Main

During Drosophila gastrulation, apical constriction of ventral cells facilitates the formation of a ventral furrow and the subsequent internalization of the presumptive mesoderm. Although myosin is known to localize to the apical cortex of constricting ventral furrow cells8,9,10,11, it is not known how myosin produces force to drive constriction. Understanding this mechanism requires a quantitative analysis of cell and cytoskeletal dynamics. We therefore developed methods to reveal and quantify apical cell shape with Spider–GFP, a green fluorescent protein (GFP)-tagged membrane-associated protein that outlines individual cells (Fig. 1a, b, Supplementary Fig. 1 and Supplementary Video 1)12. Ventral cells were constricted to about 50% of their initial apical area before the onset of invagination and continued to constrict during invagination (Fig. 1c, e). Although the average apical area steadily decreased at a rate of about 5 μm2 min-1, individual cells showed transient pulses of rapid constriction that exceeded 10–15 μm2 min-1 (Fig. 1d, f, g, and Supplementary Video 2). During the initial 2 min of constriction, weak constriction pulses were often interrupted by periods of cell stretching. However, at 2 min, constriction pulses increased in magnitude and cell shape seemed to be stabilized between pulses, leading to net constriction (Fig. 1d). These two phases probably correspond to the ‘slow/apical flattening’ and ‘fast/stochastic’ phases that have been described previously13,14. Overall, cells underwent an average of 3.2 ± 1.2 constriction pulses over 6 min, with an average interval of 82.8 ± 48 s between pulses (mean ± s.d., n = 40 cells, 126 pulses). Constriction pulses were mostly asynchronous between adjacent cells (Fig. 1h and Supplementary Video 3). As a consequence, cell apices between constrictions seemed to be pulled by their constricting neighbours. Thus, apical constriction occurs by means of pulses of rapid constriction interrupted by pauses during which cells must stabilize their constricted state before reinitiating constriction.

Figure 1: Apical constriction of ventral furrow cells is pulsed.
figure 1

a, Diagram of the imaging approach used to show apical constriction of the ventral furrow cells. We selected tangential Z-slices 2 μm below the apical surface (red slices) to show cell outlines. b, Z-slices (top) and YZ cross-sections (bottom) of cell membranes revealed with Spider–GFP. Scale bar, 10 μm. c, d, Apical areas (c) and constriction rates (d) for individual cells of a representative embryo. Each row represents data (see colour bars) for an individual cell. e, Mean apical area (red) and furrow depth (black). Dotted line indicates when tissue invagination initiates. Error bars indicate s.d. (n = 41 cells). f, g, Quantification (f) and time-lapse images (g) of the constriction of an individual cell. The red arrows (c, d) and red dots (g) mark the cell that is quantified in f. C, contraction. S, stabilization. Scale bar, 4 μm. (h) Pulsed constriction is asynchronous in neighbouring cells. Constriction rate is colour-coded (see colour bar) and mapped onto the corresponding cells in images at different time points.

PowerPoint slide

Full size image

To determine how myosin might generate force during pulsed constrictions, we simultaneously imaged myosin and cell dynamics by using myosin regulatory light chain (spaghetti squash, or squ) fused to mCherry (Myosin–mCherry) and Spider–GFP. Discrete myosin spots and fibres present on the apical cortex formed a network that extended across the tissue (Fig. 2a and Supplementary Fig. 2a). These myosin structures were dynamic, with apical myosin spots repeatedly increasing in intensity and moving together (at about 40 nm s-1) to form larger and more intense myosin structures at the medial apical cortex (Fig. 2c, Supplementary Fig. 2b, c, and Supplementary Video 4). This process, which we refer to as myosin coalescence, resulted in bursts of myosin accumulation that were correlated with constriction pulses (Fig. 2b–e and Supplementary Video 5). The peak rate of myosin coalescence preceded the peak constriction rate by 5–10 s, suggesting that myosin coalescence causes apical constriction (Fig. 2e). Between myosin coalescence events, myosin structures, including fibres, remained present on the cortex, possibly maintaining cortical tension between constriction pulses (Fig. 2c). Contrary to the purse-string model, we did not observe significant myosin accumulation at cell–cell junctions. To confirm that constriction involved medial myosin coalescence and not contraction of a circumferential purse-string, we correlated constriction rate with myosin intensity at either the medial or junctional regions of the cell. Apical constriction was correlated more significantly with medial myosin (Fig. 2f), suggesting that, in contrast to the purse-string model, constriction is driven by contractions at the medial apical cortex.

