Two contractile pools of actomyosin distinctly load and tune E-cadherin levels during morphogenesis

Epithelial tissues are highly dynamic. During embryonic morphogenesis cell contacts are constantly remodeled. This stems from active contractile forces that work against adhesive forces at cell interfaces. E-cadherin complexes play a pivotal role in this process as they both support inter-molecular adhesive forces and transmit mechanical tension due to their coupling to the cortical contractile actomyosin networks. In this context, it is unclear how tensile forces affect E-cadherin adhesion complexes and junction dynamics. Addressing this calls for methods to estimate the tensile forces (load) experienced by adhesion complexes themselves. We address this during the early morphogenesis of the Drosophila embryonic ectoderm. Tensile forces generated by Myosin-II in the apico-medial cortex (medial Myosin-II) and in the junctional cortex (junctional Myosin-II) are responsible for junction remodeling. We combined mechanical inference and laser ablations to map tension at cell junctions. We also established Vinculin as a force sensor whose enrichment with respect to E-cadherin measures the load on adhesion complexes within each junction. Combining these tools, we show that the tension generated in both medial and junction pools of Myosin-II imposes load on E-cadherin adhesion complexes. Medial Myosin-II loads adhesion complexes uniformly on all junctions of a cell and increases levels of E-cadherin. Junctional Myosin-II, on the other hand, biases the distribution of the load between junctions of the same cell and exerts shear forces, which correspond to a decrease in the levels of E-cadherin. This work highlights the difference between medial Myosin-II and junctional Myosin-II in regulating E-cadherin levels during junction remodeling and suggest opposite effects of shear versus tensile stresses on E-cadherin complexes and on the dynamics of adhesive cell contacts. adhesion and to open a study of energy dissipation at E-cadherin adhesion complexes in the study of cell-cell adhesion 4 .


Introduction
Tissue scale morphogenetic movements are driven by the dynamic remodeling of cell-cell adhesion and contractile actomyosin cytoskeleton at cell interfaces 1-4 . E-cadherin based cell adhesion machinery is not uniformly distributed at the cell interfaces. Within adherens junctions, E-cadherin forms cis-and trans-homophilic clusters whose size, density and lateral mobility (flow) depend, in part, on coupling to F-actin [5][6][7] . E-cadherin cell adhesion complexes are physically linked to the actomyosin cytoskeleton by α-Catenin and Vinculin, two F-actin binding proteins [8][9][10][11][12][13] . Such coupling to F-actin is essential for determining E-cadherin cluster size and number, underlying adhesion maturation, cell-cell cohesion 14,15 , epithelial integrity in vivo 9 and cell sorting behavior 16 . Importantly, E-cadherin coupling to F-actin via α-Catenin is dependent on force: α-Catenin's interaction with F-actin can be modeled as a two-state catch bond, where force shifts the complex to a strongly bound-state 11 potentially by a tensioninduced conformational change 10,[17][18][19][20] . This argues that actomyosin-generated tension reinforces coupling to E-cadherin complexes in vivo 21,22 . By virtue of trans-homophilic interactions, E-cadherin complexes transmit these tensile forces across actomyosin cortices of neighboring cells 16 . The link between actomyosin contractility and E-cadherin may promote the regulation of cell adhesion by actomyosin contraction during tissue morphogenesis, though this possibility has not yet been directly addressed in a developmental context. E-cadherin based cell adhesion plays a dual role by both maintaining tissue cohesion and by facilitating tissue remodeling under biochemical and mechanical regulation [23][24][25][26] . Contractile forces can affect cell adhesion, as they can directly influence the recruitment or turnover of Ecadherin molecules 27 . However, the evidence is sometimes contradictory, in some instances tension stabilizes E-cadherin, while in others tension appears to have the opposite effect.
Mammalian cell culture experiments have demonstrated that cells respond to cell extrinsic tensile forces through local reorganization of F-actin cytoskeleton and increased recruitment or stabilization of E-cadherin 12,13,18,20 . However, other experiments have demonstrated that Ecadherin levels are reduced due to signaling downstream of Src and that the contractile activity of Myosin-II is the transducer of this reduction 28 . In addition, higher junctional tension correlates with increased turnover rate of E-cadherin molecules in MDCK cells 29 , which in turn depends on the endocytosis/exocytosis of E-cadherin 30 , arguing for tension reducing Ecadherin levels. Whether mechanical load regulates E-cadherin based adhesion in vivo has been comparatively little explored 31,32 .
We addressed whether Myosin-II controls E-cadherin levels at cell contacts in the Drosophila early embryonic ectoderm, during its convergent-extension movements 33 . These movements rely on cell intercalation, which involves disassembly of junctions oriented along the dorsalventral axis (DV, vertical junctions) of the embryo, followed by elongation of new junctions along the anterior-posterior axis (AP, transverse junctions) [34][35][36] . Ectodermal cells have two distinctly regulated pools of Myosin-II that are both responsible for persistent junction shrinkage 37,38 . First, a pulsatile pool of medial Myosin-II produces semi-periodic contractions in the apical cortex, and junctional Myosin-II produces anisotropic contractions along vertical junctions. It is believed that these two networks cumulatively generate polarized tension at cell junctions. Indeed, tension at cell junctions is planar polarized as measured by focal laser ablation of actomyosin cortices 39 and by optical tweezers 40 . However, the measured tension is defined at the scale of a whole junction. It is unclear how this junction-level tension translates at the level of E-cadherin molecules to which actomyosin networks are coupled.
Further, it is unknown whether the respective actomyosin contraction contributions from the medial and the junctional networks is transmitted to E-cadherin molecules in a different manner. Lastly, it remains unresolved whether the impact of the respective contributions of the medial and junctional actomyosin on E-cadherin recruitment is distinct.
