical confocal settings used included: a 40× objective, an optimal pinhole, a scan speed of 400 Hz, a line average of three, optimal step size and a resolution of 512 × 512 or 1,024 × 1,024 pixels. For quantification, images where acquired with a digital zoom of 2× or 3×.

Images were reconstructed from optical sections using Volocity 5.3.3 software (PerkinElmer) and ImageJ 1.51w. KLF4, EdU, Fucci2a (mCherry and mVenus) and YAP intensity measurements (Fig. 6b,c) were performed on Volocity 5.3.3 using semi-automated nuclear segmentation. For the analysis of KLF4 bright cells (KLF4+), the threshold was set to include nuclei above the 80th percentile of KLF4 intensity (Fig. 2b,f,g and Extended Data Fig. 2c–e). This was visually confirmed to be over the background level of expression for nuclear KLF4 staining. EdU was set as >0.2, as detected in representative confocal images. F-actin (phalloidin labelling) and pMLC2 were quantified using the image segmentation protocol described in Supplementary Note 1.

YAP localization was assessed by measuring the fluorescent intensity of YAP immunolabelling and DAPI staining across the major cell axis of each individual cell. Nuclear and cytoplasmic YAP intensities were assigned per pixel based on the DAPI intensity. Those within the lower 20th percentile of DAPI intensity were considered cytoplasmic, while those within the top 20th percentile of DAPI intensity were measured as nuclear (Fig. 5m and Extended Data Fig. 6j–l). A total of 240 cells were analysed from three separate animals at four different time points (P7, P14, P28 and adult; 60 cells per time point).

Collagen fibres were imaged using SHG, alongside NucRed647 (Life Technologies; R37106) in wholemounts. Imaging was performed using an inverted Leica TCS SP5 acousto-optic beam splitter multiphoton laser scanning microscope equipped with a mode-locked Ti:Sapphire laser (Chameleon Ultra II; Coherent; 680–1,080 nm tuning range; 80 MHz repetition rate; 140 fs pulse width) that allows for SHG imaging. For image acquisition, the excitation laser (Ti:Sapphire) was tuned to 900 nm, and was acquired at an emission of 450 nm for SHG signal. Typical settings included: a 40× objective, a 600-μm pinhole, a 400-Hz scan speed, a line average of two, a 0.15-µm step size and a resolution of 1,024 × 1,024 pixels. Three images per animal were captured and the collagen fibre orientation was assessed using the pipeline outlined in Supplementary Note 1.

For single-cell image segmentation of wholemounts, single 2-μm z-slices of the basal layer, acquired using a Leica SP5 confocal microscope system, were used. For single-cell image segmentation of cryosections, single 2-μm z-slices of tissue cross-sections were acquired with the criteria outlined above. A minimum of three images were captured per animal. Analysis was performed via the pipeline described in Supplementary Note 1.

Note that images from in situ oesophagi (fixed postmortem undissected) were compared with oesophagi fixed immediately after dissection to confirm that the basal cell shape was not affected by tissue harvesting (Fig. 4a versus Extended Data Fig. 5a,b).

### RNA

#### Single-cell and RNA isolation

Epithelial–stromal composites were obtained as indicated above. For each sample, oesophagi were pooled from different animals (C57BL/6 wild-type mice) to enable the capture of sufficient cell numbers (five mice for P7, four mice for P28 and three mice for adult; three independent samples for P7 and P28; six independent samples for adult).

