Cellular segregation in cocultures is driven by differential adhesion and contractility on distinct timescales

Significance

Fundamental biological processes, such as tissue morphogenesis during development, rely on the correct sorting of cells. Cellular sorting is governed by basic physical properties such as the adhesion between cells and their individual contractility. Here, we study the impact of these parameters in cocultures consisting of epithelial wild-type cells and overly contractile, less adhesive tight junction–depleted ones. We find time-dependent segregation into clusters: Differential contractility drives fast segregation on short timescales, while differential adhesion dominates the segregation over longer times.

Abstract

Cellular sorting and pattern formation are crucial for many biological processes such as development, tissue regeneration, and cancer progression. Prominent physical driving forces for cellular sorting are differential adhesion and contractility. Here, we studied the segregation of epithelial cocultures containing highly contractile, ZO1/2-depleted MDCKII cells (dKD) and their wild-type (WT) counterparts using multiple quantitative, high-throughput methods to monitor their dynamical and mechanical properties. We observe a time-dependent segregation process governed mainly by differential contractility on short (<5 h) and differential adhesion on long (>5 h) timescales. The overly contractile dKD cells exert strong lateral forces on their WT neighbors, thereby apically depleting their surface area. Concomitantly, the tight junction–depleted, contractile cells exhibit weaker cell–cell adhesion and lower traction force. Drug-induced contractility reduction and partial calcium depletion delay the initial segregation but cease to change the final demixed state, rendering differential adhesion the dominant segregation force at longer timescales. This well-controlled model system shows how cell sorting is accomplished through a complex interplay between differential adhesion and contractility and can be explained largely by generic physical driving forces.

Cellular sorting and tissue separation are essential processes in embryogenesis and tissue development studied across multiple species (1, 2). Early work has shown that cells taken from different embryonic tissues and remixed together eventually segregate again (3, 4). Sorting of cells in tissues can be governed by different biological and physical factors. Owing to our accumulated knowledge about cell–cell junctions and the cytoskeleton, a hypothesis for cellular demixing based on differential adhesion was proposed (5, 6). To accommodate different biological scenarios, this hypothesis was complemented by incorporating differential cell contractility (7, 8). Adhesion- and contractility-induced tensions act antagonistically: Contractility induces cell rounding to minimize the contact zone, whereas adhesion enlarges the cell–cell contact region. The resulting surface tension of the tissue is the ratio of adhesion and contractility (9). This view has been extended more recently by the addition of local contractile cues, for example, in the anteroposterior compartment boundary in Drosophila (10–12). Alternatively, active cell forces have been proposed to also participate in regulating cellular demixing in cocultures (13). However, it remains difficult to differentiate between the various factors that govern cell sorting. In recent years, many simulation-based studies characterized different physical driving forces of demixing, identifying many possible pathways to cellular segregation via differential physical cell properties such as cell–cell adhesion, cortical tension, contractility, and cell–substrate adhesion (9, 14–23).

Recently, it has been shown that in tight junction–depleted epithelial cells (ZO1 and 2 knockdowns; abbreviated as dKD), two distinct cell populations emerge. Some cells experience impaired ROCK signaling and contract, taking on a rounded shape; pulling on their neighbors eventually results in laterally elongated cells coexisting with the contracted cells (24, 25).

Here, we now address the driving force of segregation by studying cocultures of dKD and WT cells using high-throughput/content (de)mixing experiments in combination with quantitative mechanical single-cell measurements. We found that a time-dependent demixing process in these cocultured monolayers is governed by differential contractility on short timescales (within the first 5 h), while on longer timescales (>5 h), differential adhesion prevails. Our results are consistent with Brodland’s differential interfacial tension model, which incorporates both contributions but here leads to cell sorting in a time-shifted manner (8).

Results

Demixing of Cocultured, Highly Contractile dKD, and Compliant WT Cells.

First, live cell (de)mixing experiments were recorded directly after thorough mixing and seeding using phase contrast and fluorescence microscopy (Fig. 1A). We used WT cells with a GFP tag (named WT-GFP from here on, see Materials and Methods section) to distinguish them from dKD cells. Cell segmentation and neighbor analysis using both the fluorescence signal and phase contrast images allowed for the automatic assignment of cells as WT or dKD. This enabled us to quantify how much the cells mixed randomly or demixed into clusters, also called segregation. Therefore, we defined a segregation index SI as the number of homotypic neighbors divided by the number of all neighbors.

Fig. 1.

Demixing behavior of dKD and WT cell cocultures at an initial mixing ratio of 50:50. (A) Example overlay of phase contrast (gray scale) and fluorescence (green: WT-GFP cells) channels with corresponding segmentations (green: WT-GFP, magenta: dKD cells in WT-GFP/dKD mix or WT in WT-GFP/WT control). Samples were imaged immediately after seeding and mounting on the microscope (0 h). Scale bars: 200 µm and 50 µm (zoom-in). (B) Demixing, cell amount, and area occupancy quantification. The vertical dashed line at 5 h indicates two distinct demixing timescales. The shade in the first 3 h indicates subconfluence. The yellow color represents the segmentation analysis of the WT-GFP/dKD mixture, while the purple symbols refer to the analysis of the control sample WT-GFP/WT. (B1) The segregation index SI, defined as the average ratio of homotypic and all cell neighbors, quantifies the demixing degree. The SI is shown averaged over both cell types (Top) and separately for each cell type (Bottom). (B2) Top: Relative cell amount calculated as the ratio of the number of WT-GFP cells and the total cell amount. Bottom: Total cell amounts of each cell type. (B3) Top: WT-GFP fraction of the overall cell area calculated as the ratio of the WT-GFP area and the total cell area, indicating contractility discrepancies between the cell types. Bottom: Mean cell area of each cell type. Corresponding zoom-ins of the first 5 h are shown in SI Appendix, Fig. S1, and distributions of the individual cell areas are depicted in SI Appendix, Fig. S2. Mean values and SDs are shown. 12 separate regions from six culture dishes, acquired on three separate days (two regions per dish and two dishes per day), were measured and are shown per coculture mix. Data can be found in Dataset S1.

