Cooperative regulation of adherens junction expansion through epidermal growth factor receptor activation

In the past decade, the nature and organization of the molecular components constituting adherens junctions have been under intense investigation. Much less is understood about how cells control the dynamics of adherens junction formation and remodeling. Cadherins, and in particular E-cadherin, are capable of orchestrating the assembly of proteinaceous clusters involving actin-binding protein (e.g. catenins), scaffolding proteins (e.g. α-actinin, cortactin), and tight junction proteins (e.g. ZO proteins) (Helwani et al., 2004; Huveneers and de Rooij, 2013). E-cadherin adhesion is also recognized as a signaling hub where receptor tyrosine kinases (RTKs), such as EGFR, and GTPases (including Rac1 and Cdc42) are activated (Fedor-Chaiken et al., 2003; Kovacs et al., 2002; Nakagawa et al., 2001; Noren et al., 2001; Pece and Gutkind, 2000). We have a fair understanding of how junctional components organize the mesoscale adhesive plaque and respond to mechanical stimulations. The mechanical forces, and in particular tension, have emerged as an essential parameter to stabilize junctions. It results that myosin II and its regulatory pathways are critically involved in controlling junction size and stability (Priya et al., 2015; Ratheesh et al., 2012). In epithelia where adherens junctions have been analyzed, junction morphologies are governed by the local equilibrium of mechanical tension at the junction vertices, as demonstrated by laser ablation (Rauzi et al., 2008).

Our understanding of junction remodeling arises principally from the studies of the paradigmatic Drosophila embryo epithelia, where again actomyosin contractility plays a critical role to shrink or expand the junction (Bertet et al., 2004; Rauzi et al., 2008). Pulsatile apical contractions (Yu and Fernandez-Gonzalez, 2016) and localized E-cadherin cluster recruitment (Rauzi et al., 2010) have, for example, been proposed to drive cell–cell contact exchange, but they are often characterized with respect to the final junction size and tissue organization. This tension-centric view focuses on the driving force of the junction expansion. However, the junction expansion time is likely governed by the intrinsic dynamics of the actin cytoskeleton reorganization, which is potentially equivalent to the modulation of the junction viscosity (Clement et al., 2017).

This study uses suspended cells to show that the activation of EGFR at a junction by E-cadherin engagement slows the actin dynamics of the entire free cortex of the cell. This cortical ‘priming’ leads to a cooperative acceleration of the expansion dynamics of additional junctions that the cell is subsequently engaged in. We used minimal systems of cell doublets, triplets and quadruplets to precisely image the dynamic of expansion of the junction in stereotypical conditions. We focus on the additive effect of the distal activation of the cortex on junction expansion dynamics and not on the role of EGFR-mediated local regulation of actin rearrangements as the junction is expanding (Erasmus et al., 2015).

In order to monitor the kinetics of the formation of adherens junctions, we used suspended cell doublets. This approach offers the advantages of (1) a precise timing of the onset of junction formation, (2) the creation of junctions with stereotypical quantifiable morphologies and (3) the absence of crosstalk from extracellular signaling by matrix. In a previous study (Engl et al., 2014), we established that this approach leads to a bona fide and stereotypical distribution of junctional proteins (E-cadherin, actin, vinculin and catenins) along with the contact between two cells. Here, we adapted the approach to analyze the kinetics of junction formation of S180-E-cad–GFP cells, a murine sarcoma cell line expressing no cadherins endogenously but stably transfected with GFP-tagged E-cadherin (E-cad–GFP; Chu et al., 2004). The results obtained with S180 cells were qualitatively reproduced using Madin–Darby canine kidney (MDCK) cells (see Fig. S1). Nonetheless, MDCK cells are more difficult to handle as suspended aggregates due to rapid anoikis when manipulated in suspension.

