Forces Generated by Cell Intercalation Tow Epidermal Sheets in Mammalian Tissue Morphogenesis

Summary

While gastrulation movements offer mechanistic paradigms for how collective cellular movements shape developing embryos, far less is known about coordinated cellular movements that occur later in development. Studying eyelid closure, we explore a case where an epithelium locally reshapes, expands, and moves over another epithelium. Live imaging, gene targeting and cell cycle inhibitors reveal that closure does not require overlying periderm, proliferation or supracellular actin cable assembly. Laser ablation and quantitative analyses of tissue deformations further distinguish the mechanism from wound-repair and dorsal closure. Rather, cell intercalations parallel to the tissue front locally compress it perpendicularly, pulling the surrounding epidermis along the closure axis. Functional analyses in vivo show that the mechanism requires localized myosin-IIA and α5β1-fibronectin-mediated migration, and E-cadherin downregulation likely stimulated by Wnt signaling. These studies uncover a mode of epithelial closure in which forces generated by cell intercalation are leveraged to tow the surrounding tissue.

Introduction

Individual cells, each with the capacity to migrate autonomously, work collectively within sheets or clusters to carry out complex morphogenetic tasks of development and maintain integrity of adult tissues. Insights into these mechanisms have come from extensive studies of cell movements that occur during gastrulation. This has uncovered a conserved set of cellular and molecular mechanisms that drive the elongation, spreading, and fusion of developing tissues in various contexts and model organisms. Variations on these general mechanisms operate in organogenesis, tissue regeneration, and cancer metastasis.

Among the best-studied large-scale morphogenetic movements in development are convergent extension movements, which drive axis elongation in the dorsal mesoderm of Xenopus and zebrafish, the notochord of mice, and the germband of Drosophila. Convergent extension involves ordered rearrangement of cells via intercalation, which results in a tissue narrowing along the mediolateral axis and lengthening along the anterior-posterior axis (Keller et al., 2000). This intercalation is manifested either by cells adopting a bipolar morphology and polarizing their protrusions along the mediolateral axis, as occurs in mesodermal tissues (Keller et al., 2000), or by myosin II-dependent remodeling of intercellular adherens junctions, as occurs in some epithelial tissues (Bertet et. al., 2004; Blankenship et al., 2006).

Epithelial sheet movement during wound-repair or fusion differs from convergent extension in that it utilizes a contractile actomyosin ring at the leading edge (Kiehart, 1999). Additionally, the epithelial sheet migrates outward without substantive remodeling of cells within the tissue. In Drosophila, dorsal closure involves a series of distinct cellular movements, many of which are similar to those that occur during wound-repair. Notably, leading edge cells polarize in the direction of sheet migration and assemble a supracellular actin cable around the gap. This is accompanied by pulsed contractions in underlying amnioserosa (Solon et al., 2009; Gorfinkiel et. al., 2009). The supracellular actomyosin cable is believed to both generate a centripetal force and stabilize tension generated by apically constricting amnioserosa cells to close the gap. Epithelial cells around the opening also extend actin-rich filopodial protrusions into the gap, which are believed to actively promote contact and intercellular adhesion between cells from apposing sheets.

Little is known about how epithelial cells choose between convergent extension and dorsal closure/wound-healing movements or the extent to which these mechanisms are active in the morphogenesis of differentiating tissues in late development and adulthood. Here, we use mouse embryonic eyelid closure as a model system to understand how collective cell movements drive tissue morphogenesis late in mammalian development.

Common to all mammals, eyelids form and close during embryogenesis. An eye open at birth (EOB) defect causes severe corneal inflammation and partial blindness. Understanding the process is not only clinically important, but offers an intriguing series of morphogenetic movements that involve extension and fusion of two regions of skin epidermis over the cornea (Figure 1A–B). Like digit fusion, eyelid closure is an example of a temporary epithelial fusion. In mice, eyelid development begins at embryonic day E11.5, and is accompanied by an accumulation of rounded periderm cells at the leading edge. The eyelids extend over the cornea until they meet between E15 and E16.

Figure 1.

Eyelid closure involves epidermal cell movement rather than cell proliferation. (A) Schematic of sagittal section of the eye, illustrating juxtaposition of cell types in the region. (B) Overview of eyelid closure in planar (upper, fluorescence images) and sagittal view (lower, schematic). For reference, the horizontal and vertical midlines are indicated, as well as the corneal epithelium (Cor), eyelid front, and surrounding epidermis (Epi). Eyelid closure takes place between E15 and E16. (C) Front cells do not proliferate during eyelid closure. Left: immunolabeling for Ki67, which marks cycling cells. Note that cornea and surrounding epidermis are proliferative, while K5+ eyelid front cells are not. Right: tiled, 3D imaging of an eyelid explant pulsed for 2 h with EdU reveals no asymmetry of cell proliferation around the eye. The %EdU+ cells was calculated in 36° regions with ~600 segmented cells per region. (D) Dual BrdU-analog incorporation scheme. If a subpopulation of proliferative cells were driving asymmetric tissue growth, incorporation of the two analogs in distinct regions would be expected. Instead, they are incorporated randomly throughout the tissue. (E) Live imaging of eyelid closure. Progression of closure was quantified by measuring the average displacement of the eyelid border over the eye. Cell proliferation was quantified by manually counting the number of dividing cells in regions of equal area (comprising ~4000 cells) over the course of 20 frames, and expressed as the fraction of cells in the region/hr. Right: H2B-GFP transgenic embyros were exposed in utero to mitomycin C for 12hr prior to live imaging and quantitative analysis. Reductions of cell divisions by >75% (p = 0.0011) did not affect eyelid closure rates (p = 0.90, n=3 Mito-C treated, 2 WT embryos). Scale bars, 50 μm except where indicated. See also Figure S1.