In the case of completely random cell distribution, an average segregation index of 0.5 would be expected. However, this parameter is also impacted by natural, local processes such as cell division. To account for these deviations from randomness, we performed control experiments using a pseudo coculture consisting of WT-GFP cells and unmodified WT cells.

After detaching the cells and mixing the suspensions thoroughly before seeding, initially, both the WT-GFP/WT control and the WT-GFP/dKD mixture showed a segregation index close to 0.5 (Fig. 1B1). The slight shift to higher values was likely already introduced upon initial seeding when most cells were still sedimenting, while others were already attached. Within the first hour, both cocultures initially demixed from about 0.52 to 0.57 (Fig. 1B1, see SI Appendix, Fig. S1 for zoom-in). After this annealing time, only the WT-GFP/dKD mixture segregated further, as expected. The SI increased to about 0.63 within the first 5 h, whereas the control remained at 0.57. After this fast initial demixing, both cocultures segregated further at a similar rate to reach values close to 0.7 for the WT-GFP/dKD and 0.6 for the WT-GFP/WT cells.

In coculture with dKD, WT cells are sorted into large, less percolated clusters (Fig. 1A) with a higher average SI than their dKD counterparts (SI Appendix, Figs. S2 and S3 and Fig. 1B1, on the right). The dKD cells, with a lower SI, were arranged in elongated, string-like clusters around the WT domains. This is expected from the differential interface tension model predicting the initial formation of chains of cells (here dKD) that in later stages anneal and coalesce (8). If these later stages of complete segregation are reached depends on the difference in interfacial tension. In contrast, the WT-GFP/WT control showed an inconspicuous and less defined layer morphology. The SI of the labeled WT cells was generally higher than that of the unlabeled ones. However, this SI difference vanished over time in the WT-GFP/WT coculture, whereas in the WT-GFP/dKD mixture, it even increased. Accordingly, WT-GFP/dKD cocultures exhibited a sorting behavior into distinct clusters different from homotypic monolayers.

In SI Appendix, Fig. S3, we show the detailed shape analysis and derived statistics of dKD clusters formed in confluent monolayers of WT and dKD cells (50:50) as a function of time compared with control samples composed of WT-GFP/WT cells (50:50). Initially (<6 h), a larger number of clusters are formed compared with the WT-only sample, but they coalesce with time (> 6 h). This merging of clusters is attributed to the minimization of line tension generated by the contractility of dKD cells (8). The aspect ratio follows this trend as it first rises and eventually decreases again.

As a control/normalization parameter for the SI, we next examined the cell amount of both cell types in each coculture because a difference in the relative cell amount could influence the SI as well. However, we observed no difference in the relative cell amount (WT-GFP fraction of the cell amount in Fig. 1B2) between the WT-GFP/WT control and the WT-GFP/dKD cells. Interestingly, the total cell amount differed, with overall higher proliferation rates and larger cell amounts in the WT-GFP/dKD mixture. After a short delay in dKD cell proliferation, the dKD increased at a similar rate as the WT-GFP amount from about 3 h until 15 h after seeding. Importantly, the resulting small difference in the cell amount between the cell types was present in both the WT-GFP/dKD coculture and WT-GFP /WT control (Fig. 1B2), possibly slightly biasing the SI of both to larger values. After 15 h, dKD cells started to extrude apically out of the layer, offsetting proliferation and thereby stalling the cell amount. In the WT-GFP/WT mixture, the WT-GFP also showed slightly more proliferation until 15 h after seeding, which then leveled off.

Next, to examine the cell contractility discrepancy of these cell lines, which was described previously (24, 25), we first quantified the labeled WT fraction of the cell area (Fig. 1B3). If there were no discrepancies in contractility in the coculture, this parameter would be expected to be 0.5 because each cell type would occupy 50% of the covered area. Indeed, this was the case for the WT-GFP/WT control. In contrast, however, the WT-GFP/dKD cocultures showed a strong increase in the WT area fraction within the first 5 h, precisely correlating with the SI increase [see SI Appendix, Fig. S1 (zoom-in)]. This highlights a great differential contractility with highly contractile dKD cells occupying smaller areas and stretched WT cells covering more space on the culture dish. At the same time, as described before, the relative cell amount stayed constant, confirming that the larger area coverage of WT cells is due to lateral extension provoked by contractile dKD cells and not a consequence of an increased amount of WT cells. Notably, this effect only develops over time due to collective cell–cell interactions because the WT-GFP/dKD mixture also starts at SI = 0.5. However, the contractile discrepancy is generally underestimated here. This is because the phase contrast channel was used for analysis (the fluorescence was only used to assign the cell type, see Materials and Methods section), but the lateral stretching of bordering WT cells by dKD neighbors can be best observed in the WT-GFP-specific fluorescence channel (white arrows in Fig. 2B). This is because the WT cell body extension, even overlapping above dKD cells, is specifically seen in the GFP channel (Fig. 2B), while in phase contrast, the overlapping WT and dKD cell bodies cannot be distinguished well (Fig. 1A).

Fig. 2.

Differential actomyosin contractility of WT-GFP/dKD cocultures. (A1) Representative WT-GFP/dKD cocultures costained for phosphorylated myosin (P-myosin-2 antibody), actin (phalloidin), nuclei (DAPI, cyan), and ZO1 (ZO1 antibody, green). ZO1 was used to distinguish ZO1/2 dKD from WT cells. Magenta lines indicate the location of the XZ view. (XY scale bar, 20 µm, Z, 5 µm.) (A2) Quantification of fluorescence intensities from A1 including data from two large images. Boxes show the median and the upper and lower quartiles; whiskers indicate the 5th and 95th percentiles, while data points represent individual cells. Junction to center ratio refers to fluorescence intensity ratios obtained by a segmentation process detailed in the SI. (B) WT-GFP/dKD coculture costained for actin (phalloidin) and nuclei (DAPI, cyan). The green channel was used to identify the WT-GFP cells and to examine their morphology in 3D. (XY scale bar, 50 µm, Z, 10 µm.) (C) Apical topography of WT-GFP/dKD cocultures obtained by AFM imaging. Height profile, the corresponding 3D topography map which was upscaled vertically by 50%, and the error signal (deflection image). Scale bar: 20 µm. Cells in A were fixed after 28 h and in B and C after about 48 h of growth. Data can be found in Dataset S2.