In control conditions, the junction expansion time (τ) of doublet junctions (J₁) proved uncorrelated with the final junction radius(R_∞) (Fig. 1C). Using calibrated confocal imaging, we investigated the influence of E-cadherin density on the junction formation. We found that junctions formed provided that at least one cell of the doublet had an E-cadherin density higher than a threshold value (Fig. S1C,D). The junction expansion time (τ), however, did not correlate with the level of E-cadherin expression (Fig. S1E). The characteristic expansion time of a junction in the control case was 19.6±12.4 min with a final radius of 5.8±1.1 μm (mean±s.d.). These observations suggest that the dynamic properties of a junction might be decoupled from its static properties.

We then asked whether a second junction (J₂) forming with either cell of a doublet would expand at the same speed. Contacting doublets with singlets (see Materials and Methods), we hence formed linear triplets (Fig. 1D; Fig. S1F). We followed the junction expansion dynamics of triplets that were seeded in the same microwell. In linear triplets, the second adherens junction (J₂) formed 37% faster (11.2±7.6 min, mean±s.d.) than the first adherens junction (J₁) (Fig. 1D,E; Fig. S1F–H). It is important to note that we considered exclusively the dynamics of expansion once the initial contact had formed, hence excluding the delay needed for cells to meet and to initiate the contacts. The observed decrease of the expansion time (τ) for J₂ did not depend on the time elapsed between the initiation of J₁ and J₂ (within 2 h; Fig. S1I,J). MDCK doublets and triplets behaved similarly to those of S180 cells, indicating that our observations are not specific to a single cell type (Fig. S1K).

We then analyzed the expansion dynamics of a third junction to form cell quadruplets. When the third junction (J_3l) was formed with either end cell of the triplet, resulting in a linear quadruplet, J_3l (12.4±7.8 min, mean±s.d.) expanded at the same speed as J₂ (Fig. 1E). By contrast, when the third junction (J_3t) involved the central cell of the triplet, resulting in a tetrahedral morphology, J_3t (8.7±4.9 min, mean±s.d.) expanded 22% (on average) faster than J₂ and consequently 56% (on average) faster than J₁ (Fig. 1E). In this tetrahedral morphology, the third junction involves the central cell that is already engaged in two junctions, whereas in linear quadruplets the end cells are engaged a single pre-existing junction. These observations consistently indicate that the expansion time (τ) of a junction, in this context, decreases with the number of pre-existing junctions. Remarkably, the final radii of the junctions J₁, J₂, J_3l and J_3t were equivalent (Fig. 1F). Taken together, our data indicate that the dynamics of junction expansion are cooperative but are largely decoupled from the final junction morphology.

In the absence of extracellular matrix adhesion, we reasoned that the cooperative effect was triggered solely by the de novo junction formation. In particular, we hypothesized that the engagement of E-cadherin during the first junction formation triggered a modification of the actin cytoskeleton constituting the free cortex of the cell. We reasoned that the parameter driving the junction expansion was the dynamics of actin restructuring from the free cortex into the junctional actin. We considered the cortical tension (γ) and the actin turnover time (t_Cort.) as the two main potential cortex biophysical properties that the cells regulate to control the junction expansion dynamics. We hence monitored the tension of the free cortex using pipette aspiration and its actin turnover time using fluorescence recovery after photobleaching (FRAP), as detailed in the Materials and Methods (summarized in Fig. 2A). We compared junction and cortex properties between singlets and doublets. Fig. 2B and Fig. S2A show that upon new formation of a junction, there is no significant change in cortical tension between single cells or cells engaged in contacts (doublets). The conclusion holds true independently of the time elapsed since the doublet formation (between 10 min and 1 h). This suggests that the formation of the junction does not enhance the contractility (myosin activity) of the cell cortex, as further supported by quantitative immunostaining of phosphorylated myosin II regulatory light chain (Fig. S3A,B). In contrast, the average actin recovery time at the free cortex (t_Cort.) displayed an 87% increase (from 10.9±4.3 s to 20.4±7.3 s, mean±s.d.) between a population of singlets and doublets (Fig. 2C; Fig. S2B–D). This holds true at the single-cell level, as probed by repeated FRAP measurements of individual cortices before and after de novo junction formation. At the single-cell level, the actin recovery time (t_Cort.) systematically increased, hence strengthening the conclusions from the population study (Fig. S2E). This suggests that the E-cadherin junction formation can trigger a signaling pathway that rapidly slows down the dynamical properties of the free cortex cytoskeleton. We concluded that cell multiplets can have similar static phenotypes (number of junctions, size of junctions and cortical tension, γ) but distinct kinetypes [actin turnover time (t_Cort.) and junction expansion time (τ)]. This suggests that distinct pathways could be involved in independently regulating the dynamical properties and the static properties of adherens junctions.