In this study, we have sought to investigate the effect of actomyosin contractility on cell adhesion, through the analysis of the load exerted onto E-cadherin. We analyzed the contribution of medial and junctional actomyosin networks to the load on adhesion complexes and to the recruitment of E-cadherin during morphogenesis of the embryonic ectoderm. Our analysis leads us to consider the differential role of tensile and shear stresses exerted respectively by the medial and junctional actomyosin networks.

Drosophila Vinculin is recruited at adhesion complexes via α-Catenin in a Myosin-II dependent manner
In mammalian cells, Vinculin is recruited at E-cadherin adhesion complexes via its binding with α-Catenin 10,12,19 . We asked whether a similar phenomenon occurs in Drosophila embryonic ectoderm. We first verified that Vinculin is a component of E-cadherin-based adhesion complexes. Figure 1A shows E-cadherin clusters (arrowheads) co-localizing with Vinculin clusters. This point is qualitatively supported by the similarities in the intensity profiles for Vinculin and E-cadherin in a zoom-in view of a junction (Fig. 1B). To test whether α-Catenin is required for the recruitment of Vinculin to adhesion complexes, we injected embryos with double-stranded RNA (dsRNA) to achieve an RNAi mediated knockdown of α-Catenin (see Methods). α-Catenin knockdown significantly reduced Vinculin density at cell-cell contacts (Fig. 1C, C'), implying that α-Catenin is the primary interactor of Vinculin and facilitates Vinculin recruitment in adhesion complexes. Further, Vinculin was enriched in the apico-lateral domain of ectodermal cells similar to E-cadherin and Myosin-II ( Supplementary Fig. 1A-D). With these observations, we conclude that Vinculin is a bona fide component of adhesion complexes in Drosophila, similar to its mammalian homologs.
We observed that Vinculin was enriched on vertical junctions compared to transverse junctions in the embryonic ectoderm ( Supplementary Fig. 1H, H'). This distribution strikingly mirrored that of Myosin-II (Supplementary Fig. 1G, G'), which is known to be planar polarized 34,41,42 . This is remarkable since E-cadherin, however, is present at a lower concentration on vertical junctions 35,43 . Thus, Vinculin distribution was opposite to that of E-cadherin or α-Catenin, which are enriched on transverse junctions relative to vertical junctions (Supplementary Fig. 1E-F'). Vinculin planar polarized distribution was lost in the absence of α-Catenin ( Fig. 1C''). These results suggest a differential role of adhesion and contractility in regulating Vinculin recruitment and distribution.
The tensile forces generated by Myosin-II are known to produce structural changes in α-Catenin that expose a cryptic binding site for Vinculin and enhance the recruitment of Vinculin to adhesion complexes 10,18,19 . Thus, inhibiting Myosin-II activity can result in a reduction in the junctional recruitment of Vinculin. We tested this idea by injecting in embryos a Rok inhibitor, to block the Myosin-II activity (see Methods). Rok inhibition significantly reduced Myosin-II recruitment at junctions and abolished its planar polarity (Fig. 1D-D''). The same treatment also reduced Vinculin line densities on all junctions (Fig. 1E, E'). Noticeably, it inverted the planar polarized distribution of Vinculin, which became similar to that of E-cadherin ( Fig. 1E''). Rok inhibition also reduced E-cadherin density at junctions and amplified its planar polarity ( Fig. 1F-F''). Given that Vinculin co-localizes with E-cadherin, we asked if the inversion of Vinculin planar polarity was due to a constitutive localization of Vinculin to E-cadherin in the absence of Myosin-II activity. When we normalized junctional Vinculin density to that of Ecadherin, the Vinc/E-cad ratio, indeed, this ratio was reduced upon Rok inhibition and its planar polarity was lost (Fig. 1G-G'') in a manner similar to Myosin-II. The fact that the planar polarity of Vinc/E-cad ratio qualitatively parallels that of Myosin-II suggests that the recruitment of Vinculin to adhesion complexes is enhanced by Myosin-II activity.
We further tested this by calculating the linear correlation coefficient between junctional Vinculin intensity and E-cadherin intensity. The correlation was performed by binning junctions according to their length. We term it the 'conditional correlation' (see Methods and Supplementary Fig. 2A). Such a measurement avoids the indirect correlation between the mean junctional intensities, as they are proportional to the inverse of junctional length. The correlation was consistently strong in Rok inhibited embryos independent of junction length (Supplementary Fig. 2B and C), indicating a constitutive association between Vinculin and Ecadherin in the absence of Myosin-II activity. In the presence of Myosin-II activity, the correlation between Vinculin and E-cadherin intensities was stronger on shorter junctions (Supplementary Fig. 2B and C). This suggests that Myosin-II activity enhances Vinculin recruitment to adhesion complexes at shrinking junctions.
Taken together, these results indicate that Vinculin is recruited to adhesion complexes at low levels independent of Myosin-II activity. In the presence of Myosin-II activity, Vinculin recruitment is enhanced further. In light of these observations, we decided to normalize Vinculin intensity (line density) with that of E-cadherin to specifically focus on the Myosin-II activity-dependent recruitment of Vinculin to E-cadherin. no correlation between the recoil velocity and E-cadherin density ( Supplementary Fig. 3B), indicating the specificity of the analysis.
Laser ablation is an invasive method of tension estimation and suffers greatly from the difficulty to obtain very large data sets. The number of data points per embryo is limited, as the ablations are performed within a 10-minute time window during embryonic development and need to be done one junction at a time (see Methods). To overcome these limitations, we implemented a mechanical inference method to estimate junctional tensions [48][49][50] and correlated them with Myosin-II intensity and Vinc/E-cad ratio at a cellular level for thousands of cells across multiple embryos (see Methods) in a non-invasive, high-throughput manner. Figure 2D shows a snapshot from a wild type embryo on which we implemented the mechanical inference. Figure 2E shows the corresponding output of inferred tension from mechanical inference, where the thickness of the junction is proportional to the inferred tension. Note that the mechanical inference captures the tension cables along vertical junctions, which are reported to be under higher tension 39,46 . The planar polarity of inferred tension showed a trend similar to that of Vinc/E-cad ratio (Fig. 2F, G).