Figure 2: Constriction pulses are correlated with myosin coalescence.
figure 2

a, Merged images of Myosin–mCherry (Z-projection, 5 μm depth, green) and Spider–GFP (individual Z-slice 2 μm below the apical cortex, red). YZ cross-sections at lower magnification to illustrate furrow progression are shown at the bottom. b, Mean apical area and myosin intensity (left) and myosin intensity for individual cells (right) for a representative embryo. Error bars indicate s.d. (n = 37 cells). c, Single channel and merged time-lapse images of Myosin–mCherry (green) and Spider–GFP (red). Red arrows indicate spots that will coalesce. Blue arrow indicates myosin fibre that appears between contractions. d, Plots of apical area and myosin intensity against time (top) and constriction rate and rate of change of myosin intensity against time (bottom) for an individual cell. e, Plot of correlation between constriction rate and myosin intensity rate of change for individual cells (top) and averaged (n = 37 cells, bottom) against time offset. Correlation coefficients were calculated for various time offsets by temporally shifting the data sets relative to each other. Dotted lines indicate zero offset. Note that the maximum correlation occurs when myosin rate is shifted 5–10 s later in time; myosin coalescence therefore slightly precedes constriction rate. f, Constriction rate is more highly correlated with medial myosin than with junctional myosin. The diagram (left) illustrates the purse-string model for contraction in which we expect actin and myosin to become concentrated in the junctional region on constriction. Data points (right) represent correlation coefficients for individual cells, and the black bar indicates the mean (n = 37 cells). Asterisk, the difference between the means is statistically significant (P < 0.0001). Scale bars, 4 μm.

PowerPoint slide

Full size image

Myosin coalescence resembled contraction of a cortical actin–myosin network15,16. Therefore, to determine whether apical constriction is driven by pulsed contractions of the actin–myosin network, we examined the organization of the cortical actin cytoskeleton. In fibroblasts and keratocytes, actin network contraction bundles actin filaments into fibre-like structures16,17. Consistent with this expectation was our identification of an actin filament meshwork underlying the apical cortex in which prominent actin–myosin fibres spanning the apical cortex appeared specifically in constricting cells (Fig. 3a and Supplementary Fig. 3a). An actin–myosin network contraction model would predict that myosin coalescence results from myosin spots exerting traction on each other through the cortical actin network. To test whether myosin coalescence requires an intact actin network, we disrupted the actin network with cytochalasin D (CytoD). Disruption of the actin network with CytoD resulted in apical myosin spots that localized together with actin structures and appeared specifically in ventral cells (Supplementary Fig. 3b, c). Myosin spots in CytoD-injected embryos showed more rapid movement than those in control-injected embryos, suggesting that apical myosin spots in untreated embryos are constrained by the cortical actin network (Supplementary Fig. 3d). Although myosin movement was uninhibited in CytoD-treated embryos, myosin spots failed to coalesce and cells failed to constrict (Fig. 3b and Supplementary Fig. 3e). Because myosin coalescence requires an intact actin network, we propose that pulses of myosin coalescence represent contractions of the actin–myosin network.