Hence, we searched for pathways that can alter the kinetype of the cell junctional dynamics while preserving the phenotype. We were guided by previous reports suggesting that in the context of a cell monolayer, E-cadherin engagement likely leads to the phosphorylation of several kinases (McLachlan et al., 2007; Pece and Gutkind, 2000). Hence, using phospho-specific antibodies, we screened several RTKs. We found that EGFR at the junction edge was phosphorylated at Y845 in the Src-dependent domain (Fig. S3C–E). We then analyzed the cortical actin dynamics in singlets and doublets after pharmacological inhibition of EGFR (erlotinib at 500 nM; see Materials and Methods for specificity). Inhibition of EGFR activity (1) did not alter the cortical tension (γ) of singlets and doublets (Fig. 2B), (2) fully inhibited the increase of the actin recovery time (t_Cort.) upon junction formation (Fig. 2C; Fig. S2E), (3) blocked the cooperativity of junction expansion dynamics between J₁ and J₂ (also confirmed by siRNA-mediated partial knockdown; Fig. 2D and Fig. S2F) and (4) did not impact the final radius of the junctions (Fig. 2E; Fig. S2G). Identical conclusions were applied when the inhibitor was used during formation of cell quadruplets (J_3l and J_3t; Fig. S2H,I). Importantly, E-cadherin recruitment was unchanged upon knockdown of EGFR or the addition of EGF in the medium (Fig. S3F). As a whole, EGFR inhibition did not change the junction phenotypes in the cellular multiplets but abolished the difference in their kinetypes and hence the cooperative effect of junction expansion.

We then examined whether a modification of cortical actin dynamics could cause an increase or a decrease in junction expansion time (τ) independently of its molecular origin. We used a series of pharmacological inhibitors (250 nM latrunculin A, 100 μM CK-666, 20 μM SMIFH2 and 100 nM jasplakinolide) that target different mechanisms of actin regulation (including polymerization, branching and depolymerization). We performed systematic FRAP and cortical tension measurements on singlets and correlated them with the expansion time of the doublet junctions (J₁). We compared all kinetypes and phenotypes under the various treatment conditions. Fig. S4 displays all six combinations of correlative graphs relating the possible pairs of the cortical tension (γ), cortical actin recovery time (t_Cort.), final junction radius (R_∞) and junction expansion time (τ) parameters.

The actin dynamics of the free cortex (t_Cort.) proved largely decorrelated from the final junction radius (R_∞) and cortical tension (γ) (Fig. S4A,C). The junction expansion time (τ) displayed its strongest correlation (R²=0.655) with actin recovery time (t_Cort.), irrespective of the drug treatment (Fig. S4B). It also displayed a weaker correlation with the cortical tension (γ) (Fig. S4D). These results concur with what we described for doublets, triplets and quadruplets, and confirm that the slowing down of actin dynamics alone can cause a change in the expansion time of a junction, either directly (mechanosensitive effect) and/or indirectly (reduction of protrusion rate, specific signaling). These data reinforce our hypothesis that the slowdown of actin turnover at the free cortex upon EGFR activation can causally increase the junction dynamics. However, we note that the interpretation of these data is limited by the multifaceted roles of the proteins involved, given that the inhibitors often can only block a particular function of each protein.

