Cytoskeletal dynamics and supracellular organisation of cell shape fluctuations during dorsal closure

The term morphogenesis refers to the dynamic reshaping of tissues that results in the stereotyped three-dimensional forms observed during embryonic development. Insight into the mechanics of morphogenetic processes is vital to understanding the construction of organs and tissues and to their assembly into organisms. However, as genetics and microscopy allow us to focus on single cells a paradox arises. Embryonic development is a reproducible deterministic process, but the individual cells that configure an organism exhibit remarkable behavioural heterogeneity when observed with enough time and spatial resolution (Blanchard et al., 2009; Martin et al., 2009; Mateus et al., 2009; Solon et al., 2009). How and when these heterogeneities are averaged into the stereotypical morphogenetic deformations are important questions of modern biology. The process of dorsal closure (DC) during Drosophila embryogenesis lends itself to quantifying behaviour at both the single-cell and tissue levels, and in combination with genetic perturbations provides a unique model with which to address the principles underlying averaging across length and time scales.

DC is a morphogenetic process that results in the closing of a gap that exists half way through embryogenesis in the dorsal epidermis (Harden, 2002; Jacinto and Martin, 2001; Martinez Arias, 1993). During DC, the epidermis elongates and converges towards the dorsal midline as the amnioserosa (AS) contracts and eventually disappears inside the embryo. DC is driven by forces generated by both epithelia and by their mechanical interactions (Gorfinkiel et al., 2009; Gorfinkiel and Martínez Arias, 2007; Grevengoed et al., 2001; Hutson et al., 2003; Kaltschmidt et al., 2002; Kiehart et al., 2000). The AS, in particular, provides much of the contractile force of DC (Franke et al., 2005; Kiehart et al., 2000; Lamka and Lipshitz, 1999; Scuderi and Letsou, 2005). We have previously shown that the contraction of the AS results from cell-autonomous apical contraction modulated by the mechanical constraints the tissue experiences (Gorfinkiel et al., 2009). In contrast to the collectively smooth movement of the whole tissue, individual AS cells display pulses of alternating contraction and expansion (Solon et al., 2009). These fluctuations are damped during DC, coincidentally with the appearance of a supracellular actin cable at the interface between the AS and the epidermis (Jacinto et al., 2002; Kiehart et al., 2000; Rodriguez-Diaz et al., 2008; Solon et al., 2009). This correlation has led to the proposal that the actin cable behaves as a ratchet, preventing expansion of fluctuating AS cells and promoting net cellular contraction of the tissue (Solon et al., 2009).

In the mesoderm of Drosophila, asynchronous pulsed apical contractions are associated with invagination and are correlated with bursts of apical actin-myosin accumulation, followed by a phase of stabilisation of the contracted state, and it has been suggested that the progressive enrichment of the actin-myosin apical meshwork acts as an intracellular ratchet to stabilise contractions (Martin et al., 2009). This actin-myosin meshwork forms a supracellular structure that allows the transmission of cellular forces to the whole tissue (Martin et al., 2010). The fluctuations that have been observed in the AS cells during DC provide another example of cellular dynamics associated with a morphogenetic process and provide a good system with which to understand the molecular basis of the fluctuations and their role in the behaviour of sheets of cells. Here, we present a detailed quantitative analysis of AS cell fluctuations during DC and their relationship to the cytoskeleton.

The following stocks were used: ubiECad-GFP (Oda and Tsukita, 2001), Moe-GFP (Edwards et al., 1997; Kiehart et al., 2000), Spaghetti squash (Sqh)-GFP (Royou et al., 2002), Sqh-mCherry (Martin et al., 2009), c381Gal4 (an AS driver), UAS-ctMLCK [a constitutively active form of myosin light chain kinase (Kim et al., 2002)], UAS-Dia^CA [a constitutively active form of the formin Diaphanous (Somogyi and Rorth, 2004)] and UAS-RhoN19 [a dominant-negative form of Rho (Strutt et al., 1997)]. tkv⁸ is an amorph allele for the Dpp receptor (García Fernández et al., 2007). UAS lines were expressed using the Gal4 system (Brand and Perrimon, 1993).

