Extensive analysis of fixed embryos indicates a link between PRG and TLR expression patterns, on the one hand, and anisotropic actomyosin organization, on the other hand (Zallen and Wieschaus, 2004; Paré et al., 2014; Munjal et al., 2015; Paré et al., 2019). The mechanism how TLR interaction recruits cytoskeletal components to adherens junctions however remains unclear (Paré and Zallen, 2020). Here, we extend this body of work by a quantitative, hypothesis-driven analysis of the dynamics of both myosin and PRGs at the whole embryo level during GBE. We digitally stitched data gathered from multiple live and fixed embryos to create a dynamic atlas comprising components of the anterior-posterior (AP) patterning system as well as myosin (Figure 1c and d), measured across the entire embryo (Mitchell et al., 2022). The time course of these gene products is provided with ∼1 min temporal resolution, starting from cellularization until the end of GBE (see Appendix for detail). We define $t=0$ to be the initiation of ventral furrow (VF) formation.

Throughout embryogenesis, egg geometry remains static (Figure 1b), defining a fixed reference frame that is described by a coordinate system parallel to the anterior-posterior (AP) and dorsal-ventral (DV) axes. All 3D renderings and whole-embryo cylindrical projections in this paper are oriented as in Figure 1b. We focus on junctional myosin at the apical surface (Paré and Zallen, 2020, Figure 1c), together with the PRGs Even-Skipped (Eve), Runt, Fushi-Tarazu (Ftz), Hairy, Paired (Prd), and Sloppy-Paired (Slp), to create a dynamic atlas of gene expression during GBE (Figure 1d). As reported previously, MRJs mainly align with the DV axis (Figure 1c, Bertet et al., 2004), while all of the PRGs we analyzed were expressed in a series of stripes occurring at regular intervals along the AP axis (Figure 1d, Irvine and Wieschaus, 1994; Clark and Akam, 2016). We mine this expression atlas to quantitatively test the relationship between the accumulation of myosin on junctions and cumulative PRG expression (see ‘Quantitative analysis of junctional myosin’).

Differences of PRG expression levels in adjacent cells have been proposed to underlie anisotropic myosin accumulation (Zallen and Wieschaus, 2004; Paré and Zallen, 2020). This provides a testable prediction relating PRGs to the accumulated signal and orientation of MRJs. We first computed local differences (gradients) of PRG expression levels, focusing on Runt and Eve – which are known to have the strongest impact on GBE (Irvine and Wieschaus, 1994) – as well as Ftz. Gradients were steepest along the AP axis, pointing towards the center of the stripes (Figure 1e). Across the germband, the magnitude of differences showed 14 regularly spaced stripes. While all PRGs are slightly out of register with one another (Clark and Akam, 2016), the six PRGs we analyzed had the following characteristics: (i) along the AP axis, gradients were strongest at the first and last stripe (Petkova et al., 2019), and (ii) expression levels reduced towards the dorsal pole (Figure 1f).

The profile of junctional myosin also demonstrated a DV gradient similar to the one observed for the PRGs, with a minimum on the dorsal pole (Streichan et al., 2018). However, along the AP axis, the myosin profile did not match any individual PRG gradient because they all have gaps in intensity between stripes (Figure 1c), while myosin does not. Therefore, we investigated whether combining the gradient profiles of multiple PRGs could produce a pattern consistent with the observed myosin accumulation. We used linear regression to compare the observed magnitude of junctional myosin with a weighted sum of PRG gradient patterns. The weighting parameters are adjusted to achieve the best possible agreement with the myosin profile (Equation 1).

Each PRG had its own parameter, which did not change over space or between different stripes. It was permitted to be positive or negative, representing a promotion or inhibition of myosin accumulation (for details, see ‘PRG gradient regression’). The PRG regression can also be seen as an analysis of the large-scale correlation between junctional myosin and PRG gradients, without making any a priori assumptions about the individual effects of each PRG.

The best fit produced in this way captured the observed DV modulation of myosin (Figure 1h and i). Along the AP axis, however, the patterns were quantitatively and qualitatively different. The model of local PRG differences predicted strong myosin accumulations at the anterior and posterior ends of the germband and a minimum in the center (Figure 1i). This did not fit the observed myosin localization pattern (Figure 1h). Equation 1 is the simplest possible way to model a potential relationship between myosin and PRG gradients. In the appendix (‘Nonlinear PRG-based model’), we analyze more complex, nonlinear regression models, and show that they also fail to account for the observed myosin pattern.

Our analysis suggests that local differences of PRGs cannot be linearly related to the amount of myosin accumulating on junctions, without postulating currently unknown additional complexity.

Anisotropic myosin distributions are characterized by both a local intensity – which did not correlate with PRG patterns – and a local orientation, to which we turn next. If myosin recruitment is instructed genetically according to time-independent rules, one would expect the angle between the myosin orientation and the gene stripes to be constant. Therefore, we analyzed the local orientation of MRJs across the whole embryo surface over time, which we compared to the concurrent orientation of local PRG stripes.

During GBE, both MRJs and PRG orientations are under the influence of significant tissue flow. As a consequence, their patterns could deform with the flow (i.e., passive advection behavior). We characterized the time course of the instantaneous flow field, which quantifies the global pattern of cell movements. We compared this quantification of flow with the temporal evolution of the localization patterns of PRGs and TLRs (Figure 2a–c’).

Figure 2

Download asset Open asset

Just after the initiation of gastrulation (7 min post VF initiation), Runt localization in the germband is characterized by seven stripes at different AP positions (Figure 2a and b, Appendix 1; Clark and Akam, 2016). Since this pattern of expression is stereotypic for the PRGs known to have the largest individual effects on GBE (Clark and Akam, 2016; Irvine and Wieschaus, 1994), we adopted Runt as our representative PRG. In Appendix 1—figure 10 and Figure 1d, we show that during GBE, the stripes of the PRGS Runt, Eve, Ftz, Paired, Sloppy-Paired, and Hairy remain parallel throughout GBE flow, and are transported by tissue flow in the same way. Therefore, it is sufficient to study the advection behavior of only one of them. Similarly, as a proxy for the TLRs, we chose Tartan (Paré et al., 2019), a leucine-rich-repeat receptor downstream of the PRGs that has been implicated in directing myosin anisotropy in concert with the TLRs 2, 6, and 8, due to the availability of a high-quality antibody.

15 minutes later (22 min post VF initiation), the PRG and Tartan stripes were strongly deformed, with the posterior-most Runt stripe (stripe 7) almost completely moved onto the dorsal side of the embryo. We also tested whether the dynamic expression of Tartans matched what is observed for PRGs. We found that the orientation of Tartan stripes (Paré et al., 2019), closely mirrored that of the Runt stripes (Figure 2a’ and b’, Appendix 1).