We first tested whether the cooperative expansion dynamics were sensitive to the basal amount of EGF in the medium. Fig. 3 and Fig. S5 compare the formation of doublets and triplets in serum-containing medium (10% FBS; Fig. 3A, condition1), serum-starved medium (Fig. 3A, condition 2), serum-starved medium with 20 ng/ml EGF (Fig. S5A, condition 2) and serum-starved medium with 100 ng/ml EGF (Fig. S5A, condition 3). In all cases, the expansion of J₂ was significantly faster than that of J₁ (τ for J₁ and J₂, respectively: with serum, 17.4±7.2 min and 8.4±3.0 min; without serum, 19.6±12.4 min and 11.0±7.0 min; 20 ng/ml EGF, 27.4±18.6 min and 14.4±7.4 min; 100 ng/ml EGF, 32.4±15.3 min and 10.6±8.2 min; mean±s.d.) (Fig. 3B; Fig. S5B). Cells were left at least 6 h in the medium prior to testing (see Materials and Methods for the detailed protocol). The modulation of actin dynamics was also similar between serum-containing and serum-starved cells (t_Cort. for singlets and doublets, respectively: with serum, 10.8±4.6 s and 18.0±7.6 s; without serum, 11.6±4.6 s and 16.4±7.9 s; mean±s.d.) (Fig. 3C). These observations clearly suggest that the effect is not dependent on the basal level of activation of EGFR. The addition of an EGF-neutralizing antibody (10 µg/ml) in the serum-starved medium did not modify the cooperative effect, hence likely excluding paracrine signaling by the release of EGF during junction expansion (Fig. S5A, condition 1). Discriminating between fast paracrine signaling and direct ligand-independent activation of EGFR is always ambiguous; however, we believe that these observations advocate for an EGF ligand-independent mechanism. We favored the hypothesis of an activation triggered by the engagement of E-cadherin in de novo cell–cell contacts. In the context of cell monolayers, similar claims have previously been reported (Fedor-Chaiken et al., 2003; Pece and Gutkind, 2000; Shen and Kramer, 2004), although alternative mechanisms have not been fully ruled out.

Albeit likely EGF independent, we reasoned that the cooperativity in junction expansion could be mimicked by a step activation of EFGR following a rapid change in EGF concentration in the medium. To test whether a step activation of EGFR could increase the junction expansion dynamics, we added EGF at 20, 100 or 900 ng/ml to serum-starved singlets or doublets and measured the junction expansions just after the addition of the soluble factor (Fig. 3A, condition 4; Fig. S5A, conditions 4 and 5). The expansion dynamics of the first junction J₁ increased as compared to control (9.2±7.2 min versus 19.6±12.4 min; mean±s.d.) to reach the same level as J₂ in the presence or absence of EGF burst (10.6±5.5 min and 11.0±7.0 min, respectively) (Fig. 3B). The actin recovery time in these conditions also adopted the slow value of t_Cort. (Fig. 3C). As a whole, our data indicate that EGFR inhibition abolishes the cooperative effect on junction expansion, leading to all junctions having the slow kinetype of J₁. By contrast, step activation of EFGR by EGF also abolished the cooperative effect, resulting in all junction kinetypes resembling the fast kinetype of J₂. Our data hence strongly support that step activation of EGFR by junction formation leads to a slowdown of actin dynamics in the free cortex and, consequently, a decrease in junction expansion time (τ).

The EGFR phosphorylation site Y845, which we found activated at adherens junction, is a Src-dependent activation site. Both Src and Rac have been reported to be activated by E-cadherin adhesion during junction formation, where the Rac activity required the activation of EGFR (Betson et al., 2002; McLachlan et al., 2007). We hence studied the role of Src and Rac activity in regulating the cooperative effects. We found that treatment with pharmacological inhibitors that inhibit Src and Rac1 activity blocked both the cooperative effect and the slowdown of cortical actin dynamics after junction formation (Fig. 4). Upon inhibition of Src and Rac1, junction expansion time (τ) of doublets remained unchanged, but the junction expansion time of triplets was increased by 115% (from 11.0±7.0 min to 23.7±10.2 min; mean±s.d.) and 105% (from 11.0±7.0 min to 22.6±17.0 min), respectively (Fig. 4A). Actin recovery time (t_Cort.) was reduced by 22% (from 16.4±7.7 s to 12.8±5.9 s; mean±s.d.) and 34% (from 16.4±7.7 s to 10.9±4.7 s) in doublets while having no change on singlets (Fig. 4B). In agreement with previous reports that Src is required for E-cadherin adhesive interaction in CHO cells (McLachlan et al., 2007), we also found that Src kinase activity was involved in regulating junction size and cortical tension (Fig. S5C,D). We further elaborated the pathway by focusing on Arp2/3, a well-known downstream effector of Rac1. Consistent with our earlier results, inhibiting Arp2/3 reduced the actin recovery time (t_Cort.) by 18% in doublets while having no effect on singlets (Fig. 4B). Arp2/3 inhibition also abolished the systematic increase of junction expansion time between J₁ and J₂ (Fig. 4A). These results suggest that the slowdown of cortical actin dynamics after de novo junction formation and the cooperative effect likely depended on Src activity and the Rac1–Arp2/3 pathway.