Following initial preparation, wholemounts were cut into 3 mm2 pieces and incubated in 0.5 mg ml−1 dispase (Sigma–Aldrich; D4818) in PBS containing 1 U µl−1 RNAse Inhibitor (Life Technologies; AM2696) at 37 °C for 10 min. The epithelium was then peeled away from the stroma and thoroughly minced before placing back into dispase for a 5-min incubation at 37 °C. EDTA was then added to a final concentration of 5 mM and the suspension was diluted 1/5 by adding fluorescence-activated cell sorting (FACS) staining buffer (2% heat-inactivated foetal bovine serum (Life Technologies; 26140079), 25 mM HEPES (Life Technologies; 15630056) and 1 mM EDTA in PBS) in order to reduce the dispase activity. Samples were then filtered through a 30-µm cell strainer and centrifuged at 300g for 10 min at 4 °C, then resuspended in staining buffer. Staining with primary antibodies or isotype controls was performed for 15 min at 4 °C, followed by a final wash in FACS buffer (FACS staining buffer without EDTA) at 300g for 5 min at 4 °C. The final cell suspension was placed in FACS buffer containing 1 U µl−1 RNAse inhibitor.

The following immunoglobulins were used (Supplementary Table 5) for sorting: 20 µg ml−1 EpCam-AF647 to stain epithelial cells; and 8 µg ml−1 CD45-APC-Cy7 to exclude immune contamination. Isotype controls were used at same concentration as specific antibodies: AF647 rat IgG2a and APC-Cy7 mouse IgG2a.

Work was carried out using RNAse-free and/or PCR-grade sterile reagents and plasticware wherever possible.

Viable epithelial (EpCam+/CD45/DAPI) cells were sorted on a BD FACSAria III cell sorter with BS FACSDiva 8.0.1 software, utilizing DAPI incorporation (4 µg ml−1) to exclude dead cells. Approximately 5,000 cells per sample followed the 10x Chromium pipeline for scRNA-seq. FlowJo version 10.6.2 software was used to generate the sorting plots (Extended Data Fig. 3a).

#### Library preparation and sequencing of RNA from single cells

Single-cell RNA-seq libraries were generated and sequenced using 10x Genomics kits (Single Cell 3′ v3) at the Cancer Research UK (CRUK) Cambridge Institute Genomics Core Facility. Libraries were generated in two different batches on different dates. Batch 1 consisted of three biological replicates for P7 and three for adult, used for six libraries. Batch 2 consisted of three biological replicates for P28 and three for adult, used for six libraries. The presence of adult samples in both batches provided common samples to control for any batch effect. The cells for each biological replicate were loaded into one 10x Chromium microfluidics chip channel to package them into one library. In total, 12 libraries were sequenced on Illumina NovaSeq 2 SP and 2 S2 flow cells (Supplementary Table 1).

### Data processing

The raw sequencing data from the 10x Genomics platform were processed using CellRanger (version 3.0.2) (Supplementary Table 1). CellRanger aligned reads, filtered empty dropouts and counted unique molecular identifiers (UMIs) to generate a count matrix. We used Ensembl GRCm38/mm10 (release 92) as the reference genome for read alignment. To filter out low-quality cells, basic quality control metrics were calculated using the R package scater (1.12.2)106, and cells with <1,251 genes were removed. Cells with mitochondrial proportions of >15% and genes expressed in fewer than three cells were also discarded from analysis. Read counts were normalized via a deconvolution method using the R package scran (version 1.12.1)107. Cells expressing fibroblast marker genes, including Col1a2, Col1a1, Fn1 and Pdgfra, were considered to be non-epithelial contaminants. To filter out non-epithelial cells, the normalized read count values for the genes above were averaged and the cells with averaged normalized values for the four genes above the 99th percentile (>0.4301828) of the distribution were filtered out. As a result, a total of 26,482 cells from three to six samples per time point were selected for further analysis. For basic statistics and quality control metrics, see Supplementary Table 1.

### Dimension reduction and data visualization

Principal components analysis, combined with technical noise modelling, was applied to the normalized data for dimension reduction, implemented by the denoise_PCA function in the R package scran. The data were then projected using two-dimensional UMAP or t-distributed stochastic neighbour embedding, with default parameter settings, using the R package scater. The biological replicates for each time point overlapped well with each other, confirming negligible batch effects between samples and conditions; therefore, no batch effect correction was necessary. The R package Seurat (version 3.0.2)108 was used to visualize cells in the dimension reduction space.

