Engineered intestinal crypt geometry uncovers YAP1-dependent fetal-to-adult transition

Summary

During morphogenesis, the intestine undergoes significant structural remodeling, transitioning from a simple tube of immature epithelium into a complex crypt-villus architecture housing mature cell types. However, the relationship between these structural changes and epithelial maturation has remained enigmatic. Using engineered scaffolds that replicate crypt-like geometries, we establish a robust platform for guiding the morphogenesis and differentiation of fetal intestinal cells into mature engineered tissues that mimic their in vivo counterparts. Mechanistically, tissue maturation is driven by cell crowding, leading to reduced YAP1 activation. Modulating YAP signaling in both engineered tissues and the developing mouse intestine alters epithelial lineage specification. These findings uncover a geometry-dependent mechanism that links tissue architecture to cell fate transitions. Our work provides a platform for modeling aspects of intestinal development and offers insights for refining stem cell differentiation protocols and regenerative strategies for intestinal disorders.

Graphical abstract

Highlights

• •Crypt-like geometry promotes maturation of fetal intestinal epithelial cells
• •Engineered tissues display multi-lineage differentiation and epithelial organization
• •Cell crowding reduces YAP1 activity to drive epithelial maturation
• •YAP activity modulates cell lineage specification in vivo

Using engineered scaffolds mimicking intestinal crypts, Maimets et al. show that tissue geometry drives epithelial maturation by regulating YAP activity. Crypt-like structures promote differentiation of fetal cells, providing insight into how form guides function and offering a platform for modeling intestinal development and repair.

Introduction

During organ morphogenesis, the acquisition of defined size and shape is intricately linked to the emergence of tissue-specific functions.¹ The intestine initially forms as a simple tubular structure, lined by a pseudostratified epithelium of immature progenitor cells.² As development proceeds, the tissue undergoes extensive architectural remodeling, initiated by the emergence of rudimentary villi that protrude into the lumen while maintaining a continuous intervillous domain.^2,3,4 Epithelial cells within the intervillous region exhibit high proliferative capacity and express distinct molecular signatures, including markers associated with adult stem cells.^4,5 Despite this, the tissue remains in a state of active morphometric transformation, characterized by dynamic cellular exchange and plasticity, wherein all epithelial cells retain the potential to adopt stem cell identity.⁴ Crypt formation marks a critical transition in intestinal maturation, occurring during the first post-natal week in mice⁶ and between gestational weeks 11 and 12 in humans, signaling the onset of homeostatic differentiation.^7,8,9 At this stage, renewal becomes driven by epithelial stem cells residing at the base of the crypts of Lieberkühn, accompanied by lineage-specific differentiation.¹⁰ However, whether these maturation events are mechanistically linked to tissue geometry remains unresolved.

Intestinal epithelial cells can be maintained as organoids long-term in tumor-derived hydrogels when supplemented with essential growth factors.^11,12 Adult small intestinal organoids cultured under these conditions form budding structures that recapitulate crypt-like domains and differentiated regions. By contrast, fetal intestinal epithelial cells cultured under identical conditions form simple, unstructured spheroids.^13,14 Recent bioengineering approaches have demonstrated that spatial control over organoid architecture is achievable by seeding cells onto microfabricated scaffolds that mimic crypt topography.^15,16,17 Tubular constructs with crypt-like structures enable epithelial cells to sustain balanced cellular turnover and multi-lineage differentiation, highlighting the influence of form on function.¹⁶ Yet, it remains unclear whether such engineering strategies can guide immature epithelial cells toward a mature and functional state.

Here, we leverage engineered scaffolds with adult crypt-like geometries to investigate how spatial constraints regulate the fetal-to-adult epithelial state transition. By seeding fetal cells onto scaffolds with crypt-like invaginations, we demonstrate that geometric cues drive cellular maturation and lineage specification. We identify cell density as a key determinant of the maturation process, mechanistically linked to YAP inactivation. Through gain- and loss-of-function experiments in vivo, we further show that YAP signaling is a fundamental regulator of epithelial maturation during development. Furthermore, we reveal that adult epithelial cells in a regenerative state undergo accelerated damage resolution when confined to engineered scaffolds. Together, our findings establish tissue geometry as a deterministic regulator of intestinal maturation and repair with broad implications for regenerative medicine and epithelial engineering.

Results

Immature cells form engineered tissues

In the mouse small intestine, crypts form during the first two weeks of post-natal life when the intervillus compartment is reorganized from a flat sheet of cells into dispersed u-shaped invaginations.¹⁸ This architectural reorganization precedes the appearance of differentiated cells responsible for food digestion and nutrient absorption.^19,20,21,22 To obtain insights into geometrical elements that regulate intestinal differentiation, we investigated parameters that change during crypt morphogenesis. For this, we first isolated small intestinal samples from post-natal day (P)0, P2, P5, and P16 and captured crypt progenitor units by performing whole-mount immunofluorescence for CD44v6,¹⁸ E-cadherin, and DAPI (Figure S1A). Using three-dimensional (3D) crypt reconstructions, we observed that the diameter of the crypts converges to a specific size during the critical time span of cell lineage specification (Figures S1B–S1D).^19,20

At embryonic day (E) 16.5, the small intestinal epithelium consists of equipotent progenitor cells.⁴ When these epithelial cells are cultured as organoids using standard culture methods, they give rise to primitive spheroids that lack the capacity to spontaneously convert into budding structures, a hallmark for organoids derived from the adult small intestinal epithelium.^11,13,14 To assess the impact of crypt diameter on the maturation of fetal epithelium, we cultured purified small intestinal epithelial cells from E16.5 on bioengineered hydrogel scaffolds comprising arrays of microcavities simulating the in vivo crypt diameters as they form (Figures 1A, 1B, S1C, and S1D).^15,16 The hydrogels consisted of a 1:3 Matrigel-type I collagen mixture, allowing free diffusion of nutrients and growth factors. Initially randomly dispersed, fetal cells established connections with each other and the surrounding matrix and, over a 72-h period, organized into an engineered tissue conforming to the shape of the pre-existing cavities (Figure 1C). Across a range of microcavity diameters, we observed the formation of three distinct morphologies in engineered tissues (Figures 1C and 1D). In small and medium diameter microcavities (up to 90 μm), cells adopted a columnar morphology, but a central lumen was clearly discernible only in the latter (Figure 1D). Conversely, in microcavities with the largest diameter (more than 120 μm), cells surrounded a large lumen without any columnar morphology (Figure 1D). Quantification of the percentage of tissues with observable lumen and columnar cell shapes revealed that these features aligned in microcavities with a size of 50 μm (Figures 1E and 1F), akin to the in vivo context (Figure S1D). Time-lapse microscopy over 3 days concluded that immature intestinal epithelial cells, when seeded on microfabricated scaffolds with 50 μm diameter, developed into a confluent epithelial monolayer with consistent crypt-like structures (Figures 1G and 1H; Video S1). Thus, we have identified conditions under which immature intestinal epithelial cells populate crypt-mimicking bioengineered scaffolds as a single-layered columnar epithelium. From this point onward we performed experiments using scaffolds with 50 μm diameter unless otherwise noted.

Figure 1.

Immature intestinal epithelial cells form engineered tissues when cultured in microwells with in vivo crypt size

(A) Strategy for culturing engineered tissues. W, WNT; E, epidermal growth factor (EGF); N, Noggin; R, R-spondin 1.

(B) Polydimethylsiloxane (PDMS) template used for fabricating bioengineered hydrogel substrates with a range of microwell sizes imitating in vivo crypt sizes.

(C) Bright-field image of engineered tissues at day 3 cultured on a scaffold containing a range of microwell sizes. Scale bar, 100 μm.

(D) Bright-field (left) and mTomato fluorescence (right) experiments depicting the formation of tissues with distinct morphologies. Scale bars, 20 μm (outset) and 5 μm (inset). Results in (C) and (D) are representative of at least three independent experiments.

(E and F) Quantification of lumen formation (E) and columnar cell shapes (F) across tissues with a range of microwell sizes. Data presented in (E) and (F) are means ± SD from three independent experiments.

(G) Bioengineered scaffold with 50 μm diameter size. Scale bars, 50 μm (outset) and 20 μm (inset).

(H) Bright-field image of an engineered tissue at day 3 cultured on a scaffold with a 50 μm diameter size. Scale bars, 50 μm (outset) and 20 μm (inset).

See also Figure S1.

Emergence of functional intestinal cells

We next investigated whether engineered tissues exhibit aspects of in vivo crypt organization. In the intestine, canonical Wnt signaling regulates adult stem cell function.²³ In the developing intestine, most epithelial cells exhibit activation of the Wnt-signaling pathway when assessed based on a sensitive reporter mouse strain (Top.CFP) (Figures 2A and S2A left).²⁴ This pattern of Wnt activity is altered with the emergence of CFP-negative cells appearing in the upper luminal epithelium post-natally (Figures 2A and S2A middle). Interestingly, a similar pattern of Wnt-active and inactive cells emerged in the engineered tissues after 3 days of culture (Figures 2A and S2A right). Consistent with observations of early crypt architecture in vivo, the engineered tissues developed into a polarized columnar epithelium that lacked identifiable differentiation features, such as secretory granules, when evaluated by scanning electron microscopy (SEM) (Figure 2B). Thus, within 3 days, immature intestinal epithelial cells initiate programs of tissue patterning when cultured on crypt-mimicking bioengineered scaffolds.