The existence of a crosstalk between adherens junctions and EGFR has been documented in many contexts. Discriminating between direct activation by E-cadherins or by a crosstalk with integrin or paracrine signaling has never been entirely conclusive, has led to apparently contradictory results and is likely to be context dependent. However, the general understanding is that prolonged EFGR activation by EGF locally destabilizes adherens junctions in epithelial cells, promotes E-cadherin endocytosis and triggers mesenchymal transition of epithelial cells (Barrandon and Green, 1987; Lu et al., 2003). At the same time, the polarized epithelium is less sensitive to EGF due to the sequestration of EGFR at the lateral pole beneath the zonula adherens, leading to reduced exposure to the soluble ligand (Kim et al., 2009; Perrais et al., 2007; Qian et al., 2004). This mechanism contributes to the resistance of the epithelium to metastasis. In cancer, the soluble ectodomains of E-cadherin, cleaved from cancer cells into the surrounding medium, are also thought to promote activation of EGFR and, consequently, destabilization of the junction (Rodriguez et al., 2012). In contrast, E-cadherin-dependent activation of EGFR signaling has previously been shown to influence epithelial tissue formation, homeostasis and integrity. For instance, E-cadherin has been shown to promote EGFR-mediated differentiation of lung epithelial cells in a cell density-dependent manner (Kim et al., 2009). Similarly, activated EGFR recruitment through E-cadherin is known to regulate cell proliferation and differentiation through mitogen-activated protein kinase (MAPK) activation (Pece and Gutkind, 2000).

How E-cadherin–EGFR signaling modulates cytoskeletal dynamics and, in turn, junction formation is not well characterized. Recent works have highlighted and unraveled an interdependent relationship between RTKs (especially EGFR) and cellular cytoskeletal networks. For example, Roth et al. have identified LAD1 as a target of EGFR activation that plays a role in regulating actin dynamics critically involved in cell migration (Chiasson-MacKenzie and McClatchey, 2018; Roth et al., 2018). Furthermore, magnetic twisting cytometry has been used to reveal that E-cadherin-based force transduction leads to cell stiffening through the activation of EGFR (Muhamed et al., 2016). Modulation of the actin cytoskeleton by EGFR signaling, in turn, regulates tissue morphogenesis and homeostasis. For example, Rübsam et al. have shown that E-cadherin-mediated localization and activation of EGFR regulates the cortical properties of cells and plays a deterministic role in tight junction positioning in the stratified epithelium (Rübsam et al., 2017). Another report has demonstrated the role of EGF signaling in the transition of epithelial monolayers from an interdigitated jammed cell sheet to a more regular mobile cell layer with tense cell–cell boundaries (Tang et al., 2014).