Generally, the cell area of both WT and dKD cells in coculture or control samples decreases over time due to the compaction and jamming of cells in the confluent state [Fig. 1B3 (Bottom)]. After 18 to 20 h, jamming has finished, and due to the strong contractile forces, extrusion of primarily dKD cells leads to an overall loss of cells. After approximately 20 h, the number of dKD cells decreases due to preferential exclusion from the cell monolayer, a consequence of apical constriction. Notably, extruded cells are sometimes not correctly captured by the segmentation algorithm, leading to an overall decrease of occupied area.

Impact of Proliferation on the Segregation Index.

Besides contractility and adhesion, proliferation might also be an important factor to foster demixing by enlarging the clusters in the demixed state. Albeit we could not find an enlargement in dKD cluster size (SI Appendix, Fig. S3), we conducted experiments in the presence of mitomycin C which effectively suppresses proliferation in the separation process (SI Appendix, Fig. S4). Switching off proliferation essentially had no impact on initial cell sorting (SI Appendix, Fig. S5). However, cell sorting at later time points is slowed down, suggesting missing dKD cells that were removed from the cell layer due to apical constriction in the untreated WT-GFP/dKD (SI Appendix, Fig. S4) as also discussed above.

Differential Actomyosin Contractility and 3D Cell Morphology of WT-GFP/dKD Cocultures.

To further study the differential contractility of WT-GFP/dKD cocultures on a molecular and cell morphological level, we applied confocal fluorescence microscopy and AFM imaging (Fig. 2).

In previous work (24), we observed a strong actomyosin upregulation at the apical–lateral cell periphery of dKD cells (Fig. 2A). Particularly, activated (phosphorylated) myosin accumulated at the apical cell–cell junctions. A thick perijunctional actomyosin ring was formed, constricting the ZO1/2-depleted cells apically. Conversely, the WT cells did not show any upregulation of phosphorylated myosin-2 or of the actin cytoskeleton. To conserve the cellular volume, dKD cells were forced to bulge out apically. Since all dKD cells were still connected to their neighbors, adjacent WT cells were stretched out and flattened by the apical pull of the dKD cells. Strikingly, WT cells at the WT-GFP/dKD interface were partially pulled across their direct dKD neighbors toward the center of the dKD cluster (Fig. 2B, white arrows). Note that this lateral pulling translocates certain cell components, such as ZO1 or myosin in Fig. 2 A1 and B, relative to the nucleus. Both for actin and p-myosin, it was found that dKD cells display a larger junction to center ratio of fluorescence intensity compared with WT cells at the apical plane (Fig. 2A2). The lateral elongation of WT and apical contraction of dKD cells was confirmed by AFM imaging (Fig. 2C). Interestingly, the bordering junctions at the interface between a WT and dKD cluster are particularly pronounced on the apical side (Fig. 2C). This was reflected by increased myosin accumulation in this region (Fig. 2B), which overall highlights the mechanical discrepancy between cell types.

Differential Mechanics of dKD and WT Cells in Coculture.

To directly quantify the mechanical consequences of the described contractile, molecular, and morphological disparities in WT-GFP/dKD cocultures, we examined their mechanical phenotypes by AFM indentation–relaxation, traction force microscopy, and laser ablation (Fig. 3).

Fig. 3.

Differential mechanical properties of dKD and WT cells in coculture. (A) AFM map showing the slope of the force curve during contact, which locally reflects the apparent mechanical stiffness. (Scale bar, 10 µm.) (B) Site-specific viscoelastic properties of the central cell cortex in proximity to the WT-GFP/dKD interface. The cortex indentation geometry considered in the so-called Evans model includes the contact angle $\varphi$ and base radius R₁ of the spherical cell cap, the indentation depth δ, and the contact radius r₁. Importantly, dKD cells had a pronounced cap with larger $\varphi$ and smaller R₁ than WT cells (vide supra), yielding a 5.7-fold surface area difference. Upon fitting, the fluidity β was plotted against the decade logarithm of the scaling factor representing the area compressibility modulus K_A⁰, and histograms for the latter and the prestress T₀ are shown. Small transparent data points represent individual indentations. Large symbols and error bars are binned means and SDs. Lines indicate linear fits (in log space) of the binned means. (C) Laser ablation examples of individual cell junctions. In WT cells, ZO1 in the tight junctions was stained, and in dKD cells, myosin was stained. (D) Tensile junction properties were obtained by tracking the distance (magenta lines) between two opposing junction vertices (magenta circles) upon recoil. Temporal means and SDs are shown. (E) The initial recoil velocity was calculated between the last point before (0.00 s) and the first one after ablation (0.18 s). The boxes show the median and the upper and lower quartiles. Whiskers indicate the 5th and 95th percentiles, while data points represent a single cut of one junction. Scale bar: 10 µm. All measurements were repeated in at least three independent experiments and performed on multiple WT-GFP/dKD clusters. Data can be found in Datasets S3–S5.

First, we acquired AFM indentation maps (Fig. 3A) and examined the apparent local stiffness, which is reflected in the slope of the force–distance curve. Here, we observed a similar picture as in pure dKD monolayers (24); dKD cells were softer at the central cortex and extremely stiff at the perijunctional actomyosin ring (vide supra). In contrast, neighboring WT cells showed only slightly pronounced cell boundaries but an increased stiffness at the center in comparison with dKD neighbors.

To further characterize this stiffness difference at the center of the two cell types, we performed site-specific indentation experiments followed by force relaxation and applied a recently introduced viscoelastic fitting model (24, 26, 27). In brief, this model fits the stress relaxation of the composite viscoelastic shell upon indentation according to a power law of the area compressibility modulus assuming constant volume during the experiment (SI Appendix). Importantly, the cell geometry (area and angle of the apical cap), which differs tremendously between both cell lines, can be adjusted in this model (Fig. 3B). We limited the indentation depth to about 1 µm to minimize the impact of the nucleus on the cellular response to deformation. Three parameters are obtained: the prestress T₀ corresponding to the actomyosin cortex tension, the apparent area compressibility modulus (scaling factor) K_A⁰, which besides the shell’s stiffness also mirrors the excess cell surface area, and the fluidity β representing the viscous behavior (energy dissipation) of the cortex. β = 1 corresponds to a Newtonian fluid, whereas β = 0 refers to an elastic solid.