### Data clustering and cluster annotation

Next, a Louvain community detection method was used for clustering109. First, a shared nearest-neighbour graph was constructed using k = 20 nearest neighbours of each cell (the buildSNNGraph function in the R package scran). In this graph, two cells were connected by an edge if they shared nearest neighbours, with the edge weight determined by the highest average rank of the shared neighbours. Then, the Walktrap method from the R package igraph (version 1.2.4.1) (with steps = 4 as the default option) was used to identify densely connected communities that were considered to be cell clusters.

Cell clusters were annotated based on differentially expressed genes (DEGs) and known marker genes for cell types. If a few neighbouring clusters in the dimension reduction spaces shared key expression patterns, they were merged into one cell type manually. First, based on basal cell markers such as Krt14, Itgb1 and Itga6 and differentiation markers such as Tgm3, Krt13 and Grhl3, all clusters were classified into basal cells (clusters 1–3, 5–7, 10–13 and 15–17) and differentiated cells (clusters 4, 8, 9 and 14) (Extended Data Fig. 3b,d–f). The predominant expression of basal progenitor markers throughout the entire dataset (including Krt14, Itgb1 and Itga6) reflects an enrichment of basal cells under the experimental conditions used (Extended Data Fig. 3d,e). Accordingly, increased expression of differentiation markers (such as Krt13, Krt4, Tgm3, Grhl3 and Sbsn) was restricted to a few clusters (Extended Data Fig. 3d,e).

Then, we classified all cells into different cell cycle phases based on cell cycle genes: clusters 1, 2, 4, 6, 7, 8, 9, 12 and 14 for G0/G1; clusters 11 and 13 for G1/S; and clusters 3, 5, 10, 15, 16 and 17 for S/G2/M (Extended Data Fig. 3h,i). As a result, all cells were annotated as one of three cell types as follows: clusters 3, 5, 10, 11, 13, 15, 16 and 17 for basal cycling cells; clusters 1, 2, 6, 7 and 12 for resting basal cells; and clusters 4, 8, 9 and 14 for differentiated cells (Extended Data Fig. 3d–i). Of note, although suprabasal cells at early time points do not proliferate (Extended Data Fig. 3j,k), they tend to express key markers of basal identity (Fig. 1i and Extended Data Fig. 3d–f,i). For this reason, clusters containing resting basal cells at P7 probably represent a combination of both basal and basal-like suprabasal cells.

To confirm that the annotated three cell types matched well between different time points, we integrated the data for P7, P28 and adult using a standard workflow of Seurat (version 3.0.2) (Extended Data Fig. 3c). Specifically, we normalized the data for each time point separately using SCTransform and then integrated them by applying the following built-in functions sequentially with default parameters: SelectIntegrationFeatures (nfeatures = 3000); PrepSCTIntegration; FindIntegrationAnchors; and IntegrateData. For cluster annotation, we visualized gene expression in the dimension reduction and in violin plots using built-in functions of Seurat (version 3.0.2).

### Classifying gene expression for basal cells into distinct dynamic patterns along the time course and Gene Ontology analysis of the different expression patterns

After annotating cell types, we focused on molecular change for basal cells over time. To this end, we first identified DEGs for each time point using the function findMarkers of scran (version 1.12.1) by defining them as the genes having a false discovery rate (FDR) of <0.05 and an absolute value of log2[fold change] > (95th percentile of all log2[fold change]) in the three comparisons of P7 versus P28, P28 versus adult and P7 versus adult. The total number of DEGs was 1,738. We then scaled the normalized expression value of DEGs for all cells and calculated the average of the scaled expression value across the cells for each time point. Then, the scaled log2[fold change] for each time was calculated by subtracting the averaged scaled value at P7 from those at all time points (thus, the scaled log2[fold change] at P7 = 0).