The doublet system used here establishes four main points that differ from previous literature reports: (1) EGFR can be activated directly or indirectly by the de novo formation of E-cadherin junctions in absence of integrin signaling. (2) This fast activation distally slows down the properties of the whole cell cortex and not only the junction itself. (3) The free cortex in cells with activated EGFR displays slower actin turnover, probably originating from the activation of Arp2/3, which leads to a faster junction expansion. (4) A short burst of EGFR activation by soluble EGF (but not a prolonged exposure to EGF) can phenocopy the priming by de novo adherens junction formation (summarized in Fig. 5). Moreover, our data strengthen observations that Arp2/3 activity leads to a decrease in actin turnover whereas formin activation increases actin dynamics. These results are in line with a previous report that actin assembly speed is decreased in mDia1 (DIAPH1)-depleted cells while perturbation of Arp2/3 leads to an increase (Bovellan et al., 2014). This phenomenon is caused by the competitive regulation of actin dynamics by formin and Arp2/3. The authors found that actin assembly speed (dynamics) increases if a slower-than-average nucleator (such as Arp2/3) is depleted, whereas depleting a faster-than-average nucleator (such as mDia1) decreases actin assembly speed (Bovellan et al., 2014). In addition to EGFR-mediated regulation of Rho GTPases, EGFR directly interacts with the actin cytoskeleton, which in turn mediates a reciprocal regulation of EGFR both in terms of spatial localization and activation–inactivation kinetics (Tang and Gross, 2003).

To our knowledge, this study is the first to establish that junction expansion dynamics can be controlled by cells through an RTK activation that impinges on global cortex properties, at least in the context of suspended cells. It is important to note that the step activation of EGFR increases the junction expansion dynamics but is not necessary to trigger junction formation. This could be due to either alternative or redundant regulation of the cortex by other RTK or regulatory pathways (Bedzhov et al., 2012). It could also be the case that EGFR activation is not involved in the early stage of junction formation.

Last but not least, we would like to emphasize that our study shows that the dynamical properties of the junctions in our systems (kinetypes) are decoupled from the final size of the contact (phenotype), which is governed by the level of contractility of the cortex. Testing a similar effect on adherent cells is technically more challenging and will be the topic of a subsequent publication.

At this stage, we can only speculate as to why the junction expansion dynamics depend on actin turnover more than on contractility. Our data suggest that, at least in the context of suspended cells, cortical tension determines the size of the contact, in line with the current understanding that junction size results from the mechanical balance of cortical and junctional tension. The simple assumption that the junction expansion results from cortical tension acting against a dissipative viscosity of the created interface and free cortex fails to account for the exponential relaxation that robustly describes all our data. At the moment we do not have a satisfying model to account for the central role of free cortex dynamics in the expansion dynamics of the junction; however, we hypothesize that the expansion time is controlled either by (1) the dynamics of rearrangement of actin at the edge of the junction or by (2) the local protrusive activity of the free cortex in the close vicinity of the contact edge. Junction expansion in this context would echo, at least locally, the zippering mechanism invoked in keratinocytes and in fibroblasts. Further theoretical studies are needed to precisely model the observations.

Our study raises the interesting point that suspended cells can independently regulate the dynamics and the static outcome of their contact. Actomyosin contractility can thus not be considered as the sole parameter that controls the junction evolution in this context.

Antibodies and pharmacological inhibitors were as follows: rabbit anti-phospho-EGFR (Y845) polyclonal antibody (Invitrogen, 44-784G; 1:200), rabbit anti-EGFR antibody (Cell Signaling Technology, 2232S; 1:1000), mouse anti-β-actin loading control monoclonal antibody (BA3R; Invitrogen, MA5-15739; 1:5000), anti-rabbit Alexa Fluor 555-conjugated secondary antibody (Thermo Fisher Scientific; 1:200), anti-phospho-myosin light chain 2 (Ser19) antibody (Cell Signaling Technology, 3671; 1:200), Alexa Fluor 405-coupled phalloidin (Invitrogen; 1:100), and horseradish peroxidase (HRP)-coupled anti-mouse and anti-rabbit IgGs (Invitrogen; 1:2000) were used for immunoblotting; recombinant human EGF Protein (R&D Systems); EGF neutralizing antibody (Sino Biological; 10 μg/ml); erlotinib hydrochloride (500 nM; Sigma-Aldrich); dasatinib (500 nM; Selleckchem); NSC-23766 (200 μM; Tocris); CK-666 (100 μM; Sigma-Aldrich); latrunculin A (250 nM; Sigma-Aldrich); SMIFH2 (20 μM; Sigma-Aldrich); jasplakinolide (100 nM; Sigma-Aldrich).