Because K_A⁰ was previously found not to be independent of but to scale with the fluidity β (27), we plotted β against K_A⁰ (Fig. 3B). Interestingly, we found that the fluidity was not significantly different between WT and dKD cells (P = 0.53, with the same median of 0.6). However, we found a shift to larger K_A⁰ values for WT cells compared with their dKD neighbors in coculture. This increase can be attributed to the removal of excess surface area A_ex compared with that of the geometrical surface area A₀ of the cells via ${K_{\mathrm{A}}}^0={{\tilde{K}}_{\mathrm{A}}}^0 \frac{A_0+A_{\mathrm{ex}}}{A_0}$ (28).

The picture which therefore emerges suggests that surface area is sacrificed to mitigate the external stress from adjacent dKD cells. This occurs at the expense of cell stiffening while preserving fluidity. On one hand, we observed an unchanged fluidity and only a relatively small difference in prestress within the range of the SD [0.49 ± 0.22 mN m⁻¹ for WT cells compared with 0.31 ± 0.14 mN m⁻¹ for dKD (median ± SD)]. On the other hand, WT cells exhibited a substantially larger scaling factor K_A⁰, with an increase of more than one order of magnitude [0.061 ± 0.084 mN m⁻¹ for WT compared with 0.004 ± 0.016 N m⁻¹ for neighboring dKD cells (median ± SD)]. From K_A⁰, we were able to estimate that about six-fold as much excess surface area was stored in dKD as in WT cells. This fits the theoretical 5.7-fold surface area difference between the different cap geometries, indicating that the apical surface material is conserved upon stretching, i.e., dKD cells contract laterally and store the membrane/cortex material apically, i.e., relaxing membrane tension, while WT cells sacrifice apical excess area to cope with the external areal strain. Although a tremendous amount of excess surface material is sacrificed by WT cells, the mechanics of the cortex is largely unaffected, with small differences in prestress. In agreement, we also did not observe an obvious change in the actin signal at the central cortex in Fig. 2A (vide supra). In consequence, the observed lateral contraction of the dKD cells did not originate simply from cortex mechanics but most likely from the perijunctional actomyosin ring (Fig. 2A).

To confirm this assumption and to characterize how the differential contractility translates into interfacial tension in the layer, we specifically examined the junctional tension using laser ablation, severing the cell junctions, and measuring the recoil velocity (Fig. 3 C–E). We cut cell junctions between WT/WT, dKD/dKD, and WT-GFP/dKD cells, respectively. We also compared the mechanical equilibrium in cocultures with WT and dKD monocultures. For this purpose, we analyzed the recoil dynamics of the opposite apex nodes of the ablated compound over time (Fig. 3D) and recorded the initial recoil velocity which depends on the tensile force and the compliance of the cell (Fig. 3E).

We found a significant, four- to six-fold increase in recoil velocity for all junctions bordering a contractile dKD cell (10 to 12 µm s⁻¹ compared with 2 to 3 µm s⁻¹ without any direct contact with a dKD cell). dKD/dKD junctions in coculture were comparable with dKD monocultures (P = 0.83). Interestingly, the WT-GFP/dKD interface had slightly smaller recoil velocities than dKD/dKD junctions (P = 0.29), while WT/WT junctions displayed slightly but significantly higher velocities than WT monocultures. This highlights the establishment of a mechanical equilibrium in cocultures based on a tug-of-war between highly contractile dKD cells and compliant WT neighbors; in cocultures, tension from dKD cells is accommodated by WT cells, while in dKD monocultures, all cells exhibit increased tension. Forces are transmitted over long distances within WT clusters to distribute the load, and the whole cell monolayer is under extensile tension as all cut junctions display the same recoil direction (retraction toward the junction knots). dKD cells pull at their neighbors, thereby generating elevated tension at the junctions. Notably, the whole cell monolayer is under moderate extensile tension, like a fluid wetting a surface. Overall, the data show that the increased contractility of dKD cells translates into increased junctional tension of all direct neighbors in the layer, while WT/WT junctions more far away from dKD cells display lower initial recoil velocities.

A further indication that the observed segregation is based on interfacial tension was another set of experiments in which we varied the mixing ratio between dKD and WT cells before seeding (SI Appendix, Fig. S2). We observed that the pattern of elongated dKD cell stripes which surrounded the predominately roundish WT clusters persisted independent of the mixing ratio. This is indicative of interfacial energy minimization in accordance with the tension-based sorting hypothesis and is in contrast to demixing driven by active forces as reported recently (13).

Traction force microscopy was carried out to measure the impact of the cytoskeletal remodeling in response to ZO1/2 depletion on the cell–substrate interaction (SI Appendix, Figs. S7 and S8). We found that the traction forces (per unit area) exerted by confluent WT cells were more than twice as high [84.0 ± 19.5 Pa (mean ± SD), n = 7 monolayers] as those observed for confluent dKD cells [38.5 ± 5.2 Pa (mean ± SD), n = 8 monolayers]. In the coculture, this trend is preserved (SI Appendix, Fig. S8). This suggests that the remodeling of the actin cytoskeleton also involves the basal side, creating a strong imbalance between the apical and basal sides and goes hand in hand with the reduced cell–cell adhesion measured between pairs of dKD cells (vide infra). Although single-cell force spectroscopy measurements of dKD and WT MDCK-II cells in contact with glassy substrates showed no significant difference (SI Appendix, Fig. S9), it should be mentioned that, in principle, altered cell–cell adhesion is often accompanied by altered cell–substrate adhesion. Therefore, differential cell–substrate adhesion could contribute to the segregation process either indirectly through a change in cell–cell adhesion or through changes in friction with the substrate and thus cell motility (23, 29).

Differential Cell–Cell Adhesion of WT and dKD Cells.

While the increased contractility of dKD cells is well documented and could induce segregation via energy minimization, changes in intercellular adhesion might also be expected due to the loss of the adhesion-mediating junctional ZO proteins.