We calculated the linear correlation between Vinc/E-cad and inferred tension by binning junctions according to cell. We term it the 'local correlation' (see Methods and Supplementary Fig. 2A), which accounts for temporal and inter-embryo variations. This allowed us to combine data for all cells across multiple time points from many embryos (n≥5 embryos) to perform high-throughput statistics and gave us access to information at a cellular level. Figure 2H shows the distribution of the local correlation coefficient of junctional Myosin-II intensity and inferred tension. The median correlation coefficient for wild-type embryos is 0.6 and drops to 0.2 for Rok inhibited embryos. This showed that the inferred tension strongly correlates with Myosin-II activity, validating the mechanical inference once again. We performed the same analysis for Vinc/E-cad ratio and Vinculin intensity. Figure 2I shows the distribution of the local correlation coefficient of the Vinc/E-cad ratio and inferred tension. The median correlation coefficient for the wild-type embryos is 0.6 and drops to 0.14 after Rok inhibition, showing that Vinc/E-cad ratio strongly correlates with junctional tension in the presence of Myosin-II activity. Figure 2I shows the distribution of the local correlation coefficient of the Vinculin intensity and inferred tension. Even though the median correlation coefficient for the wild-type embryos is 0.56, the correlation is negative (-0.25) for Rok inhibited embryos. This is consistent with the inversion of Vinculin planar polarized distribution upon Rok inhibition ( Fig. 1E''). This indicated, once again, that Vinc/E-cad ratio correlates better with tension as compared to Vinculin intensity itself.
Further, we estimated the 'conditional correlation' between inferred tension and junctional intensities. In such analysis, we showed that Vinc/E-cad ratio and Myosin-II intensity correlate better with inferred tension than does Vinculin intensity (Supplementary Fig. 4B and C). This observation is consistent with the laser ablation analysis, where we found that the estimated tension has a stronger correlation with Vinc/E-cad ratio or Myosin-II density as compared to Vinculin density itself (Fig. 2B, C and Supplementary Fig. 3A). This further established Vinc/Ecad as a better indicator of junctional tension than Vinculin intensity alone.
Altogether, our data indicated that Vinculin is a bona fide force sensor at adhesion complexes in Drosophila and that the distribution of Vinc/E-cad ratio can be used as a proxy for the distribution of junctional tension.

The ratio of Vinculin to E-cadherin densities reflects mechanical load on adhesion complexes
Mechanical inference and laser ablations provide an estimate for the junctional tension, a macroscopic quantity that is assumed to be uniform along the junction. E-cadherin adhesion complexes, on the other hand, are distributed in clusters along the junction ( Fig. 1A and Supplementary Fig. 1E, F). Adhesion complexes, composed of E-cadherin, β-Catenin and α-Catenin, mechanically resist the contractile forces from actomyosin. Thus, adhesion complexes could be under differently oriented contractile forces and resist different magnitude of mechanical loads as they couple independently to the actomyosin network. Vinculin can be an estimate of the mechanical load [I still find this word confusing. By tensile forces on an a complex do we mean the load along the length of the complex. If so, we should state that clearly] experienced by each adhesion complex, as individual molecules of Vinculin are recruited to α-Catenin, in a load dependent manner 17,51 . Given that Vinc/E-cad ratio correlates with 'junctional tension' (a macroscopic quantity), we asked if it can be a readout of the mechanical load at adhesion complexes (a microscopic quantity), potentially providing access to forces at a sub-junctional level.
We address this question by over-expressing E-cadherin to increase its junctional level. The Ecadherin over-expression is expected to reduce the number of Myosin-II molecules per Ecadherin molecule, thereby reducing the load per adhesion complex. Vinculin level is hence expected to decrease relative to E-cadherin due to a reduction of tension supported by each E-cadherin molecule in the adhesion clusters (Fig. 3A). Indeed, E-cadherin over-expression produced a mild, but significant, increase in its junctional density (Fig. 3B, B'), while the distribution of Myosin-II was unchanged ( Fig. 3E-E''). The junctional tension was also unchanged as shown by recoil velocities after laser ablations ( Supplementary Fig. 5A). Concomitantly, there was a reduction in Vinculin density on all junctions (Fig. 3C, C''), leading to an even stronger decrease in Vinc/E-cad ratio (Fig. 3D, D'). We suggest that the decrease in Vinculin levels is not due to junctional tension or Myosin-II, as both quantities are unaffected by E-cadherin over-expression, but a response to the decrease of the load per adhesion complex (see Fig. 3A).
Moreover, Vinc/E-cad ratio can be calculated at a sub-junctional scale, even at a scale as small as individual adhesion clusters. Therefore, we further asked if we can see a consistent change in Vinc/E-cad ratio at a sub-junctional scale. Pixels in an image are the smallest possible spatial scale available in our analysis. So, we estimated Vinc/E-cad ratio at individual pixels (see Methods and Supplementary Fig. 6). In such an analysis, we observed a consistent reduction in Vinc/E-cad ratio upon E-cadherin over-expression, across all E-cadherin pixel intensity bins (Fig. 3F).
These quantifications suggest that, Vinc/E-cad ratio at each adhesion complex can be used as a proxy for the 'load on adhesion complex'. We note that Vinc/E-cad is independent of junctional length, thus can be estimated at a microscopic scale of adhesion clusters and will act as a molecular force sensor.
. CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; Medial and junctional Myosin-II distinctly load E-cadherin adhesion complexes E-cadherin adhesion complexes are mechanically coupled to two spatially separated and distinctly regulated pools of Myosin-II, the medial pool and the junctional pool 37,52 . The relaxation kinetics of actomyosin cortex in laser ablation experiments suggest that medial Myosin-II exerts tension that is predominantly orthogonal to cell contacts, whereas junctional Myosin-II exerts tension that is predominantly parallel to cell contacts 37,39,43 . The forces produced by these two pools are differently oriented towards the junctions, so we asked whether they distinctly load adhesion complexes. First, we inhibited Myosin-II activity globally (Rok inhibition). This treatment reduced the levels of Vinc/E-cad ratio, and suppressed its planar polarity (Fig. 4A-B'), consistent with the idea that Myosin-II activity is required to load the adhesion complexes. Next, we tuned the Myosin-II activation in the medial pool only. A recent study demonstrated that a Gα12/13-RhoGEF2-Rho1-Rok pathway phosphorylates and activates Myosin-II in the medial pool downstream of GPCR signaling 38 . Using RhoGEF2-RNAi, we reduced the activation of Myosin-II only in the medial pool, without affecting Myosin-II recruitment in the junctional pool ( Fig. 4C-C''). This treatment decreased the Vinc/E-cad ratio without affecting its planar polarized distribution (Fig. 4D, D'). This reduction in the Vinc/E-cad ratio could be due to a reduction of the load generated by the medial Myosin-II or a global reduction in junctional tension itself. We ruled out the latter possibility by laser ablation experiments ( Supplementary Fig. 5B) and the fact that junctional Myosin-II intensity is unchanged in RhoGEF2-RNAi embryos (Fig. 4C"). To complement this observation, we increased the recruitment of Myosin-II in the medial pool using Gα12/13 over-expression, without affecting Myosin-II recruitment in the junctional pool ( Fig. 4E-E''). Consistently, this treatment increased the Vinc/E-cad ratio without affecting its planar polarized distribution (Fig. 4F, F').
Thus, a global decrease (increase) in the levels of medial Myosin-II decreases (increases) the load on adhesion complexes uniformly on all junctions of a cell, as indicated by the uniform decrease (increase) of Vinc/E-cad ratio. In contrast to a global inhibition of Myosin-II activity, a specific inhibition of medial Myosin-II activity did not affect the planar polarized distribution of the junctional Myosin-II and preserved the planar polarized distribution of Vinc/E-cad ratio ( Fig. 4B', D'). Thus, we conclude that the planar polarized junctional Myosin-II imposes a larger amount of load on vertical junctions than transverse junctions and determines the planar polarity of Vinc/E-cad ratio.

Medial and junctional Myosin-II have opposite effects on junctional E-cadherin concentration
Given that medial and junctional pools of Myosin-II load adhesion complexes differently, we asked whether these two pools had distinct impacts on E-cadherin levels at cell junctions. We found that a global inhibition of Myosin-II activity (Rok inhibition) decreased the E-cadherin density at junctions (Fig. 1F, F'). Interestingly, a specific inhibition of the medial Myosin-II using RhoGEF2-RNAi without perturbation of the junctional Myosin-II ( Fig. 4C-C'') leads to a reduction in E-cadherin levels (Fig. 5A, A') that is comparable to the Rok inhibited embryos. This indicated that the global levels of E-cadherin are regulated by medial Myosin-II and the presence of junctional Myosin-II alone did not restore the global levels of E-cadherin. To further test this, we increased the levels of medial Myosin-II using Gα12/13 over-expression while preserving the levels of junctional Myosin-II ( Fig. 4E-E''). We observed an increase in E-. CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; cadherin density at junctions (Fig. 5B-B'). These results suggest that the contractile medial Myosin-II regulates the global junctional recruitment of E-cadherin.
Planar polarized junctional Myosin-II ( Supplementary Fig. 1G') is important for junction shrinkage 34,41 . It is hypothesized that the shear stress generated by junctional Myosin-II may affect the stability of adherens junctions by stretching the trans-cellular E-cadherin dimers 48 . However, this hypothesis has never been tested with experimental data due to the difficulty of measuring shear stress. Mechanical inference provides a unique way to approximate junctional shear stress from inferred tensions (Fig. 6A) 48 . Hence, we tested this hypothesis by estimating the 'conditional correlation' of E-cadherin density with inferred shear stress on junctions. The shear stress displayed a negative correlation with E-cadherin density on vertically shrinking junctions (Fig. 6C, D). In contrast, this correlation is reduced when pooling all junctions with different orientations (Fig. 6B) and vanishes for transverse junctions and Rok inhibited embryos (Fig. 6E, F). It is interesting to note that the transverse junctions show the same extent of this correlation, irrespective of Myosin-II activity. This further emphasizes that the shear forces are specifically active on vertical junctions. In addition, the correlation with inferred tension is much weaker (Fig. 6B, C, D). Combined together, these results indicate that shear stress rather than tension enhances the dissociation of E-cadherin from vertically shrinking junctions by shearing the adhesion complexes during junction remodeling.
Together these experiments suggest that the medial Myosin-II increases the levels of junctional E-cadherin by uniformly loading the adhesion complexes on cell junctions, while planar polarized junctional Myosin-II regulates junction remodeling by exerting shear forces on the adhesion complexes at vertically shrinking junctions.

Discussion
How contractile forces generated by Myosin-II activity regulate junction remodeling during morphogenesis is still an open question. In this study, we have used Vinculin as a molecular force sensor on E-cadherin complexes, whose recruitment to adhesion complexes is modulated by the contractile activity of Myosin-II and the resulting tensile forces; hence its ratio with E-cadherin provides a potential readout of mechanical forces on E-cadherin adhesion complexes at cell junctions. Using mechanical inference and laser ablation, we found that the enrichment of Vinculin relative to E-cadherin can be used to estimate the distribution of load on E-cadherin at cell junctions. While we have established Vinc/E-cad ratio as a tension sensor, it remains to be determined how this ratio depends on junctional tension explicitly. The conditional correlation revealed a length-dependent correlation between Vinc/E-cad ratio and inferred tension, with reduced correlation for short and long junctions ( Supplementary  Fig. 4C). This could suggest a non-linear dependence of Vinc/E-cad ratio on junctional tension with saturated response at short junctions under large tension and the presence of a tension threshold for activation at long junctions under small tension. Experiments with quantitatively controlled tension could determine the response curve of Vinc/E-cad ratio to the magnitude of tension.