The reorientation of PRG and Tartan localization during GBE is reflected in a rapid decline of the autocorrelation of local stripe orientations (Figure 2f). To understand whether tissue flow could account for this deformation, we calculated six inter-stripe positions of the Runt pattern (Figure 2a, magenta lines, Appendix 1—figure 4), which we advected along cell trajectories computed from the velocity field, as measured by particle image velocimetry (PIV). The resulting advected inter-stripe positions (Figure 2a’, cyan lines) remained between the position of Runt stripes at 22 min. Manual tracking (Appendix 1—figure 5) confirms that cells that initially express Runt retain expression after 20 min. Together, this suggests that both PRG and Tartan patterns flow with the tissue frame of reference: their reorientation is quantitatively accounted for by advection due to tissue flow.

Next, we asked whether the velocity field behaved in the same manner. 7 min post VF initiation, a characteristic tissue flow pattern emerged, with four vortices and two hyperbolic fixed points (Streichan et al., 2018, Figure 2c). This characteristic flow pattern persisted during GBE, as highlighted by a very similar pattern observed 15 min later (Figure 2c’). While the magnitude of the flow clearly changed, its local direction was nearly constant, and vortices shifted only slightly over time. The autocorrelation of local tissue flow direction remained high throughout GBE (Figure 2e). Thus, the instantaneous flow field was nearly stationary for the entire duration of GBE: although cells travel long distances across the embryo surface, different cell neighborhoods that pass through a common spatial coordinate will move in the same direction, irrespective of the precise time point during GBE.

Tissue flow is known to be driven by anisotropic myosin (Bertet et al., 2004; Blankenship et al., 2006; Rauzi et al., 2008; Paré and Zallen, 2020). In fact, the flow field can be quantitatively predicted from the myosin distribution using a simple hydrodynamic model (Streichan et al., 2018). This suggests that a static myosin pattern should be required to produce the observed, nearly static, tissue flow field. Yet directional myosin recruitment by cell-intrinsic PRG patterning should lead to a continuous reorientation and advection of myosin anisotropy.

To resolve this discrepancy, we measured the orientation of MRJs and compared them to local Runt stripe orientations (Figure 2h–k’). The latter is defined by the direction of the Runt gradient, rotated by ${90}^{\circ }$. MRJs were detected using a segmentation-free method based on the Radon transform (Figure 2j and l, Streichan et al., 2018; see ‘Quantitative analysis of junctional myosin’). This method is robust even at low signal-to-noise ratios. The local myosin anisotropy direction (Figure 2k–k’) was defined by the intensity-weighted mean of MRJ orientations in a three-cell radius (see ‘Quantitative analysis of junctional myosin’). To link local angles with the organ-scale geometry, we measured angles with respect to the DV axis, which we defined geometrically by the direction of maximal curvature of the embryo surface.

The degree to which a given PRG stripe deforms over time is dependent upon its position along the AP axis. In general, the closer a stripe is to the posterior pole of the embryo, the more it will deform due to the strong posterior vortices. Therefore, we analyzed the relationship between myosin and PRG orientations on a per-stripe basis (Figure 2j, SI), using Runt as a representative PRG.

For Runt stripe 3, MRJs are parallel to the DV axis (Figure 2h, h’ and k), in accordance with previous reports (Zallen and Wieschaus, 2004; Paré et al., 2014). As highlighted by Runt stripe 6, in more posterior stripes we observed an increasing myosin/DV axis angle (Figure 2i, i’ and k’). In this region, 22 min into GBE, MRJ orientation was less streamlined than at the onset (Figure 2i’). Quantitatively, we found that the myosin/DV axis angle in Runt stripe 3 was small throughout GBE and aligned with MRJs (Figure 2k). In contrast, in Runt stripe 6, which rotated away from the DV axis due to its proximity to the posterior flow field vortex, MRJs reoriented away from the stripe (Figure 2k’). The global autocorrelation of myosin anisotropy orientation remained consistently high throughout GBE, indicating a nearly stationary pattern (Figure 2g), akin to instantaneous flow (Figure 2e). Finally, in Appendix 1—figure 15 we show that in accordance with the above analyses, the rate of change of the Runt pattern is much lower in the Lagrangian frame of reference that flows with the tissue than in a static frame of reference, whereas the opposite is true for the myosin orientation.

How to establish an instructive link between a continuously reorienting PRG pattern that moves with the tissue frame of reference, and the nearly stationary direction of both the myosin anisotropy and the flow field is not clear. The dynamic, orientational mismatch is independent of what type of linear or non-linear form of PRG-instructed myosin recruitment is posited. Our observations raise the question: what intracellular dynamic rules for myosin recruitment are required to establish the observed global myosin pattern?

To decode the nearly stationary myosin orientation and understand its residual dynamics, we studied the interplay of two dynamic effects: (i) at the subcellular level, the cytoskeleton can dynamically rearrange, due to binding and unbinding of molecular motors to the actin meshwork (Agarwal and Zaidel-Bar, 2019), and (ii) at the tissue level, advection will reorient junctions. We first studied myosin turnover using fluorescence recovery after photobleaching (FRAP). We photobleached individual junctions and measured the signal recovery for about 5 min (Figure 3a). In the first frame after bleaching (1.5 s), we measured approximately 50% signal reduction, indicating the presence of a fully mobile myosin subpopulation, possibly cytoplasmic (Wachsmuth, 2014). The recovery curve of $N=25$ junctions reflected multiple time scales and had a high standard deviation, possibly due to myosin oscillations on a time scale of ∼60 s (Figure 3a’, Gustafson et al., 2021). During the first 30 s after photobleaching, fluorescence recovery was rapid, as previously measured (Fernandez-Gonzalez et al., 2009; Munjal et al., 2015). After 30 s, the rate of recovery slowed, and pre-bleach levels were reached by 210 s, suggesting myosin molecular motors on junctions are dynamically recruited into the cortex and only bind transiently. See Appendix 1—figure 13 for a a mathematical model of the FRAP recovery curve.