Figure 3: Pulsed myosin coalescence and adherens junction bending require an actin–myosin network.
figure 3

a, Cortical myosin (green), cortical F-actin (red), and F-actin 2 μm below the apical cortex (white, to illustrate cell shape) were revealed in fixed embryos. b, Time-lapse images of Myosin–GFP in control-injected (DMSO) and CytoD-injected embryos. Arrows indicate individual myosin spots. Note that myosin spots move, but do not coalesce, in CytoD-treated embryos. c, Single-channel and merged time-lapse images of Myosin–mCherry (green) and E-Cadherin–GFP (red). Red arrows indicate myosin coalescence. Blue arrows indicate the site where adherens junctions bend inwards beneath a myosin spot. Scale bars, 4 μm.

PowerPoint slide

Full size image

Because actin–myosin contractions occurred at the medial apical cortex, it was unclear how the actin–myosin network was coupled to adherens junctions. We therefore imaged E-Cadherin–GFP and Myosin–mCherry to examine the relationship between myosin and adherens junctions. Before apical constriction, adherens junctions are present about 4 μm below the apical cortex18. As apical constriction initiated, these subapical adherens junctions gradually disappeared and adherens junctions simultaneously appeared apically at the same level as myosin8,19. This apical redistribution of adherens junctions occurred at specific sites along cell edges (midway between vertices; Supplementary Fig. 3f). As apical constriction initiated, these sites bent inwards. This bending depended on the presence of an intact actin network, which is consistent with contraction of the actin–myosin network generating force to pull junctions (Supplementary Fig. 3f). Indeed, myosin spots undergoing coalescence were observed to lead adherens junctions as they transiently bent inwards (Fig. 3c). Thus, pulsed contraction of the actin–myosin network at the medial cortex seems to pull the cell surface inwards at discrete adherens junction sites, resulting in apical constriction.

The transcription factors Twist and Snail regulate the apical constriction of ventral furrow cells20,21,22,23. Snail is a transcriptional repressor whose target or targets are currently unknown, whereas Twist enhances snail expression and activates the expression of fog and t48, which are thought to activate the Rho1 GTPase and promote myosin contractility8,10,19,21,24. To examine the mechanism of pulsed apical constriction further, we tested how Twist and Snail regulate myosin dynamics. In contrast to wild-type ventral cells, in which myosin was concentrated on the apical cortex (Fig. 2a), twist and snail mutants accumulated myosin predominantly at cell junctions, similarly to lateral cells (Fig. 4a). These ventral cells failed to constrict productively, which supported our cortical actin–myosin network contraction model, rather than the purse-string model, for apical constriction. twist and snail mutants differentially affected the coalescence of the minimal myosin that did localize to the apical cortex. Although myosin coalescence was inhibited in snail mutants, it still occurred in twist mutants, as did pulsed constrictions (Fig. 4a and Supplementary Video 6). This difference was also observed when Snail or Twist activity was knocked down by RNA-mediated interference (referred to as snailRNAi or twistRNAi) (Supplementary Fig. 4a and Supplementary Video 7). However, the magnitude of constriction pulses in twistRNAi embryos was greater than that of twist mutant embryos, suggesting that the low level of Twist activity present in twistRNAi embryos enhances contraction efficiency by activating the expression of snail or other transcriptional targets. Myosin coalescence was inhibited in snail twist double mutants, demonstrating that the pulsed constrictions in twist mutants required snail expression (Fig. 4a and Supplementary Video 6). Thus, the expression of snail, not twist, initiates the actin–myosin network contractions that power constriction pulses.

Figure 4: Snail and Twist function at distinct phases of pulsed constriction.
figure 4

a, Time-lapse images of Myosin–GFP Z-projections. Blue arrows indicate myosin spots that do not efficiently coalesce in snail mutants. Red arrows indicate myosin coalescence in twist mutants that seems to pull cell junctions. At least one coalescence event that pulled cell junctions occurred over a 6-min period for 53% of cells in the twist mutant, in contrast with 4% of cells in snail and snail twist mutants (n = 60 cells, three embryos per mutant). Scale bar, 4 μm. b, Time-lapse images of Spider–GFP in snailRNAi or twistRNAi embryos. P[sna] indicates twist-independent snail expression. Scale bar, 4 μm. c, Quantification of apical area (red) and constriction rate (blue) for individual cells in snailRNAi (left) and twistRNAi (right) embryos. d, Ratchet model of apical constriction. Myosin (green) contracts an apical actin network (red) that is coupled to adherens junctions (blue) driving constriction. Contractions are pulsed, interrupted by a phase in which the constricted state of the cell is stabilized.