The following genotypes were analysed:

• ubiECad-GFP;

• sqh^AX3; Sqh-GFP;

• Moe-GFP;

• Moe-GFP; Sqh-mCherry;

• ubiECad-GFP, Sqh-mCherry;

• c381Gal4/ubiECad-GFP, Sqh-mCherry; UAS-ctMLCK;

• c381Gal4/ubiECad-GFP, Sqh-mCherry; UAS-Dia^CA;

• c381Gal4/ubiECad-GFP, Sqh-mCherry; UAS-RhoN19;

• tkv8, ubiECad-GFP.

Stage 12-13 Drosophila embryos were dechorionated, mounted in coverslips with the dorsal side glued to the glass and covered with Voltalef oil 10S (Attachem). The AS was imaged at 24°C using an inverted LSM 510 Meta laser-scanning microscope with a 40× oil-immersion Plan/Fluor (NA 1.3) objective. The whole AS expressing ECad-GFP was imaged with an argon laser using the unmixing tool from the Zeiss software to remove autofluorescence from the vitelline membrane. Fifteen or sixteen z-sections 1.5 μm apart were collected every 30 seconds. GFP and mCherry cytoskeletal reporters were simultaneously imaged with an argon laser and a 561 nm diode laser and 5-6 z-sections 1 μm apart were collected every 10 seconds.

Embryos were fixed and stained as described previously (Kaltschmidt et al., 2002). The following primary antibodies were used: rat monoclonal against Drosophila E-Cadherin, Shotgun (Dcad2, 1/20; Developmental Studies Hybridoma Bank, University of Iowa, developed by T. Uemura), rabbit antisera against phospho-Myosin light chain 2 (Cell Signaling). Alexa 488- and Alexa 647-conjugated antibodies (Molecular Probes) were used as secondary antibodies.

Cell shapes during DC were tracked from confocal stacks, projections of which are shown in Fig. 1A, as previously described (Blanchard et al., 2009; Gorfinkiel et al., 2009). Cell shape fluctuations were analysed using the cell radius time series of individual cells (Fig. 1B,C). Though initially we used cell area, we converted all area measures to radii [using radius=π(area/π)] because we considered that the actin-myosin contraction of cells was more appropriately a radial process, contracting to a focal point, and hence a linear process. Using cell area would have resulted in larger cells having disproportionately large fluctuations. Radial measures represent the average radius of a cell, irrespective of its shape anisotropy.

The trend in each cell radius was calculated using a boxcar average with a window of 5 minutes (red lines in Fig. 1B,C), and subtracted from raw cell radius to produce a detrended cell radius time series (see Fig. S1A,B in the supplementary material). The peaks and troughs in the detrended time series were identified where there was a change in gradient sign. We further removed peaks and troughs that were separated by very small amplitude radial changes (less than 0.13 μm). This threshold was set at a compromise level, maximising the removal of high-frequency noise while minimising the removal of genuine fluctuations. Using the remaining peak (π/2 radians) and trough (3π/2 radians) phase markers (colour-coded in Fig. S1A,B in the supplementary material) we calculated the amplitude (half the difference in radius between peak and trough) and cycle length (the duration of the cycle from peak to subsequent peak) of each cycle and the durations of the contraction and expansion parts to each cycle (see Fig. S1C,D in the supplementary material; Fig. 1D). Combinations of low amplitude and long cycle lengths were the signature of time periods when there were no discernible fluctuations, so we used cell radius fluctuation amplitude divided by cycle length as a measure of fluctuation strength, in μm per minute. Below a fluctuation strength threshold of 0.05 μm per minute (see Fig. S1C,D in the supplementary material, red bars; Fig. 1E, vertical dashed white line), cells were considered to have no discernible fluctuations and cycle length was registered as invalid.

For ubiECad-GFP and Sqh-mCherry double-labelled embryos, we classified and quantified fluctuations in average apical medial fluorescence intensities using the same methods as described above. We removed pixels at and within a distance of two pixels from the cell membranes from the analysis to exclude regions with fluorescence associated with the junctional pool. We used a threshold amplitude change of 1.3 greyscale for removing minor peaks and troughs.

Throughout, 95% confidence intervals (Cumming et al., 2007) have been calculated and drawn in dashed lines on all relevant graphs or as error bars in figures. Non-overlapping confidence intervals indicate significant differences.