An EOB phenotype is easy to spot, and consequently has been linked to a number of genes that encode regulators of growth factor signaling, epithelial-mesenchymal crosstalk and contractility. Most studies that specifically deal with eyelid closure have focused on the upstream signaling pathways involved rather than the downstream cellular and molecular mechanisms that physically drive the process. Models have been largely limited to cell migration and actomyosin contraction analogous to wound-healing or dorsal closure, adding possible roles for periderm and cell proliferation.

We combine live imaging and quantitative analyses of tissue deformations with genetic and laser ablation of specific cell populations to probe the dynamics of this process in mice. In contrast to classical models of epithelial sheet movements, a contractile actomyosin cable is not utilized. Rather, the process involves a multicellular layer of epidermal cells at the eyelid front that undergoes localized cell intercalations perpendicular to the axis of closure, reminiscent of convergent extension movements in gastrulation. Mathematical analysis shows that such movements create a region of active shear specifically at the eyelid front, supporting a model in which forces generated by cell intercalation tow the surrounding epidermal sheets over the eye. We test and confirm this model by laser-ablation and functional genetics, revealing that unlike other examples of epithelial intercalations, these movements depend on integrin-fibronectin adhesion and myosin II-dependent cell motility rather than cell-cell junction remodeling. Our study sheds important light on how a complex tissue, in this case a vertically stratifying epithelium, can tailor several types of well-known coordinated behaviors to accomplish a complicated morphogenetic task in late development.

RESULTS

The periderm is not required for eyelid closure

Temporary epithelial fusions that occur on the embryo surface are thought to be mediated by periderm, a transient layer of superficial keratin K8⁺K18⁺ cells that forms by E11.5 to protect the embryo until K5⁺K14⁺ epidermal progenitors have stratified and differentiated (Maconnachie E.,1979; Findlater et al., 1993; Figure S1). To test this hypothesis, we first performed lineage tracing to either selectively label periderm or basal epidermal progenitors. By monitoring subsequent development, it was clear that eyelid epidermis is derived from epidermal progenitors and not periderm (Figure S1A–B).

To directly test the periderm’s importance to the process, we infected E11.5 Rosa26^DTA embryos with a Cre lentivirus. The virus selectively transduces only the outermost layer, which at this stage is periderm, thereby abolished upon diphtheria toxin activation (Figure S1C). Notably, however, eyelids closed normally, indicating that the elaborate changes in periderm that accompany closure are not essential to the mechanism.

Eyelid closure is driven by cell motility rather than proliferation

Tissue morphogenesis often requires cell proliferation and migration, as exemplified by vertebrate gastrulation, where axis elongation requires not only cell intercalation, but also oriented cell divisions regulated by the Wnt/planar cell polarity (PCP) pathway (Wei and Mikawa, 2000; Gong et al., 2004). To explore this in eyelid closure, we administered nucleotide analogue 5-Ethynyl-2′-deoxyuridine (EdU) at E14.5 and analyzed at E15.5. Although cornea and surrounding embryonic skin were highly proliferative, epithelial cells at the eyelid front displayed little or no proliferation, as judged by EdU or Ki67 labeling (Figure 1C).

Since front cells did not appear to proliferate, we next addressed whether asymmetric growth of surrounding tissue might push eyelids over cornea. Embryos were pulsed for 2 h with EdU prior to performing tiled 3D imaging. Automated counting of EdU+ cells in 30° regions around the eye revealed uniform proliferation (Figure 1C). To test whether spatially-organized proliferation might contribute to asymmetric tissue growth over time, we injected BrdU at the start of eyelid closure and EdU 12 h later (Figure 1D). If asymmetric tissue growth occurs, the analogs should be incorporated in distinct bands around the eye. Instead, random incorporation was observed, consistent with uniform tissue growth.

Finally, to directly test whether proliferation is required for closure, we treated E15 living embryos with 10 μg/ml mitomycin-C (Mito-C), which quantitatively blocks keratinocyte mitoses. Eyelid explants were taken 12 h later and imaged over a 6 h period. Despite ~75% reduction in cell divisions compared to untreated embryos, eyelid closure progressed normally (Figure 1E). Closure also occurred when Mito-C treated embryos were allowed to progress to E16 in vivo (Figure S1D). Taken together, these results suggest that eyelid closure is accomplished primarily by cell migration rather than proliferation.

Epidermal cells at the eyelid border acquire mesenchymal features and adopt a bipolar morphology

Observing little role for periderm or proliferation in eyelid closure, we focused on the non-proliferative cells at the eyelid front. Immunofluorescence of sagittal sections revealed prominent actin fibers at the eyelid tip, an observation which has led researchers to posit that eyelid closure is driven by assembly and constriction of an actomyosin cable in a purse-string fashion (Shimizu et al., 2005). Supracellular actin cables are polarized structures that span multiple junctions in cells bordering epithelial gaps and play an essential role in would healing and Drosophila dorsal closure. However, by whole-mount imaging, we found that instead of a polarized, supracellular actin cable in a single row of cells at the leading edge, the eyelid front consisted of highly elongated cells that organized their actomyosin cytoskeleton mediolaterally and packed together vertically in multiple layers (Figure 2A). This differed markedly from epidermal cells behind the front, which formed a honeycomb-like lattice with a characteristic cortical actomyosin network.

Figure 2.

Cells in the eyelid front adopt an elongated, bipolar morphology, and pack in a multicellular sheet. (A) Analysis of eyelid morphology by whole-mount phalloidin staining and myosin II-GFP expression. Quantification of fluorescence intensity shows that actomyosin is enriched in eyelid front cells (middle panels). Note that intensity is greatest at elongated bipolar tips rather than along eye border (rightmost panel). In contrast, surrounding epidermal cells remain honeycomb in architecture and show uniform, cortical localization of actin and myosin. (B) Left: elongated, mesenchymal-like cells of the eyelid front are derived from epidermis, as determined by analyzing K14-Cre X Rosa^mT/mG embryos. Eyelid front cells are mGFP+, indicating that they are epidermal, while periderm and dermal cells are mTomato+. Right: sparse-labeling of eyelid front cells permit morphological analysis of individual mGFP cells. Quantification of cell elongation by aspect ratio reveals a gradient from eyelid front to the base (n = 116 cells, 11–26 per bin). (C) Live imaging of individual cytoplasmic YFP+ labeled epidermal cells within the eyelid. Occasional capturing of actively elongating cells near the surrounding epidermis suggests that they undergo an active shape change (right: quantification of relative elongation, n = 3 cells). Scale bars, 50 μm. See also Figure S2.