Next, we tuned the loading forces on adhesion complexes by increasing the E-cadherin levels and revealed that Vinc/E-cad ratio can be a load sensor at the adhesion complex scale. Given that the stoichiometry between Vinculin and E-cadherin (proportional to Vinc/E-cad ratio) is a dimensionless quantity, we argue that the Vinc/E-cad ratio estimates the load experienced by . CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; individual (diffraction limited) adhesion clusters. It is interesting to note that the distribution of Vinc/E-cad ratio is not homogeneous along a junction (e.g. Fig. 1G, 2D, 3D, 4B, 4D and 4F) and that Vinc/E-cad ratio is greater at brighter E-cadherin pixels (Fig. 3F). This indicates that load distribution is inhomogeneous along the junction and that the junctional subdomains with higher E-cadherin density experience greater load. This observation is consistent with a recent study 53 , which reported that the mechanosensitive conformational changes in α-Catenin can be observed predominantly in larger E-cadherin clusters.
We used Vinc/E-cad ratio as a load estimate to study the effect of contractile forces from two distinct pools of Myosin-II, the medial and the junctional pool. The two pools are distinct in terms of their upstream regulation and have been studied in the Drosophila embryonic ectoderm 38 . They are mechanically coupled to adhesion complexes to exert forces on cell-cell contacts. In this study, we showed that these two pools of Myosin-II have distinct impact on the distribution of load on E-cadherin. Medial Myosin-II uniformly loads adhesion complexes across all junctions within a cell. In contrast, the planar polarized junctional Myosin-II biases the load towards vertical shrinking junctions, thus regulating the planar polarity of load. We have quantitatively demonstrated that both pools of Myosin-II exert forces on E-cadherin complexes and cell contacts.
The load generated due to activity of medial Myosin-II increases the levels of junctional Ecadherin. This observation is consistent with a study in the Drosophila mesoderm 32 , where it is observed that the activity of medial Myosin-II protects E-cadherin from a Snail mediated downregulation. A change in junctional Rho signaling can also change E-cadherin levels through its impact on the F-actin organization and Myosin-II activity [54][55][56] . We think that this is not the case as the junctional Myosin-II levels and presumably junctional Rho signaling is unchanged when we specifically tuned medial Myosin-II. In fact, the changes in junctional Ecadherin levels correlated with the changes in medial Rho signaling downstream of activation by the Gα12/13-RhoGEF2 signaling module. Given that the inhibition of medial Rho signaling (RhoGEF2-RNAi) and Rok inhibition have similar impact on E-cadherin levels, we argue that the effect of medial Rho signaling on E-cadherin levels is through its effect on Myosin-II activation. It remains to be determined if the effect of medial Myosin-II activity on the levels of junctional E-cadherin is a mechanosensitive response or not.
We used a correlative analysis with mechanical inference to study the effect of junctional Myosin-II on junctional E-cadherin levels. We constructed a model to estimate shear stress based on inferred tensions. In this model, an asymmetric distribution of inferred tension on opposite sides of the junction generates shear stress that stretches E-cadherin trans-dimers on shrinking junctions 48 . Strikingly, we observed a negative correlation between the inferred shear stress and the junctional E-cadherin levels. Particularly, the negative correlation was specific to vertical junctions (the category to which shrinking junctions belong) and vanished on either the transverse junctions or the junctions from Rok inhibited embryos. This observation argues that junctional Myosin-II promotes dissociation of E-cadherin on vertically shrinking junctions during junction remodeling via a shear effect on E-cadherin complexes. However, we cannot rule out the possibility that medial Myosin-II also contributes to the shear stress to some extent, as the mechanical inference does not specify the source of the forces. The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; We hereby propose a mechanical model for cell junction remodeling, where we highlight the importance of the subcellular origin of contractile forces and their mechanical effect, namely tensile versus shear stress, in promoting a change in the levels of E-cadherin at cell contacts and on junction dynamics. The mechanisms that generate the different responses in Ecadherin levels remain unknown. We have established Vinculin as a molecular force sensor, but it remains to be determined whether Vinculin is involved in the stabilization of adherens junctions by regulating E-cadherin levels as a mechanotransducer. Vinculin is not essential for survival in Drosophila 57 , raising questions about the necessity of its function as a mechanotransducer. To reveal the mechanism by which actomyosin contractility regulates Ecadherin levels, it is essential to study the magnitude and orientation of contractile forces, the spatial distribution of mechanical coupling between the adhesion complexes and the actomyosin network, and the different modes of energy dissipation at adhesive complexes under mechanical forces. Given this distinction between tensile and shear stress in the regulation of E-cadherin at cell contacts, it will also be important to consider the dynamics of E-cadherin complexes. We speculate that adhesion mediated by E-cadherin has evolved to stabilize complexes under tensile stress and to constantly remodel them under shear stress. Tensile (i.e. normal) stresses reinforce cell-cell coupling to induce tissue deformation such as tissue invagination. This mode maintains adhesion below a threshold where junctions may collapse at once: this condition is observed when adhesion is mildly compromised using partial knock-down of E-cadherin or αcatenin in the mesoderm and endoderm where medial Myosin-II accumulates at very high levels. The shear mode also maintains adhesion but dynamically, thereby allowing tissue remodeling such as during cell intercalation in the ectoderm: on average the density of complexes could remain constant but the turnover of homophilic bonds would be increased. The differential effect of tensile and shear stress on E-cadherin dynamics has the potential of reconciling conflicting evidence on the role of contractile forces on adhesion and to open a study of energy dissipation at E-cadherin adhesion complexes in the study of cell-cell adhesion 4 .
. CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018;

Embryo preparation, RNAi and drug injections
Embryos were prepared as described before 34,60,61 . Briefly, embryos were dechorionated using bleach, for about 40 seconds and then washed thoroughly with distilled water. The embryos were then aligned on a flat piece of agar and then glued to a glass coverslip. These embryos can be submerged in water and can be imaged directly. Alternatively, glued embryos were kept in an airtight box containing Drierite for about 7 minutes, then covered in halocarbon oil and then injected with either RNase free water itself or RNase free water containing dsRNA or drugs.
α-Catenin RNAi was achieved by injecting dsRNA in embryos (previously used in 9 ). dsRNA probes against α-Catenin were made using PCR products containing the sequence of the T7 promoter targeting nucleotides 101-828 of α-Catenin sequence (GenBank accession . CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; D13964). dsRNA prepared as already described were diluted for injection at 5μM concentration and injected within the first hour of embryonic development to achieve maximum knockdown. As a control, separate set of embryos of the same stage were injected with similar volume of RNase free water.
Rok inhibition was achieved through drug injections. H1152 is a membrane permeable pharmacological inhibitor that has high specificity for Rok and blocks its kinase activity. This drug was dissolved in RNase free water @20mM and injected at the end of cellularization. As a control, separate set of embryos of the same stage were injected with similar volume of RNase free water. The reduction in Myosin-II recruitment acts as a direct readout of Rok inhibition. The effect of Rok inhibition on Vinc/E-cad ratio could not be assessed directly, as Myosin-II couldn't be imaged simultaneously. Thus, first the inhibition was performed in embryos expressing MyoII-GFP and Vinculin-mCherry. The reduction in Vinculin intensity can then be used to assess the extent of Rok inhibition in embryos expressing Vinculin-mCherry and E-cadherin-GFP, while also estimating the Vinc/E-cad ratio. The reduction in E-cadherin was cross-checked with another set of embryos expressing E-cadherin-GFP and MyoII-mCherry, where reduction in MyoII-mCherry recruitment acted as a direct readout of Rok inhibition.

Imaging
Time-lapse images were acquired to encompass stage 7 to 8 of the embryonic development 62 , which needs ~15min at room temperature (~22 0 C). Embryos were imaged for 20-30min depending on the experiment, on a Nikon spinning-disk Eclipse Ti inverted microscope using a 100x 1.45 NA oil immersion objective. MyoII-mCherry and E-cadherin-GFP signals were captured every 30s or higher, on 11 Z-planes, separated by 0.5μm. Vinculin-mCherry and Ecadherin-GFP signals were captured every 30s or higher, on 7 Z-planes, separated by 0.5μm. A Nikon spinning-disc Eclipse Ti inverted microscope using a ×100, 1.4 N.A oil immersion objective was used for imaging α-Catenin-YFP. Both systems acquire images using the MetaMorph software. Laser power and exposure settings were optimized to get best signal while minimizing photo-bleaching, and were kept constant between control and perturbation embryos.

Laser ablation experiments
Ablations were performed in a 10-minute time window around stage7b (stage7b ± 5min) on an inverted microscope (Eclipse TE 2000-E, Nikon) equipped with a spinning-disc (Ultraview ERS, Perkin Elmer) for fast imaging. Time lapse at a single z-plane was acquired using a 100x 1.4NA oil immersion objective. Two color images were acquired in sequence on the same camera, when necessary. Ablations were performed in parallel with image acquisition. Ablation events were obtained by exposing the junctions, for duration of 2-3ms, to a nearinfrared laser (1,030 nm) focused in a diffraction-limited spot. Laser power at the back aperture of the objective was ~800mW.

Image analysis and Statistics
All image processing was done using FIJI freeware. Raw images were processed using a custom written macro. First, it generated a 'signal image' by using the StackFocuser plug-in to determine the plane of best focus, followed by a maximum-intensity projection of only 3 zplanes (1 z-plane in focus determined by StackFocuser + 2 z-planes basal to it). The macro also . CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; generated a 'background image', first, using a maximum-intensity projection of basal-most 3 planes, followed by applying a 50pixel radius median filter. The macro then subtracted the 'background image' from 'signal image' to produce 'processed image'. The images were independently segmented using 'Packing analyzer v2.0' (described in 63 ), which was implemented as a plugin in FIJI, to get segmented junctional networks. E-Cadherin, Vinculin, Myosin-II or α-Catenin intensities were used, depending on the genotype of the embryos (and in that order of preference), for image segmentation in order to identify cell-cell contacts in a semi-automated manner. Using another custom written macro, the segmentation was used to demarcate the junctional ROIs of about 5 pixel width, such that the vertices (tricellular junctions) are excluded. Line densities were measured for the protein of interest to calculate 'mean junctional intensity'. Junctions were categorized based on their angle relative to AP axis into 6 'angle bins' (0-15, 15-30 . . . 75-90 degrees). An average of the junctional line densities was calculated within each 'angle bin' to get 6 values for every embryo. A further average of these 'averaged line densities' acts as a data point per embryo to estimate 'mean junctional intensity' for the protein of interest. The ratio between the 'averaged line density' for '0-15' (AP) and '75-90' (DV) categories produces the 'amplitude of polarity', either AP/DV or DV/AP as mentioned in Y-axis labels. Alternatively, the planar polarity of a protein was represented as 'relative intensity', where the 'averaged line density' values from the 'angle bins' were normalized by the 'average line density' of either '0-15' or '75-90' category, whichever is smaller. In case of Vinc/E-cad ratio, similar calculations were performed after having calculated the ratio between Vinculin and E-cadherin line densities for every junction. The same segmentation was also used to identify the medial ROIs (at least 2 pixels away from any junctional ROI and tri-cellular junctions), for medial Myosin-II intensity estimates. An average of medial area densities was calculated to get one data point per embryo, to estimate 'mean medial intensity' for Myosin-II across multiple embryos.