To quantify cell–cell adhesion, we performed AFM experiments with one cell attached to the AFM cantilever serving as the probe and the other one adhered to the petri dish. The two cells were brought into conformal contact and separated after a short dwell time in contact (Fig. 4A). The separation forces between the two cells measured here are not obtained under equilibrium conditions and are therefore referred to as the dynamic adhesion strength. We found decreased adhesion forces for all dKD cells (two dKD cells as well as a dKD adhering to a WT-GFP cell) as shown in Fig. 4B. As a control, we also compared WT cells and GFP-tagged WT cells. While the WT-GFP cells displayed slightly lower adhesion forces than pure WT cells, they are still consistently more adhesive than dKD cells (P < 0.001 compared with WT-GFP/dKD and P < 0.01 with dKD/dKD). Interestingly, the adhesion between two cells was always dominated by the respective weaker binding partner, i.e., the dKD cells, indicative of largely immobile receptor–ligand pairs. Accordingly, in WT-GFP/dKD cocultures, differential adhesion and contractility together determined the differential interfacial tension during cell segregation.

Fig. 4.

Differential intercellular adhesion of WT and dKD cells. (A) AFM-based adhesion measurements: 1. Before or after an experiment, one cell is connected to the cantilever and one adheres to the culture dish substrate, without contact with each other. 2. Adhesive contact between cells is established at 2 nN for 5 s. 3. As the cantilever is retracted, the cells are pulled apart and bonds rupture. Schematic retraction curves depict small (dark red) and large (light red) adhesion forces. (B) These adhesion forces are compared between different important cell combinations. Violins represent kernel density estimation with horizontal, dashed lines showing the quartiles and median. Violins are scaled to have the same area. Single data points represent individual adhesion peak forces. Three consecutive indentation–retraction cycles were performed for each cell pair. For each combination, at least four individual cell-cantilever probes and at least eight cells on the substrate were measured, with experiments repeated on at least 4 d. Dataset S6.

The loss of ZO proteins perturbs tight junction–associated signaling by redistributing ROCK to the adherens junction (25), where it leads to apical constriction in dKD cells that coexist with outstretched WT cells to maintain force balance and avoid bending of the 2D monolayer away from the surface. Myosin-2 is also redistributed and predominately found in the apical region (Fig. 2A1) where it fosters constriction, while concomitantly adhesion to the substrate is diminished.

Timescale Dependency: Contractility Drives Early, Adhesion Final Sorting.

While we established that there is differential contractility and differential adhesion in WT-GFP/dKD cocultures, it remains unclear, which one dominates over the other. Therefore, we performed the demixing experiments shown in Fig. 1 again, however, this time in the presence of the Rho kinase (ROCK) inhibitor Y27632, to reduce cell contractility, switching off one of the contributions to demixing (Fig. 5 and SI Appendix, Figs. S10 and S11). Y27632 mainly affects the actomyosin contractility of cells, while the difference in cell–cell adhesion remains the same (SI Appendix, Fig. S12). This should restore the apical conformity lost due to the increased ROCK activity in response to ZO1/2 depletion. We also impaired all cell–cell contacts by reducing the calcium content of the medium to induce a similar effect (SI Appendix, Fig. S13). Here, the effective differential contractility was reduced concomitantly (SI Appendix, Fig. S13B, Bottom). Importantly, upon completely abolishing both contractility and adhesion (at 0 mM Ca²⁺), no sorting was possible neither in WT-GFP/WT nor in WT-GFP/dKD cultures, leading to SI levels even lower than that of WT-GFP/WT controls in normal medium (SI Appendix, Fig. S13B, Top).

Fig. 5.

Contractility drives early, adhesion final sorting. Demixing behavior of highly contractile dKD and wild-type cell cocultures at an initial mixing ratio of 50:50 treated with 50 µM Y27632. Experiments and figure panels are set up analogously to Fig. 1, and, for comparison, untreated WT-GFP/dKD from Fig. 1 were included. (A) Example overlay of phase contrast (gray scale) and fluorescence (green: WT-GFP cells) channels with corresponding segmentations (green: WT-GFP, magenta: dKD cells in WT-GFP/dKD mix or WT in WT-GFP/WT control). Samples were imaged immediately after seeding and mounting on the microscope (0 h). Images from the start of the experiment (t = 0 h) can be found in SI Appendix, Fig. S10A. Scale bars: 200 µm and 50 µm (zoom-in). (B) Demixing, cell amount, and area occupancy quantification. The vertical dashed line at 5 h indicates two distinct demixing timescales thought to be determined by contractility and adhesion. The shade in the first 3 h indicates subconfluence. Yellow symbols represent data from untreated WT-GFP/dKD mixtures, purple symbols refer to the ROCK inhibitor Y27632-treated WT-GFP/WT control, and green symbols refer to WT-GFP/dKD cell mixtures exposed to Y27632. (B1) The segregation index SI, defined as the average ratio of homotypic and all cell neighbors, quantifies the demixing degree. The SI is shown averaged over both cell types (Top) and separately for each cell type (Bottom). (B2) Top: Relative cell amount calculated as the ratio of the number of WT-GFP cells and the total cell amount. Bottom: Total cell amounts of each cell type. (B3) Top: WT-GFP fraction of the overall cell area calculated as the ratio of the WT-GFP area and the total cell area, indicating contractility discrepancies between the cell types. Bottom: Mean cell area of each cell type. Corresponding zoom-ins of the first 5 h are shown in SI Appendix, Fig. S10B, a comparison of untreated and Y27632-treated WT-GFP/WT cultures as well as later experiment times (24 h to 35 h) can be found in SI Appendix, Fig. S11, and distributions of the individual cell areas are depicted in SI Appendix, Fig. S2. Velocity and persistence analyses can be found in SI Appendix, Fig. S14. Mean values and SDs are shown. Six separate regions from three culture dishes (two per dish), acquired on separate days, were measured and are shown per coculture mix. Dataset S1.