During DC, AS cells exhibited periodic apical shape fluctuations in the plane of the zonula adherens (see Movie 1 in the supplementary material). We calculated the amplitude, cycle length and phase of cell radius fluctuations for all time points of all cells (see Materials and methods, Fig. 1A-C and Fig. S1 in the supplementary material). A time series analysis revealed consistent differences in the amplitude and cycle length of fluctuations as DC progressed, patterns that were repeated across multiple embryos (Fig. 1D-F). We observed three different modes of AS cell fluctuation, which can be correlated with three developmental epochs: early, slow and fast phases of DC. The early phase lasts ∼45 minutes from the end of germ-band retraction to the start of DC and the beginning of the slow phase [time zero, defined as the onset of net AS contraction (Gorfinkiel et al., 2009)]. The fast phase starts 80 minutes later when the net AS contraction rate accelerates dramatically. In the early phase, AS cells fluctuated with large amplitudes (8-10% of cell radius) and long cycle lengths (4-minute cycles). At the onset of the slow phase, the amplitude decreased (6-7% of cell radius) and cycle length decreased (3-minute cycles). This change in cycle length at zero minutes was abrupt in all cells of the AS (Fig. 1F). By contrast, the decrease in proportional amplitude was slightly delayed relative to this transition (Fig. 1D). Finally, during the fast phase, only some cells displayed measurable fluctuations, and these were of low amplitude and short cycle length. The spatial and temporal pattern of decrease in both the amplitude and cycle length of fluctuations as DC progressed correlates with an increasing rate of contraction of AS cells as a whole (Gorfinkiel et al., 2009). More specifically, patterns of cycle length were mimicked by those of the rate of change in the trends in cell radius (Fig. 1G) and there was a strong linear relationship between the change in cell radius trends and cycle length (Fig. 1H). Although, overall, cell radius was correlated with cycle length, the detailed pattern of cycle lengths did not match that of cell radius, and cells of the same area could exhibit very different fluctuating behaviour (Fig. 1D,F and see Fig. S2A in the supplementary material). Thus, of the various measures of fluctuating AS cells, fluctuation frequency and the rate of contraction were most intimately related.

To better understand the mechanics of cell fluctuations, we investigated how they were organised in time and across the tissue. We compared the duration of the contraction and expansion phases (half cycles) for each fluctuation cycle. The duration of contraction was shorter than that of expansion throughout DC, significantly so for early and slow phases of DC (Fig. 2A), suggesting that cell fluctuations might be driven by a robust cell-autonomous contractile phase. To investigate the organisation of fluctuations in space, we first broke the measures of cell radius fluctuation down into an elliptical description of the change in shape of each cell. We calculated the average invariant strain rates (Blanchard et al., 2009) for both the contraction and expansion phases of each cell fluctuation cycle (an example snapshot is shown in Fig. 2B), having first removed trends in cell shape strain rates. Major (longer) and minor (shorter) perpendicular strain rates describe the elliptical deformation rates of each cell. Contraction and expansion were both often anisotropic, with the major strain rate accounting for much of the change in shape of each cell. We therefore quantified the frequency of different combinations of strain rate pairs for each half cycle, and found that there was a significant positive relationship (Fig. 2C). However, the regression gradient was shallow (0.154), indicating that the majority of fluctuations in cell shape were indeed anisotropic. We then asked of the anisotropic fluctuations (those with a correlation ratio below 0.5, Fig. 2C) whether the orientation of contraction was followed by expansion in the same orientation, and vice versa, by calculating the change in orientation angle (Fig. 2D). The orientations of subsequent half cycle deformations were very similar, with 70% of orientation difference angles being less than 30 degrees. Orientations were significantly more similar for the transition from contraction to expansion than for the reverse (contraction to expansion n=4345; expansion to contraction n=4196, D=0.0736, P<<0.001; Kolmogorov-Smirnov two-sample test). That expansion is more likely to return an apical cell shape to its previous expanded shape than vice versa suggests that the equilibrium shape of a cell apex is closer to its expanded shape, and that active contraction drives cell apices away from this equilibrium.