In case of ablation experiments, images were first processed using the 'rolling ball' background subtraction method implemented in FIJI (rolling ball radius 50). Junctional ROIs were drawn manually (5 pixels wide) on the ablated junction and on 20+ neighboring junctions. Then, the line density of the ablated junction was normalized by the averaged line density of the neighboring junctions. This yielded the 'normalized junctional intensity' for protein of interest. Such normalization is necessary to reduce embryo-to-embryo variability. In case of Vinc/E-cad ratio, similar calculations were performed after having calculated the ratio between Vinculin and E-cadherin line densities for all marked junctions. The vertices of the ablated junction were tracked manually to estimate the recoil velocity in 2 seconds after the ablation. Spearman correlation gave an estimate of the extent of correlation between 'pre-ablation normalized junctional intensity' and corresponding 'post-ablation initial recoil velocity' values.
For 'pixel scale analysis' of Vinc/E-cad ratio, we identified E-cadherin positive pixels by estimating the Signal-to-Noise Ratio (SNR) at all pixels ( Supplementary Fig. 6A, B) and measured Vinc/E-cad for the pixels with SNR>1. We empirically decided the range of Ecadherin pixel intensities to span an order of magnitude, such that each intensity range hosts statistically meaningful number of pixels ( Supplementary Fig. 6C).
The 'mean values' and 'standard errors on mean' were calculated from 'n' data points. The same data points were used for testing statistical significance. In planar polarity and junctional . CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; intensity measurements, 'n' is the number of embryos. Error bars indicate s.e.m. (i.e., s.d./√ n). The p-values were estimated using student's t-test, wherever applicable. In laser ablation experiments, 'n' is the number of ablated junctions that are pooled from many embryos. The p-values were estimated using Mann-Whitney U-test. The sample sizes were not predetermined using any statistical methods. The experiments were not randomized, and the investigators were not blinded to allocation during experiments and outcome assessment.
All measurements were performed on 5-25 embryos spread over at least 3 rounds of embryo collections and preparation.

Mechanical Inference
Mechanical inference is an image-based force inference technique that takes a segmented cellular network as the input and estimates relative tensions along cell junctions by assuming force balance at each vertex [48][49][50]64,65 . We implement mechanical inference on segmented images of cell network based on the E-cadherin channel. We collect 30 images at a time interval of 30 seconds for each embryo. The E-cadherin channel images are processed using the freeware ilastik for pixel classification. The resulting probability maps of pixels are processed using a customized MATLAB script for cell segmentation using a watershed algorithm. The mechanical inference is performed on the segmented image by imposing force balance at each vertex of the cell where junctional tensions add up to zero. We assume a homogeneous pressure distribution across the tissue based on the observation that the junctional curvatures are negligible in the ectoderm during the time window of observation, hence pressure does not enter the force balance equation.
The relative junctional tensions are obtained by fitting a tension triangulation network perpendicular to the corresponding cell network. This is termed the variational mechanical inference as the optimal tension network is obtained by minimizing an energy functional determined by the inner product of the tension network and the cell network 49 . Our analysis has shown that tensions inferred with this method correlate better with Vinc/E-cad ratio or Myosin density as compared to those estimated by a pseudo-inverse of the force balance matrix. The higher accuracy of the variational mechanical inference is due to the global fitting of both the magnitude and the orientation of the tension as opposed to the pseudo-inverse method, which only fits the magnitude of the tension while constraining the direction of the tension along the junctions. The flexibility in the orientation of the tension compensates for the potential errors in the segmentation of the cell, yielding higher accuracy for the mechanical inference. To guarantee the tension network to be a triangulation network, we kept only cells with three-fold vertices, which make up most of the cells in the population. We normalized the average inferred tension to be one. Shear stress on the E-cadherin clusters was obtained from a model of the junction (Fig. 6A) resulting from the transfer of tension from one side of the junction to the other. The shear stress takes the form τ=|(T1+T3)−(T2+T4)|/L, where 'L' is the length of the junction 48 .

Local correlation and conditional correlation
The correlations were performed by binning the junctions according to cell, termed local correlation ( Supplementary Figure 2A), or according to junctional length, termed conditional correlation (Supplementary Figure 2A). These two types of correlation are special cases of partial correlation, defined as the correlation between two random variables X and Y while . CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; holding the third variable Z constant, whose correlation coefficient is given by is the standard deviation of X (Y) at fixed Z, and | ( | ) is the average value of X (Y) at fixed Z. The partial correlation removes the spurious correlation between X and Y due to the confounding variable Z which is related to both X and Y.
The local correlation avoided the temporal and spatial variations of tension and fluorescence intensity, and yielded the correlation coefficient for each cell. We implemented the local correlation by computing the Pearson correlation coefficient between inferred tension and either the Vinc/E-cad ratio, Myosin intensity or Vinculin intensity among junctions within each cell. The resulting correlation coefficients were combined across all cells and multiple embryos to yield a distribution as shown in Figure 2H-J.
The conditional correlation avoided the spurious intensity-intensity and intensity-tension correlation resulting from the variation of junctional length, because both the intensities and the inferred tension are proportional to the inverse of junctional length ( Supplementary Fig.  4A). It also yielded the correlation coefficient as a function of junctional length ( Figure 6D and F, Supplementary Fig. 2C and 4C). We implemented the conditional correlation by sorting 10 junctions of the same length into the same bin. A linear correlation coefficient was computed among these 10 junctions with the same length for Vinculin and E-cadherin intensity ( Supplementary Fig. 2C), E-cadherin intensity and inferred shear ( Fig. 6D and F), junctional intensity and inferred tension ( Supplementary Fig. 4C). The binning was performed independently for each snapshot to avoid temporal and inter-embryo variations. Finally, we obtained the distribution of the conditional correlation coefficient by combining all the bins across time points and embryos ( Fig. 6C and E, Supplementary Fig. 2B and 4B). Quantifications in B and C come respectively from embryos expressing MyoII-GFP or embryos co-expressing Vinculin-mCherry and E-cadherin-GFP. Images/Quantifications in D-G come from embryos co-expressing Vinculin-mCherry and E-cadherin-GFP. Inferred tension junctional Myosin-II correlations were estimated in appropriately injected embryos co-expressing MyoII-mCherry and E-cadherin-GFP (data in H). Inferred tension Vinc/E-cad ratio correlations were estimated in appropriately injected embryos co-expressing Vinculin-mCherry and E-cadherin-GFP (data in I and J). The boxes in H, I and J represent 25 th percentiles to 75 th percentiles; the whiskers represent 5 th percentiles to 95 th percentiles; and the red lines represent the medians.