The epidermal origins of these front cells was confirmed in K14-Cre X Rosa^mT/mG embryos, which activated Cre in epidermal cells prior to eyelid closure, switching them from membrane-bound tdTomato to membrane-bound mGFP expression. Like their parents, front cells expressed mGFP, while dermal and peridermal cells did not (Figure 2B).

Sparse transduction with YFP enabled live imaging of individual front cells by live imaging. During the closure process, epidermal cells reaching the eyelid front elongated along their mediolateral axis, indicating that these shape changes occurred in response to cues from their local microenvironment (Figure 2C, Movie S1). These analyses suggested that this collective reorganization of epidermal cells was an active process.

Live imaging of eyelid closure reveals cell intercalations

Immunolabeling of extracellular matrix (ECM) components in sagittal sections showed no signs of an organized basement membrane, suggesting that eyelid epithelium extends directly over the single-layered corneal epithelium (Figure S2). To understand how cells were moving, we optimized conditions for imaging the closure process ex vivo by culturing embryonic eye explants at an air-liquid interface (Li et al., 2011) (Figure 3A). Under these conditions, closure proceeded reliably ex vivo for up to 12 h, with minimal apoptosis and no obvious morphological defects (Figure S3A–C). To visualize and track collective cell movements and mitoses, we utilized eyelid explants from K14-H2BGFP-expressing embryos; to observe the dynamics of individual cells, we utilized explants from E15.5 Rosa^{mT mG} or Rosa^YFP embryos that had been transduced in utero at E9.5 with low-titer LV-Cre.

Figure 3.

Live imaging of the eyelid closure reveals front cell intercalations. (A) Schematic of live imaging setup. Eyelid explants are removed from K14-H2BGFP embryos, and placed on an air-liquid interface in an incubation chamber. With kinetics comparable to those in vivo, eyelid closure is then imaged for up to 16hr on an inverted confocal microscope. (B) 3D tracking of eyelid and surrounding epidermal cells reveals cells with significant tangential movements despite an overall radial trajectory. The average speed of front cells is substantially greater than surrounding epidermal cells (P < 0.0001, n = 7000–10000 tracks), suggesting an active role in closure. (C) Live-imaging of eyelids sparsely-labeled with mGFP reveals mediolateral protrusive activity and antiparallel cell intercalations (see accompanying Movie S5). (D) Left: a strong velocity component of the eyelid front, but not surrounding epidermis, is oriented perpendicularly to the closure axis. Average tangential velocity of cells in 256 μm² regions are plotted as contours. Arrow length is proportional to speed. Right: histograms of the overall orientation of cell movements relative to the eye center. Far from the center, movements are primarily radial (blue), while at the front, most cells move completely orthogonally (red). (E) Orientation of cell movements at early and mid/late eyelid closure. Left: Color-coded plots of the transverse displacements of front cells from live imaging data indicates that cells with opposing movements are distributed throughout the front. Middle: Heat maps of overall front cell speed. The fastest moving cells are nearest the eye border. Right: Plot of residual azimuthal velocity after averaging in 32×32μm regions throughout the front. Although domains of movement in a single direction appear as closure progresses, these are restricted to peripheral regions. Near the center, corresponding to the fastest-moving, most elongated cells, movements almost completely oppose each other as observed in early closure. (F) Left: In vivo whole-mount immunolabeling for golgi marker GM130 was used to create a map of eyelid front cell polarities, revealing that eyelid front cells are strongly polarized along their mediolateral axis, but that cells with opposing polarities are distributed throughout the front. Right: Immunolabeling of GM130 and aPKC in sagittal sections of eyelids reveals clear apicobasal polarity of cells in surrounding epidermis, but loss at the eyelid front. Scale bars, 50 μm (B, F), 10 μm (D). See also Figure S3.

When imaged in this way, the overall rate of closure ex vivo (~10 μm/hr) was similar that observed in vivo (~300 μm/24hr). We then performed live, 3D imaging of closure and tracked cell movements (Figure 3B, S3D). Interestingly, although the overall movement of eyelids was centripetal, individual front cells appeared to move perpendicularly to this axis (Movies S2–3). This was visualized by imaging clones of mGFP-labeled front cells, revealing mediolateral protrusive activity and intercalation of cells with their neighbors (Figure 3C and Movie S4). Moreover, when the average velocity of these cells was contrasted with the movement of surrounding epidermis, it was clear that front cells moved considerably faster, suggesting that their movement was at least in part independent of the other epidermal cells (Figure 3B).

These key features were more readily visualized by averaging the azimuthal component of cell movements in regions around the eye. This revealed substantial movements perpendicular to the closure axis within the eyelid front but not surrounding epidermis (Figure 3D). Likewise, histograms of the overall orientation of cell movements relative to the eye center revealed that cell movements far from the center were primarily radial, while most cells at the front moved orthogonally.

Although transverse movements were observed throughout the eyelid front, the tendency of the surrounding tissue to be drawn toward the center gave the appearance of domains of movement in a single direction. We explored this further by examining the spatial orientation of cell movements in early and mid/late closure (Figure 3E). Color-coding the azimuthal movements of cells based on their overall direction revealed that in early closure, cells migrating in opposing directions were distributed throughout the eyelid front (top left panel). By averaging velocities of front cells such that movements with equivalent speeds but opposite directions negated each other, it was clear that these movements largely opposed each other (top middle and right panels). Although a similar analysis of mid/late closure revealed distinct domains of movement, these regions were peripheral to the fastest-moving and most-elongated cells near the center (bottom panels). There, cell movements largely opposed each as observed at earlier stages. These analyses suggested that as closure progressed, active cell intercalation and stratification at the front was driving passive movement behind it.