Four SMARTtpool siRNAs against EGFR (siEGFR) were purchased from Dharmacon (SO-2728829G). Targeted sequences were: siRNA1, 5′-GCAUAGGCAUUGGUGAAUU-3′; siRNA2, 5′-GCCAGGUCUUCAAGGAUGU-3′; siRNA3, 5′-CCUUUAUGCUCCUCGGAAG-3′; siRNA4, 5′-GAUUGGUGCUGUGCGAUUC-3′. Four non-targeting pool siRNAs (siCtrl; Dharmacon, SO-2736448G) were used as a non-target control.

S180-E-cad–GFP cells (Chu et al., 2004), a murine sarcoma cell line, were thawed from a low-passage frozen batch and tested for mycoplasma within 6 months. MDCK cells expressing GFP-tagged E-cadherin (MDCK-E-cad–GFP) were kindly provided by Dr Benoit Ladoux's lab (Mechanobiology Institute, National University of Singapore) and were tested for mycoplasma. Cells were cultured at 37°C, 5% CO₂ in high-glucose Dulbecco's Modified Eagle Medium (DMEM, Thermo Fisher Scientific), supplemented with 10% fetal bovine serum (FBS, Thermo Fisher Scientific), 100 units/ml penicillin and 100 μg/ml streptomycin. For serum starvation, the cells were serum-starved in a growth medium containing high-glucose DMEM and lacking FBS for more than 16 h before any experiments. Cells at 80% confluence were transiently transfected with 3 μg DNA using a Neon electroporation system (Invitrogen). EGFR–mEOS (Mechanobiology Institute, National University of Singapore) and actin–mApple (gift from the lab of M. W. Davidson, Florida State University) were used for Neon transfection. Knockdown of EGFR was performed using SMARTpool siRNA. For siRNA transfection, S180-E-cad–GFP cells were plated on 6-well plates and transfected using Lipofectamine RNAimax transfection reagent (Invitrogen) according to the manufacturer's instructions.

The microwell was fabricated on both 12 mm and 27 mm glass bottom Iwaki dishes and visualized under the high-resolution spinning-disc confocal microscope. Master moulds with an array of 20- to 30-μm-diameter holes accompanied with 7×7 position markers were manufactured in SU8-3050 resist on silicon wafers using standard lithography techniques. The thickness of the SU8-3050 resist layer sets the height of the microwells; here the total height was 46 μm. The details of the microwell preparation steps were as described previously (Engl et al., 2014; Gao et al., 2018).

Cells at 80% confluence were detached by flushing medium over the Petri dish to obtain a single-cell suspension. For serum starvation, cells were serum-starved for at least 16 h before detaching. Cell seeding inside the microwell was achieved by adding one drop of prepared single-cell suspension. A glass-bottom dish (Iwaki) with suspended single cells inside microwells was mounted onto the microscope stage and immersed in serum-containing or serum-starved medium without Phenol Red. Another drop of the single-cell suspension was added into the microwells for the second seeding (Fig. S1A). Floating cells were removed after seeding. Images were acquired every 1 min, up to 60 min, using a spinning-disc confocal microscope (Yokogawa CSU-W1) attached to a Nikon Eclipse Ti-E inverted microscope body, with a 60× NA1.3 water lens.

As doublets constantly reorient during spinning-disc confocal microscopic imaging, E-cadherin at junctions always presents at a certain inclination and is seldom in the focal plane. Therefore, a volume of 5–10 μm thickness was imaged to ensure a thorough recording of the contact (Fig. S1B). The image stacks were processed with custom MATLAB codes to re-orientate the cell contact in a fixed referential. The MATLAB code was developed to determine the 3D coordinates (x₀, y₀, z₀) of the center of the contact and the two angles (θ₁ and θ₂) for the inclination (Engl et al., 2014). With the two angles, the stack was re-orientated to allow the contact to be horizontal. A reference circle was automatically fitted onto the contact ring, where the junction radius was provided from the x₀ and y₀ coordinates. By restricting the thickness for integration, we managed to isolate only the junction region with minimum noise from the surrounding plasma membrane. With this processing, deconvoluted image stacks could be interpolated individually, enabling direct visualization of contacts between cell doublets and measurement of junction radius.