These observations raised interesting questions about how cells accommodate the half cycle fluctuations of their neighbours. To assess this, we investigated the correlation ratio of the major strain rate of each cell with the strain rate in the same orientation of all of its neighbours, calculated as a function of the angle of the neighbouring cell (Fig. 2E). In this example, neighbour 1 is nearly in line with the major orientation of contraction of the focal cell and is expanding in that orientation and so is strongly anti-correlated. Neighbour 3 is in a perpendicular position, with a contraction that is strongly correlated with the focal cell. Neighbours in line with the orientation of contraction had anti-correlated strain rates (–0.35; Fig. 2F, black line). This means that about one-third of the contraction of a cell is accommodated by cells beyond its immediate neighbours. The strain rates of neighbours perpendicular to the major strain rate of the focal cell were positively correlated (0.2), suggesting that cell fluctuations are coordinated in rows. These patterns can be seen in Fig. 2B (orange ribbons), in which some cells with parallel anisotropic strain rates are arranged in rows. For expansion half cycles, neighbours in line with the orientation of expansion of the focal cell had strain rates that were again anti-correlated (–0.15), but significantly less so than for contracting half cycles. This means that about two-thirds of cell expansion is accommodated by cells beyond its immediate neighbours. Neighbours perpendicular to the orientation of expansion had strain rates that were significantly more strongly correlated (0.31; Fig. 2F, red line). Rows of expanding anisotropic neighbours can also be seen in Fig. 2B (cyan ribbons). These differences again suggest that contraction is the more active process, having the most dramatic effect on immediate neighbours in line with the orientation of contraction. Expansion is better organised into rows of cells that expand in parallel, often accommodating contraction at a distance beyond immediate neighbours.

Fluctuating behaviour similar in scale to that described here has been correlated with the activity of the actin-myosin cytoskeleton (Martin et al., 2009; Salbreux et al., 2007), and this led us to image actin and myosin dynamics in AS cells using fluorescent protein reporters. We observed that the actin-binding domain of Moesin fused to GFP (Moe-GFP), and the myosin regulatory light chain, Spaghetti squash (Sqh), fused to GFP or mCherry (Sqh-GFP or Sqh-mCherry) were present at the apices of cells in two distinct structures: as a dynamic apical medial network, and in a cortical ring at cell-cell adherens junctions. During the early and slow phases of DC, we observed accumulations, or foci, of Moe-GFP and Sqh-GFP that appeared transiently and often moved as a wave across the apical surface of cells (see Movies 2 and 3 in the supplementary material). Foci coalesced and disappeared both at the centre of the apical cortex and near the lateral membranes.

The formation of Moe-GFP and Sqh-GFP foci coincided with local displacements of the membrane towards the foci, suggesting that they might be contributing to the cell shape fluctuations (Fig. 3A). We used double-labelled Sqh-mCherry and E-Cadherin-GFP (ECad-GFP) embryos (see Movie 5 in the supplementary material) to quantify simultaneously the intensity of the apical medial myosin (excluding the junctional myosin) and changes in the apical cell radius. Average apical medial myosin fluorescence intensity was inversely correlated with cell radius (Fig. 3B-D). The rapid increase in myosin fluorescence intensity reached a peak that preceded the trough of cell radius (Fig. 3D), reflecting the subcellular nature of the fluctuating behaviour, in which expansion at one end of a cell can occur simultaneously with contraction at the other end, and where the greatest extent of myosin fluorescence occurred mid-wave (Fig. 3A).

Simultaneous imaging of Moe-GFP and Sqh-mCherry (see Movie 6 in the supplementary material) revealed that both proteins co-localised in time and space at these dynamic foci (Fig. 3E), although Sqh-mCherry localisation was sharper than the more diffuse Moe-GFP. We quantified simultaneously the intensity of Moe-GFP and Sqh-mCherry fluorescence at a subcellular scale. The average fluorescence intensity over time within a circle at the centre of each cell (diameter 4.2 μm) was measured (Fig. 3F,G). Peaks in fluorescence were identical for Moe-GFP and Sqh-mCherry, and although myosin fluorescence was lost slightly before actin, these differences were not significant (Fig. 3H). Interestingly, fluorescence gain was faster than fluorescence loss, suggesting differences in the mechanism of foci assembly and disassembly. In contrast to these similarities at cell centres, we sometimes observed an increase in the localisation of Moe-GFP, but not of Sqh-mCherry, at the cell edges.

Altogether, these observations suggest that cell radius fluctuations are driven by active apical actin-myosin contraction. The increase in Moe-GFP and Sqh-mCherry fluorescence at a contractile focus [peaks were frequently double the trough values in non-detrended (see Materials and methods) data] was much greater than would have been expected simply from an increase in the density of existing myosin fluorescence due to cell contraction (maximum 20%; see Fig. S1C,D in the supplementary material). Combined with the divorce between myosin peak and area trough, and the similarity of the Moe and Sqh patterns and dynamics, these results suggest that large-scale actin-myosin structures were being assembled and disassembled in each cell contraction cycle.