Upon first visual inspection after ROCK inhibition (Fig. 5A), at early time stages, no difference was discernible between the WT-GFP/dKD mixture and the WT-GFP/WT control. Only at later times, stronger demixing was observed in the WT-GFP/dKD coculture as mirrored in the segregation index (Fig. 5B1). Here, we plotted the untreated WT-GFP/dKD mixture from Fig. 1 to serve as a reference together with contractility-inhibited WT-GFP/dKD and WT-GFP/WT cocultures. While the WT-GFP/WT control did not change its segregation behavior upon Y27632 administration, the very fast, early segregation of WT-GFP/dKD cocultures (<5 h) was substantially diminished. Instead of this fast initial behavior, segregation of the WT-GFP/dKD mixture was slowed down. Nevertheless, after about 15 h, the contractility-inhibited WT-GFP/dKD mixture reached approximately the same SI of approximately 0.7 as the untreated counterpart. Accordingly, the up-regulated contractility of dKD cells was critical for early segregation, while the adhesion differential was still able to induce cellular demixing upon longer timescales.

As a control parameter, we also inspected the ratios of cell area and amount (Fig. 5 B2 and B3) as in Fig. 1. The WT fraction of the cell amount (Fig. 5B2) again served to provide context for the SI values and relative area coverage. While the untreated WT-GFP/dKD cell amount ratio was slightly shifted toward more WT cells, both drug-treated cocultures remained at a 0.5 ratio (Fig. 5B2). Interestingly, the proliferation in the WT-GFP/WT control was increased by Y27632 to the same level present in treated and untreated WT-GFP/dKD (except for the dKD extrusion after 15 h) as shown in Fig. 5B2 and SI Appendix, Fig. S14A, while the SI remained much lower. To further rule out that local clustering due to proliferation dominates the segregation, we investigated the relationship between the SI and the cell amount (SI Appendix, Fig. S14A) upon exposure to Y27632. The SI generally increased with increasing cell amounts but with a lower slope at higher cell amounts. However, for both treated cultures and the WT-GFP/dKD mixture the proliferation rate was approximately constant over time, while the scaling of the SI with the total cell amount was much different. At the same cell amount, the SI remained lower in the treated WT-GFP/WT control than in the untreated WT-GFP/dKD mixture. In addition, the difference in proliferation between the treated and untreated WT-GFP/WT samples did not translate into an increase in segregation. Note that the cell amount is essentially equivalent to cell density in our experiments because the size of the field of view was always the same.

To assess cell contractility, the area ratio once again served as a broad-scale readout (Fig. 5B3). Here, we did observe the expected drop upon contractility inhibition for the WT-GFP/dKD mixture, while the WT-GFP/WT control was unaffected. Importantly, this drop in contractility remained over the whole duration of the experiments, confirming that the effect of the drug did not wear off over time. Moreover, since switching off proliferation by administration of mitomycin C (vide supra and SI Appendix, Fig. S4) neither changed cell sorting dynamics, we can safely conclude that first contractility and later adhesion dominate the segregation process.

To rule out that the drug acts on cell motility (e.g., due to effects of the focal adhesions on the substrate) influencing demixing, we quantified the velocity and persistence via cell tracking (SI Appendix, Fig. S14B). We investigated this, particularly for the first 5 h, where the impact of the drug on segregation is the strongest. If higher motility was a driving factor for random mixing, we would expect an increase in the motility parameters, particularly of the WT-GFP/dKD mixture upon drug treatment. However, this was not the case, but, to the contrary, the motility parameters even decreased slightly or remained the same (SI Appendix, Fig. S14B). The WT-GFP/WT control showed a slight drop in both parameters, while its (de)mixing behavior was largely unaffected. Accordingly, the drug provoked the delay in WT-GFP/dKD sorting not by affecting motility but indeed via inhibiting cellular contractility.

Additionally, we also confirmed that inhibition of ROCK does not significantly alter cell–cell adhesion. For this purpose, we cultured WT-GFP cells on an AFM cantilever facing the apical part to the opposing monolayer on the petri dish (µ-Dish, low; ibidi) and measured separation forces in the presence and absence of Y27632 (SI Appendix, Fig. S15). We found that adhesion between WT-GFP and dKD cells is only slightly reduced by the ROCK inhibitor [330 ± 130 pN down to 300 ± 120 pN with Y27632 (mean ± SD)] at 5-s contact time between the cells. This is, however, also true for the separation force measured between two WT-GFP cells [435 ± 350 pN to 370 ± 200 pN with Y27632 (mean ± SD)], which means that the overall impact of Y27632 on cell–cell adhesion is small and the gradient of adhesion strength between the two cell types remains unchanged in the presence of ROCK inhibitor.

A similar effect was obtained with reduced calcium concentration in the culture medium, slowing down the contractility-based cell sorting (SI Appendix, Fig. S13, 0.07 mM Ca²⁺) without inhibiting demixing, i.e., the same final SI is reached after >20 h. However, further withdrawal of calcium from the culture medium completely abolished both contractility and adhesion-based cell sorting. As expected, the two cell types do not display any segregation anymore.

Discussion

Our goal was to identify and scrutinize the driving forces for the demixing of cocultures consisting of WT and dKD MDCKII cells displaying both different cell–cell adhesion due to the knockdown of ZO1/2 and differential contractility due to actomyosin upregulation in the apical domain. Tight junction–associated signaling is pivotal for maintaining epithelial sheet morphology, integrity, and function (30–34). In planar epithelial cell sheets, the apical, contractile forces are typically balanced to avoid deformation and bending of the entire sheet. The Par polarity proteins Par-3 and Willin, a FERM domain protein, are involved in this regulation by suppressing the junctional localization of ROCK through its phosphorylation by the protein kinase aPKC (25, 30). This ensures uniformly shaped apical domains and balanced contractility. Loss of ZO1/2, however, perturbs Par-3 localization and therefore leads to apical constrictions and atypical apical morphology (25). ZO proteins are required for epithelial polarization, and it was shown previously that depletion results in a tug-of-war between adjacent cells that could, however, be largely resolved by inhibition of ROCK (24, 25). It was therefore of great interest to examine the consequences of ZO depletion for cell sorting and layer morphology.