Figure 3

Download asset Open asset

Next, we characterized the orientation of myosin anisotropy at the tissue scale. The angle between the local myosin anisotropy and the DV axis at every point of the germband over time is a 3D dataset that we broke down in several ways. We first analyzed the orientation of detected MRJs with respect to the DV axis as a function of time (Figure 3b). Consistent with the high autocorrelation of myosin orientation (Figure 2f), we found only a small change in the spatially averaged myosin/DV axis angle. Before the initiation of GBE, the myosin/DV axis angle was 2°. After the onset of GBE flow, measured at 10 min post VF initiation, the spatial average increased and reached a maximum of 8° at 25 min. Strikingly, after 25 min post VF initiation, the average myosin/DV axis angle began to decreases again, as the orientation of myosin anisotropy realigned with the DV axis. We refer to this phase of myosin behavior as the ‘recovery phase'.

We also carried out a regional, stripe-specific analysis to account for the fact that the tissue flow field responsible for advection varies across the germband (Figure 3c and Appendix 1—figure 9). Since PRGs advect with cells, we used the angle of individual stripes with respect to the DV axis to characterize local tissue-level reorientation, allowing us to test the degree to which myosin orientation is affected by advection. For Runt stripe 5 (Figure 3c), the Runt/DV axis angle was approximately 3° at the beginning of GBE and then increased monotonously starting at 10 min, exceeding 20° by 30 min post VF formation. The myosin/DV axis angle measured in the region of stripe 5 also increased after 10 min. This increase was notably less than that of the Runt/DV axis angle from which it clearly diverges by 15 min. By 28 min, myosin deflection away from the DV axis reached a maximum of 12°, followed by a recovery phase. The same analysis for runt stripe 3 showed neither of the two quantities accumulated an angle with the DV axis that exceeded 4° (Appendix 1—figure 9).

The two stripes, 3 and 5, are in spatially distinct regions of the tissue flow field. Runt stripe 3 coincides with the central region of the flow, where instantaneous flow and thus cell trajectories were mainly parallel to the DV axis (Figure 3d and e). Consequently, we expected the local sense of orientation to be conserved. Runt stripe 5 was passing through a vortex that we expected to rotate cell junctions (Figure 3d). To test this idea, we first characterized the rate of local rotation in degrees per minute due to tissue flow (Figure 3e) by calculating the vorticity (Figure 3e’, Landau and Lifshitz, 1987). At the whole embryo level, the spatial pattern of the vorticity had a broad peak of about 5° per minute around the domain of the posterior vortex and a small peak below 2° per minute on the anterior, vanishing elsewhere. We next characterized the spatial pattern of the orientation of MRJs with respect to the DV axis (Figure 3f). Across most of the embryo surface, MRJs were parallel with the DV axis, with a clear exception in a domain around the posterior vortex, where individual junctions deflected as far as 20° from the DV axis (Figure 3f’). This spatial correlation between vorticity and DV axis deflection also extends across time, with a 5 min delay between the time course of vorticity and the myosin angle defect (Figure 3b).

Next, we asked if the vorticity and the myosin/DV axis angle could be causally linked. Myosin stayed bound on junctions for an extended but finite amount of time (Figure 3a’). We postulated that myosin motors preferentially bind to junctions when they are parallel to the DV axis. Once these junctions are rotated by tissue flow, the myosin orientation is rotated with them, until motors begin to detach (Figure 3d). The resulting deflection angle is given by the product of rotation rate and myosin lifetime, which we reasoned could quantitatively account for the orientation defect of myosin in regions of high vorticity. We implemented this geometric source hypothesis in a mathematical model that describes the time course of the myosin concentration $m$ on a junction with orientation $\theta$:

The model assumes that (i) myosin binds to junctions that are parallel to the DV axis (the ‘static source’ is peaked around $\theta =0$), (ii) myosin unbinds with rate $1/\tau$, and (iii) that junctions rotate proportional to local vorticity. It is also possible to consider a model where the myosin recruitment rate is constant and, instead, the detachment rate is modulated according to junction orientation. We find that such a model cannot account for the observed deflection of myosin orientation (see Appendix, ‘Effects of modification of myosin lifetime by static source’). Further, in addition to tissue rotation, junctions could in principle be rotated by tissue strain. This is however not the case in the germ band because the majority of tissue strain is due to cell rearrangement. Junction shortening or lengthening also does not affect junction orientation. For a detailed discussion, see the Appendix , ‘Additional effects in static source model’, where we also discuss the influence of tissue curvature on junction rearrangement . Finally, related ideas have previously been proposed in Farrell et al., 2017 on a qualitative level.

Equation 2 for a single junction can be transformed to describe the entire angular distribution of junctional myosin (see Appendix I.7) and used to predict the local average deflection angle $\overline{\theta }$ of the myosin anisotropy:

Crucially, $\overline{\theta }$ only depends on the direction around which recruitment is maximal and is independent of its magnitude, potential spatial modulation, and functional form (see Appendix I.7), in agreement with the reasoning presented above. The only free parameter is the effective myosin lifetime $\tau$, which captures the duration that myosin motors remain bound on junctions (see Appendix I.7). Our model makes several quantitative predictions about junction dynamics that we tested (Figure 3b–c and e–i).

First, we computed the predicted angle between the orientation of myosin anisotropy and the DV axis by solving Equation 4 (see Appendix for mathematical definition). We assume that the parameter $\tau$ is constant in time and across the embryo. Using an ensemble of $N=5$ embryos, we then fitted the parameter $\tau$ by minimizing the average difference between the myosin orientation predicted from Equation 4 and the observed orientation across the germband during convergent extension. We refer to the angle between the vorticity-based prediction and the actual myosin angle as the ‘rotation-rate adjusted’ myosin/DV axis angle. We found that the vorticity-adjusted junction orientation aligned well with the DV axis when the duration $\tau$ of myosin binding was 5 min (Figure 3c and g). Throughout GBE – even in regions of highest vorticity – the rotation-rate-adjusted angle remained close to 0° , while the discrepancy between the Runt and myosin orientations continuously increased (Figure 3c and Appendix 1—figure 9). Figure 3b shows the predicted spatial average myosin/DV axis angle (see Appendix I.7.1), which is in good agreement with the observed data.

The value of $\tau$ found by our fit is qualitatively similar, but somewhat larger, than the FRAP-measured myosin lifetime. As discussed in the Appendix, this is because FRAP measures the time individual motors remain on a junction, while $\tau$ measures how long the total myosin level on a junction persists. The latter time can be larger if factors that affect myosin levels, such as kinases, are longer-lived than individual motors, or if there is a positive feedback of current myosin levels on myosin recruitment.