To better understand the role of the apical medial network of myosin in the apical contraction of AS cells, we analysed its behaviour as DC progressed. We quantified the average apical medial myosin at the start of short movies covering the early, slow and fast phases of DC and found that fluorescence levels of apical medial myosin increased significantly over time (Fig. 3J). Furthermore, apical medial myosin progressively occupied the whole apical surface of AS cells and eventually formed a continuous sheet spanning the whole tissue (see Movie 4 in the supplementary material). These changes in the localisation of myosin were also evident in fixed embryos stained with an anti-phospho-myosin antibody (Fig. 3I). During the early and slow phases of DC, phospho-myosin accumulated in foci at the apical surface of AS cells, similar to the Sqh-mCherry foci observed in live embryos. During the fast phase of DC, there was a continuous labelling of phospho-myosin at the apical surface of AS cells (Fig. 3I). Thus, during DC there is the progressive formation of an apical actin-myosin meshwork that starts as discrete and dynamic foci moving along the apical cell surfaces and gradually forms a continuous, although still dynamic, network.

Together with the increase in apical medial myosin, we observed that the average levels of junctional myosin also increased as DC progressed (see Movie 3 in the supplementary material; Fig. 3J). Immunostaining with an anti-phospho-myosin antibody also showed a moderate increase in the levels of phospho-myosin at cell-cell junctions (Fig. 3I). The change in shape of AS cellular membranes during DC from wiggly to straight interfaces (Fig. 3I) is indicative of an increase in cortical tension (Lecuit and Lenne, 2007) and is thus in agreement with the observed increase in the levels of junctional myosin.

Altogether, our results suggest that actin-myosin activities, both at the apical surface of AS cells and at the level of cell-cell junctions, are responsible for the apical contraction of these cells. In support of this, laser-ablation experiments in AS cells have shown the presence of tensile stress at both cell-cell interfaces and cell centres (Ma et al., 2009).

In order to test whether there was an underlying causality to the correlation between actin-myosin activity and the macroscopic cell shape fluctuations, we modified cytoskeletal dynamics exclusively in the AS cells using the UAS-Gal4 system. The overactivation of myosin through the ectopic expression of a constitutively active myosin light chain kinase (ctMLCK) has been shown to induce the premature contraction of AS cells (Homem and Peifer, 2008). Live imaging of ASGal4/UAS-ctMLCK embryos confirmed that there was a premature contraction of the whole tissue, evident in the elongated shape of the AS in the anteroposterior axis soon after germ-band retraction (see Movie 7 in the supplementary material and Fig. 4A). This shape at such an early stage is evidence that contraction of the AS is enhanced and dominates over zippering (Hutson et al., 2003). ASGal4/UAS-ctMLCK cells exhibited fewer shape fluctuations than the wild type and their mean amplitude remained remarkably constant over time and space (Fig. 4B). Cycle lengths also appeared relatively constant over time, at 3-3.5 minutes on average (Fig. 4C,D). However, cell shapes were unusual, with smaller cell apices (see Fig. S2B in the supplementary material), more isotropic shapes (see Fig. S3 in the supplementary material) and straighter apicolateral membranes (Fig. 4A) than wild-type cells of the same stage. Importantly, in these embryos, AS cells produced frequent apical blebs (see Fig. S4A,B in the supplementary material) (Charras et al., 2006; Paluch et al., 2006).

To alter the actin network in a different manner, we used an activated form of Diaphanous (Dia^CA), an actin regulator that promotes filament elongation. As reported previously, the ectopic expression of Dia^CA induces a premature contraction of the AS (Homem and Peifer, 2008). Live imaging of ASGal4/UAS-Dia^CA embryos (see Movie 8 in the supplementary material; Fig. 4E) showed that AS cells exhibit negligible cell area fluctuations (Fig. 4F). Cell apices appeared small (see Fig. S2C in the supplementary material) and significantly more isotropic than in the wild type (see Fig. S3 in the supplementary material), as was the case for the ectopic expression of ctMLCK. Cell membranes were noticeably more wiggly than in the wild type (compare Fig. 4E with Fig. 1A).