Figure3: Vinc/E-cad ratio represents the load on adhesion complexes.
Mild over-expression of E-cadherin reduces Vinc/E-cad ratio levels without changes in either the distribution of Myosin-II or the tension on junctions.
. CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; (A) Schematics showing the effect of E-cadherin over-expression on Vinculin recruitment. In the WT scenario, Vinculin recruitment is driven by the amount of tension generated by actomyosin contractility loaded on each adhesion complex. After E-cadherin overexpression, the same tension is supported by more adhesion complexes, leading to less Vinculin recruitment, which would result in an overall decrease of Vinc/E-cad ratio. All scale bars represent 5μm. Statistical significances were estimated using 'Student's t-test'. For the data in B', C', D', E', E'' and F n=7 embryos for both, control and UAS-E-cad::GFP. Images/quantifications in B, B' and E-E'' come from embryo co-expressing MyoII-mCherry and E-cadherin-GFP. Images/quantifications in C-D' and F come from embryo co-expressing Vinculin-mCherry and E-cadherin-GFP. In all cases, Junctions/pixels marked based on Ecadherin localization. ns, p>0.05; *, p<0.05; **, p<0.01; ***, p<0.001 The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; images showing the distribution of Myosin-II and Vinc/E-cad ratio, respectively, in the genetic outcross control embryos (left) and Gα12/13 over-expressing embryos (right). (E' and E'') Quantifications showing the increase in the levels of medial Myosin-II, without a significant change in the distribution of junctional Myosin-II. Data from n=7 embryos for both Gα12/13 over-expression and control. (D') Quantifications showing an increase in the mean junctional Vinc/E-cad ratio, without a significant effect on its planar polarized distribution. Data from n=7 embryos for control and n=6 embryos for Gα12/13 over-expression.
All scale bars represent 5μm. Statistical significance estimated using 'Student's t-test'. Images/quantifications in A-A'', C-C'' and E-E'' come from embryo co-expressing MyoII-mCherry and E-cadherin-GFP. Images/ quantifications in B-B', D-D' and F-F' come from embryo co-expressing Vinculin-mCherry and E-cadherin-GFP. In all cases, Junctions marked based on E-cadherin localization. Images/quantifications in B-B' are from the same set of embryos as those presented in Figure 1G-G''. ns, p>0.05; *, p<0.05; **, p<0.01; ***, p<0.001; ****, p<0.0001 All scale bars represent 5μm. Statistical significance estimated using 'Student's t-test'. Images/quantifications in all panels come from embryo co-expressing MyoII-mCherry and Ecadherin-GFP. In all cases, Junctions marked based on E-cadherin localization. ns, p>0.05; *, p<0.05; **, p<0.01; ***, p<0.001 The error bars in D and F represent the standard error across 100 different bins with the same length of junction. The boxes in B, C and E represent 25thpercentiles to 75thpercentiles; the whiskers represent 5thpercentiles to 95thpercentiles; and the red lines represent the medians. All scale bars are 5μm. Images/quantifications in B, E and E' come from embryos expressing E-cadherin-GFP. Images/quantifications in C, G and G' come from embryos expressing MyoII-GFP. Images/quantifications in D, H and H' come from embryos expressing Vinculin-GFP. Images/quantifications in F and F' come from embryos expressing α-Catenin-YFP.

Supplementary Figure2: Correlations analysis and Vinculin-E-cadherin conditional correlation
(A) Schematic representation to describe 'conditional correlation' and 'local correlation'. Various junctions are numbered as shown. Similarly colored junctions belong to the same length category and will be binned together for 'conditional correlation' analysis. Junctions of the same cell will be binned together for 'local correlation' analysis. Also, see Methods.
(B, C) Conditional correlation of Vinculin intensity with E-cadherin intensity conditioned on the length of the junction for wild-type and Rok inhibited embryos. The statistics are based on 35180 junctions across 6 embryos with 10 junctions in one conditional bin for both wild-type and Rok inhibited embryos.
The boxes in B represent 25 th percentiles to 75 th percentiles; the whiskers represent 5 th percentiles to 95 th percentiles; and the red lines represent the medians. The error bars in C represent the standard error across 100 different bins with the same length of junction. Vinculin-E-cadherin correlations in B and C were estimated for appropriately injected embryos co-expressing Vinculin-mCherry and E-cadherin-GFP. The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018; (similar to Supplementary Fig. 1B). The corresponding SNR images show all pixels with intensity>1. In such a representation, all cytoplasmic pixels are blank. Scale bar 5μm. (C) Quantifications showing the number of pixels in various E-cadherin pixel intensity bins. Each bin is 25 intensity units wide. The number of pixels is comparable between control and E-cadherin over-expressing embryos, across an order of a magnitude of pixel intensities. Number of pixels in a bin are separately estimated for individual embryos. Mean and SEM are calculated across embryos. Arrow indicates the minimum average number of pixels required to have meaningful statistics for individual embryos in both control and E-cadherin overexpressing embryos.
. CC-BY-NC-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/255158 doi: bioRxiv preprint first posted online Jan. 27, 2018;