Second, we analyzed the angular distribution of MRJs, i.e., the range of orientations of MRJs detected in a small tissue patch, and measured its spread (Figure 3h–h” and Appendix 1—figure 16). The dynamics of the angular distribution depended on location within the embryo. The standard deviation remained nearly constant in regions of low vorticity, for example, around Runt stripe 3 (Figure 3h, Figure 2h–h’). In regions of high vorticity, the standard deviation rapidly changed, giving rise to a much broader distribution (Figure 3h’, Figure 2i–i’). This feature is accurately captured in our model: without vorticity there will be no reorientation, and only junctions parallel to the DV axis will recruit myosin (Figure 3h”). With vorticity, junctions that recruited myosin while aligned with the DV axis will rotate. Since myosin stays bound for a extended but finite lifetime, the distribution widens (Figure 3h”).

Third, we analyzed the time course of vorticity. It increased with the onset of GBE, reached a plateau by 20 min, and slowed down by 25 min (Figure 3b). Strikingly, MRJs realigned with the DV axis once vorticity decreased (Figure 3b and i), as predicted by Equation 3. The time delay between vorticity and deflection (Figure 3b) is expected from our model: it takes time to deflect junctions in response to vorticity and, correspondingly, for myosin to detach from deflected junctions once vorticity decreases. At 25 min, MRJs in a posterior region on the ventrolateral side of the embryo were strongly deflected, with a median angle of about 20° to the DV axis (Figure 3i). 10 min later, the median of the distribution shifted by 21° , aligning with the DV axis. The dynamics of the distribution during recovery – i.e., while vorticity decreases – mirrors the dynamics during the onset of vorticity, supporting the notion that the myosin angular distribution is mainly determined by the strength of the local vorticity. Indeed the simulation in Figure 3h” matches the observations in both Figure 3h’ and Figure 3i (of course, the magnitude of vorticity change in Figure 3i and h’ is different, leading to quantitative differences in the myosin orientation). These observations were consistent with the geometric source hypothesis, both qualitatively and quantitatively (Figure 3c).

Our mathematical model accounts for the dynamics of myosin orientations in terms of the rotation due to flow, and the extended but finite binding lifetime of myosin motors to junctions. The model accurately describes key features of the spatiotemporal dynamics of the mean as well as standard deviation of the distribution of MRJ orientations. The model’s single parameter is the same across the embryo, constant in time, and agrees qualitatively with recovery kinetics as measured on individual junctions experimentally. Our global, comprehensive analysis confirms and extends previous local, tracking-based observations that showed that junctions can remodel their myosin levels as they rotate (Farrell et al., 2017). These results point towards control of myosin orientation by static geometric cues, as opposed to the passively advected PRGs and Tartan stripes.

Harnessing the genetic toolkit available in the Drosophila model system, we show that we can account for the behavior of mutants by modulating parameters of our mathematical model – vorticity, myosin kinetics, and geometry (Figure 4a–a”). Twist is expressed in the VF and ${\displaystyle {twist}^{ey53}}$ mutants have a defect in VF formation, accompanied by reduced kinetics across the entire embryo (Martin et al., 2009; Butler et al., 2009; Streichan et al., 2018; Gustafson et al., 2021). We found the average speed of GBE in ${\displaystyle {twist}^{ey53}}$ mutants was reduced by a factor of two compared to WT (Figure 4b). This was accompanied by a corresponding reduction of the vorticity (Figure 4a and c). The myosin/DV axis angle was likewise smaller (Figure 4a and c). Fitting the model to this data revealed a similar myosin binding lifetime as in WT, and the rotation rate-adjusted angle of MRJs closely aligned with the DV axis.

Figure 4

Download asset Open asset

Analysis of fixed samples has demonstrated that during GBE, myosin polarization in PRG and TLR mutants is reduced (Paré et al., 2014). GBE tissue flow in these mutants is impaired as well, particularly in later phases (Irvine and Wieschaus, 1994). Our dynamic data indicates that myosin and PRGs/TLRs are regulated in different frames of reference. This raises the question of how patterning gene expression can be quantitatively linked to myosin anisotropy and – by extension – tissue flow. To study this, we performed live imaging of myosin in ${\displaystyle {eve}^{R13}}$ mutants (Figure 4a’, b, d and e). We found that the initial kinetics of GBE closely match that of WT embryos. However, at 10 min post VF formation the kinetics of ${\displaystyle {eve}^{R13}}$ mutants change abruptly (Figure 4b). Additionally, vorticity was reduced compared to WT (Figure 4d). We detected anisotropic myosin in the germband during GBE (Figure 4e), although anisotropy was significantly reduced in comparison to WT (Appendix 1—figure 18). We found that MRJs were mainly aligned with the DV axis, except in the region of the posterior vortex (Figure 4a’). The best fit of the static source model to this data suggests that myosin binding lifetime is significantly reduced in ${\displaystyle {eve}^{R13}}$ mutants (to $\tau =2-3\text{min}$). We therefore carried out FRAP experiments on junctional myosin in ${\displaystyle {eve}^{R13}}$ mutants (Appendix 1—figure 11). The FRAP data shows complex modifications to the myosin dynamics, but is compatible with overall more rapid myosin dynamics in ${\displaystyle {eve}^{R13}}$ compared to WT embryos (see Appendix 1—figure 11). A joint reduction of vorticity, and myosin lifetime can account for the near constant time course of the MRJs angle defect with the DV axis (Figure 4d).

Finally, by knocking down the atypical cadherin Fat2 (Fat2-RNAi) in somatic ovarial cells in female flies, we created nearly spherical embryos (Chan et al., 2019) with a shorter, but variable length, and up to 30% extended DV circumference (Figure 4f, top). We originally planned to use Fat2-RNAi to modify the direction of the proposed static source, the geometric DV axis. However, since Fat2-RNAi eggs remain highly rationally symmetric about the AP axis, the direction of the geometric DV axis is not significantly affected. Instead, this mutant provided an opportunity to test if PRG stripes would change in the same way as MRJs around the ectopically extended DV circumference. The AP patterning system remains intact in Fat2-RNAi embryos (Chan et al., 2019): PRGs were expressed in seven stripes along the AP axis, with a decrease in expression at the dorsal pole (Figure 4f, right). Similar to WT, myosin recruitment to junctions was strongest in ventral regions and dropped markedly on the lateral side. In both WT and Fat2-RNAi embryos, MRJs were detected up to ∼175 μm away from the VF in the lateral ectoderm (Figure 4g–g’). Strikingly, since the absolute length of the DV circumference is larger in Fat2-RNAi embryos than it is in WT embryos, and the PRG stripes extended normally to the dorsal pole, there is a substantial region on the lateral surface of Fat2-RNAi embryos where PRG stripes are clearly visible but no myosin anisotropy could be detected (Figure 4f). The magnitude of tissue flow in Fat2-RNAi was reduced (Appendix 1—figure 19). Moreover, MRJ orientations changed little due to low spatial overlap between the regions of high vorticity and myosin recruitment (Appendix 1—figure 20). These observations suggest that presence of striped PRG expression is not sufficient to set up myosin anisotropy.