Consistent with a mechanism of cell intercalation in which small transverse displacements contribute to substantial overall tissue movements, the polarity of front cells was strongly oriented along their mediolateral axis, but with cells in a given region frequently displaying opposite polarities (Figure 3F). This was in contrast to cells of surrounding epidermis, which maintained a clear apicobasal polarity. These opposing polarized movements at the eyelid front more closely resembled mesodermal convergent extension movements than epithelial fusion mediated by constriction of a supracellular actin cable or individual cell migration in wound-repair. However, unlike classical convergent extension movements, this behavior was restricted to a small population of cells.

Quantitative analyses of tissue deformations correlate with peak actomyosin intensity and favor a model of localized forces at the front driving epithelial sheet movement

A priori, cell intercalations could drive epithelial sheet closure either by expanding and spreading over the eye or by actively pulling on surrounding tissue. To distinguish between these possibilities, we first examined changes in tissue dimensions by live imaging (Figure 4A). By measuring the axial length of the front as the eyelid moved, we learned that it did not substantially expand over the eye. Rather and remarkably, the eyelid front appeared to compress. Consistent with this was an increase in cell density as closure progressed.

Figure 4.

Quantitative analysis of tissue deformation in the eye. (A) As the eyelid moves over the eye, the axial length of the layer of front cells remains relatively constant, and the density of cells in a fixed region increases linearly (n = 3 eyelids). This is accompanied by an overall thickening of the eyelid front between E15 and E16 in vivo (p = 0.0044; n = 8 early, 5 late eyelids). (B) View of the eyelid along the xz axis with three individual cells highlighted, illustrating convergence of cells along this axis. The slower speed of movement along the z axis relative to xy axis suggests that thickening is a consequence of tissue compression by mediolateral intercalation. (C) Quantitative analysis of tissue deformation in the eyelid. Top: particle image velocimetry (PIV) was used to measure the tissue flow field and to quantify deformation and shear in the eyelid and surrounding epidermis. The eigenvectors of the symmetrized velocity gradient tensor (strain rate tensor) identify the principal axes of deformation during flow, while the eigenvalues specify the rates of tissue deformation along these principal directions (right panel). Middle: the profile of average tissue compression and extension displays a sharp peak in the region of active cell intercalation, indicating that the eyelid front is maximally compressed perpendicular to, and extended along, the closure axis. Bottom: the region of maximum tissue compression and extension coincides precisely with peak actin and myosin intensity within the eyelid front.

Interestingly, the front thickened vertically during closure. Visualizing individually-labeled cells in 3D similarly revealed convergence along the xz as well as xy axis (Figure 4B), although quantification of cell velocities revealed that movements in z-direction were significantly smaller and unlikely to be an active part of the process. Importantly, live imaging showed no signs of concerted cell contraction, indicating that the observed changes in tissue dimensions were a direct result of cell intercalation (Movie S5).

If eyelid closure is driven by this region of active cell intercalation, we would expect to observe significant shear in the front, but little in the surrounding tissue. To test this, we used particle image velocimetry (PIV) to measure the tissue flow field and quantify the rate and direction of tissue deformations (Figure 4C and Supplementary Procedures). Notably, both the average rate of tissue compression and extension displayed sharp peaks at the eyelid front, indicating that in this region, the tissue was maximally compressed perpendicular to and extended along the axis of closure. Importantly, this region coincided with peak actomyosin intensity and cell elongation in the eyelid (Figures 4D, E). These results favored a model in which forces generated by cell intercalation tow the surrounding epidermis.

These tissue deformations differ fundamentally from those mediated by a contractile actin cable. The eyelid front exhibited local shear deformations, where compression and extension rates were similar. This is particularly evident in Figure 4D, where the two strain rates have similar magnitude but opposite sign throughout the front. In contrast, Drosophila dorsal closure is characterized predominantly by tissue compression alone, where the activities of the actin purse string and amnioserosa effectively compress the tissue along the closure axis (Blanchard et al., 2009).

A towing mechanism of eyelid closure

If cell intercalations drive large-scale tissue movements of eyelid closure as our quantitative analyses suggested, we might expect to observe deformation of cells in epidermis directly behind the front, but little movement of cells relative to each other. As illustrated by the colored row of cells live-imaged and shown in Figure 5A, cells maintained their relative positions despite significant translational movement toward the eye. Moreover, by comparing the velocity of epidermal cells to their nearest-neighbors and to cells at increasing distances, we learned that cells within surrounding epidermis are highly coordinated, while eyelid front cells were relatively uncoordinated, reflecting their more mesenchymal-like behavior.

Figure 5.

Intercalating cells are required to tow eyelids over the eye. (A) Behind the eyelid front, cells in the epidermis and their associated underlying dermis move as a coherent unit. Left: individual frames from a timelapse of basal epidermal cells located behind the eyelid front (arrow denotes direction, yellow bar denotes spatial reference). A row of epidermal cells are colored to appreciate that epidermal cells surrounding the eye maintain their relative positions and move as a unit during closure. Right: cells throughout the eye were tracked individually and their velocity vectors were compared to their nearest-neighbors (left panel), and to cells at increasing distances (right panel). These measurements underscore the high degree of coordination over large distances that epidermal cells behind the front display in comparison to the intercalating cells (p = 0.001, Wilcox test). (B) Analysis of cell elongation in the surrounding epidermis. Near the eye, cells become increasing elongated along the axis of closure (left). This elongation becomes more pronounced between E15 and E16 (right). (C) Compared to controls, epidermal cells are significantly less elongated when intercalating cells are laser-ablated (p < 0.0001; n > 1000 cells pooled from 5 eyelids per condition). Shown are representative images of these shape changes. (D) Left: when a region ~40 μm from the eye is ablated, the epidermis in front of the gap is pulled toward the eye. Frames from a timelapse sequence and quantifications illustrate that the tissue band broadens (yellow line) (n = 3 eyelids). Right: quantifications showing how the size and shape of the ablated region changes during closure. The gap widens toward the eye border until a wound healing response is initiated. (E) Ablation of a region (in red) of eyelid front cells prevents it from translocating. Top: scheme to measure the effects of a laser ablation by plotting translocation of the eyelid at each position along the eye perimeter. Bottom: plot of the average distance moved by each position along the eyelid border. The epidermis adjacent to a region of ablated front cells fails to move over the eye (red plot), while surrounding, unablated tissue is unaffected (gray plot). (F) Summary of additional ablation experiments and their effects on eyelid closure. See also Figures S4–5.