Cells were serum-starved for 16 h and pre-incubated with erlotinib hydrochloride (500 nM, 2 h), dasatinib (500 nM, 1 h), NSC-23766 (200 μM, 2 h), CK-666 (100 μM, 1 h), latrunculin A (250 nM, 30 min), SMIFH2 (20 μM, 1 h) or jasplakinolide (100 nM, 30 min) before imaging junction expansion, measuring cortical tension and performing FRAP. For EGF treatment, cells were serum-starved for 16 h and pre-incubated with EGF at 20 ng/ml or 100 ng/ml for 6 h before imaging. Cells were kept in the same concentration of inhibitors or EGF during the whole imaging duration. For EGF burst addition, cells were serum-starved for 16 h prior to imaging. EGF at 20 ng/ml or 100 ng/ml was added into the serum-starved medium 5 min before imaging junction expansion.

Single S180-E-cad–GFP cells were blocked with 5% BSA and were deposited on a glass channel. Micropipettes were pulled with a Flaming pipette puller (P-2000, Sutter Instruments) and forged into a 5–7 μm diameter. The cortical tension here was computed with the law of Laplace: ⁠, where R_P is the radius of the pipette, R_c is the radius of the cell and ΔP is the critical pressure when the extension of the surface of the cell into the pipette (L_P) was equal to the radius of the pipette (R_P). Thus, measuring cortical tension γ is converted to measuring ΔP. Before each measurement, the pressure in the pipette was equilibrated with the outside pressure. When the pressure in the pipette was gradually decreased, cells are aspirated into the pipette immediately. Once the cell forms a hemispherical protrusion in the pipette, which means L_P is equal to R_P, the threshold pressure is reached.

Confluent cells were lysed in a RIPA buffer (Sigma) with added protease and phosphatase inhibitors (Roche) for 20 min at 4°C. Proteins extracted were quantified by BCA assay (Bio-Rad). Then, 4–20% SDS polyacrylamide gel electrophoresis (PAGE) (Bio-Rad) and electro-transfer at 100 V for 2 h were performed to separate protein extracts and transfer them to PVDF membranes (Bio-Rad). Non-specific sites were blocked with 5% BSA in Tris-buffered saline (TBS) containing 0.05% Tween-20. Membranes were incubated with primary antibodies (1:1000 dilution) overnight at 4°C and then followed by an incubation of secondary HRP antibodies (1:2000) for 1 h at room temperature. Results were visualized using a ChemiDoc chemoluminescence detection system (Bio-Rad). Quantification was performed using Image Lab (Bio-Rad). β-actin was used as a loading control to normalize the quantification.

At 1 h after the onset of the junction expansion, we fixed S180-E-cad–GFP cells with pre-warmed 4% paraformaldehyde in phosphate-buffered saline (PBS) at 37°C for 15 min and then permeabilized with 0.2% Triton X-100 in TBS for 30 min at room temperature. Samples were blocked with 1% BSA in TBS for 1 h and incubated with primary antibodies overnight at 4°C. After washing with PBS three times, every 10 min, cells were incubated with secondary antibodies in the dark at room temperature for 1 h. Cells were rinsed with PBS again and were ready for image acquisition. Three-dimensional stacks of confocal images were acquired using a spinning-disc confocal microscope (Yokogawa CSU-W1) attached to a Nikon Eclipse Ti-E inverted microscope body, with a 60× NA1.3 water lens.

Prism (GraphPad Software) and MATLAB (Math Works) were used for data analysis and graph plotting. Graphs were mounted using Adobe Illustrator. ANOVA tests and paired or unpaired Student's t-tests were carried out to analyse the levels of significant difference (ns, not significant, P>0.05; *P≤0.05; **P≤0.01; ***P≤0.001; ****P≤0.0001).

We acknowledge Ying Bena Lim from C.T. Lim's lab (Mechanobiology Institute, National University of Singapore) for helping with the cortical tension measurements.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.258929.