Figure 2.

Engineered tissues contain intestinal cells that exhibit phenotypic maturity

(A) Fluorescence-activated cell sorting (FACS) profiles showing distribution of CFP⁺ and CFP⁻ cells at E16.5 (left), P2 (middle), and engineered tissues at day 3 (right). Results in (A) are representative of at least three independent experiments.

(B) SEM image of an engineered tissue at day 3. Scale bar, 5 μm. Data in (B) are representative of two independent experiments, and at least 10 different crypt-like structures were analyzed.

(C) Detection of Lgr5-eGFP (green, left), Krt20 (red, middle), lysozyme (green, right), and DAPI (white) in engineered tissues at day 6. Scale bars, 20 μm. Results in (C) are representative of at least three independent experiments.

(D) SEM image of an engineered tissue at day 6. Red arrows indicate Paneth cells (left) and a brush border of microvilli (right). Scale bars, 5 μm. Data in (D) are representative of two independent experiments, and at least 10 different crypt-like structures were analyzed.

(E) Lgr5 lineage tracing strategy within engineered tissues.

(F) Lgr5 lineage tracing experiments at day 6. Green (mG) ribbons, representing tracts of labeled cells descendant from recombined Lgr5⁺ cells, can be traced from the bottom of the crypt to the exterior surfaces. Scale bars, 100 μm (left, middle) and 20 μm (right).

(G) Quantification of microwells containing cell patches of Lgr5-lineage at days 3 and 6.

(H and I) Normalized mTomato expression and quantification in parent Lgr5-eGFP⁺ (blue) and mGFP⁺ (red) cells at 0 h (H) and 48 h (I).

(J) Krt20 lineage tracing strategy within engineered tissues.

(K) Krt20 lineage tracing experiments at day 6. Green (mG) patches of labeled cells are descendant from recombined Krt20⁺ cells, which can mostly be detected on the exterior surfaces. Scale bars, 100 μm (left, middle) and 20 μm (right).

(L) Quantification of microwells containing cell patches of Krt20-lineage at days 3 and 6.

(M and N) Normalized mTomato expression and quantification in mGFP^- (all non-mGFP⁺ cells) (mTomato, blue) and mGFP⁺ (red) cells at 0h (M) and 48h (N).

Results in (F), (G), (K), and (L) are representative of four biologically independent experiments and presented as means ± SD. Results in (H), (I), (M), and (N) are representative of three biologically independent experiments where seven (H and M), ten (I), and five (N) engineered tissues were analyzed and presented as means ± SD. Statistical significance was assessed using unpaired t test.

See also Figure S2.

Next, we extended cultures of engineered tissues from 3 to 6 days (Figure S2B). Temporal cell marker analysis demonstrated that cells expressing Lgr5, a marker of adult stem cells in vivo,¹⁰ emerged at the base of the crypt-like structures between days 4 and 5 (Figures 2C, S2C, S2D, and S2G) and by day 6 reached a frequency comparable to the mouse small intestine (0.41% ± 0.11%; Figure S2C). Conversely, keratin (Krt) 20 expression, which in vivo marks differentiated cells on villi,⁴ was primarily detected on cells located on the flat exterior surfaces connecting crypt-like structures (Figure 2C). Moreover, by day 6, Paneth cells could be detected in crypt-like structures based on the expression of Lysozyme1 and the presence of electron-dense secretory granules by SEM (Figures 2C, 2D, and S2E–S2G). Lastly, at day 6, a uniform brush border of microvilli was positioned along the outer surfaces (Figures 2D and S2E). Thus, over the course of 6 days, engineered tissues gain important features recapitulating in vivo form and cellular phenotypes.

In the fetal epithelium, Krt20⁺ villus cells can generate Lgr5⁺ stem cells that persist into adulthood.⁴ However, soon after birth, the long-term self-renewal potential of intestinal cells becomes restricted to Lgr5⁺ stem cells.^4,25 To determine whether Lgr5⁺ and Krt20⁺ cells in engineered tissues function comparably to their adult counterparts, we performed lineage tracing experiments using tamoxifen-inducible Lgr5- and Krt20-driven CreER models (Figures 2E and 2J). When labeling Lgr5-expressing cells, a fluorescent label could be detected in 76% ± 5% of crypt-like structures at day 4, which was maintained in 51% ± 2% of crypt-like structures for 2 days (Figures 2F and 2G). The decline in the fraction of Lgr5⁺ crypt-like structures is consistent with neutral drift dynamics observed in vivo, whereby clonal loss within the stem cell niche permits the expansion of remaining clones.^26,27,28 At day 6, labeled ribbons extending from the bottom to the top of the crypt-like structures could be observed (Figure 2F). Furthermore, flow cytometry analysis exhibited a rapid reduction in mTomato signal in the labeled cells, reflecting the replicative role of Lgr5⁺ cells in the formation and maintenance of the engineered tissues (Figures 2H and 2I). By contrast, and consistent with the observation that Krt20 was mostly expressed on the exterior surfaces, only 9% ± 4% and 22% ± 7% of the crypt-like structures contained labeled cells from the Krt20 lineage on days 4 and 6, respectively, consisting of patches of a few cells with limited proliferative capacity (Figures 2K–2N, S2H, and S2I). Interestingly, ablation of Lgr5+ cells using Lgr5^-eGFP-DTR model results in the failure to form or maintain crypt-like structures (Figure S2J). Together, these data demonstrated that culturing immature intestinal cells as engineered tissues enables not only the correct positioning of cellular markers but also the appropriate state-specific cellular behaviors observed in the mature epithelium in vivo.

Geometry promotes multilineage commitment

Having established that bioengineered scaffolds induce maturation of fetal intestinal cells, we next investigated how geometry specifically influences the cellular composition of the epithelium. To address this, we first performed gene expression analysis for markers of the absorptive and secretory lineages on patterned areas, flat surfaces, and conventional 3D organoid cultures, where immature intestinal cells are cultured with the same growth factors (Figures 3A and 3B). The levels of the absorptive lineage marker Fabp2 were similar across the three conditions (Figure 3B). This was in stark contrast to all analyzed markers of secretory cells (Lyz1, Defa6, Muc2, and ChgA), which were detected at significantly higher levels on the patterned versus the flat areas (Figure 3B). This suggests a geometry-driven emergence of these cell lineages that also mark the transition into a mature epithelium.

Figure 3.

Cellular diversity in engineered tissues reveals an abundance of secretory cell types

(A) Bright-field images of fetal cells cultured in a conventional organoid setting (3D) and engineered tissues divided into flat (F) and patterned (P) regions. Scale bars, 100 μm.

(B) mRNA expression levels of Fabp2, Defa6, Muc2, ChgA, and Lyz1. F, flat exterior surface, P, patterned area. Results are representative of eight independent experiments and presented as means ± SD. Statistical significance was assessed using the one-way ANOVA test.

(C–E) UMAP visualization of scRNA-seq clusters (C), shown separately for fetal cells cultured as 3D organoids (D) or as engineered tissues (E).

(F) Cell similarity score between fetal cells cultured in 3D or as engineered tissues and a reference dataset of adult intestinal epithelium.¹⁹ The p value was calculated using the Wilcoxon rank sum test.

(G) Atoh1 lineage tracing strategy within engineered tissues.

(H) Atoh1 lineage tracing experiments at day 6. Green (mG) cells representing secretory descendants from recombined Atoh1⁺ cells. Scale bars, 20 μm.

(I) FACS profiles showing distribution of GFP⁺ cells in 3D cultures (left) and engineered tissues (right) at day 6.

(J) Quantification of the percentage of Atoh1-lineage cells in 3D cultures or engineered tissues (ET) by flow cytometry.

Results in (H), (I), and (J) are representative of five biologically independent experiments and presented as means ± SD. Statistical significance was assessed using unpaired t test.

See also Figures S3 and S4.

To investigate whether the emergence of crypt-like features depends on a specific microwell size, we cultured engineered tissues on microfabricated scaffolds with larger microwells (d = 200 μm) (Figure S3A). After 6 days, differentiation markers emerged in the patterned regions compared with the flat exterior surfaces (Figure S3B), highlighting the role of geometry in tissue maturation. However, confocal imaging and 3D reconstruction of the resulting engineered tissues revealed that 87% ± 12% of the large crypt-like structures exhibited aberrant patterning, with domains either growing inward toward the lumen or invading outward into the surrounding matrix (Figures S3C and S3D). This phenomenon was never observed in the 50 μm diameter microwells, which match the average size of a small intestinal crypt in vivo.¹⁷ These findings further underscore the importance of optimal cavity size in tissue patterning. We excluded the condition of culturing cells on flat surfaces from further analyses, as it displayed no significant differences when compared with fetal cells cultured in conventional 3D organoid cultures.