Quantification of cell elongation in vivo by tiled imaging of whole-mounts revealed that epidermal cells in close proximity to front cells elongated along the closure axis (Figure 5B). This became more pronounced as closure progressed. If active forces at the closure front are responsible for pulling on the surrounding epithelium, then specifically ablating the front cells should abrogate these epidermal cell elongations. After determining a two-photon laser power and dwell time that did not damage surrounding tissue (Figure S4A), we targeted the front cells. As shown in Figure 5C, their ablation resulted in a relaxation of epidermal cells behind this zone, returning them to the shape of epidermal cells more distant from the eye.

To evaluate whether forces generated by front cells not only stretch cells but also pull the eyelid epidermis forward, we ablated a region ~40 μm behind the front cells, and measured the effects both at the wound site and in the tissue located between the wound and eye (Figure 5D). Following laser ablation, the epidermis between the laser gap and the eye border widened over time, while the wound site widened towards the eye until a wound healing response was initiated to close it (see also Figure S8). These results are consistent with front cells generating a pulling force.

We directly tested the importance of front cells to eyelid closure by specifically ablating a region of intercalating cells along one eyelid and measuring the immediate effect on the translational movement of adjacent epidermis. In contrast to the control eyelid, epidermis directly behind the ablated front cells failed to move effectively over the eye (Figure 5E and Movie S6).

A variety of additional tissue ablations followed by live imaging further differentiated between multiple possible modes of eyelid closure (Figure S4B–C, summarized in Figure 5F). Ablating cells on the underside of the eyelid had little or no effect on closure. We ruled out contributions from underlying dermal tissue by imaging K14-Cre × Rosa^{mT mG} embryos (Figure S5 and Movie S7). PIV analysis revealed that average flow velocities in dermis (red) and surrounding eyelid epidermis (green) were essentially equivalent (2.3 μm/hr vs. 2.8 μm/hr, p = 0.16); moreover, these velocities were substantially lower than those of green front cells (4.3 μm/hr, p < 0.001). These findings provided additional evidence against the notion that closure is driven by cell shape changes that bend the eyelid over the eye or by streaming of dermal cells into the region.

Finally, whereas ablation of front cells effectively halted eyelid closure, ablation of corneal epithelial cells or eyelid canthi had little or no effect, arguing against contributions from the cornea or a zippering mechanism. The lack of corneal effects differed markedly from similar studies of Drosophila dorsal closure, where ablation of underlying amnioserosal tissue has dire consequences to the process (Hutson et al., 2003; Solon et al., 2009). Overall, eyelid closure shared a few characteristics with dorsal closure but also exhibited striking differences.

Molecular mechanisms underlying eyelid closure

If intercalation of front cells were truly the driving force for eyelid closure, as our mathematical analyses and laser ablation studies suggest, then we should be able to halt the process by genetically interfering with their movement. Although eyelid closure differed from conventional epithelial convergent extension movements, the behavior of front cells bore certain similarities to mesodermal convergent extension movements, where integrin-fibronectin interactions orient protrusive activity and organize ECM (Davidson et al., 2006; Marsden and DeSimone, 2003; Skoglund and Keller, 2010). This parallel seemed worth pursuing since α5 integrin had been shown to be upregulated specifically in these cells (Caroll et al., 1998).

We first showed that fibronectin and its receptor α5β1 is enriched in the eyelid front (Figure 6A; see also Figure S2). Given the well-established opposing actions of adherens junctions and integrins in epidermis (Livshits et al., 2012 and references therein), we asked whether this was accompanied by a corresponding downregulation in P- and/or E- cadherins. While functionally redundant, P-cadherin was nearly absent in eyelid front cells (Figure 6B). E-cadherin was also diminished in front cells, and Cdh1 transcriptional downregulation was verified by FACS-isolation and RT-PCR analysis of these cells (Figure S6).

Figure 6.

Eyelid closure requires myosin-IIA and integrin-fibronectin adhesion for cell intercalation. (A) Fibronectin and its receptor α5β1 integrin are specifically enriched in the eyelid front. Left: E15.5 eyelid whole-mount immunolabeling illustrates localized expression in the region of intercalating cells. Right: sagittal view, showing that expression is not in corneal epithelium underlying the front cells. (B) Expression of core adherens junction components in the eyelid (sagittal views). Note that P-cadherin is present in the periderm and surrounding epidermis but absent in the front cells. (C) Lentiviral-mediated Cdh1-shRNA/H2BRFP expression results in near-complete loss of E-cadherin in transduced regions of the eyelid (arrowheads denote rare untransduced cells). Note that α5 integrin is still restricted to the eyelid front, eyelids still close and cell intercalation speed is equivalent to embryos transduced with a scrambled-sequence control hairpin. (D) Loss of α5 integrin, fibronectin, or myosin-IIA prevents eyelid closure. Left: Quantifications of shRNA-mediated knockdowns in vitro by real-time PCR (top) and eyelid closure defects at E16.5 (bottom). (E) Injection of lentiviral K17-Cre (LV-K17-Cre) into Rosa^YFP embryos and analysis at E15.5 illustrates the specificity of Cre-expression in eyelid front cells and co-localization with K17 protein (top). Same injections of LV-K17-Cre into Myh9^fl/fl embryos results in open eye phenotype and loss of hair follicle bumps (4/4 injected embryos vs. 6 WT littermates) (bottom). (F) Despite eyelid closure defects, the morphology of front cells in all these KDs is intact, illustrated here by a whole-mount phalloidin staining in an Fn1 KD embryo. (G) Left: quantification of the velocity of front cells following KD. For all KDs, intercalation speeds are reduced compared to a scrambled-shRNA. Middle: the overall rate of eyelid closure, measured by following the eye border over time, is similarly reduced upon knockdown (p < 0.0001). Right: comparable to laser ablation of intercalating cells, cell elongation in the surrounding epidermis is reduced when eyelid front cells lack fibronectin or α5 integrin. Scale bars, 50 μm. See also Figures S6–7.