PowerPoint slide

Full size image

Net apical constriction was inhibited in both snailRNAi and twistRNAi embryos (Supplementary Fig. 4b). We therefore wondered why the pulsed contractions that we observed in twistRNAi embryos failed to constrict cells. Using Spider–GFP to visualize cell outlines, we found that although constriction pulses were inhibited in snailRNAi embryos, constriction pulses still occurred in twistRNAi embryos (Fig. 4b, c, Supplementary Fig. 4c and Supplementary Video 8). However, the constricted state of cells in twistRNAi embryos was not stabilized between pulses, resulting in fluctuations in apical area with little net constriction (Fig. 4b, c). This stabilization defect was not due to lower snail activity, because these fluctuations continued when snail expression was driven independently of twist by using the P[sna] transgene (Fig. 4b)20. Although the frequency and magnitude of constriction pulses in such embryos were similar to those in control embryos, stretching events were significantly higher in twistRNAi; P[sna] embryos, suggesting a defect in maintaining cortical tension (Supplementary Fig. 4d). This defect might result from a failure to establish a dense actin meshwork, because both twist mutants and twistRNAi embryos had a more loosely arranged apical meshwork of actin spots and fibres than constricting wild-type cells did (Supplementary Fig. 4e). twist expression therefore stabilizes the constricted state of cells between pulsed contractions.

Thus, we propose a ‘ratchet’ model for apical constriction, in which phases of actin–myosin network contraction and stabilization are repeated to constrict the cell apex incrementally (Fig. 4d). In contrast to the purse-string model, we find that apical constriction is correlated with pulses of actin–myosin network contraction that occur on the apical cortex. Pulsed cortical contractions could allow dynamic rearrangements of the actin network to optimize force generation as cells change shape. Because contractions are asynchronous, cells must resist pulling forces from adjacent cells between contractions. A cortical actin–myosin meshwork seems to provide the cortical tension necessary to stabilize apical cell shape and promote net constriction. The transcription factors Snail and Twist are critical for the contraction and stabilization phases of constriction, respectively. Thus, Snail and Twist activities are temporally coordinated to drive productive apical constriction. Despite the dynamic nature of the contractions in individual cells, the behaviour of the system at the tissue level is continuous, in a similar manner to convergent extension in Xenopus25. Pulsed contraction may therefore represent a conserved cellular mechanism that drives precise tissue-level behaviour.

Methods Summary

Image acquisition and analysis

Two-colour imaging was performed at 22–25 °C with a Leica SP5 confocal microscope, a 63×/1.3 numerical aperture glycerine-immersion objective, an argon ion laser and a 561-nm diode laser. Spider–GFP images represent confocal slices 2 μm below the apical cortex, whereas myosin images represent maximum-intensity Z-projections of an apical section 5 μm in depth. Image stacks for Spider–GFP movies were acquired every 6 s, and image stacks for two-colour movies were acquired every 5 s. Using MATLAB (MathWorks) we developed methods to track cells and measure apical area and myosin intensity. Data points were smoothed with a Gaussian smoothing filter with σ = 15–18 s (three time points). Myosin intensity was measured from maximum-intensity Z-projections as the sum intensity of all pixels in a cell. Mean myosin intensity was calculated for junctional and medial pools of myosin by creating masks that selected regions less than 0.3 μm or more than 0.3 μm from the cell boundary, respectively.

Embryo fixation and staining

Heat fixation and staining with anti-myosin heavy chain (MHC) antibody did not preserve the normal organization of apical myosin observed in live squ–GFP26, squ–mCherry (Myosin–mCherry), and GFP-zipper (GFP–MHC)27 embryos. We therefore used endogenous GFP fluorescence to reveal myosin. squ–GFP embryos were fixed with 10% paraformaldehyde/heptane for 20 min, devitellinized manually, stained with Alexa-568 phalloidin (Invitrogen) to reveal actin, and mounted in AquaPolymount (Poysciences, Inc.).