To understand how these changes in the cell area fluctuations of AS cells correlated with cytoskeletal dynamics, we simultaneously imaged AS cells with ECad-GFP and Sqh-mCherry under these two perturbed conditions (see Movies 9 and 10 in the supplementary material). In ASGal4/UAS-ctMLCK embryos, most AS cells still showed dynamic foci of myosin accumulation, but with a different signature from the wild type. Apical myosin fluorescence intensity was again inversely correlated with apical cell radius, but the peak of myosin was now coincident with the trough of cell radius (Fig. 4G,H). We think that bleb contraction results in a knot of high myosin concentration that persists after bleb contraction is complete (for example, see Fig. S4B in the supplementary material, red arrow) (Charras et al., 2006; Paluch et al., 2006). By contrast, in AS cells from ASGal4/UAS-Dia^CA embryos, we rarely observed foci of myosin accumulation. Neither apical myosin fluorescence nor cell shape fluctuated (Fig. 4I).

Immunostaining of mutant embryos with an anti-phospho-myosin antibody showed changes in the levels and the subcellular localisation of phospho-myosin in AS cells compared with equivalent stages in the wild type. Similar observations have been reported in the embryonic epidermis (Bertet et al., 2009; Fernandez-Gonzalez et al., 2009). In ASGal4/UAS-ctMLCK embryos, there was a clear increase in phospho-myosin levels at cell-cell junctions, in a disrupted apical medial population (Fig. 4J) as well as sub-apically. Total apical myosin was significantly greater than in wild-type embryos during both early (wild type, n=5; ctMLCK, n=3; t=8.16, P<0.001; t-test) and fast (wild type, n=6; ctMLCK, n=3; t=2.68, P=0.0365; t-test) phases. By contrast, in ASGal4/UAS-Dia^CA embryos there was no discernible increase in junctional or apical medial myosin (Fig. 4K). However, by the fast phase, we observed a relative increase in sub-apical myosin levels. Compared with wild-type fast-phase embryos, in which there was greater total apical than sub-apical myosin, in ASGal4/UAS-Dia^CA embryos the reverse was true (Fig. 4L) and this difference was significant (wild type, n=5; ASGal4/UAS-Dia^CA, n=4; t=4.45, P<0.003; t-test). This suggests that whereas in ASGal4/UAS-ctMLCK embryos there was an increase in tension throughout AS cells, in ASGal4/UAS-Dia^CA there was no increase in junctional tension but an increase in tension in the sub-apical and lateral cortex. These observations could explain the differences in the shape of the membranes in these two perturbed conditions, with straight membranes in the former and wiggly membranes in the latter.

To complement the above mutants with enhanced myosin or actin activity, we analysed the effect of reducing actin and myosin activity on the behaviour of AS cells. Both myosin phosphorylation mediated by Rho kinase and actin polymerisation mediated by Dia are major effectors of the small GTPase RhoA (Rho1 – FlyBase) (Bishop and Hall, 2000; Wheeler and Ridley, 2004). AS cells expressing a dominant-negative form of RhoA did not fluctuate, nor did they consistently reduce their apical surface area, and as development progressed the tissue tore apart (see Movie 10 and Fig. S5A in the supplementary material). They did not form Sqh-mCherry foci, nor was there localisation at cell-cell junctions (see Fig. S5B in the supplementary material). Residual movements of AS cells seemed to be a passive consequence of other morphogenetic movements occurring in the embryo.

Altogether, these results show that the presence and signature of cell fluctuations and the shape of cellular membranes depend on the relative strengths, dynamics and localisation of cellular actin-myosin. We suggest that the progressive accumulation of myosin as DC proceeds increases cell contractility.

The results presented above raise the important question of what causes the changes in actin-myosin localisation over time, leading to an increase in cell contractility. One possibility is that these changes are the result of an autonomous programme of AS cell contraction. Alternatively, signals from the epidermis could activate myosin and/or actin activities.

It has been proposed that the damping of AS cell fluctuations over time is a consequence of the assembly of the supracellular actin cable at the leading edge (LE), which would act in a ratchet-like mechanism (Solon et al., 2009). In this model, contraction of the cable would prevent the relaxation of AS cells after a contraction cycle. We tested this by asking how the amplitude and cycle length of fluctuations progress in a mutant in which the supracellular actin cable and zippering processes are defective. Dpp, a member of the BMP/TGFβ family of secreted signals, is expressed in the dorsal-most epidermal (DME) cells during stage 11, and also later on, during stage 13. Zygotic mutants for the Dpp receptor thickveins (tkv) exhibit defects in DC (Riesgo-Escovar and Hafen, 1997); specifically, DME cells have defects in dorsoventral elongation, the LE does not maintain an actin-myosin cable and zippering movements are absent (García Fernández et al., 2007) (Ricos et al., 1999). The proposal for the role of the actin-myosin cable suggests that these embryos should not show a damping of fluctuations.