To resolve the cellular diversity and the relative abundance of cell types in engineered tissues grown on standard 50 μm microwells versus conventional 3D cultures, we performed single-cell RNA sequencing (scRNA-seq) and profiled the transcriptomes of 1,660 cells across both systems. Based on previously described cell-type markers in the fetal⁴ as well as adult²⁹ intestinal epithelium, the single-cell profiles were separated into 6 transcriptionally distinct clusters consisting of enteroendocrine cells (EECs), goblet/Paneth cells, highly proliferative stem/progenitor cells, fetal cells, and two distinct populations of enterocytes (ECs) 1 and 2, expressing both bottom and tip EC markers³⁰ (Figures 3C and S4A–S4H). Interestingly, the 3D cultures retained a substantial proportion (approx. 80%) of stem/progenitor cells, indicative of their undifferentiated state, whereas only a small fraction (7.6%) of these cells were detected in the engineered tissues (Figures 3D, 3E, and S4I). Strikingly, engineered tissues contained a substantially higher proportion of EECs and goblet/Paneth cells, suggesting that the engineered tissue architecture promotes differentiation of these terminal cell populations that are otherwise difficult to produce in vitro.³¹ Importantly, comparison with reference datasets from adult intestinal epithelium,²⁹ adult organoids,³² and fetal progenitors (FEnS)³² revealed that engineered tissues align closely with adult epithelium and adult organoids, whereas 3D cultures show a higher degree of similarity to FEnS, indicating a more mature state in engineered tissues (Figures 3F, S4J, and S4K).

To corroborate the observation of increased numbers of secretory cells in engineered tissues, we performed Atoh1-CreERT²;Rosa26-mTmG (Atoh1;mTmG) fate mapping (Figure 3G), labeling cells that differentiate into the secretory lineage.³³ Notably, while in 3D cultures only a small portion (6%) of cells were labeled with GFP, 23% of the cells within engineered tissues were derived from the Atoh1 lineage (Figures 3H–3J). Collectively, these findings demonstrate that the cellular diversity in engineered tissues is very similar to adult intestinal epithelium and includes cell types that are rare or absent in conventional 3D cultures of fetal cells.^13,14

Cell crowding triggers differentiation

Time-resolved analyses of engineered tissues revealed that by day 3, cells had colonized the surface but were still undifferentiated, whereas by day 6, a considerable number of differentiated cells could be detected. To determine the mechanisms governing this cell state transition, we investigated the morphological changes that occur between days 3 and 6. We conducted whole-mount imaging of engineered tissues, followed by 3D segmentation of nuclei and cell volume, enabling quantitative assessment of tissue packing and cell density (Figures 4A–4C, S5A, and S5B). This analysis revealed that the number of cells in the crypt-like structures increased by 1.4-fold between days 3 and 6, accompanied by a decrease in the distance between neighboring nuclei (Figure 4D) and a significant reduction of cell volume throughout crypt-like structures, but not in the surrounding flat surfaces (Figure 4E). Collectively, this suggested a localized increase of cellular crowding within a crypt-like structure at day 6. Importantly, a similar relationship between cellular crowding and differentiation could also be identified in vivo, with a decreased distance between neighboring nuclei correlating with the emergence of secretory Paneth cells (Figures S5C–S5G).

Figure 4.

Cellular crowding impacts differentiation in engineered tissues

(A) Strategy for segmentation of nuclei in engineered tissues from Vil^Cre;Rosa26-mTmG mice. Scale bars, 20 μm.

(B and C) Representative images of 3D rendering of the nuclei (B) and total cells (C) in engineered tissues at days 3 and 6. In (C), Numbers 1–4 indicate positions within engineered tissue. Scale bars, 20 μm.

(D) Quantification of nuclei (left) and average distance to three nearest neighboring nuclei (right) in crypt-like structures at days 3 and 6. Results are representative of three biologically independent experiments where fifteen crypt-like structures were evaluated and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(E) Quantification of cell volume in the base (position 1), middle (position 2), top of the crypt-like structure (position 3), and the flat surface (position 4). Each data point represents a single cell’s volume. Results are collected from three biologically independent experiments and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(F) Atoh1 lineage tracing strategy within engineered tissues.

(G and H) Atoh1 lineage tracing experiments at day 4 at two seeding densities: standard (G) and 5× diluted (H). Green (mG) ribbons, representing tracts of labeled cells descendant from recombined Atoh1⁺ cells. Scale bars, 100 μm (outset) and 20 μm (inset).

(I) Quantification of microwells containing Atoh1-lineage cells. Results are representative of three biologically independent experiments and presented as means ± SD.

(J) Representative XY, XZ, and YZ projections from Atoh1-lineage traced engineered tissues. Scale bar, 20 μm.

(K) Quantification of GFP⁺ cell number per crypt-like structure in engineered tissues cultured under standard seeding conditions (control) or 5-fold lower seeding density (5× dilution). Each data point represents a single crypt-like structure.

(L) Quantification of distance to three nearest neighboring nuclei in crypt-like structures from standard seeding conditions (red) or 5× diluted (blue) encircled in (K).

See also Figure S5.

To functionally test whether cell density affects the emergence of differentiated cells, we performed fate mapping using the Atoh1;mTmG model (Figure 4F) while culturing fetal cells at two initial seeding densities (standard [control] and a 5-fold lower [5× dilution]; Figures 4G and 4H). By day 4, 53% ± 6% of crypt-like structures contained labeled cells under standard conditions, compared with only 8% ± 7% at the lower seeding density (Figure 4I). Confocal microscopy of individual crypt-like structures revealed a clear correlation between total cell number and the emergence of GFP⁺ cells (Figures 4J and 4K). Structures containing GFP⁺ cells exhibited shorter inter-nuclear distances and a more uniform level of compaction compared with non-labeled counterparts (Figure 4L). These findings suggest that a critical cell density is required to trigger differentiation within engineered tissues, potentially through density-dependent cues linked to tissue architecture.

YAP1 regulates intestinal maturation

YAP1 acts as a nuclear relay of cell density, with increased cell crowding resulting in its sequestration within the cytoplasm.^34,35 Moreover, the fetal intestinal epithelium is characterized by high levels of YAP signaling.³⁶ Consistent with the described regulation of YAP1 activity,³⁶ we observed widespread nuclear YAP1 across entire crypt-like structures at day 3, which was markedly reduced by day 6, irrespective of microwell size (Figures 5A, 5B, and S5H). To test whether crowding-induced differentiation is mediated through YAP, we again modulated initial seeding density while lineage-tracing Atoh1⁺ cells (Figure 4F). Lower-density cultures (5× dilution) retained a higher fraction of nuclear YAP⁺ cells specifically in the microwells (Figures 5C and 5D) and displayed increased inter-nuclear distances (Figure 5E). This observation links nuclear YAP1 with an undifferentiated cell state in engineered tissues, consistent with its role in vivo. To assess whether forced activation of YAP1 rescued the effects of geometry-induced maturation and differentiation, we established engineered tissues from TetO-hYAP^S127A;R26-rtTA (YAP1-OE) mice and overexpressed a constitutively active mutant of YAP1 (S127A) in a doxycycline-inducible manner (Figure 5F). Importantly, YAP1 activation did not impair the formation of engineered tissues (Figure 5G) or final cellular density (Figures 5H and 5I). It did, however, abolish the geometry-induced upregulation of differentiation genes (Lyz1, Defa6, Muc2, ChgA, and Fabp2; Figure 5J) and increase expression of fetal-associated genes (Ly6a and Anxa1; Figure S5I).

Figure 5.

In vitro differentiation is governed by YAP1 activity

(A) Detection of YAP1 (red) and DAPI (white) in sections of engineered tissues at days 3 and 6. Red arrows indicate nuclear YAP1 expression. Scale bars, 10 μm.

(B) Quantification of nuclear YAP1 expression in top (above), middle (middle), and bottom (below) layers of the crypt-like structure. Results are representative of three independent experiments where in total 5 crypt-like structures were evaluated and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(C) Detection of Atoh1-lineage cells (green), YAP1 (red), and DAPI (white) in sections of engineered tissues at day 4 at two seeding densities: control (top row) and 5× diluted (bottom row). Scale bars, 20 μm.

(D) Quantification of nuclear YAP1 expression at two seeding densities. S, surface; M, microwell. Results are representative of three biologically independent experiments where three crypt-like-structures were evaluated and presented as means ± SD. Statistical significance was assessed using one-way ANOVA.

(E) Quantification of distance to nearest neighboring nuclei in crypt-like structures at two seeding densities. Results are representative of three biologically independent experiments where three crypt-like-structures were evaluated and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(F) Strategy for forced activation of YAP1 in engineered tissues.

(G) Bright-field images of engineered tissues forming with or without the presence of doxycycline (DOX). Scale bars, 100 μm (outset) and 20 μm (inset).

(H) Detection of DAPI+ (white) cells in TetO-hYAP^1S127A;R26-rtTA engineered tissues treated with or without doxycycline. Scale bars, 20 μm.

(I) Quantification of the number cells in TetO-hYAP^1S127A;R26-rtTA engineered tissues treated with or without doxycycline. Results are representative of six crypts-like structures collected from three independent engineered tissues and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(J) mRNA expression levels of hYAP1, Lyz1, Defa6, Muc2, and ChgA from cells cultured on flat (F) and patterned (P) regions of engineered tissues with or without doxycycline (DOX) treatment. Results are representative of at least three independent experiments and presented as means ± SD. Statistical significance was assessed using one-way ANOVA test.