Previous studies of convergent extension movements suggest that fibronectin can have multiple roles in regulating cell intercalation movements. On the one hand, integrins and fibronectin could be required actively for the necessary shape changes and movements, either for cells to gain traction on their neighbors and the surrounding ECM or to orient protrusive activity. On the other, cell intercalation could primarily be mediated by remodeling of cadherin-based junctions, with integrin-fibronectin signaling functioning passively to reduce cell-cell adhesion to a permissible level (Marsden, et. al., 2003).

Preventing adherens junction formation in the epidermis is known to compromise eyelid closure, likely due to overall lack of sheet integrity (Vasioukhin et al., 2001). We therefore exploited the redundancy of P- and E-cadherin in the skin and transduced E9.5 embryos with Cdh1 shRNAs to selectively reduce cell-cell adhesion in the eyelid front. Despite markedly diminishing E-cadherin, cell intercalation and eyelid closure progressed similarly to wild-type (Figure 6C; S7). This was in contrast to the broader-scale intercalations of convergent extension movements in Xenopus, which require cadherins (Delarue, et. al., 1998). Importantly, reducing E-cadherin did not cause surrounding epidermis to gain α5 integrin expression or adopt morphological and motile properties of eyelid front cells, indicating that other signaling cues drive these events.

We next examined the consequences of knocking down α5 and fibronectin expression. To avoid potential indirect roles of integrins and fibronectin in regulating cell-cell adhesion or generating local migration signals, we also targeted myosin-IIA (Figure 6D). In all three cases, surrounding epidermis was largely intact and overall cell architecture, adhesion to basement membrane, and cell-cell junctions were not noticeably perturbed (data not shown). Myosin IIA depletion did result in occasional errors in cell divisions, and a block in hair follicle formation.

The most striking defect for all three KDs was failed eyelid closure at E16.5 (Figure 6D). This was even true when we used a K17-Cre lentivirus to selectively knockout myosin-IIA in the front cells of Myh9^fl/fl embryos (Figure 6E). Eyelid front cells diminished for α5, fibronectin or myosin-IIa still displayed their elaborate array of actin fibers and adopted their characteristic bipolar, elongated morphology, indicating that their specification was unaffected (Figure 6F). Interestingly, however, these cells migrated significantly more slowly than control cells (Figure 6G). This decrease in cell movement was accompanied by a significantly reduced overall rate of eyelid closure, based on measurements of eyelid border translocation over time (Figure 6H). Notably, inhibiting intercalation movements of front cells by KD of either α5 integrin, fibronectin or myosin-IIa also led to reduced cell elongation of surrounding epidermal tissue (Figure 6I). These data suggested an active contribution to the forces generated at the front.

Depletion of β1 integrin had a broader range of defects, consistent with β1’s partnership with the major epidermal integrin α3 (data not shown; see Raghavan et al., 2003). Underscoring the specific importance of α5, however, no eyelid closure defect was detected upon KD of αv, a fibronectin receptor that has been reported to be expressed in migrating but not stationary epidermal cells (Marchisio et al., 1991). Together, these results suggest a direct reliance on α5 integrin/fibronectin for the cell intercalations that drive eyelid closure, and that they are sufficient to drive the process.

Discussion

Investigating how cells move collectively in morphogenetic processes has yielded many insights into the diversity of cellular behaviors that contribute to the shaping of tissues and organs, in homeostasis, wound-repair and disease. Elegant studies in Drosophila, Xenopus, and zebrafish have revealed both significant commonalities and important differences in the mechanisms and regulation of such movements as convergent extension and epithelial fusions. By developing eyelid closure as a genetically and morphologically tractable and accessible model to study the dynamics of epithelial morphogenetic movements in mice, we have uncovered a mechanism that differs not only from these previously described collective cell movements, but also from prevailing theories as to how eyelid closure might take place.

Eyelid closure defects have been reported for mice harboring mutations in a diverse array of signaling proteins, including EGF and TGFα (Luetteke et al., 1993; Miettinen et al., 1995), and both canonical and non-canonical Wnt signaling (Gage et al., 2008; Huang et al., 2009; Wu et al., 2012). By activating the MAPK-ERK signaling pathway, growth factors have been implicated in epidermal proliferation as well as migration (Minn et al., 2005; Tao et al., 2005; Huang J et al., 2009). This has led to a model whereby epithelial cells at the eyelid front originate from proliferation of periderm cells, which then migrate across the cornea and fuse. Despite the attractiveness of this hypothesis, the role of the periderm in eyelid closure has remained correlative. Our lineage tracing and ablation studies show that epidermis rather than periderm is the origin of front cells and that accumulation of periderm cells around the developing eye is a consequence rather than an essential component of the process. Likewise, our studies with mitomycin-C show that if proliferation is impaired just prior to eyelid closure, the process proceeds largely unscathed. While this is not to say that proliferation is dispensable for eyelid formation, it is dispensable for the closure process.

Actin polymerization has been implicated as the downstream effector of these pathways in eyelid closure. Substantiating evidence comes from genetic studies, which show essential roles for Rho-associated kinases 1 and 2 (ROCKs 1 and 2) and their relative, LIMK2 (Shimizu et al., 2005; Thumkeo et al., 2005; Rice DS et al., 2012). Further strengthening the notion that actomyosin bundle formation is essential for eyelid closure is that EGF cannot stimulate myosin light chain (MLC) phosphorylation when ROCK1 is absent (Shimizu et al., 2005). Moreover, LIMK2-deficient keratinocytes in vitro show reduced actin filaments (Rice DS et al., 2012). Together, these results have led to the hypothesis that eyelid closure might be analogous to wound healing in mammals and/or dorsal closure in Drosophila, and driven either by a) constriction of a supracellular actin cable at the epidermal border in a mode of purse-string closure, and/or b) forward migration of a growing epithelial sheet across the cornea.