One possibility is that PRGs play a role in directing myosin anisotropy in an initial phase, with control over myosin orientation transferred to a static cue once flow starts. However, this hypothesis is not supported by the lack of linear correlation between PRGs and the initial myosin pattern, the results in Fat2-RNAi embryos, and the results of Gustafson et al., 2021, which found that mechanical cues can explain the early myosin pattern. Taken together, our results suggest that instead of directly instructing anisotropic myosin recruitment, PRGs might influence the myosin anisotropy by regulating retention of myosin to junctions.

Images recorded by the lightsheet microscope were registered based on the position of fiduciary beads embedded in the agarose (Fluoresbrite multifluorescent 0.5 μm beads 24054, Polysciences Inc, as described in Gustafson et al., 2021) using the Multiview reconstruction plugin (Preibisch et al., 2014) in Fiji (Schindelin et al., 2012). We used all-to-all registration, mapping all perspectives at all time points to a common reference frame using an affine transformation. Images were then deconvolved and fused using the algorithm introduced in Preibisch et al., 2014, yielding images with an isotropic resolution of 0.2619 μm.

The embryo surface is detected within the resulting volumetric data using an Ilastik detector (Berg et al., 2019), to which a surface was fitted using the ImSAnE software (Heemskerk and Streichan, 2015), which was used for tissue cartography as described in Heemskerk and Streichan, 2015. To improve accuracy, we applied two iterations of the Ilastik + ImSAnE workflow. The resulting ‘onion’ layers normal to the embryo surface, spaced 1.5 μm, were used to generate maximum-intensity projections.

Imaging for the FRAP experiments shown in Figure 3 and the ${\displaystyle {eve}^{R13}}$ data shown in Appendix 1—figure 17 was done using a Leica SP5 confocal microscope and ×63/1.4 NA oil immersion objective at a frame rate of one frame per 1.78 s. Junctions were tracked manually using Fiji (Schindelin et al., 2012). We bleached regions of size $5\mu \text{m}\times 5\mu \text{m}$ for approximately 8 s using 50 mW laser power.

A full stock list is presented in Table S1. The fluorescent fusion proteins used in this study include Myosin::GFP (II or III, sqh::GFP, Royou et al., 2002), Myosin::mCherry (II, sqh::mCherry, Martin et al., 2009), Runt::LlamaTag-GFP (Bothma et al., 2018, gift from H. Garcia), Eve::YFP (III, Ludwig et al., 2011), Gap43::mCherry (III, Membrane::mCherry, Martin et al., 2010).

Recombinant chromosomes containing the chromosomal deficiency Df(2L)dpp[s7-dp35] 21F1–3;22F1–2 (halo) and either ${\displaystyle {eve}^{R13}}$, or ${\displaystyle {twist}^{ey53}}$ were balanced with CyO. The chromosome containing ${\displaystyle {eve}^{R13}}$ was recombined with Myosin::mCherry (II). Halo, ${\displaystyle {twist}^{ey53}}$ embryos were balanced with a version of CyO that also contains Myosin::GFP. Homozygous ${\displaystyle {eve}^{R13}}$ and ${\displaystyle {twist}^{ey53}}$ embryos were identified based on visualization of the halo phenotype while heterozygous control embryos did not show the halo phenotype.

The following stocks were used to generate reduced aspect ratio (Fat2-RNAi) embryos: w; Traffic jam-Gal4; Myosin::GFP; Gap43::mCherry, w; Myosin::GFP; UAS-Fat2-RNAi (Chanet et al., 2017).

For heat fixation, embryos were dechorionated with 50% bleach and then fixed using heat and methanol as described previously (Müller and Wieschaus, 1996). Primary antibodies for immunohistochemistry were Runt (guinea pig, 1:500, Wieschaus Lab), Even-Skipped (rabbit, 1:500, gift from M.Biggin), Fushi-Tarazu (rabbit, 1:1000, gift from M.Biggin), Paired (mouse, 1:100, gift from N.Patel), Sloppy-Paired (rabbit, 1:500, gift from M.Biggin), Hairy (rat, 1:100, Wieschaus Lab), and Tartan (rabbit, 1:100, this study, GenScript, based on full-length peptide). Donkey and goat secondary antibodies conjugated to Alexa Fluor 488, 561, and 647 were used (1:500, Thermo Fisher Scientific). Embryos were mounted in 1.5% low gelling temperature agarose (Millipore Sigma-Aldrich) for light-sheet imaging, and mounted in 50% PBST 50% Aqua-Poly/Mount (Polysciences) for confocal imaging.

Image processing, described in detail in the SI, used a combination of custom Python scripts using the Scientific Python (Virtanen et al., 2020) and scikit-image (van der Walt et al., 2014) packages and custom MATLAB scripts. These scripts are available at Claussen and Streichan, 2022. Tissue cartography was performed using the ImSAnE software (Heemskerk and Streichan, 2015). Surface detection, cell tracking, and segmentation of Runt stripes were performed using Ilastik (Berg et al., 2019).

In this section, we collect some additional data on the behavior of the genetic pattern of PRGs and TLRs as well as on the myosin orientation. Appendix 1—figure 3 shows a quantification between the angle of the PRG Runt and the TLR Tartan, complementing Figure 2A in the main text. Appendix 1—figure 4 shows that the PRG Eve is advected by tissue flow in the same way as Runt (see Figure 2A). Appendix 1—figure 5 shows single-cell tracking data confirming the Lagrangian behavior of Runt.

Appendix 1—figures 6–8 show additional data on the local angular distribution of myosin and Runt (see also Appendix 1-figure 16). Appendix 1—figure 6 complements Figure 2F by showing histograms of both the local myosin and the Runt orientation in the ventro-lateral region of the germband. In comparison to Figure 2F, where for simplicity all myosin junctions in the region under investigation were combined in a histogram, here, we proceeded in a two step fashion. The region was scanned with a $30\mu \text{m}\times 30\mu \text{m}$ window, for each window the mean orientation was calculated and subtracted from the angles of the junctions in the window, and the resulting distributions were joined. This process permits separating local spread from larger scale spatial gradients. Appendix 1—figure 6 also shows the corresponding distributions for the Runt stripe orientations in the same region. Not only is the spread much smaller, the distribution also changes much less. The total spread of the Runt stripe distribution depends on the smoothing parameters used to compute the Runt gradient, but the dynamics of the distribution does not. Appendix 1—figure 7 shows the spatial pattern of the local Runt angular distribution across the germband. Note the strong dissimilarity with the corresponding figure for myosin, Appendix 1—figure 16. Finally, Appendix 1—figure 8 carries out a embryo-scale comparison of the local Runt and myosin distributions, as illustrated in Appendix 1—figure 6, using the entire ensemble of 10 embryos available to us. We compute the Kolmogorov–Smirnov, a measure of distance between two probability distributions (the maximal area between the two probability densities), of the local angular distributions across the germband, finding high values strongly indicative of disagreement between the two. Each window typically contains more than 100 junctions.