Drug/RNAi injection

CytoD was injected laterally at mid–late cellularization with 0.5 mg ml-1 CytoD in 10% dimethylsulphoxide (DMSO; Calbiochem). Double-stranded RNAs against snail and twist (2 mg ml-1) were injected laterally into freshly laid eggs that were incubated 2.5–3.0 h before gastrulation was imaged.

Online Methods

Fly stocks and genetics

Fluorescent fusion protein stocks Spider–GFP (III) (95-1)12, Myosin–GFP (II or III) (squAX3 ; squ–GFP, a gift from R. Karess)26, E-Cadherin–GFP (II) (ubi-DE-cad–GFP)13, and Moesin–GFP (III)28 are described in the indicated references. Myosin–mCherry (III) (squ–mCherryA11 ) was recombined with Spider–GFP or Moesin–GFP to generate Myosin–mCherry Spider–GFP/TM3 and Myosin–mCherry Moesin–GFP. Homozygous Myosin–mCherry Spider–GFP flies could not be maintained as a stock. Therefore, Myosin–mCherry Spider–GFP/Myosin–mCherry flies were used for two-colour imaging. The dynamics of Myosin–mCherry and Spider–GFP observed in the two-colour strain are indistinguishable from the dynamics of Myosin–GFP and Spider–GFP alone. This suggests that the behaviour of myosin we observe with two-colour imaging reflects normal myosin dynamics. E-Cadherin–GFP Myosin–mCherry flies used for imaging were ubi-DE-cad–GFP shgR69 /CyO; squ–mCherryM1 .

Strategies were used to reveal both homozygous and hemizygous twist and snail mutants. First, the squ–GFP transgene was jumped onto the CyO balancer, generating CyO–Myosin–GFP. Chromosomes containing snaIIG05 , twiey53 and snaIIG05 twiey53 were marked with Df(2L)dpp[s7-dp35] 21F1–3;22F1–2 (halo) to allow homozygous embryos to be distinguished from their heterozygous siblings. We then rebalanced twist and snail mutants to obtain halo twiey53 /CyO–Myosin–GFP, halo snaIIG05/CyO–Myosin–GFP, and halo snaIIG05 twiey53/CyO–Myosin–GFP. Homozygous mutants were selected by identifying the halo mutant phenotype. Myosin accumulation in ventral cells was delayed in twist and snail mutants, occurring after cephalic furrow initiation, in contrast to wild-type embryos, in which myosin appears before the cephalic furrow is observed. The results presented are representative from nine halo snaIIG05 movies, six halo twiey53 movies and three halo snaIIG05 twiey53 movies.

Alternatively, a compound second chromosome stock homozygous for Myosin–GFP (III) (C(2)v; Myosin–GFP) was used. In the C(2)v stock, the right arms (containing twist) and the left arms (containing halo and snail) of chromosome 2 assort independently. For snail mutants, C(2)v; Myosin–GFP was crossed to halo snaIIG05/CyO and halo embryos were identified. For twist mutants, C(2)v; Myosin–GFP was crossed to halo twiey53 /CyO and one-third of the non-halo progeny were twist mutants. Both strategies identified the distinct phenotypes for snail and twist mutants described in Fig. 4a.

To provide twist-independent snail expression we used the P[snac ] (referred to as P[sna]) transgene, which contains the snail cDNA downstream of two copies of the proximal element of the twist promoter20. One zygotic copy of P[snac ] was sufficient to induce invagination in a snail mutant background, indicating that it provides enough activity to rescue snail-dependent contraction.