By combining high resolution imaging of eyelid whole-mounts with a quantitative analysis of proliferation, cell movements, laser ablations, and genetics, we have provided the necessary context to evaluate and refine these possibilities. Our studies lend compelling support for the notion that actomyosin dynamics and cell migration are essential for eyelid closure. However, our studies also show that neither individual cell migration nor a traditional model of actomyosin cable contraction account for the surprising way in which they are utilized by the process.

In striking contrast to traditional models of epithelial fusion, in which cells bordering the gap retain their epithelial characteristics and assemble an actomyosin ring whose contraction drives closure without substantial cell rearrangements within the tissue, epidermal cells in the eyelid front adopt some mesenchymal properties and actively migrate perpendicular to the closure axis. Our quantitative analysis of tissue deformations indicates that these movements maximally compress and extend the tissue at the eyelid front. Moreover, the presence of two distinct regions in the eyelid—one of active shear at the eyelid front and the other a passive region in the surrounding tissue—distinguishes it from processes driven by a contractile actin cable, in which deformations are compressive throughout. Thus, by generating a localized contractile force through active cell intercalation, eyelid front cells achieve the functional equivalent of purse string closure, but by a fundamentally distinct mechanism.

Intercalations are often associated with convergent extension movements (Keller et al., 2000). In epithelial tissues such as the Drosophila germband, this occurs by a myosin II-dependent remodeling of specific cell junctions, and critically depends on an asymmetric distribution of myosin-II and ROCK activity (Fernandez-Gonzalez et al., 2009). In mouse skin epidermis, myosin-II polarization is not obvious (Figure 3), and the cell cortex appears to be under uniform tension as evidenced by its response to laser-ablation of cell-cell junctions (Luxenberg, Heller, and Fuchs, unpublished).

Rather, intercalations of eyelid front epidermal cells bear more resemblance to the mediolateral intercalations of dorsal mesoderm in Xenopus or mouse notochord, where mesenchymal cells elongate perpendicular to the axis of tissue extension and spawn mono- or bipolar actin-based protrusions. Like eyelid front cells, these mesenchymal cells are thought to exert tractive forces on surrounding cells, pulling cells between each other and mediating the ordered rearrangement of tissue (Keller et al., 2000; Nishimura et al., 2012; Yamanaka et al., 2007). Finely-tuned levels of cell-cell adhesion and cell-ECM interactions have also been implicated in this process, either by polarizing cell protrusive activity or by enabling cells to exert traction on each other (Davidson et al., 2006; Marsden and DeSimone, 2003).

Our genetic analyses illustrating the essential nature of α5β1 integrin, fibronectin and myosin-IIA in eyelid closure further underscore these similarities. Absence of eyelid closure defects upon further reduction of cadherins suggests that, fundamentally, the cell intercalation that occurs is integrin- rather than cadherin-dependent. Importantly, eyelid front cells still adopt their unique morphological and motile properties upon loss of fibronectin and integrins, suggesting that they are regulated by other signals in the region. Finally, inhibiting cell motility by targeting myosin-IIA, the downstream effector of integrin signaling, implicates a role for integrins and fibronectin in active cell migration. That said, eyelid front cells both maintain and require their epidermal character, ultimately sealing the eyelid sheets. Moreover, in contrast to mesenchymal convergent extension, the eyelid front tissue doesn’t actually converge or extend; rather it generates contractile forces to tow the epidermis behind it over the cornea, thickening vertically in the process.

Whether epithelial or mesenchymal, convergent extension movements typically involve relatively homogenous tissue, so that the cell intercalations reshape the entire tissue. In this regard, it was striking that cell intercalation was restricted to the eyelid front. Our live imaging and ablation experiments clearly show that surrounding epidermal cells maintain their intercellular partners and do not themselves undergo intercalations during closure, instead elongating along the axis of closure. Most importantly, our analysis of tissue deformation rates suggests that this elongation, and indeed the movement of the eyelid as a whole, occurs as a passive response to forces generated by intercalation of front cells. In total, our results support a model in which contractile forces generated by cell intercalation tow the eyelid epidermis over the eye (Figure 7). While passive tissue movements accompany nearly all examples of cell intercalation in development, this offers a unique example in which a population of intercalating cells is specified to carry out a process of epithelial fusion.

Figure 7.

A towing mechanism of eyelid closure. Proposed towing mechanism of eyelid closure. As the eye emerges and local (likely Wnt) signals are transmitted, a population of surrounding epidermal cells downregulate Cdh1, activate Fn1 and Itga5 and acquire mesenchymal features (top). They produce an elaborate actomyosin network and initiate intercalation movements, leading to localized compression and enabling them to exert a net force on the surrounding epidermis. They elongate the epidermal cells within the sheet and pull it over the eye (bottom).

In summary, our results suggest that the forces underlying eyelid closure are front cell intercalations mediated by fibronectin, α5β1 and myosin IIa-based movements and supported by a concomitant reduction in cadherins. Finally, although beyond the scope of the present study, which focuses on collective movements in tissue morphogenesis, it is intriguing that changes in actomyosin dynamics are well-known consequences of non-canonical Wnt signaling, while other key changes, Cdh1 downregulation and Fn1 upregulation, are consequences of canonical Wnts (Wu CI et al., 2012; Jamora et al., 2003; Ten Berge et al., 2008). As shown in Figure S6A, canonical Wnt signaling is indeed active in the eyelid, particularly at the junction where the front meets the surrounding epidermis, and likely plays a role in signaling upstream of the cellular mechanisms we’ve unearthed in this study. Given the localized action of these morphogens, this would also explain why the behavior of front cells differs so markedly from the surrounding epidermal cells. As an example in which cells derived from a differentiating epidermis become migratory and undergo cell intercalations, eyelid closure becomes a paradigm for understanding how well-described mechanisms of collective cell movement can be tailored and combined to achieve morphogenetic processes in increasingly complex tissue environments.