We found that the main driving forces for creating clusters of dKD cells coexisting with WT clusters are timescale separated. On short timescales (within the first 5 h), differential contractility prevails, while on longer timescales (>5 h), cell sorting is driven predominately by differential adhesion. Clusters of dKD cells are shaped by elongated chains of cells that shorten in later stages. This dynamic behavior is expected from the general rule that cell sorting occurs in three distinct steps that include the formation of elongated chains that shorten and smooth and finally, when the tension between the different cell types is large enough compared to the homotypic tensions, anneal into large round clusters and minimize the line tension (8). Our data suggest that if differential contractility is abolished, differential adhesion alone is sufficient for cell sorting but considerably slower and less efficient. Since altered cell–cell adhesion is often accompanied by a modified cell–substrate adhesion, it is, in principle, also conceivable that differential friction with the substrate and thus motility of the cells lead to segregation. Here, however, no obvious effects of altered cell–substrate adhesion on the single-cell level were found (SI Appendix, Fig. S9), which does not exclude such influence on the level of confluent cell monolayers.

The envisioned mechanism comprising adhesion- and contractility-based cell segregation is summarized in Fig. 6. While in randomly mixed WT cultures adhesion between all cells is the same and they display similar contractility, in WT-GFP/dKD cocultures, the adhesion and contractility between the cell types are considerably different, inducing segregation into clusters. In particular, the difference in apical contractility between dKD and WT cells results in a substantial tension difference that initially favors rapid but partial sorting. The apical forces are balanced by neighboring cells, so that large, outstretched WT cells coexist with small, contractile dKD cells. The apical forces exerted by the dKD cells are balanced by WT neighbors via their higher traction forces on the substrate. This force balance prevents bending of the cell layer into the third dimension. Cell–substrate adhesion forces are stronger for WT cells than for dKD cells, as also suggested by phosphorylated myosin-2 staining (Fig. 2). In response to the contraction of adjacent dKD neighbors, WT cells are stretched and therefore sacrifice a large amount of excess surface area to prevent lysis. If contractility is balanced again by inhibiting ROCK, segregation based on differential contractility is strongly delayed. The same was found for moderate calcium depletion slowing down initial cell sorting by decreasing the effective differential contractility (SI Appendix, Fig. S13B). It was shown that Y27632 restores normal apical area distribution in ZO1/2-depleted Eph4 cells (25). The authors found an excess of contractility due to aberrant ROCK activation at the adherens junctions. However, even after switching off contractility, the remaining tension difference due to stronger WT-GFP/WT adhesion compared with dKD/dKD adhesion still induces the same amount of segregation as in untreated layers over a longer time, highlighting a redundant but time-dependent role of contractility and adhesion. Considering that adhesion complexes mature progressively over time (35), whereas contractility is a property of individual cells, it is conceivable that differential contractility promotes sorting immediately, while adhesion acts on longer timescales. Depletion of calcium to an extent that prevents cells from ablation led to full suppression of demixing as it switches off both cell–cell adhesion and contractility. Using only moderate calcium depletion, we found that only the contractility-based cell sorting was affected, i.e., slowing down initial cell sorting [SI Appendix, Fig. S13, presumably via calcium’s role in promoting cell contraction (36, 37)]. Proliferation plays only a role in later stages when jamming occurs, and dKD cells are extruded from the cell layer due to the strong apical forces (SI Appendix, Fig. S4).

Fig. 6.

Proposed model of the interplay between adhesion and contractility in WT-GFP/dKD cell layers. (A) In WT-GFP/WT control layers, adhesion is the same between all cells, and they are equally contractile; hence, random mixing takes place. (B) Adding dKD cells induces differences in both adhesion and contractility between the cell types. dKD cells lose some adhesive contact and contract excessively, yielding tremendous apical excess surface area. As a consequence, compliant neighboring WT cells are stretched out and respond by surface area dilatation. (C) To test the relative impact of adhesion and contractility, the latter was balanced again by drug addition, revealing a temporal dependency: Balanced contractility restores random mixing at early stages, but differential adhesion is still able to promote cell sorting into clusters on long timescales.

It is well established that tight junction–depleted cells show increased contractility (30, 33, 34, 38–40). However, so far, the implications of increased contractility of dKD cells for the behavior of the monolayer were only studied with emphasis on impaired migration dynamics and signaling (24, 25). Here, we showed that epithelial cells, which are stretched by their contractile neighbors, respond primarily by apical area dilatation instead of adaption of cortex mechanics. By comparing the apparent area compressibility modules of dKD and WT cells, we found a sixfold larger excess area for highly contractile dKD cells compared with dilated WT neighbors, equivalent to the change in geometric surface area, indicating the conservation of excess material instead of its recycling. This is consistently observed for cocultures as well as dKD monocultures, which were described previously, where two populations emerged in a tug-of-war: a contractile population that stretches out the neighboring cell population to conserve the amount of occupied surface area (24). However, in that work, it remained unclear if recruiting excess surface area indeed dominates the stretch response. During development, a generation of two mechanical cell populations among the same cell type was identified as an emergent property upon collective cell interactions (41). A recent study implicated asymmetric ROCK signaling in inducing these two populations to interact in confluent dKD monocultures (25). This tug-of-war might intuitively favor segregation into clusters to decrease the number of WT cells that are subject to dilatation by adjacent dKD neighbors.