To assess the in vivo relevance of our model, we employed both gain- and loss-of-function mutants (Figures 6A and 6B). Consistent with our findings in engineered tissues and previous reports in adult intestine,^37,38 YAP activation during crypt formation (P2-P5) reduced the number of Atoh1⁺ secretory progenitors by 2.9-fold and eliminated Lyz1⁺ secretory cells, as measured by RNA in situ hybridization (Figures 6C and 6D). Conversely, dual loss of YAP and TAZ in VilCre;YAP^fl/flTAZ^fl/fl mice increased these populations 1.7 and 2.9, respectively (Figures 6E and 6F). Neither genetic activation nor inactivation altered cellular density, mirroring the results from engineered tissues (Figures 6G and 6H). Collectively, this showed a strong concordance between cellular crowding, reduction in nuclear YAP expression, and differentiation that can be modulated by YAP signaling activity. Thus, YAP1 regulation serves as a link between tissue architecture and a key cellular function, differentiation, during intestinal maturation.

Figure 6.

Intestinal maturation is governed by YAP1 activity

(A and B) Strategies for forced activation of YAP1 (A) and deletion of YAP and TAZ (B) during intestinal maturation in vivo.

(C) Co-detection of Atoh1 (red) and Lyz (green) in TetO-hYAP^1S127A;R26-rtTA small intestine at P5. Scale bars, 20 μm (outset) and 5 μm (inset).

(D) Quantification of Atoh1⁺ and Lyz⁺ cells in TetO-hYAP^1S127A;R26-rtTA small intestine at P5. Results are collected from three (−TetO) and four (+TetO) biological replicates and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(E) Co-detection of Atoh1 (red) and Lyz (green) in VillinCre;Yap^fl/fl;Taz^fl/fl small intestine at P5. Scale bars, 20 μm (outset) and 5 μm (inset).

(F) Quantification of Atoh1⁺ and Lyz⁺ cells in VillinCre;Yap^fl/fl;Taz^fl/fl small intestine at P5.

Results are collected from four independent experiments using material from two biological replicates and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(G) Quantification of the number of cells in crypt-forming units of TetO-hYAP^1S127A;R26-rtTA mice. Results are collected from three (−TetO) and four (+TetO) biological replicates and presented as means ± SD. Statistical significance was assessed using unpaired t test.

(H) Quantification of the number of cells in crypt-forming units of VillinCre;Yap^fl/fl;Taz^fl/fl mice. Results are collected from four independent experiments using material from two biological replicates and presented as means ± SD. Statistical significance was assessed using unpaired t test.

Modeling epithelial response to PGE₂-induced inflammation

During regeneration, intestinal epithelial cells reactivate developmental gene programs that converge upon the activation of YAP.^39,40,41 However, how this regenerative state is resolved during the return to homeostasis remains unresolved. To investigate how geometry influences the epithelial response to inflammatory cues, we treated adult small intestinal organoids with prostaglandin E₂ (PGE₂), a key inflammatory mediator elevated in inflammatory bowel disease and associated with YAP-mediated activation in animal models of regeneration^42,43,44 (Figures 7A–7C). PGE₂ treatment induced strong upregulation of Sca1, a marker of tissue damage^39,40 (Figure 7D). Next, we purified cells by flow cytometry based on Sca1 expression and examined the resolution of the YAP^high state in engineered tissues versus 3D cultures. By day 4, cells in engineered tissues displayed a columnar morphology, whereas 3D cultures retained a smooth fetal-like phenotype¹⁴ (Figure 7E). Consistently, the expression of regeneration-associated markers (Anxa1, Clu, and Ly6a) and YAP target genes (Ctgf and Cyr61) was suppressed in engineered tissues when compared with 3D cultures (Figure 7F). These findings demonstrate that engineered scaffolds impose geometric cues that actively drive resolution of the inflammation-associated YAP^high state, providing a deterministic platform to study cell state transitions.

Figure 7.

Modeling epithelial response to PGE2-induced inflammation

(A) Strategy for investigating epithelial restitution using engineered scaffolds. W, Wnt; E, EGF; N, Noggin; R, Rspondin 1.

(B and C) Bright-field images of adult small intestinal organoids (B) before mechanical dispersion and (C) following 3 days of culture without (top) or with (bottom) PGE₂. Scale bars, 100 μm (inset) and 200 μm (outset).

(D) FACS profile showing distribution of Sca1⁺ cells following vehicle (top) or PGE2 (bottom) treatment.

(E) Bright-field images of PGE₂-challenged Sca1⁺ cells cultured as 3D (top) or as engineered tissues (bottom). Scale bars: 3D culture 100 μm (inset) and 200 μm (outset), engineered tissue 20 μm (inset) and 100 μm (outset).

(F) mRNA expression levels of Anxa1, Clu, Ly6a, Ctgf, and Cyr61. 3D, 3D culture; ET, engineered tissue. Results are representative of 4 independent experiments and presented as means ± SD. Statistical significance was assessed using unpaired t test.

Discussion

Understanding the relationship between organ architecture and function has been a longstanding challenge,⁴⁵ often described by the principle of “structure follows function.”⁴⁶ This concept implies that a tissue’s shape is tailored to support its physiological role. In line with this, previous studies have shown that tissue geometry can instruct the fate of organoid-derived intestinal cells.¹⁵ Here, we present evidence that mucosal geometry acts as a fundamental template for initiating cellular differentiation during intestinal maturation. As the organ develops, these geometric features become essential, creating a temporal window that enables the transition from an immature fetal structure to a fully mature tissue capable of executing functions such as digestion, nutrient absorption, and response to the environment. Importantly, we can precisely control this transition using microstructured hydrogel scaffolds that emulate the in vivo crypt topography.

Local tissue structure and signaling interactions between the mesenchyme and epithelium collectively govern the progression of tissue morphogenesis toward a homeostatic state in the intestine.^2,24 However, as the tissue approaches its homeostatic size and architecture and begins to perform its required functions, cells start to experience spatial constraints within their immediate microenvironment, imposing limitations on further growth. The compaction leads to a reduction in local YAP activity, liberating cells from their YAP1-driven immature state, thereby enabling them to initiate terminal differentiation.⁴⁷ Interestingly, compaction has been demonstrated to induce changes in epigenetic cell states in other epithelial systems.^48,49 It is tempting to speculate whether similar effects play a role in safeguarding intestinal maturation.^32,50

Previous work by Liberali and colleagues identified a YAP-dependent symmetry-breaking mechanism during adult organoid formation following single-cell passaging,⁵¹ underscoring the essential role of YAP in adult organoid formation.^52,53 This process parallels the transient activation of YAP-driven regenerative programs.^36,39,40,41 Similarly, fetal epithelial cells, which have not yet acquired mature functional properties, are maintained in a YAP1-high state.^14,36 Through a combination of bioengineering and in vivo approaches, we show that the fetal-to-adult transition is regulated by cellular confinement within emerging crypts, which suppresses YAP-mediated signaling. This mechanism can also be exploited to resolve inflammation-associated epithelial states. Thus, spatial control of YAP activation enables epithelial maturation and the acquisition of functional properties. The engineered scaffolds provide a deterministic platform to guide this process, illustrating that tissue geometry is not merely a passive structural feature but an active regulator of specific signaling networks. Manipulating tissue architecture may therefore represent a complementary approach to control cell state transitions during development, homeostasis, and repair.

Regenerative medicine holds promise for generating cells for disease modeling and cellular therapies using readily available sources. However, current differentiation protocols from pluripotent stem cells often result in immature cell populations exhibiting characteristics akin to either first- or second-trimester phenotypes.⁵⁴ While such models are well-suited for investigating congenital disorders that impact fetal development, it can be contended that they have limitations in modeling aspects of adult tissues. Considering our observations, it is tempting to speculate whether substrates designed to replicate the spatial organization of an in vivo niche could facilitate more efficient maturation processes.

Limitations of the study

The engineered tissue system is epithelial in composition and does not incorporate mesenchymal or immune components that contribute to crypt formation and intestinal homeostasis in vivo. As a result, paracrine and inflammatory signaling events that influence epithelial maturation cannot be modeled directly in this simplified model. Nevertheless, this reductionist design enables exquisite control of the signaling environment through its defined composition of growth factors and matrix components. While the engineered scaffolds provide precise geometric control of epithelial organization, additional biophysical parameters, such as matrix stiffness and growth factor gradients, remain interlinked and may further modulate signaling dynamics and cellular differentiation.

The composition of the matrix also imposes practical constraints. Because Matrigel alone lacks sufficient mechanical integrity to sustain patterned structures, it was combined with type I collagen to generate stable scaffolds. However, a 1:3 mixture of Matrigel to type I collagen does not support the growth of fetal intestinal cells, precluding a perfectly matched material comparison between the two culture systems. We therefore used pure Matrigel for 3D organoid culture, consistent with current standards in the field. Notably, adult intestinal epithelial cells tolerate both type 1 collagen matrices alone and a mixture with Matrigel, highlighting that fetal and adult intestinal cells are functionally distinct and respond differently to matrix-derived biochemical and mechanical cues.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Kim B. Jensen (kim.jensen@sund.ku.dk).