Appendix 1—figure 4

Download asset Open asset

Appendix 1—figure 5

Download asset Open asset

Appendix 1—figure 6

Download asset Open asset

Appendix 1—figure 7

Download asset Open asset

Appendix 1—figure 8

Download asset Open asset

Appendix 1—figure 9 completes the argument of Figure 3 by showing the time course of PRG and myosin orientation as well as our prediction for the myosin orientation in the regions of all seven Runt stripes.

Appendix 1—figure 9

Download asset Open asset

Appendix 1—figure 10

Download asset Open asset

To obtain the FRAP curves shown in Appendix 1—figure 3 from the raw confocal images, we used a two-step normalization procedure (Wachsmuth, 2014). The average raw intensity in the region of interest (ROI) containing the bleached junction will be denoted ${\displaystyle I_t}$. ${\displaystyle t}$ denotes time, with $t=0$ being the first frame after bleaching. The bleached region is monitored for a total time $T$. In addition to the ROI containing the bleached junction, we also monitor three additional $5\mu \text{m}\times 5\mu \text{m}$ control regions that are neither bleached nor contain any junctions. Their average defines the background intensity $I_{\text{bg},t}$.

In a first step, we normalize the ROI intensity by the background, which also corrects for photo-bleaching, defining ${\tilde{I}}_t=I_t/I_{\text{bg},t}$. Next, the pre-bleach intensity ${\tilde{I}}_{\text{pre}}$ is the average normalized intensity in the ROI before bleaching. We define the FRAP signal as

Note that the FRAP signal at $Y_{t=0}$ at the first frame recorded after bleaching is not zero. This is for two reasons: incomplete bleaching and the inevitable delay between the first frame and the end of bleaching. During this 1.5 s delay, non-bleached cytosolic myosin diffuses into the bleached region. Based on the FRAP curve $Y_t$, the myosin signal can be divided into three fractions:

• The fully mobile/incompletely bleached fraction Y₀.

• The slowly mobile fraction $Y_T-Y_0$ (signal recovered by the end of observation).

• The immobile fraction $1-Y_T$ (signal not recovered by the end of observation).

Previous work Munjal et al., 2015 used a different definition of the FRAP signal, namely

This definition leads to systematic underestimation of fluorescent recovery and overestimation of the immobile fraction, as can be seen from an example. Consider 90% effective photo-bleaching so that 10% of the pre-bleach signal is still present at the first time point post-bleach. To obtain a complete recovery using the ${\widehat{Y}}_t$-measure would therefore require a 10% increase in fluorescent intensity over pre-bleach levels.

We carried out FRAP measurements (Wachsmuth, 2014) as described above in two of the mutant genotypes described in the main text, ${\displaystyle {eve}^{R13}}$ and Fat2-RNAi.

In ${\displaystyle {eve}^{R13}}$, myosin dynamics deviates from WT in several respects. First, we observe increased variability in FRAP recovery across junctions. Because of this high variability, we are not able to resolve features in the FRAP curve later than 3 min. post bleaching, limiting the time over which we can compare it to WT. Second, we carried out two subsequent FRAP experiments for each embryo. When we stratify FRAPed junctions according to whether they were imaged first or second, we see that myosin recovery is incomplete in ‘late’ embryos, potentially because myosin recruitment has ceased (Appendix 1—figure 12). This is consistent with the slowdown of flow in ${\displaystyle {eve}^{R13}}$ after the onset of tissue flow in Figure 4B. We therefore only include the ‘early’ experiments in Appendix 1—figure 11. Finally, in the ${\displaystyle {eve}^{R13}}$ FRAP experiments, we achieved more complete bleaching (down to ~35% vs. 55% in WT). The ${\displaystyle {eve}^{R13}}$ and WT experiments were carried out 1 year apart, during which our experimental facilities have improved. When comparing the recovery kinetics, we are however interested in the ratios of fluorescent intensities at different timepoints (which then yields the recovery rate), so that the bleaching completeness drops out of the final result.

With all these caveats in mind, Appendix 1—figure 11 shows the FRAP curve obtained for ${\displaystyle {eve}^{R13}}$ embryos. We observe that within 2 min post bleach, myosin intensity has recovered from 35% of pre-bleach intensity to 90%. This corresponds to a recovery rate of ~$2.4{\text{min}}^{-1}$ (interval corresponding to a 50% increase in fit error vs. the minimum: $1.95-2.9{\text{min}}^{-1}$). To obtain the recovery rate $r$, we model the FRAP curve by a single exponential:

where $t$ is the time in minutes. A direct comparison to the WT FRAP curve is however problematic since the WT FRAP curve does not appear to follow a simple exponential recovery (Appendix 1—figure 13). Instead, the WT recovery appears biphasic, with a very rapid initial and a slower final phase, as described by the following equation:

Here, c₀ and c₁ parameterize the initial fast recovery. The second, slow recovery occurs with a rate of ~$1.4{\text{min}}^{-1}$, and therefore slower than in ${\displaystyle {eve}^{R13}}$. Since junction rotation happens at a rate of ~1°/min, it is the slower recovery rate that is relevant for the myosin orientation behavior. Along the same vein, if we fit the single-exponential model to the WT data, ignoring the initial fast recovery for the purpose of the fit, we obtain a recovery rate of $1.9{\text{min}}^{-1}$. This suggests that the data is compatible with somewhat faster myosin kinetics in ${\displaystyle {eve}^{R13}}$ than in WT. However, we caution that the necessary difference modelling of the FRAP curve between the two genotypes renders the recovery rates hard to compare directly.

Finally, we also recorded FRAP curves in Fat2-RNAi mutants, which we present in Appendix 1—figure 14. Here, the FRAP curve resembles the biphasic behavior seen in the WT, with rapid early recovery. This is in accordance with the fact that the anterior-posterior patterning in these mutants remains intact (Chanet et al., 2017). The measurement error in Fat2-RNAi mutants was higher than in WT since the embryo only carried a fluorescent myosin marker, and not also a membrane marker.