Live imaging (see Movie 11 in the supplementary material) revealed that in these mutants, AS cells exhibited a spatial and temporal pattern of shape fluctuations that shared strong similarities with that of the wild type (Fig. 5A-D). Importantly, AS cells displayed a decrease in the amplitude and cycle length of fluctuations as they transited from the early to the slow phase of DC. AS cells exhibited a wild-type evolution in cell shape, from anisotropic during the early phase to a more isotropic shape during the slow and fast phases (see Fig. S3 in the supplementary material), indicative of an increase in apical tension. Eventually, AS cells in tkv mutants contracted completely. In spite of these similarities, fluctuation amplitude and cycle length during the early phase were lower in tkv mutants than in wild-type embryos (Fig. 5B,D).

Thus, the damping of cell shape fluctuations, the changes in cell shape that are a signature of increased apical tension, and the eventual contraction of AS cells still occur in tkv mutants, even though these embryos were able to maintain neither an actin cable at the LE nor epidermal cell elongation, and did not develop zippering movements. Although these results do not rule out the possibility that segments of actin cable remain functional in tkv mutants, nor that in wild-type embryos the epidermis, the actin cable and zippering movements contribute to the patterned contraction of AS cells, they suggest to us that AS cells have a robust intrinsic programme of contraction.

The process of DC relies on the coordination of the elongation of epidermal cells and contraction of the AS. Throughout the early stages of this process the AS cells exhibit fluctuations of their apical area (Solon et al., 2009) (this work). We have shown that this fluctuating behaviour is driven by transient actin-myosin accumulations at the apical cortex of cells, involving the assembly, contraction and disassembly of large-scale actin and myosin structures. The formation and disassembly of foci could result from the self-organising properties of myosin and actin and the presence of specific regulators, which are known to spontaneously form active networks, asters and rings in vitro (Backouche et al., 2006; Haviv et al., 2006). Tension generated by the contraction of neighbouring cells (Solon et al., 2009) (this work), the sudden loss of tension due to detachment of actin from the cell membrane, spontaneous catastrophic collapse (Howard, 2009), and the stretching of the apical membrane resulting in the influx of ions such as calcium (Salbreux et al., 2007), could all contribute to the timing and location of fluctuations.

Differences in the absolute and relative strengths of actin-myosin structures in distinct subcellular populations provide an explanation for the patterns that we have observed in wild-type and mutant embryos. Our observations show that throughout DC the amounts of myosin increase in two subcellular populations: in a cortical ring and in a medial apical network. The amount of myosin at cell-cell junctions determines the shape of cell membranes: low levels lead to wiggly membranes and increasing levels of junctional myosin lead to the straightening of the membranes. The amount of actin-myosin in the medial population determines the fluctuating behaviour of cells: low levels lead to low-frequency fluctuations and no tissue contraction, as seen in the early phase of DC; intermediate levels lead to high-frequency fluctuations and slow tissue contraction, as seen in the slow phase of DC; and high levels lead to a coherent, but dynamic, sheet of actin-myosin across cells that was strongly contractile, as seen in the wild-type fast phase.

Increasing the amounts of both junctional and medial myosin in ASGal4/UAS-ctMLCK embryos led to premature apical contraction, precocious straight cell membranes and isotropic cell shapes, which are a signature of high cortical tension (Lecuit and Lenne, 2007). By contrast, in ASGal4/UAS-Dia^CA embryos, low junctional myosin levels throughout DC gave rise to wiggly apical membranes. However, these cells showed precocious apical contraction and no discernible apical shape fluctuations. We observed an increase in myosin levels at a sub-apical position, which suggests to us that in these cells contraction is not only apical but spans the sub-apical and lateral axis. This sub-apical and lateral contraction of AS cells could be an impediment to apical shape fluctuations, as apical contraction must be accommodated by basal or lateral expansion.

Overactivating myosin led to apical blebbing in AS cells. Blebbing results from an increase in the ratio of cortical tension to cortex-membrane adhesion (Charras and Paluch, 2008) and is associated with transient actin-myosin accumulations similar to those observed here (Charras et al., 2008). We looked for, but could find no evidence of, blebs in AS cells in wild-type embryos. We suggest that dynamic actin-myosin structures are the more general property of cells, which in combination with high cortical tension or weak cortex-membrane adhesion can lead to blebs.