Materials availability

Unique reagents generated in this study are available from the lead contact upon request.

Data and code availability

Single-cell RNA-seq data have been deposited at GEO: GSE247255. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank K. Christopher Garcia, Wim de Lau, and Hans Clevers for providing sequence-optimized surrogate WNT; H. Clevers (Lgr5-eGFP-ires-creERT2), F. de Sauvage (Genentech; Lgr5-iDTR-eGFP), D. Winton (Atoh1-CreERT2), S. Robine (VilCre), J. Larsson (tetO-hYAPS127A), D. Pan (YAPfl/fl), and S. Piccolo (Tazfl/fl) for gifts of mice; the animal caretakers at the MGU, 10–3, and Biocenter; G. Dela Cruz and the reNEW Flow Cytometry Platform for support; J. Bulkescher, J. Viswalingam Bagge, and the reNEW Imaging Platform; and members of our respective labs for comments and technical assistance. We thank all engineers of the Center of Micronanotechnology (CMi, EPFL) for support in microfabrication. We thank Irina Kolotuev from the Electron Microscopy Facility of the University of Lausanne for help with sample preparation and SEM imaging. We thank Lingbing Zhang for designing the graphical abstract. K.B.J. funding: this work was supported by the European Union’s Horizon 2020 research and innovation program (grant agreements INTENS 668294 and STEMHEALTH ERCCoG682665) and the Novo Nordisk Foundation (NNF17OC0028730, NNF18OC0034066, and NNF20OC0064376, K.B.J.). J.G. was supported by the Marie Curie Fellowship (656099/H2020-MSCA-IF-2014). M.L. funding: this work was funded by support from the Swiss National Science Foundation (SNSF) research grant 310030_179447, the EU Horizon 2020 Project INTENS (#668294-2), and Ecole Polytechnique Fédérale de Lausanne (EPFL). S.A.W. funding: this work was supported by the European Union’s Horizon 2020 research and innovation programme (grant agreement 770877–STEMpop), the Academy of Finland Center of Excellence BarrierForce, and the Max Planck Society. The Novo Nordisk Foundation Center for Stem Cell Medicine is supported by Novo Nordisk Foundation grant number NNF21CC0073729.

Author contributions

M.M., M.N., M.P.L., and K.B.J. designed the experiments. M.M. and M.N. performed the experiments with assistance from H.L.L., R.B.B., M.Z., E.F., S.L.H., A.M.B., I.B., and J.G. M.M., C.L., F.B., R.B.B., and A.G. analyzed the data. M.M., M.N., M.P.L., and K.B.J. interpreted the results. M.P.L., S.A.W., and K.B.J. provided resources. M.M. and K.B.J. wrote the manuscript with contributions from all authors.

Declaration of interests

The Ecole Polytechnique Fédérale de Lausanne has filed for patent protection (EP16199677.2, PCT/EP2017/079651, and US20190367872A1) on the scaffold-guided organoid technology described herein, and M.N., N.G., and M.P.L. are named as inventors on those patents; M.N., N.G., and M.P.L. are employees of F. Hoffmann-La Roche.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-E-Cadherin Clone 36/E-Cadherin (RUO)	BD Bioscience	610181; RRID: AB_397580
CD44var (v6) Monoclonal Antibody (9A4)	Invitrogen	BMS145; RRID: AB_10597738
Lysozyme 1(EC.3.2.1.17)	DAKO	A0099; RRID: AB_2341231
EpCAM (Clone G8.8)	ThermoFisher	#17-5791-82; RRID: AB_2716944
BD Pharmingen™ PE-Cy™7 Rat Anti-Mouse CD45	BD Bioscience	552848; RRID: AB_394489
CD31 (PECAM-1) Monoclonal Antibody (390), PE-Cyanine7	ThermoFisher	25-0311-82; RRID: AB_2869771
Keratin K20	Nordic Biosite/Progen	318-GP-K20; RRID: AB_1541036
YAP (D8H1X)	Cell Signaling	#14074; RRID: AB_2650491
Alexa Flour 488 Donkey anti Rabbit IgG (H+L)	ThermoFisher	A-21206
Alexa Fluor 488 Donkey anti Rat IgG (H+L)	ThermoFisher	A-21208
Alexa Fluor 555 Goat anti mouse IgG2a	ThermoFisher	A21137
Chemicals, peptides, and recombinant proteins
4-hydroxytamoxifen	Sigma-Aldrich	H6278
RapiClear 1.52	Sunjin Lab	RC152002
Y-27632 dihydrochloride	Sigma-Aldrich	Y0503
N-Acetyl-L-cysteine	Sigma-Aldrich	A9165
Next-Generation Surrogate Wnt	K. Christopher Garcia	Miao et al.55
Human EGF	Peprotech	AF-100-15
Murine Noggin	Peprotech	250–38
R-spondin 1 conditioned medium	in-house production	N/A
N2	Gibco	17502048
B27	Gibco	12587010
Diphtheria Toxin	Merck	D0564
Doxycycline hyclate	Sigma-Aldrich	D9891
Critical commercial assays
TotalSeq hashtag antibodies	Biolegend	A0301, A0302, A0303, and A0304
Rneasy Micro Kit	Qiagen	74004
Chromium Next GEM Single Cell 3′ Kit v3.1	10× Genomics	PN-1000263
Library Construction Kit, 16 rxns	10× Genomics	PN-1000190
Chromium Next GEM Chip K Single Cell Kit	10× Genomics	PN-1000287
Dual Index Kit TT Set A	10× Genomics	PN-1000215
RNAscope Multiplex Fluorescent Reagent Kit v2	ACD Biotechne	323100
RNAscope probe mm-Atoh1	ACD Biotechne	408791-C3
RNAscope probe mm-Lyz1	ACD Biotechne	415131
Deposited data
Single cell RNA-seq data	This study	GSE247255
scRNA-seq dataset from adult intestinal epithelium	Haber et al.29	GSE92332
scRNA-seq dataset from adult organoids	Hansen et al.32	GSE221278
scRNA-seq dataset from fetal progenitors	Hansen et al.32	GSE221278
Segmentation dataset and model	This study	https://zenodo.org/records/17565548
Experimental models: Organisms/strains
C57BL/6J	Taconic	17758-F
Top.CFP	Palle Serup	Serup et al.56
Lgr5-eGFP-IRES-CreERT2	Hans Clevers	Barker et al.10
Lgr5-iDTR-eGFP	Frederic de Sauvage	Tian et al.57
VilCre	Sylvie Robine	el Marjou et al.58
tetO-hYAPS127A	Jonas Larsson	Jansson et al.59
Atoh1-CreERT2	Doug Winton	Tomic et al.60
Tazfl/fl	Stefano Piccolo	Azzolin et al.52
Yapfl/fl	Duojia Pan	Zhang et al.55
R26-TTA/TTA:Col1a1-tetO-H2B-mCherry	Jackson Laboratory	014602
Krt20-T2A-CreERT2	Jackson Laboratory	30600
Rosa26mT/mG	Anne-Grapin Botton	Muzumdar et al.61
Software and algorithms
FlowJo	BD Life Sciences	https://www.flowjo.com/flowjo/overview
GraphPad Prism software	Dotmatics	https://graphpad.com/features
Fiji/ImageJ	ImageJ	https://imagej.net/Fiji
R Studio	R Studio	https://rstudio.com/
Adobe Creative Cloud	Adobe	https://www.adobe.com/dk/creativecloud.html
LAS X	Leica	https://www.leica-microsystems.com/products/microscope-software/p/leica-las-x-ls/
ZEN Microscopy Sofrware	Zeiss	https://www.zeiss.com/microscopy/en/products/software/zeiss-zen.html
Imaris	Oxford Instruments	https://imaris.oxinst.com/
MATLAB	MathWorks	https://www.mathworks.com/products/matlab.html
Cell Ranger	10× Genomics	https://www.10xgenomics.com/support/software/cell-ranger/latest

Mice

All animals were housed in specific pathogen-free animal facilities, in either open or individually ventilated cages always with companion mice, and cages were placed under a 12-h light–dark cycle. Food and water were provided ad libitum. Randomized cohorts including both male and female animals were distributed in an unblinded manner into the experimental time points for analysis. Sample sizes were selected to provide sufficient statistical power for analysis.

Unless otherwise specified, C57BL/6J mice (purchased from Taconic) were used for all the experiments. Top.CFP,⁵⁶ Lgr5-eGFP-IRES-CreER^T210, Krt20-T2A-CreER^T2 (Jackson laboratories, stock no. 30600), Lgr5-DTR-eGFP,⁵⁷ ROSA^mT/mG62, Atoh1-CreER^T260 (kind gift from D. Winton), Vil^Cre58 (kind gift from S. Robine) tetO-hYAP^{S127A 55} (kind gift from J. Larsson) Rosa26-rtTA (Jackson laboratories, stock no. 014602), Yap^fl/fl55 (kind gift from D. Pan) and Taz^fl/fl52 (kind gift from S. Piccolo) murine lines used in the experiments have been described. For in vivo YAP activation, mice were injected with 40 μg doxycycline per 1 g of mouse followed by supplementation of doxycycline in food pellets (2 g/1 kg, Inotiv) and in drinking water (2 g/1 l). None of the animals used in these studies had been subjected to prior procedures and all were drug- and test-naive. The Danish animal inspectorate reviewed and approved all animal procedures.