The scaled log2[fold change] profiles for DEGs (1,738 genes) were clustered based on k-means clustering (k = 10) using pheatmap (1.0.12), and the resulting clusters were grouped into four major patterns: pattern 1 (148 genes); pattern 2 (545 genes); pattern 3 (153 genes); and pattern 4 (892 genes) (Extended Data Fig. 4a,b and Supplementary Table 2). The genes in each major pattern were used for Gene Ontology analysis by DAVID (https://david.ncifcrf.gov/)110. The resulting Gene Ontology terms were manually curated and selected based on experimental observations and biological knowledge (Extended Data Fig. 4b and Supplementary Table 3).

We then examined the transcriptional heterogeneity in resting basal cells over time. First, we calculated the proportion of resting basal cell clusters at each time point (Fig. 3d,e): clusters 2 and 6 at P7; and clusters 1, 7 and 12 at P28 and adult. This was marked by the emergence of cluster 12 over time, which represents a subset of resting basal cells expressing higher levels of Klf4. To understand temporal changes in resting basal cells from P7 through P28 to the adult stage at the molecular level, we then characterized gene expression changes from cluster 2 over cluster 7 to cluster 12. To this end, we calculated marker genes for the resting basal cell clusters (2, 7 and 12) using the function FindMarkers, as described above: marker genes for each cluster were defined as the genes having FDR < 0.05 and log2[fold change] > (97.5 percentile of all log2[fold change]) in the comparison of each cluster with the rest of the resting basal cell clusters. The total number of all marker genes was 1,729. Next, based on our observation in Fig. 3c–e, we grouped clusters of resting basal cells into three groups to represent each time point as follows: clusters 2 and 6 to represent P7; clusters 1 and 7 for P28/adult; and cluster 12 for P28/adult. For the 1,729 marker genes, their normalized expression value was scaled among resting basal cells and then averaged across the cells of each group. The average expression for the three groups were clustered based on e-means clustering (k = 6) using pheatmap (1.0.12) and the resulting six clusters were grouped into three major patterns (Fig. 3f): pattern 1 (748 genes); pattern 2 (163 genes); and pattern 3 (818 genes). Then, we performed Gene Ontology analysis using DAVID for each of the three major patterns to understand underlying biological processes (Supplementary Table 4). As a result, we found that enriched biological processes are very similar between basal cells (Extended Data Fig. 4b) and resting basal cells only (Supplementary Table 4). Of note, well-known Klf4 and Yap-related genes, among other genes associated with a response to mechanical stimuli, belonged to the three major patterns from the resting basal cell clusters, showing dynamic expression across the clusters (Fig. 3f).

We then explored a potential regulatory relationship between genes denoting a response to mechanical stimuli, including YAP targets) and Klf4-related genes in resting basal cells. We examined the temporal order of expression of YAP and Klf4 target genes at the single-cell level from the resting basal cell state and along the differentiation trajectory. Thus, we performed pseudotime analysis using the R package monocle (version 2.12.0) for resting basal cells and differentiated cells for P7 and P28/adult separately to project the expression profile of relevant genes along differentiation. To this end, we first identified DEGs for resting basal cells and differentiated cells at P7 and P28/adult separately by applying the criteria of FDR < 0.05 and an absolute value of log2[fold change] > (95th percentile of all absolute value of log2[fold change]). As a result, the number of DEGs was 1,018 for P7 and P28/adult, respectively. Based on the DEGs, dimension reduction was performed using the DDRTree algorithm implemented in the R package monocle. Next, the pseudotime order was calculated using the orderCells function of monocle with default parameters. We then displayed the expression profiles for relevant mechanics and Klf4-related genes along the pseudotime for P7 and P28/adult, separately (Fig. 3h and Extended Data Fig. 4f); for each gene, the auto-scaled gene expression was plotted using a rolling mean along its trajectory with a window size of 5% of the cells.

### Whole oesophageal organ culture and tissue stretching

A 3D-printed stretcher was developed from the prototype in ref. 80 using Autodesk Fusion 360 and Ultimaker Cura software and printed using an Ultimaker 3 3D printer in acrylonitrile butadiene styrene (biocompatible plastic). A millimetric scale was included in the printing of the device to guide the stretch. Once printed and sterilized, a 10-cm adjustable screw was inserted into the lower clamp to allow for controlled tissue stretching (Fig. 7b). The system was then placed in a sterile plastic container with a lid, allowing for the adjustable screw to be accessible from the outside of the container.