Construction of Myosin–mCherry

Myosin regulatory light chain (spaghetti squash, or squ), including its native promoter, was tagged at the carboxy terminus with mCherry. A 2-kilobase genomic fragment containing the squ open reading frame (ORF) and promoter was inserted into the KpnI and SalI sites of pBluescript. The squ 3′ untranslated region (800 base pairs) was then cloned into the downstream BamHI and XbaI sites. The sequence for mCherry, including a short linker region, was cloned into the ClaI and BamHI sites in between the squ ORF and 3′ untranslated region. The 3.5-kilobase KpnI/XbaI fragment containing Myosin–mCherry was then cloned into pCasPer4 and sent to BestGene Inc. to make transgenic flies. Myosin–mCherry complemented the null squAX3 allele, demonstrating that it is functional.

Time-lapse image acquisition

Egg collections were performed in plastic cups covered with apple-juice plates. Flies were allowed to lay eggs for 2–4 h at 25 °C before the plate was removed and embryos undergoing cellularization were collected. Embryos were dechorionated with 50% bleach, washed with water, and then mounted on a slide with embryo glue (Scotch tape resuspended in heptane), with the ventral side facing upwards. A chamber was made with two no. 1.5 coverslips as spacers and was filled with Halocarbon 27 oil for imaging. Embryos were not compressed. Mesoderm invagination occurred with a timeframe similar to that deduced from fixed embryos, and embryos imaged under these conditions could subsequenty hatch, demonstrating that our imaging conditions had minimal impact on development.

Single-colour images of Spider–GFP and two-colour images of Myosin–mCherry Spider–GFP were obtained with a Leica SP5 confocal microscope, a 63×/1.3 numerical aperture glycerine-immersion objective, an argon ion laser and a 561-nm diode laser. Images were acquired with a pinhole setting of two Airy units. For simultaneous two-colour images, we set the excitation bandpass to 495–550 nm to detect GFP, and to 578–650 nm to detect mCherry. There was minimal bleedthrough between the two channels. Images were acquired at a resolution of 141 nm per pixel. Myosin–GFP images were obtained with a PerkinElmer Ultraview spinning disk confocal microscope, a 60×/1.2 numerical aperture water-immersion objective, an argon/krypton laser and an Orca ER 4 charge-coupled device camera (Hamamatsu).

Image processing and analysis

The images presented were processed with ImageJ (http://rsb.info.nih.gov/ij/) and Photoshop CS (Adobe Systems, Inc.). Unless otherwise noted, Spider–GFP images represent a single confocal slice 2 μm below the apical cortex, whereas Myosin–GFP, Myosin–mCherry and Moesin–GFP images represent maximum-intensity Z-projections 5 μm in depth. A Gaussian smoothing filter with a radius of one pixel was used to reduce noise in published images.

To show apical cell shape, we manually selected Z-slices at a depth that was about 2 μm below the apical surface and the apical myosin. Cell outlines at 2 μm depth were very similar in dimensions to more apical cell outlines; however, they were easier to reveal and lacked membrane irregularities (namely in Fig. 3c) that complicated image analysis. Shifts in Z-position did not result in discontinuities in the data (Fig. 1e), and the pulsed behaviour that we describe was also observed when a single Z-slice was used.

We developed methods to measure apical area, constriction rate and myosin intensity with MATLAB (MathWorks). Raw images were bandpass-filtered with effective cutoff wavelengths of 1.4 μm (low pass) and 17.9 μm (high pass). After thresholding, a series of morphological operations were applied to reduce the width of the membranes to one pixel and to remove spurious background labelling. Examples of extracted cell outlines are shown in Supplementary Fig. 1. Indexed cells were automatically tracked on the basis of distances between cell centroids at subsequent time points. Cell properties were measured at each time point. We manually removed cells with errors in the segmentation to ensure that all cells in the data set were correctly identified. Data for apical area and myosin were smoothed with a Gaussian smoothing filter (σ = 15–18 s, three time points), and constriction rates and myosin rates of change were calculated from the smoothed data. Unless otherwise stated, a constriction pulse was defined as an event in which the constriction rate exceeded one standard deviation above the mean (more than 10.8 μm2 min-1). To measure myosin intensity, we first clipped intensity values below two standard deviations above the mean to separate myosin structures from unselective background labelling. Myosin intensity was then measured from maximum-intensity Z-projections (two highest values averaged) as the sum intensity of all pixels in a given cell. The correlation between constriction rate and the myosin intensity rate of change was determined by calculating the correlation coefficients between these two data sets for individual cells using the time interval from 1.7 min to 4.8 min. To examine the time dependence of this correlation, the data for constriction rate and myosin rate of change were shifted in time relative to one another.