Method details

Whole-mount immunostaining of in vivo intestinal tissue

Proximal small intestinal samples from P0, P2, P5 and P16 mice were fixed with 4% PFA overnight at 4 °C followed by dehydration in methanol. Samples were stored in methanol at −20 °C. The tissue was rehydrated in a PBS series before staining with all steps being performed at 4 °C. Samples were blocked and permeabilized in 1% BSA, 0.5% Triton X-100 (Triton X-100 Surfact-Amps Detergent Solution, Thermo Fisher) in PBS for 24 h. Primary antibodies (listed in key resources table) were incubated in 1% BSA, 0.5% Triton X-100 in PBS for 48 h. Tissue was subsequently washed in 0.5% Triton X-100 in PBS overnight. Secondary antibodies (listed in key resources table) were incubated in 1% BSA, 0.5% Triton X-100 in PBS for 48 h and the tissue was subsequently washed overnight with 0.5% Triton X-100 in PBS. DAPI was used to counterstain nuclei in the indicated experiments. Samples were then dehydrated in methanol and kept at −20 °C. Samples were cleared using 1:2 benzyl alcohol:benzyl benzoate (BABB) (Sigma). In brief, samples

were processed through 5 changes (1 min each) of BABB:methanol 1:1, followed by 5 changes (2 min each) of BABB. The samples were subsequently mounted within a Fast well (FW20-FastWells 20-mm diameter × 1.0-mm depth or 25 × 25 mm, Grace-Biolabs).⁴ Z-stack images were acquired using laser-scanning confocal microscopy (Leica Stellaris 5).

Image analysis to capture crypt progenitor units (CPUs)

Following antibody staining for E-cadherin, CD44v6 and DAPI, the images were imported into Imaris v10.0.1 (Bitplane, Zurich, Switzerland) where CPUs were segmented for qualitative analysis using surface rendering function. First, the surface of the epithelium was masked using E-cadherin staining. Subsequently, CD44v6 staining was displayed on epithelium surface allowing the masking of CPUs. All CPUs were manually processed to ensure each unit was an individual object, allowing us to perform quantitative measurements. The diameter was defined as the longest ellipsoid axis, which could be contained within the crypt progenitor unit. Structure below 20 μm were manually removed from quantification.

Hydrogel holders and microtopography stamps fabrication

For hydrogel casting on the bottom of the 3.5 cm dish, special ring-like PDMS holders were fabricated (see below - Microfabrication of engineered scaffolds with controlled geometry). These hydrogel holders defined resulting hydrogel thickness to 1 mm and had overhanging features to hold the hydrogel block attached to the bottom of the plate. Holders were fabricated using conventional soft-lithography methods established at the Center of Micronanotechnology (CMi, EPFL). In brief, the array of circular rings (∅ = 10 mm) was drawn using CleWin (Phoenix Software). The designed layout was written with a diode laser onto a fused silica plate coated with chrome and positive photoresist (Nanofilm) using an automated system (VPG200, Heidelberg Instruments). Exposed photoresist was removed with a developer (DV10, Süss MicroTec), and the chrome layer underneath was etched with an acid–oxidizer solution of perchloric acid, cerium ammonium nitrate, and water. The resulting mask was developed with TechniStrip P1316 (Microchemicals) to remove the residual resist and extensively washed with ultrapure water. The mold was made from double-layered epoxy-based negative photoresist SU8. First, a 250 μm thick layer of SU8 GM1075 (Gerlteltec) photoresist was cast onto a dehydrated silicon wafer using a negative resist coater (LMS200, Sawatec). After baking at 110°C for 4 h, a second 250 μm thick layer of the SU8 GM1075 was coated, resulting in total thickness of a 500 μm. After second bake, this wafer was aligned and exposed to ultraviolet (UV) radiation through the mask (MA6/BA6, Süss MicroTec). After the postexposure bake at 85°C for 5 h, the wafer was developed with propylene glycol monomethyl ether acetate (Sigma) and baked at 135°C for 4 h. The wafer was then plasma-activated and silanized with vapored trichloro (1H,1H,2H,2H-perfluorooctyl) silane (Sigma-Aldrich) overnight. This wafer was then used for polydimethylsiloxane (PDMS) molding (Sylgard 184, Dow Corning). Ten weight-parts of elastomer base were vigorously mixed with one part of curing agent and poured onto the mold. After degassing under vacuum, PDMS was baked for 24 h at 80°C. The resulting PDMS replica was plasma-bonded to another 300 μm thin PDMS sheet prepared in advance and left at 80°C overnight for curing. At the end, this PDMS sandwich was cut and punched with appropriate size biopsy punchers (5 mm for inner hydrogel area, 18 mm outside diameter). Resulting hydrogel holders were sterilized with UV and kept sterile until further use. The stamps featuring micropillars topography were fabricated at the CMi, EPFL. In brief, the 3D models of the micropillars were designed in Autodesk Inventor. The STL model was further processed in DeScribe 1.7 (Photonic Professional) to optimize it for printing 2PP lithography (NanoScribe GT2, Photonic Professional). The printing parameters were defined to have a slicing distance of 1 μm and a hatching distance of 600 nm. Writing was performed in the galvo scan mode and the 3D model was divided in 400 × 400 μm subblocks, corresponding to the FOV of 25× objective. The model was printed in acrylic photopolymer resin IP-S (Photonic Professional), which was deposited on the surface of indium-tin-oxide (ITO)-coated soda lime slides. Once the printing process was finalized, the slides were developed for 7 min in propylene glycol monomethyl ether acetate (PGMEA, Sigma-Aldrich), followed by 5 min rinsing with ultrapure isopropanol and gentle blow-drying. Then, samples were UV cured for 10 min (MA6/BA6, Süss MicroTec) and baked at 80 °C for 3 h. To fabricate a master mold featuring inverted topography, 3D printed models were then plasma-activated and silanized with vapored trichloro (1H,1H,2H,2H-perfluorooctyl) silane (Sigma-Aldrich) overnight. Then, polydimethylsiloxane (PDMS, Sylgard 184, Dow Corning) was used for molding. Ten weight-parts of elastomer base were vigorously mixed with one part of curing agent and poured onto the mold. After degassing under vacuum, PDMS was baked for 12 h at 80 °C. The resulting PDMS replica was cut, plasma-activated, and silanized with vapored trichloro (1H,1H,2H,2H-perfluorooctyl) silane (Sigma-Aldrich) overnight. This replica was used multiple times as a master for molding stamps, following the same protocol (PDMS Sylgard 184, 1:10 ratio, baked at 80°C for 12 h).

Microfabrication of engineered scaffolds with controlled geometry

Elastomeric stamps containing the desired geometries were coated with BSA in PBS (1% w/v in PBS; Thermo Fisher Scientific) overnight to prevent hydrogel adhesion. Before use, stamps were washed once with distilled water and dried with the gentle air blowing. For hydrogel casting, PDMS ring holders were placed on the bottom of the 3.5 cm dish and UV sterilized prior use. To make hydrogel precursor, type I Collagen Solution, (Bovine, 6 mg/ml, Advanced Biomatrix) was neutralized according to manufacturer protocol and then mixed with Matrigel (8-10 mg/ml, Corning) in 75/25 v/v ratio. Then, a drop of liquid hydrogel precursor was injected inside of the ring spacer and a stamp with microtopography was placed atop. After 15–30 min polymerization in the incubator (37 °C), stamps were removed and hydrogels were covered with PBS. Hydrogels were UV sterilized and could be used either the same day or stored up to 3 months.

Isolation of epithelial cells from the fetal small intestine by flow cytometry

Proximal halves of the small intestine from E16.5 fetuses were dissected and incubated with collagenase (Sigma) (125 μg/ml) in 0.1% BSA in PBS for 45 min at 37 °C and subjected to vigorous pipetting every 15 min using a P1000 pipette. Released cells were pelleted, resuspended in PBS containing 1% BSA, and incubated with fluorescent-conjugated primary antibodies (listed in key resources table) for 15 min at room temperature. After washing, DAPI (1 μM) was added to exclude dead cells. Purified cell populations were subsequently isolated using a FACSAria III or FACSymphony S6 (BD Bioscience), 70u nozzle, 1.0 ND filter. FACS data was analyzed using FlowJo (BD Bioscience) software.

