Dynamics of supracellular keratin bundling and nuclear uncaging in stretched epithelia

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

There is broad consensus that intermediate filaments, such as keratin, play a key role in protecting cells and tissues from large deformations. However, little is known about how they fulfil this function. Here we show that epithelial cells slowly adapt to stretching through a coupling of a star-bundling transition of keratin filaments with the escape of the nucleus from its keratin cage. The bundling transition begins with a depletion of keratin filaments at tricellular junctions followed by a progressive accumulation in thick bundles that bisect cell, cell junctions. Bundling is a cooperative process that initiates in a few scattered cells and propagates to their neighbours, leading to the growth of multicellular clusters that contain a percolated network of thick keratin bundles. Bundling dynamics are slow and strongly influenced by the interaction between actin and keratin. Informed by a computational model, we provide evidence that keratin bundling generates a compressive stress on the nucleus, which is relaxed by nuclear escape from the keratin cage. The topological transitions identified here provide epithelia with a multiscale mechanism to adapt to sustained stretching.

Intermediate filaments, such as keratin and vimentin, constitute an essential class of cytoskeletal polymers widely expressed in animal cells. Owing to their hierarchical structure, individual filaments and their networks exhibit exceptional mechanical properties: they can stretch several times their original length but undergo pronounced stiffening at higher strains 1 , 2 , 3 , 4 , 5 . At the single cell level, intermediate filaments support cell integrity by resisting both tensile and compressive stresses and by shielding the nucleus from its mechanical microenvironment 6 , 7 , 8 , 9 , 10 , 11 , 12 . In epithelial tissues, keratin intermediate filament networks are responsible for strain-stiffening and are crucial for maintaining tissue cohesion during large deformations 13 , 14 , 15 , 16 , 17 . Collectively, these findings have led to the growing consensus that, although actin filaments largely determine cell and tissue mechanics at small deformations, intermediate filaments play a key role in governing cellular responses to larger strains 17 , 18 , 19 , 20 .

Although extensive functional and mechanical evidence now supports this consensus, the reorganization of intermediate filament networks in response to mechanical stretching and their specific role in providing mechanical protection remain poorly understood. In relaxed epithelial monolayers, keratin filaments form two interconnected networks, one lining the cell cortex and the other enveloping the nucleus. The cortical keratin network, commonly referred to as the rim, connects adjacent cells through desmosomes and is linked to the nuclear-enveloping network through short filament bundles known as spokes. This rim-and-spoke structure relies heavily on the crosslinking protein plectin, which bridges the actin and keratin cytoskeletons and couples keratin to the nucleus via nesprins 21 , 22 . Upon mechanical stretching, the organization of these networks is transformed 23 , 24 , 25 , 26 , 27 , 28 . In MDCK monolayers, cells exposed to large strains display long, thick keratin cables that withstand tension and form supracellular networks 19 , 29 . Long keratin cables spanning several cells have also been found in developing embryos, which indicates that they have a protective and stabilizing role 30 . How such keratin structures form under stretching, how they organize into supracellular networks and how they modulate the mechanical properties of the tissue are not known.

Here we combine experimental and theoretical approaches to study the dynamic reorganization of the keratin cytoskeleton and its coupling to the nucleus during sustained stretching of epithelial monolayers. Using a new device to stretch curved cell monolayers, we found that the keratin network adapts to prolonged mechanical loading by undergoing a slow bundling transition. The kinetics of this transition is largely governed by the actin, keratin interaction. Strikingly, we show that the bundling transition leads to the full escape of the nucleus from its keratin cage. A computational model explains this behaviour in terms of a compression-by-tension mechanism in the entangled keratin network. Finally, we demonstrate that the drastic structural remodelling of the keratin cytoskeleton is a collective supracellular process driven by a nucleation-and-growth mechanism. Our study unveils a new multiscale response of cellular monolayers to sustained stretching, wherein the nucleus is ultimately unshielded from the keratin cytoskeleton.

We first studied epithelial domes that form spontaneously in MDCK monolayers due to the osmotic pressure generated by directional ion pumping 31 , 32 . These domes undergo autonomous cycles of inflation and deflation lasting several hours and reaching areal strains of 200, 300%. Using micropatterning techniques, we generated arrays of spontaneous domes with a defined footprint radius 19 and fixed them to visualize the spatial distribution of F-actin and keratin-18 filaments (Fig. 1a and Extended Data Fig. 1a ). Although the distribution of F-actin was largely cortical in all cells, keratin displayed two markedly distinct organizations depending on the extent of stretching. Cells within the unstretched monolayer showed a prominent keratin mesh enveloping the nucleus that was connected to the cell surface through thin and short bundles (Fig. 1b and Extended Data Fig. 2 ). These unstretched cells also displayed a thin keratin network lining the cell cortex, consistent with the rim-and-spoke arrangement commonly reported in epithelial cells 33 , 34 . By contrast, stretched cells within the dome had a star-like arrangement composed of a dense central node and long, thick bundles radiating towards the cell periphery (Fig. 1b ). Each bundle was mirrored by a corresponding bundle in the neighbouring cell, leading to the emergence of a supracellular percolated network of interconnected bundles. This network defines supracellular clusters bounded by transitional cells that exhibited partial bundling restricted to specific subcellular regions.