Appendix 1—figure 11

Download asset Open asset

Appendix 1—figure 12

Download asset Open asset

Appendix 1—figure 13

Download asset Open asset

Appendix 1—figure 14

Download asset Open asset

Here, we present some additional data on the advection behavior of Runt stripes and myosin orientation. Instead of plotting the accumulated angle between Runt stripes/the myosin orientation and the DV axis, we can also analyze their rates of change. In Appendix 1—figure 15 ,we compare the convective derivative, which measures the rate of change in the Lagrangian reference frame (flowing with the tissue), and the partial time derivative, which measures the rate of change in a fixed reference frame. This comparison shows that the Runt changes much less when viewed in the Lagrangian frame of reference, whereas for the myosin orientation the opposite is true.

Appendix 1—figure 15

Download asset Open asset

In this section, we give details on the model used to test for correlations between the tissue-scale myosin pattern and the PRGs in Appendix 1—figure 1. As explained in the main text, previous work suggests that myosin is recruited to junctions between cells with differing levels of PRG expression. These cell-cell differences can be measured across the entire embryo by computing the gradient of the measured fluorescent intensity. The gradients are used as inputs in a linear regression model with the goal of predicting the myosin pattern.

Here, we give a step-by-step description of this process. We enumerate the different PRGs (e.g., Eve, Runt, and Ftz) with an index $i=1,2,\mathrm{\dots }$. The level of expression of a particular gene at a position $\mathbf{x}$ on the embryo surface is a scalar function ${\varphi }_i(\mathbf{x})$. Before computing the gradients, the functions ${\varphi }_i$ are convolved with a Gaussian kernel $K_{{\sigma }_1}$ with standard deviation ${\displaystyle {\sigma }_1\approx 1}$ cell diameter. This ensures that when computing the gradient, we measure differences between cells and not spurious differences between the levels within and without a cell’s nucleus. We denote the result $K_{{\sigma }_1}*{\varphi }_i$. Next, we compute the magnitude of the gradient of the PRG fields, ${|\nabla (K_{{\sigma }_1}*{\varphi }_i)|}^2$. Since we consider the germband away from the posterior pole, where the embryo is effectively cylindrical, $\nabla$ is computed using ordinary partial derivatives. The gradient magnitude field is then smoothed again, with a second kernel of standard deviation ${\sigma }_2\approx$ 4 cell diameters. This is done for two reasons: First, we are interested in predicting the embryo-scale myosin distribution. Second, smoothing over small scales will reduce errors, in particular those due to imperfect alignment of data from different embryos. Smoothing makes the model more generous, as can be seen from the limit case where both the myosin and the PRG patterns are completely smoothed out - the linear regression would report perfect correlation.

The linear regression model for the myosin magnitude takes the following form:

Myosin magnitude is modeled by a linear superposition of PRG gradients with weights, each gene acting independently. The weights ${\alpha }_i$ represent the strength of each gene and are chosen by least-squares minimization. Note that the both the background levels of the PRG fluorescent signal and their image contrast are irrelevant since the former produces 0 gradient and the latter is absorbed by the fit weights. The model can be extended to allow cross terms:

Cross terms model recruitment of myosin specifically to edges between two different PRG gene stripes. In our model, we consider three PRGs, Eve, Runt, and Ftz and allow for both linear terms and an Eve/Runt cross term.

Both the myosin and the PRG patterns depend on time $t$. This raises a potential complication since the myosin pattern is would be expected to react to the PRGs with a delay whose duration is unknown. We carried out the linear regression to fit the myosin pattern at time $t_{\mathrm{myosin}}=10$ min post VF formation when the characteristic GBE myosin pattern has already been established, but no significant deformation due to tissue flow has yet occurred. The PRG gradients are evaluated at an earlier time ${\displaystyle t_{\mathrm{P}\mathrm{R}\mathrm{G}}<t_{\mathrm{m}\mathrm{y}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{n}}}$ that we can vary as an additional fit parameter. However, the PRG pattern changes little in the 20 min preceding t_myosin and we do not find a good agreement between the PRG-based model and the myosin distribution for any value of $t_{\mathrm{PRG}}$.

In this section, we supply a quantitative model describing the behavior of junctional myosin under the hypothesis of recruitment by a static source, supporting the argument made in section ‘Results.’ During tissue flow with velocity $\mathbf{v}$, cell junctions rotate with a rate equal to one-half of the vorticity $\omega =\nabla \times \mathbf{v}$ (Landau and Lifshitz, 1987). Note that the tissue shear does not significantly reorient cell junctions: since GBE proceeds primarily by cell intercalation, cell shapes are not changed by the tissue-level shear. Further, junction lengthening and shrinkage do not affect junction orientation. We weight junctions by their average myosin intensity, i.e., total myosin signal divided by junction length, so that junction length does not affect the weighting.

We assume that myosin is recruited to a static source whose orientation of the source is defined geometrically: it is parallel to the direction of maximal curvature of the embryo surface. This defines the DV axis for the purpose of the measurements reported in Appendix 1—figures 2–4. Myosin is assumed to be recruited to edges aligned with the source and to detach from all edges with a constant rate $1/\tau$. These assumptions can be encoded into equations in different, equivalent ways. In main text Equation 2, we considered a single junction, below we consider the local myosin nematic tensor constructed from detected MRJ in a small tissue patch, and in Sect. I.7.1 we consider the entire angular distribution of myosin.

This yields the following equations for the local myosin tensor $Q_m$ as defined in Sect. I.1.4:

Here, $\mathrm{\Omega }$ is the vorticity matrix ${\mathrm{\Omega }}_{ij}=({\partial }_i{v}_j-{\partial }_j{v}_i)/2$ and $\mathrm{\Gamma }$ is the source tensor, parallel to the DV axis. Advection has been neglected (see Sect. I.7.2). In the steady state, this equation is solved by

where $R_{\omega \tau /2}$ represents a rotation by angle $\omega \tau /2$. The steady state is a good approximation for most of GBE, since after flow onset, the velocity pattern flow is relatively steady compared to the myosin lifetime $\tau \approx 5\text{min}$.

Crucially, Equation 16 shows that the dominant eigenvector of $Q_m$, i.e., the direction of myosin anisotropy, is given by the dominant eigenvector of $\mathrm{\Gamma }$, rotated by an angle of $\omega \tau /2$. This prediction is completely independent of the magnitude of the source $\mathrm{\Gamma }$!