Quantification of Lgr5⁺ cells in vivo by flow cytometry

10 cm fragments of proximal small intestine were dissected and incubated with EDTA buffer (5 mM EDTA, 5% fetal calf serum, 10 mM HEPES and 1 mM DTT) for 30 min at 37 °C to isolate whole epithelium. Single-cell suspensions were then prepared by TrypLE (Gibco) digestion supplemented with Y-27632 (Sigma; 10 μM), N-acetyl-L-cysteine (Sigma; 1mM) and DNase I (Roche; 100 μg/ml). Released cells were pelleted, resuspended in PBS containing 1% BSA, and incubated with fluorescent-conjugated primary antibodies (listed in key resources table) for 15 min at room temperature. After washing, DAPI (1 μM) was added to exclude dead cells. The fraction of Lgr5-eGFP⁺ cells was quantified using FACSymphony S6 (BD Bioscience) flow cytometer, and data were analyzed with FlowJo software (BD Bioscience).

Cell culture

Purified EpCAM⁺ cells were pelleted and resuspended in seeding medium consisting of surrogate Wnt⁶³ (10% of volume), human EGF (Peprotech; 50 ng/ml), murine noggin (Peprotech; 100 ng/ml), mouse R-spondin 1 conditioned medium (10% of volume) and Y-27632 (Sigma; 10 μM) (WENR) in basal medium (Advanced DMEM/F12 (Gibco) supplemented with Pen/Strep (Gibco), HEPES (Gibco), Glutamax (Gibco), N2 (Gibco), B27 (Gibco)) at a concentration of 14.000 cells/μl. Subsequently, cells were plated either on engineered scaffolds 70.000 cells/scaffold or encapsulated in growth factor reduced Matrigel (Corning) at 3.000 cells/25 μl Matrigel. Following 2 days, engineered scaffolds were released from the bottom of the culture vesicle and flipped to facilitate the removal of cellular debris from the flat surface and allow growth factor penetrance. At the same time surrogate Wnt and Y-27632 were removed from the medium. Medium was subsequently changed every 2 days. For lineage tracing experiments, 4-hydroxytamoxifen (Sigma; 1 uM) was added to the medium. For Lgr5+ cell ablation experiments, 40 ng/ml Diphtheria Toxin (Millipore) was added to the medium. For transgene induction in the rtTA;tetO-YAP^S127A model, DOX (Sigma; 1μg/ml) was added to the medium.

Flow cytometry

Epithelial cell isolation from E16.5 or P2 mice has been previously described.²⁴ For releasing cells from engineered tissues, engineered scaffolds were dissolved using incubation in Collagenase XI (Sigma; 5 mg/ml) solution containing Y-27632 (Sigma; 10 μM). Following cell release, cells were washed, pelleted in non-stick RNase-free microfuge tubes (ThermoFisher Scientific). In parallel, organoids in 3D cultures were released from Matrigel by washing with 0.1% BSA in PBS and pelleted in non-stick RNase-free microfuge tubes (ThermoFisher Scientific). All samples were then incubated in TrypLE (Gibco) containing Y-27632 (Sigma; 10 μM) and DNase I (Roche; 50 μg/ml) until single cell suspension was achieved. DAPI was used as viability staining. Data were collected with FACSAria III or FACSymphony S6 (BD Bioscience. FACS data was analyzed using FlowJo (BD Bioscience).

Two-dimensional immunostaining of engineered tissues

Engineered tissues were fixed with 4% paraformaldehyde (PFA) overnight at 4 °C, embedded in OCT and kept at -80 °C until sectioning. Sections (7 μm thick) were prepared with a cryotome (Cryostat CM 3050S). To perform immunofluorescence, sections were blocked and permeabilized in 10% adult bovine serum (Sigma), 5% skim milk (Sigma), 0.3% Triton X-100 in PBS (blocking buffer) for at least 1 h at 4 °C. Primary antibodies (listed in key resources table) were incubated overnight in blocking buffer at 4 °C. Alexa-Fluor-Conjugated secondary antibodies (listed in key resources table) were incubated for 1–2 h at room temperature in 10% adult bovine serum and 0.1% bovine serum albumin (BSA) (Sigma) in PBS. Diamidino-2-phenylindole dihydrochloride (DAPI) (Sigma; 1 μM) was used to counterstain nuclei in the indicated experiments. Images were acquired using laser-scanning confocal microscopes (Leica TSC SP8, Leica Stellaris 5 and Zeiss LSM800). All images were subsequently analyzed with Fiji or Imaris software.

Three-dimensional immunostaining of engineered tissues

Engineered tissues were fixed with 4% paraformaldehyde (PFA) overnight at 4 °C. Samples were blocked and permeabilized in 1% BSA, 0.5% Triton X-100 (Triton X-100 Surfact-Amps Detergent Solution, Thermo Fisher) in PBS overnight at 4 °C. Primary antibodies (listed in key resources table) were incubated in 1% BSA, 0.5% Triton X-100 in PBS for 24 h at 4 °C. Tissue was subsequently washed in 0.5% Triton X-100 for the duration of the day changing the washing solution every 2 hours. Secondary antibodies (listed in key resources table) were incubated in 1% BSA, 0.5% Triton X-100 in PBS with DAPI (Sigma, 1μg/ml) overnight at 4 °C. Tissue was subsequently washed in 0.5% Triton X-100 for the duration of the day changing the washing solution every 2 hours. Following last wash, tissues were mounted on glass slides in RapiClear 1.52 (Sunjin Lab) and imaged the next day using a laser-scanning confocal microscope (Leica Stellaris 5). All images were subsequently analyzed with Fiji or Imaris software.

RNA in situ hybridisation (RNA scope)

For in situ hybridization tissue was processed according to manufacturer’s protocols. Briefly, following fixation, the intestinal pieces were incubated in 10%, 20% and 30% sucrose (1h each) and embedded in OCT for cryo-sectioning where 5 μm sections were cut on the Leica Cryostat. co-ISH was performed using RNAscope Multiplex Fluorescent Reagent Kit v2 (323100, ACD Biotechne) by the instructions of the manufacturer with probes Mm-Atoh1 (408791-C3), and Mm-Lyz1 (415131). Diamidino-2-phenylindole dihydrochloride (DAPI) (Sigma; 1 μM) was used to counterstain nuclei in the indicated experiments. Slides were mounted using ProLong Gold Antifade mounting medium (ProLong Antifade reagent; Invitrogen). Images were acquired using laser-scanning confocal microscopy (Leica Stellaris 5) and processed using Fiji/ImageJ (NIH).

SEM, sample preparation and imaging

Samples were fixed in 1% PFA and 2.5% glutaraldehyde in 0.1 M phosphate buffer (PB) for 2 h at room temperature, followed by 1 h incubation in 2% (w/v) osmium tetroxide and 1.5% (w/v) K4[Fe(CN)6] in 100 mM PB. Samples were incubated for 1 h in 1% (w/v) tannic acid in 100 mM PB, then 30 min in 2% (w/v) aqueous solution of osmium tetroxide followed by 1% (w/v) uranyl acetate for 2 h at room temperature. At the end of gradual dehydration cycles, samples were flat-embedded in Epon-Araldite mix. Polymerized flat blocks were trimmed using a 90° diamond trim tool, and the arrays of 100-nm sections were obtained using a 35° ATC diamond knife (Diatome) mounted on a Leica UC6 microtome. Sections were transferred to wafers using a modified array tomography procedure. For the ultrastructure analysis, wafers were mounted on aluminium stubs and analyzed using a FEI Helios Nanolab 650 scanning electron microscope (Thermo Fisher Scientific). The imaging settings were as follows: MD or ICD detectors, accelerating voltage: 2 kV, current 0.8 nA, dwell time 6 μs. Images were collected manually or using the AT module of the MAPS program (Thermo Fisher Scientific).

Reverse transcription quantitative PCR

Total RNA was extracted using the RNeasy Micro kit (Qiagen) and cDNA synthesized using random hexamers and Superscript III reverse transcriptase (Invitrogen). Quantitative PCR was performed with PowerUp SYBR Green Master Mix (Thermo Fisher Scientific) in QuantStudio 7 Flex Real-Time PCR System (Applied Biosciences). Relative expression values were calculated using the ΔΔCt method. Geometrical means of Ct values for 36B4 or Gapdh and Tubb5 were used for normalization. Primer sequences are provided in Table 1.

Table 1.

Details of primers used for qPCR in the study, related to STAR Methods

Gene	Forward	Reverse
Fabp2	TGCTGTCCGAGAGGTTTCTG	AGAATCGCTTGGCCTCAACT
Muc2	ACAAGCTGGCAGTGGTGAAC	AGACCTTGGTGTAGGCATCG
ChgA	GAACAGCCCCATGACAAAAG	TCGGAGATGACTTCCAGGAC
Lyz1	GCCAAGGTCTACAATCGTTGTGAGTTG	CAGTCAGCCAGCTTGACACCACG
Defa6	AATCCTCCTCTCTGCCCTGG	ACCAGATCTCTCAATGATTCC
hYAP	CTCGAACCCCAGATGACTTC	TTCCACATTTGTCCCAGGAAT
Gapdh	TGTTCCTACCCCCAATGTGT	TGTGAGGGAGATGCTCAGTG
Tubb5	GATCGGTGCTAAGTTCTGGGA	AGGGACATACTTGCCACCTGT

In vitro tissue rendering and nuclear segmentation

Three-dimensional segmentation of DAPI-labelled nuclei was implemented using the Python package Cellpose.⁶⁴. A custom model was trained using manually curated data from sagittal and transverse sections of the 3D stack. 3D surface rendering was performed using the surface layer of the Python package Napari.