Our study confirms that segregation can be explained by the different interfacial tension models (8). However, we found a temporal separation of the dominant source for demixing, where differences in contractility determine the cell sorting process on short timescales immediately after seeding, while differential adhesion contributes less but also permanently to the sorting process on longer timescales. On the one hand, adhesion-based sorting was shown before to emerge in cell cultures, e.g., upon different expression levels of cadherins (6, 42–44). Similarly, sorting based on cadherin levels was demonstrated in follicle and retina cells of Drosophila oocytes (45, 46). Signaling-controlled cadherin turnover has also been implicated in cell segregation (47, 48). Note, in our study, adhesion differences were induced by tight junction disruption, which was also shown by previous work to decrease adhesion, in agreement with our data (32, 49). Purely adhesion-based sorting was recently confirmed via simulations and experiments in direct relation to constant contractility (18). On the other hand, differential contractility was found to aid sorting in an embryo and possibly dominate over adhesion (50, 51). In cocultures of zebrafish germ layer cells, differential contractility alone was found to be sufficient for sorting (50). However, sorting also took place on two timescales, a fast, early (<0.5 h) and a slower, later timescale, while the authors did not investigate the temporal evolution further. An interplay between adhesion and contractility was found in cancer cell line aggregates (52) and confirmed in recent studies using vertex/Voronoi models (9, 15, 16). In particular, interfacial tension was shown to be determined by the ratio of cell adhesion and contractility, governing the tissue-scale tension (9). Accordingly, the increased contractility paired with the lower adhesion of dKD cells translates well into the high tension values measured by laser ablation. However, a demixing mechanism of locally increased contractility at the boundary between two cell types, as reported in Drosophila wing discs, can be ruled out in our work (10–12). While we measured tremendous differences in line tension between the different cell types, the WT-GFP/dKD interface did not exhibit the highest tension but rather values equal to or slightly below that of dKD/dKD junctions.

Another mechanism in contrast to our data was proposed by a recent study examining the demixing of E-cadherin-depleted and wild-type MDCK cells. The authors identified active cell forces as the governing factor of sorting (13). While the demixing behavior in that study appears very similar to our data, the initial segregation was slower. Furthermore, they observed a pattern reversal at uneven mixing ratios which were absent in our cocultures. This stability of the sorting pattern is indicative of interfacial energy minimization by minimizing the contact region between heterotypic cell types upon sorting based on adhesion and/or contractility (13). E-cadherin-depleted and wild-type keratinocytes were recently shown to sort mainly based on shape disparities, and this was thoroughly explained in vertex simulations as well as observed earlier in zebrafish embryos (14, 53). Although the shape differences in our cell lines seem to be small and result from the tug-of-war between the cell types, we cannot entirely exclude their contribution (24).

We also addressed the possible crosstalk of contractility and adhesion (54). For reference, actomyosin contractility has been shown to enhance adherens junction–based adhesion (55–57). Measurement of cell–cell adhesion forces in the presence of Y27632 showed that albeit adhesion was slightly reduced due to loss of contractility, the difference between the forces measured for WT-GFP/WT and WT-GFP/dKD cell pairs remained unchanged.

However, there would still be the disruption of the tight junctions in dKD cells. In addition, if the adhesion difference had been just as abolished as the differential contractility, we would not have observed the prevailing demixing at longer timescales. Actomyosin contractility has also been shown to modulate focal adhesions and thereby cell motility (58–60). However, we observed no influence of motility on sorting, possibly due to the high cell density in our experiments (with confluence reached after only a few hours).

In addition, cell sorting could be influenced by proliferation creating local clusters. However, upon Y27632 treatment, the WT-GFP/WT control increased its proliferation to the same level present in treated and untreated WT-GFP/dKD mixtures, yet, its SI remained much lower. At the same cell amount, the SI remained lower in the treated WT-GFP/WT control than in the untreated WT-GFP/dKD mixture. In addition, the large difference in proliferation between the treated and untreated WT-GFP/WT samples did not translate into an increase in segregation. The cell amount ratio of the cell types was also consistent among the WT-GFP/dKD mixture and its respective WT-GFP/WT counterpart, both treated and untreated, whereas their SI differed. Ultimately, using mitomycin C to suppress proliferation, we found indeed that the early stage of cell sorting dominated by differences in contractility was unaltered.

Another caveat to note is that due to technical limitations, our cell adhesion measurements are on much shorter timescales than the observed mixing dynamics and the relevant cell–cell interactions in general (61). Nevertheless, in agreement with other work, it is reasonable to assume that the loss of tight junction integrity reduces intercellular adhesion on all relevant timescales (32, 49). While ZO proteins do not directly bind to the other cell, their loss destabilizes the contact zone and influences the transmembrane proteins (32, 62).

Ultimately, our data suggest that adhesion alone is sufficient but less efficient in driving cell sorting without differential contractility. This could yet be another example of how biology employs functional redundancy to ensure fundamental processes such as the sorting of different cell types.

Materials and Methods

Full materials and methods are available in the SI Appendix section. MDCKII cells were used for all experiments, and knockdowns or fluorescent protein labeling was performed using CRISPR/Cas as described by Beutel et al. (63) or Skamrahl et al. (24). For the figures, images were brightness-adjusted in Fiji to improve visibility (64). Segmentation was performed using Cellpose 1.0 (65). Cell positions and areas were obtained using OpenCV as described by Skamrahl et al. (24, 66, 67). Neighbor analysis was performed with self-written Python scripts. Trackpy was used for tracking (68, 69). Confocal microscopy (FluoView1200; Olympus, Tokyo, Japan) was performed after fixation and antibody- and phalloidin-based labeling. Analysis was performed using segmentation via Cellpose. AFM was carried out on a NanoWizard 4XP (Bruker Nano, JPK, Berlin, Germany) and calibrated by the thermal noise method (70). An 800-nm femtosecond pulsed laser was used for laser ablation, and the opposing vertex knots of a junction were tracked. Cell–cell adhesion measurements were performed on a CellHesion 200 AFM (JPK) at 0.5 µm s⁻¹ (single cells)/1.0 µm s⁻¹ (cells grown on cantilever). Proliferation was inhibited by incubation for 1 h with 10 µg/mL mitomycin C which was removed by centrifugation before seeding (71). Calcium withdrawal was performed using minimum essential medium eagle; Spinner modification (SMEM; Sigma-Aldrich Life Science, UK) and media were supplemented with chelex serum (FCS Gold neutralized chelex treated, PPA, Pasching, Germany) to better control the calcium concentration. Cluster analysis was performed on the WT-GFP fluorescence images using scikit-image in Python.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (CSV)

Dataset S02 (CSV)

Dataset S03 (CSV)

Dataset S04 (CSV)

Dataset S05 (CSV)

Dataset S06 (CSV)

Dataset S07 (CSV)

Dataset S08 (CSV)

Dataset S09 (CSV)

Dataset S10 (CSV)

Dataset S11 (CSV)

Dataset S12 (CSV)

Data, Materials, and Software Availability

 All study data are included in the article and/or SI Appendix.