Experimental Procedures

Immunofluorescence and fixed tissue imaging

For sagittal tissue sections, whole embryos were embedded in OCT, sectioned (10 μm) on a Leica cryostat, and fixed for 15 min in 4% paraformaldehyde in PBS. Sections were blocked and permeabilized for 1 h in blocking buffer (0.3% Triton X-100, 1% BSA, 1% fish gelatin, 5% donkey serum, and 5% goat serum in PBS). Primary antibodies were incubated overnight at 4°C.

For whole-mount immunofluorescence, embryos were fixed for 1 h in 4% PFA. Eyelids or skin explants were dissected and blocked/permeabilized for 5 h-overnight in blocking buffer. Primary antibodies were diluted 1:200, secondaries 1:500, and incubated for 24 h at 4°C, followed by 5 h of washing in 0.1% Triton in PBS, exchanged every ~30 minutes. F-actin was labeled using Alexa Fluor 546 or 647-conjugated phalloidin (Life), diluted 1:500 and incubated for 2 h at room temperature.

Low-magnification imaging was performed on a Zeiss Axioplan2 using a Plan-Apochromat ×20/0.8 air objective. High magnification images were collected on a Zeiss LSM 780 or 510 Meta using Plan-Apochromat ×63/1.4 oil or C-Apochromat ×40/1.2 W objectives. Tiled Z-stacks were collected using Zeiss ZEN software or PerkinElmer Volocity.

Eyelid culture and live imaging

Eyelids were explanted from mid-to-late E15.5 embryos into warm media (Defined Keratinocyte Serum-free Media supplemented with 600 μm calcium and 5% w/v penicillin-streptomycin, Life Technologies). For live imaging, we used methodology previously described for embryonic skin explants (Li et al., 2011). Using a small volume of growth factor-reduced matrigel, allowed to polymerize for 25 min at 37°C, we sealed eyes against a Lumox teflon-bottom dish (Sarstedt). Eyelid closure was imaged for periods of 6–16 h in 5% CO2 on a PerkinElmer Volocity spinning disk system equipped with a heated enclosure and gas mixer (Solent) and ×20/0.75 CFI Plan-Apo objective, or a Zeiss LSM 780 system with a stage-top incubator and Plan-Apochromat ×20/0.8 objective.

Image analysis

Basic image analysis and all manual measurements were performed in ImageJ. 3D reconstructions were made in Bitplane Imaris. Quantification of proliferation and apoptosis rates in tissue sections and whole mounts was performed using adaptive thresholding and watershed segmentation in semiautomated fashion using custom ImageJ macros. Measurement of cell elongation in tiled Z-stacks around the eye was performed using custom MATLAB scripts, in which cells are segmented based on cortical actin staining or membrane-GFP expression using a watershed algorithm. Cell elongation is defined as 1 - W/L, where W and L are the major and minor axes determined by computing central moments of segmented cells.

Global analysis of cell polarity in the eyelid in Figure 3F was performed by 3D imaging of H2B-GFP expressing whole-mounts stained with GM130 or pericentrin. Nuclei were filtered using a 3D bandpass filter, and GM130 or pericentrin spots were filtered using a median filter to merge puncta into a single region. Both channels were segmented independently in 3D, and pericentrin spots were assigned to the nearest nucleus. A cell’s axis of polarity was defined by the vector connecting the centroid of a nucleus to the location of the nearest GM130 or pericentrin spot.

Quantitative analysis of cell movements

Image stacks were deconvolved using the Richardson-Lucy algorithm and filtered using a 3D bandpass filter. Nuclei were segmented in 3D using a simple region-growing algorithm (Keller et al., 2008) and tracked using the tracking module from the Danuser lab’s μTrack software package, which features robust handling of track splitting and merging (Jaqaman et al., 2008). Segmentation and tracking efficiency were determined by manual verification in two movies (Figure S3). Downstream analysis of cell speed was performed using custom MATLAB routines.

For determining cell-cell coordination in eyelid front cells and the surrounding epidermis, we used a Delaunay triangulation to connect cells to their nearest neighbors, and computed the cosine similarity between velocity vectors of cell pairs using $\mathit{cos}\theta =\frac{{\mathrm{v}}_1·{\mathrm{v}}_2}{\mid {\mathrm{v}}_1\mid \mid {\mathrm{v}}_2\mid }$.

PIV analysis was performed in the PIVLab MATLAB package (Thielicke, et. al., 2014).

Laser Ablation

Tissue ablations were performed on a Zeiss LSM 510 NLO system using a Ti:sapphire laser (Chameleon Ultra, Coherent Scientific) tuned to 800 nm. Laser power and dwell time were calibrated per experiment, but were typically performed between 80–100% transmission using scan speed 6 and 50–75 repetitions (~90–140 μs dwell time; see Figure S4). Quantification of the effects of ablations was performed by manually tracing the border of the eyelid every 10 frames, fitting a spline to the points, and calculating the average distance traveled using numerical integration.

Lentiviral transductions

Production and concentration of lentivirus, as well as ultrasound-guided in utero injections were performed as previously described (Beronja et al., 2010). shRNAs were obtained from the Broad Institute’s Mission TRC-1 mouse library.

Statistical Analysis

All statistical analyses were performed in the R statistical environment (R Core Team, 2014). A two-tailed, unpaired t-test was used to assess the level of significance between two experimental conditions, while multiple conditions were compared using ANOVA followed by Tukey’s HSD test. All means are reported ± s.e.m. unless otherwise indicated.

• Establishes embryonic eyelid closure as a model of collective cell movements in mice

• Cells within a differentiating epidermis become motile likely through Wnt signaling

• Localized cell intercalation generates a region of active shear at the eyelid front

• Laser ablation and genetic loss-of-function suggest a towing mechanism of closure