In vivo tissue rendering and nuclear segmentation

Following antibody staining for E-cadherin, CD44v6 and DAPI, the images were imported into Imaris v10.0.1 (Bitplane, Zürich, Switzerland). First, crypt progenitor units (CPUs) were segmented by using CD44v6 channel and the ‘Surface’ function of Imaris. The option of ‘Region Growing’ was used to assist splitting of touching crypts. Based on CD44v6 segmentation, the DAPI channel was masked to eliminate any out-of-CPU DAPI signal. We selected five CPUs per biological replicate arbitrarily with the only criterion being that DAPI signal was not diffuse between nuclei. We used the function ‘Spots’ of Imaris to identify the central parts of nuclei automatically, and, consequently, their centroid. These spots-like structures were approximated using a 3D model, with z-dimension set at 1.5 times of the image acquisition z-step. The option of ‘Region Growing’ was used to assist the identification of nuclei with different size. Manual correction of centroids was performed where it was considered necessary (e.g. deletion of duplicate centroids). The measurement of distance between each spot-like structure’s centroid and its three nearest neighboring ones was extracted automatically from Imaris statistics.

Calculation of cell volume

Bottom, middle ant top areas of the crypt were selected to be evenly distributed across the crypt, with each of them including 12 z-slices (12 μm in thickness), covering roughly the depth of one cell. In addition, two flat areas of the engineered tissues were selected, with each of them including 5 z-slices (5 μm in thickness). The volume of cells was calculated using the MATLAB® (The MathWorks, Inc.) toolbox CellSegm.⁶⁵ The thresholding method was used while we adjusted the parameters of intensity inside cells (1.1), intensity of border (1.1), convexity of cell area (0.5) and convexity of border (0.4), based on our cell samples. Image stacks of cell outlines were generated in MATLAB and they were imported in Imaris® (v10.1, Bitplane, Zürich, Switzerland) for visualization, 3D rendering and extraction of representative images of the cell segmentation results.

Isolation of single cells for single cell RNA sequencing

Patterned areas were extracted from engineered tissues using a 3 mm punch biopsy (Miltex). Engineered scaffolds were dissolved using incubation in Collagenase XI (Sigma; 5 mg/ml) solution containing Y-27632 (Sigma; 10 μM). Following cell release, cells were washed and pelleted in non-stick RNase-free microfuge tubes (ThermoFisher Scientific). In parallel, organoids in 3D cultures were released from Matrigel by washing with 0.1% BSA in PBS and pelleted in non-stick RNase-free microfuge tubes (ThermoFisher Scientific). All samples were then incubated in TrypLE (Gibco) containing Y-27632 (Sigma; 10 μM) and DNase I (Roche; 50 μg/ml) until single cell suspension was achieved. Following washing and pelleting, cells were stained for 20 minutes on ice in 1% BSA in PBS, containing DAPI (Sigma; 1μg/ml) and TotalSeq hashtag antibodies (A0301-A0304, Biolegend; 1:100) for sample multiplexing. In total, four samples were multiplexed for library preparation: two biological replicates from 3D cultures and two biological replicates from engineered tissues. After washing with 0.1% BSA in PBS and pelleting, cells were re-suspended in ultra clean 0.8% BSA in PBS and proceeded to cell sorting by flow cytometry. A total of 10,000 to 15,000 events from each sample were sorted on a BD FACSAria III sorter (BD Biosciences) and pooled directly into one tube containing 3 μl ultra clean 0.8% BSA in PBS.

Preparation of libraries and single cell RNA sequencing

Single-cell libraries were prepared using the Chromium Next GEM Single Cell 3’ Reagent Kits v3.1 according to the 10X Genomics protocol v3.1 with the implementation of hashtag libraries for demultiplexing. An additive primer (hashtag-oligo (HTO) primer) was added in the step for cDNA amplification. The hashtag cDNA and endogenous cDNA were separated after cDNA amplification using SPRIselect beads (Beckman Coulter) for size separation. Here, hashtag cDNA was isolated in the supernatant containing smaller DNA fragments and collected to generate the hashtag library following the protocol from Stoeckius et al.⁶⁶ and CITE-seq (https://citeseq.wordpress.com). Briefly, the supernatant was cleaned using SPRIselect beads (Beckman Coulter) and 30 ng of cDNA was used as input for PCR amplification (10 cycles). Illumina indexes for sequencing were subsequently added using Kapa HiFi HotStart Ready Mix (Roche Diagnostics). For the endogenous cDNA library, the 10x Genomics protocol was followed. The dual index kit set TT (10x Genomics) was used for sample indexing and 12 cycles were used during the final amplification step. The hashtagged cDNA and the endogenous cDNA libraries were diluted to 2 nM and combined into two pooled libraries for sequencing (5% hashtag + 95% endogenous cDNA). The final library for sequencing was further diluted to 1.7 pM before being sequenced with the Illumina NovaSeq 6000 platform with a S1 Reagent kit v1.5 (100 cycles).

Computational analysis

Processing of raw scRNA-seq reads

FASTQ files were generated from raw base call files using 10x Genomics Cell Ranger’s mkfastq function (10x Genomics Cell Ranger 6.1.0). For quantification of reads the salmon alevin w/feature barcodes pipeline was used (https://combine-lab.github.io/alevin-tutorial/2020/alevin-features/). Specifically, salmon (v 1.4.0) was used to pseudo-quantify both HTO and RNA (with the mm10 transcriptome) reads. The downstream analysis was performed using Seurat v.4.3.0.⁶⁷ If no specific details or parameters are specified, the functions and analysis were run with default settings.

Demultiplexing of hashtag oligos

First, cells overlapping between the HTO matrix and the RNA matrix were selected for downstream analysis. The HTO matrix was further normalized with the function NormalizeData (normalization method CLR and margin=2) and scaled using ScaleData. To demultiplex samples and assign cells with hashtag oligos, the function HTODemux was used (with positive.quantile=0.96) on the HTO matrix. Based on the output from the demultiplexing, cells assigned with no hashtags (negative) or multiples were filtered out.

Quality control and filtering of cells

Cells were then assessed for their quality based on the RNA matrix. Cells with unique feature counts less than 200 or more than 6000 were filtered out as well as cells with more than 40 % mitochondrial counts. The RNA matrix was further normalized by multiplying it by a scale factor (10000) and log-transformed (with NormalizeData). Additional doublet prediction analysis was performed using DoubletFinder⁶⁸ and the predicted doublets were additionally filtered out.

Data normalization and scaling

We used cell cycle genes provided by Seurat to calculate the cell cycle score with the function CellCycleScoring. The difference between the S score and the G2M score was then calculated and regressed out when the data was scaled using the function ScaleData. In addition, to regress unwanted sources of variation, nCount_RNA, percentage of mitochondrial and ribosomal counts was added to the vars.to.regress option. No additional batch correction was run as the replicates within the two conditions mixed well (Extended Data Figure 3). Additionally, the proliferation score was calculated using AddModuleScore on a list of previously reported proliferation genes.⁶⁹

Dimensionality reduction, clustering and cell type annotations

For dimensionality reduction we ran RunPCA and RunUMAP (with dims = 1:30) followed by FindNeighbors (dims = 1:30) and FindClusters (with resolution=0.3) for clustering. This generated 7 clusters that we reduced to 6 by combining the two clusters expressing markers for stem cells and proliferation. The clusters were further annotated based on their differently expressed genes, calculated by FindAllMarkers (with min.pct=0.25 and logfc.threshold=0.25) as well as the expression of known marker genes.^4,29

Integration with adult intestinal epithelium scRNA-seq data

We downloaded published adult intestinal epithelium scRNA-seq data from the Single Cell Portal (id: SCP44) including 7216 cells from 6 mice.²⁹ The dataset was first batch corrected using the functions SelectIntegrationFeatures, FindIntegrationAnchors and IntegrateData.

Next, we used our data as query and the published data as reference and transferred the cell type labels from the reference to the query.⁶¹ Using the functions FindTransferAnchors, TransferData and GetTransferPredictions we calculated the prediction score of each cell type (in reference) for each cell (in query). In Figure 3F we show the maximum prediction score of each cell for the two conditions. Statistical significance was assessed using the Wilcoxon rank sum test. In Extended Data Figure 3J, a dataset from adult small intestinal organoids³² was used as query, and the prediction score was calculated using the same functions as described above.

Separate analysis of engineered tissue cells

After demultiplexing, the engineered tissue cells were selected and analyzed separately (Extended Data Figures 3F–H). Quality control, normalization, scaling and dimensionality reduction were performed using the same functions and parameters as for the full dataset. Signature scores for the bottom and top landmark genes from Moor et al.³⁰ were calculated using AddModuleScore.

Statistics

The number of biological and technical replicates and the number of animals is indicated in figure legends and text. All tested animals were included. Sample size was not predetermined. For all experiments with error bars standard deviation (s.d.) was calculated to indicate the variation within each experiment or sample. Statistical tests were used to assess significance as indicated in individual figure legends.

Published: February 12, 2026