We note that the effective lifetime $\tau$ in the model Equation 15 is not necessarily equal to the time ${\tau }_{\text{bound}}$ an individual myosin motor remains bound to a junction (as measured by FRAP, for example). This can be seen by considering an example scenario in which myosin motors detach extremely rapidly, but the myosin concentration on an edge is controlled by a long-lived actor up the regulatory chain setting, for example, the rate of myosin phosphorylation. Due to the possible persistence of actors upstream of junctional myosin, the effective lifetime $\tau$ is larger or equal to the bound time of individual motors. This is a possible explanation for the discrepancy between the FRAP measured ${\tau }_{\text{bound}}$ and the inferred effective $\tau$ in Figure 3.

Prediction of variance of myosin angular distribution

Based on our model, we can predict both the mean myosin orientation as well as the variance of the angular distribution. In the main text, we presented simulated histograms (according to Equation 17), which show striking qualitative agreement of the predicted behavior and the observed broadening of the angular distribution once vorticity sets in. In the tissue patch shown in the main text (Appendix 1—figure 2), we find a change of the mean of ${\mu }_2-{\mu }_1={16}^{\circ }$ and a change in variances of $\sqrt{{\sigma }_2^2-{\sigma }_2^2}={17}^{\circ }$ between the two timepoints analyzed, in line with the prediction ${\mu }_2-{\mu }_1=\sqrt{{\sigma }_2^2-{\sigma }_2^2}$ of Equation 20. In this section, we check the prediction of Equation 20 on the entire ensemble of $N=5$ embryos and over the entire germband.

However, quantitatively, the behavior of the variance is more complicated than that of the mean angle. The variance depends on the strength of the anisotropic myosin recruitment compared to the isotropic background as well as on the myosin signal-to-noise ratio, which is not the case for the mean. If the strength of the anisotropic source or the overall levels of junctional myosin (and hence the signal-to-noise ratio) decreases, as they do towards the end of GBE, the variance will increase, even in the absence of vorticity. On the other hand, a transient infcrease of the anisotropic recruitment leads to a decrease in variance, as is observed during the strong increase in myosin anisotropy due to strain generated by the ventral furrow during early GBE. Therefore, the vorticity effect Equation 20 only represents one contribution to the variance. Further, the local variance depends on the size of the local tissue patch queried. If this size is chosen too large, tissue-scale gradients (e.g., of vorticity) contribute and increase the variance. If it is chosen too small, the ensemble calculation stops making sense.

Below we show both the temporal and spatial correlation of the measured local standard deviation (in 50 μm× 50 μm windows) and the vorticity-based prediction in the germband. We exclude a strip of 50 μm width around to the VF from the analysis since here the pulling effects of the VF dominate. We take the reference time for the initial variance at $t=10$ min post VF initiation when the anisotropic myosin pattern is mostly established. The results of the prediction of Equation 20 are shown in Appendix 1—figure 16. The most important aspects of the spatial and temporal behavior of the variance are explained by the vorticity model.

Appendix 1—figure 16

Download asset Open asset

Principal axis of embryo-scale tension and turgor pressure

As mentioned in the ‘Discussion’ section, one possible candidate for the statically oriented myosin source is epithelial tension that myosin dynamics is known to be sensitive to Fernandez-Gonzalez et al., 2009. The direction of tension agrees with the direction of the inferred myosin source. Indeed, epithelial tension in the germband is known to be strongly anisotropic from laser ablation experiments, with higher tension on junctions parallel to the DV axis (Munjal et al., 2015). Noll et al., 2020, using an image-based force inference algorithm, confirmed that on the scale of the entire embryo, the epithelial tension aligns with the geometric DV axis, even after the onset of tissue flow. Noll et al., 2020 also found that the distribution of junctional myosin closely matched the epithelial tension. Strikingly, most junctional myosin is balanced: ∼80% of junctional myosin is involved in static force balance, i.e., it creates a net-zero local force.

Previous work cell-scale literature, e.g., Munjal et al., 2015, presented the anisotropic tension as a consequence of the anisotropy of junctional myosin. However, to set a static myosin orientation, the epithelial tension cannot be a pure readout of the current myosin distribution. One potential static contribution to tension anisotropy is the turgor pressure difference between the yolk within the blastoderm and the perivitlline space outside of it (e.g., visible during dorsal closure; Lu et al., 2016). This normal pressure is balanced by epithelial surface stress, much like the excess pressure in an inflated balloon. Pressure, stress, and geometry are linked by the Young–Laplace law. Due to the embryo’s cylinder-like geometry, the resulting surface stress is anisotropic: in a pressurized cylinder with closed ends, the stress along the azimuthal axis is twice the stress along the height axis of the cylinder (Audoly and Pomeau, 2010). Fairly generally, excess internal pressure leads to anisotropic stress parallel to principal axes of curvature (Audoly and Pomeau, 2010; Deserno, 2015). Interestingly, blastodermic turgor pressure leading to cortical tension has already been shown to play a crucial role in mouse blastocyst development (Chan et al., 2019)

Finally, theoretical work, Noll et al., 2020 (recently experimentally validated in Gustafson et al., 2021) has shown how epithelial tissue can support static tension even during viscous flow. Strain-rate based recruitment can drive junctional myosin to a balanced state (with zero net local forces), such as that required to balance turgor pressure.

Additional data on mutants

Appendix 1—figure 17 shows a kymograph of a contracting junction in WT and eve, illustrating that in eve mutants, myosin is still associated with junction contraction. Appendix 1—figure 18 shows that the myosin distribution in eve is significantly less anisotropic than in WT, even if the anisotropy remains clear. To test that this difference is not due to better visibility of edges in ${\displaystyle {eve}^{R13}}$ due to myosin being visualized using Myosin::mCherry instead of a Myosin::GFP, we verified that this difference between WT and eve persists if fewer and fewer eve junctions are included (filtering by myosin intensity, excluding up to 3/4 detected junctions).

Appendix 1—figure 17

Download asset Open asset

Appendix 1—figure 18

Download asset Open asset

Appendix 1—figure 19

Download asset Open asset

Appendix 1—figure 20

Download asset Open asset

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

The authors thank Eric Wieschaus, Boris Shraiman, Fridtjof Brauns, and members of the Streichan lab for valuable discussions and suggestions. We additionally wish to thank Eric Wieschaus for providing several fly lines, Sophie Streichan for aid in handling stocks, crosses, and reagents, and Cécile Regis for assistance with 3D visualizations. This research was supported by NIH grant no. R35 GM138203 and partially supported by the National Science Foundation under grants PHY-1707973 and PHY-1748958. MFL acknowledges support by NIH F31 fellowship no. HD093377. NPM acknowledges support from the Helen Hay Whitney Foundation.

© 2023, Lefebvre, Claussen et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.