Previous results have shown that neighbours oscillate mainly in anti-phase (Solon et al., 2009). Our results reveal a more subtle picture, with anti-phase correlation predominantly in one orientation and in-phase correlation perpendicular to this. This combination results in interesting patterns, with rows or patches of cells that become synchronised for short periods of time and represent an emergent property of the system. We expect that some sort of multicellular pattern is inevitable because of the requirement to maintain a coherent epithelium while cells fluctuate. However, the patterns also suggest that cell fluctuations can become entrained locally and for short periods. Our analysis of the dynamics of cell shape suggests that cell contraction is the active process, but that fluctuating behaviour of a cell can be influenced and the timing of active contraction altered by the forces generated by immediate neighbours and more distant cells. Thus, the organisation of apical contractions and expansions at the multicellular scale arises from the feedback in both directions between intrinsic cell behaviour and mechanical context.

We wanted to understand how changes in actin-myosin behaviour at a subcellular scale resulted in the patterned contraction of the AS. There is a gradual increase in the rate of contraction of individual cells that strongly accelerates with the onset of zippering behaviour (Gorfinkiel et al., 2009). These changes were correlated primarily with a shortening of fluctuation cycle length. Our results suggest that these changes result from an increase in both apical medial and junctional myosin levels. The overall increase in myosin levels and the formation of a continuous actin-myosin network could provide the molecular basis for the transition of the AS to a more solid tissue (Ma et al., 2009).

What causes the increase in apical myosin levels? One possibility is that it is induced by a chemical or mechanical signal from the epidermis. A radial gradient of fluctuations (Solon et al., 2009) and of the rate of contraction of AS cells (Gorfinkiel et al., 2009) suggests that the epidermis is providing some information for the patterned contraction of the AS. However, the analysis of tkv mutants shows that even when the mechanical properties of DME cells have been altered and Dpp signalling is compromised (B. García Fernández, PhD thesis, New University of Lisbon, 2007) (Ricos et al., 1999), AS cells change their fluctuation behaviour in a similar pattern to the wild type and finally contract. This reveals that several processes that are individually redundant ensure DC. DC could result from an AS-autonomous programme of increasing medial and junctional myosin (and/or actin), through changes in the dynamics of actin and myosin activity, of intracellular trafficking or of cell adhesion. Alternatively, apical myosin might increase as a result of the fluctuating behaviour of AS cells and the build-up of tension due to neighbour contractions. Whatever the mechanism, it is likely that an increase in myosin activity involves activation of the Rho GTPase, which has a central role both in integrating mechanical and structural cues and in regulating myosin-based tension (Ingber, 2008) (this work).

AS cells fluctuate at low frequencies for a long period during early DC without any tissue contraction. High-frequency fluctuations drive moderate cell and tissue contraction during the slow phase, before they disappear in the transition to rapid tissue contraction. This raises the question of whether fluctuations have a function, as it is at least theoretically possible that contraction could be achieved in a smooth manner. One can speculate that cell fluctuations would be a way to maintain a basic level of cell activity that could be turned easily into morphogenetically relevant behaviours (Farhadifar et al., 2007; Rauzi et al., 2008). Cell fluctuations could, more simply, be an epiphenomenon of the self-organised dynamics of actin and myosin. Alternatively, pulsatile contraction might ensure that differences in apical tension are equilibrated between neighbouring cells, ensuring coordination in the contraction of cells across the tissue. Our analysis of neighbour relations suggests that fluctuations allow for a certain degree of coordination between cells. A combination of empirical investigation and modelling will be crucial to understand the importance of fluctuations per se during morphogenesis.

Recent work shows that the formation of transient actomyosin networks correlates with the apical constriction of AS cells (David et al., 2010). The results are compatible with the frequency data measured in our study.

We thank Stephen Young, Pietro Cicuta, Alexandre Kabla, Joaquín de Navascués, Benedicte Sanson, Carlos Torroja and members of our labs for useful discussions; and D. Kiehart, A. Martin, J. Zallen, P. Rorth, M. VanBerkum, B. García Fernández, C. Klämbt and T. Lecuit for stocks. We are very grateful to Pablo Barrecheguren and Joaquín de Navascués for their invaluable help during the resubmission of the manuscript. This work was supported by a Biotechnology and Biological Sciences Research Council grant to N.G. and A.M.-A. and an Engineering and Physical Sciences Research Council grant to R.J.A.