a , Spontaneous dome of MDCK cells expressing keratin-18, GFP (green) stained for F-actin (magenta) and nuclei (Hoechst, blue). Top: lateral view. Bottom: maximum intensity projection with dome footprint marked with a white dotted circle. b , Zoomed-in views of the three regions marked with white rectangles in a , showing non-bundled, partly bundled and bundled cells. Top: lateral views. Bottom: maximum intensity projections. c , Schematic illustration of the intensity radius R i and average angle θ . d , Keratin-18, GFP images of the three cells shown in b . e , Keratin-18, GFP intensity in the spontaneous dome shown in a . The dome footprint is marked with a black dotted circle. The segmented cell boundaries are colour-coded according to the average angle θ . f , Intensity radius plotted against the average angle ( R = −0.34, P = 2.9 × 10 −15 ). g , Average angle θ plotted against cell area ( R = 0.71, P = 6.3 × 10 −79 ). n = 509 cells in f and g . R values are Pearson’s correlation coefficient.

To assess the bundling state of each cell from keratin images, we defined two metrics (Fig. 1c ). One is the intensity radius R i , which reflects the radial distribution of keratin within the cell. Values close to 1 indicate a predominantly peripheral localization, whereas values near 0 indicate a central localization. The second metric is the average angle θ , which quantifies the orientation relative to the cell tangent for keratin filaments in the outer half of the cell. This metric approaches 90° for filaments organized radially and 0° for those arranged circumferentially (Fig. 1c and Extended Data Fig. 3 ). Non-bundled cells displayed high values of R i and low values of θ , which highlights that filaments accumulated close to the cell junctions and arranged parallel to it. Conversely, cells forming star-like bundles showed low values of R i and high values of θ , reflecting the central accumulation of keratin and the radial orientation of the bundles. Partly bundled cells displayed intermediate values of θ but values of R i like those for non-bundled cells, which reflects the absence of a central knot (Fig. 1c, f and Extended Data Fig. 4a,b ). All unstretched cells outside the dome were in the non-bundled configuration, whereas in the dome, 29 ± 10% (mean ± s.d.) were partly bundled and 12 ± 7% were fully bundled (Extended Data Fig. 4c ). The analysis of these metrics across all cells in the domes established a clear dependence between cell area and its bundling state. Indeed, by plotting θ versus cell area, we found that small cells were generally non-bundled, large cells were bundled and intermediate-sized cells were partly bundled (Fig. 1g ). This result prompted us to study the mechanisms underlying bundle formation and their dynamics.

In previous theoretical work, we hypothesized that bundling could be explained as a topological phenomenon in an entangled network of filaments connected to laterally mobile desmosomes, such that the straightening of keratin filaments upon stretching is ultimately hindered by a self-organized knot 35 . To test this hypothesis, we developed a microfluidic device that formed domes with a defined geometry and with dynamic control of inflation (Fig. 2a and Extended Data Fig. 1b ). This device consists of two perpendicular channels separated by a thin polydimethylsiloxane (PDMS) layer containing evenly spaced holes with a diameter of 80 µm that define the dome footprint. This PDMS layer is covered with a permeable polycarbonate membrane featuring 400-nm-sized pores that allow the passage of liquid but not of cells. Using a photopatterning technique (PRIMO 36 ), the polycarbonate membrane was coated with a high concentration of fibronectin outside the footprint and a low concentration of fibronectin inside it. This enabled the seeding of cells and the subsequent formation of a confluent monolayer on the fibronectin-coated side. Hydrostatic pressure was then applied by injecting fluid beneath the membrane, resulting in delamination of the monolayer only at the low adhesion footprint. With this approach, we formed directed epithelial domes of defined geometry and controlled their luminal pressure over time (Fig. 2a and Extended Data Fig. 1b ).

a , Schematic of the microfluidic device showing the porous membrane (black) coated with high (dark violet) and low (light violet, footprint) fibronectin levels. Hydrostatic pressure applied from the basal side (blue and black arrows) causes the cell monolayer to delaminate at the footprint, resulting in the formation of a dome. b , Maximum intensity projections and side views of a directed dome of keratin-18, GFP cells. The dome footprint is marked with a dotted white circle, and segmented cell borders are marked in green. c , Time evolution of a tissue strain for four independent domes. Squares mark each bundling event, defined as an individual cell crossing the bundling threshold. d , Time sequence of cropped images of a directed dome. Arrows indicate the formation of a keratin-depletion zone at a tricellular junction. e , Zoomed-in views of the tricellular junction marked in d . f , Time evolution of the area of individual depletion zones ( n = 13 cells). g , Mean area of the depletion zone as a function of time rescaled by the peak time T Peak ( n = 13 cells). h , A representative individual cell undergoing the bundling tra