Furrow depth (Fig. 1e) was calculated by measuring the distance between vitelline membrane and apical cortex in YZ cross-sections.

Myosin–GFP spot velocity was measured using the Manual Tracking plugin for ImageJ. We tracked individual spots or other distinct myosin structures for the duration of their lifetime and averaged the three highest velocities to calculate maximum velocity. For wild-type movies, we calculated the velocity of myosin structures coalescing. We chose coalescence events that occurred early or occurred near the midline to minimize the effects of tissue movement during invagination.

Drug injection

Embryos were dechorionated in 50% bleach for 2 min, washed with water, mounted on the edge of a glass slide (ventral side upward) with embryo glue, and desiccated for 4–8 min. A 3:1 mixture of halocarbon 700:halocarbon 27 was used for injection. Embryos were injected laterally at mid–late cellularization (furrow canals had passed the base of the nuclei) with about 1% egg volume of control or drug solution. We injected 0.5 mg ml-1 CytoD (Calbiochem) in 10% DMSO.

RNAi

Primers for double-stranded (ds)RNA were designed with E-RNAi (http://www.dkfz.de/signaling2/e-rnai/)29. Primers included the sequence of the T7 promoter (5′-TAATACGACTCACTATAGGG-3′) followed by the following recognition sequences: Twi01-F, 5′-GCCAAGCAAGATCACCAAAT-3′; Twi01-R: 5′-GACCTCGTTGCTGGGTATGT-3′; Twi02-F, 5′-GGAGCTGCAGAACAATGTGA-3′; Twi02-R, 5′-TGCTGTTGCTGGTGGATTAG-3′; Sna01-F, 5′-CGGAACCGAAACGTGACTAT-3′; Sna01-R, 5′-GCGGTAGTTTTTGGCATGAT-3′; Sna02-F, 5′-ATCATGCCAAAAACTACCGC-3′; Sna02-R, 5′-AGCGACATCCTGGAGAAAGA-3′; Control-F, 5′-GAATGCTATGGGAGGCGATA-3′; Control-R, 5′-TCAGCTTAGGCTCTGGGTGT-3′.

Primer pairs were used to amplify a PCR product from genomic DNA. PCR products were used directly in a transcription reaction with T7 polymerase with the MEGAscript transcription kit (Ambion). Annealing was performed by adding 10 mM EDTA, 0.1% SDS and 0.1 M NaCl to the reaction and incubating this mixture in a water bath heated to above 90 °C, which was allowed to cool for several hours. The dsRNA was purified by extraction with phenol/chloroform and resuspended in injection buffer (5 mM KCl, 0.1 mM sodium phosphate, pH 7.0). We injected a 2 mg ml-1 solution of snail or twist dsRNA into the embryo. Identical results were obtained with either dsRNA fragment used to knock down twist or snail. Sna01 and Twi01 were used for all the experiments in the manuscript, with the exception of Supplementary Fig. 4e, in which Twi02 was used. Control primers amplified a portion of an unknown ORF, CG3651, which has no function during ventral furrow formation. Injections were performed as described for the drug injections, except that egg collections were performed after 30 min to inject embryos at the earliest possible stage. To provide twist-independent snail expression we crossed Spider–GFP virgin females to P[snac ]20 males and collected embryos for twist dsRNA injection.

Statistics

Statistical significance between means was determined with an unpaired t-test. P values were calculated with Prism 5 (Graphpad Software, Inc.).