Oesophageal epithelial samples were collected at the indicated time points and the muscle layer was removed while retaining the tubular structure. Samples were then placed into the upper and lower clamps under no tension. Control samples were clamped at one end (top clamp in Fig. 7b, inset) and cultured under no longitudinal tension. Stretched samples were tightened into clamps using M3 nuts and screws. The millimetric scale and adjustable screw were used to apply a 40% stretch in tissue length. Whole-organ cultures were kept for 24 or 48 h at 37 °C under 5% CO2 in minimal medium containing one part Dulbecco’s modified Eagle’s medium (4.5 g l−1 d-glucose, pyruvate and l-glutamine; Invitrogen; 11971-025) and one part Dulbecco’s modified Eagle’s medium/F-12, 5 μg ml−1 insulin, 1.8 × 10−4 M adenine, 5% foetal calf serum, 5% penicillin–streptomycin and 5 µg ml−1 apo-transferrin.

In vitro stretched tissues were treated with pharmacological compounds to inhibit myosin II motor activity (25 µM BLEBB for 48 h) or YAP activity (1 µM verteporfin for 24 h), respectively. Vehicle (dimethyl sulfoxide (DMSO)) was added to control samples.

The tissue response to different levels of tensile strain was assessed in P7 samples subjected to increasing levels of longitudinal stretching (20, 40 and 100%) for 24 h. For EdU incorporation assays, EdU was added to media at a final concentration of 10 µM for 1 h, 24 h before collection.

### BLEBB treatment of adult oesophageal organ culture

Epithelial–stromal composites were cut into pieces of ~3 mm × 4 mm and placed onto transparent ThinCert inserts (Greiner Bio-One; 657641) with the stromal side to the membrane. Tissues were dried for 5 min at 37 °C and cultured in minimal medium at 37 °C under 5% CO2 for 3 h to ensure adequate tissue adhesion to the membrane. Then, 25 µM BLEBB was added to the media. Control explants were treated with DMSO vehicle alone. Samples were collected after 45 h.

For EdU incorporation assays, EdU was added and detected as above.

### Changes in basal cell area upon tissue composite separation

Oesophagi from Rosa26-mT/mG mice were dissected and the uppermost keratinized layers of the epithelium were mechanically separated with fine forceps. Alternatively, stromal layers were removed by incubating oesophagi in 5 mM EDTA in PBS for 2.5 h at 37 °C.

Basal cells were immediately imaged by confocal microscopy. Endogenous expression of the tdTomato reporter in cell membranes allowed for imaging without fixation and immunolabeling. The cell area was measured using the manual region of interest segmentation feature of the Volocity software and compared with that of basal cells from control unprocessed tissue samples.

### Statistics and reproducibility

A minimum of three independent mice or tissue cultures were used in all cases, with the exception of the TEADi experiments (see Fig. 6d–f) due to limitations associated with the COVID-19 pandemic. The numbers of biological replicates and animals are indicated in figure captions (n refers to the number of independent replicates per time point and/or condition). All replicates showed consistent results.

For image analysis, a minimum of three independent samples were inspected in all cases, capturing at least three replicate images each. All figures show representative images of a minimum of three mice or tissue cultures. The data are expressed as median values ± s.e.m. unless otherwise indicated.

Differences between groups were assessed by two-tailed unpaired t-test, one- or two-way analysis of variance (ANOVA) or Chi-squared analysis, as indicated in the figure captions. ANOVA-based analysis was followed by Tukey’s test for multiple comparisons. The correlation between YAP and KLF4 expression levels in immunostained wholemounts was calculated by computing the Pearson correlation coefficient (r). All tests were two sided. Exact P values are indicated in the relevant figures up to four decimal places. Statistical differences between groups were assessed utilizing GraphPad Prism software. No statistical method was used to predetermine sample size. The experiments were performed without randomization or blinding.

### Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.