Upon specification of the placode of cells that will form the embryonic salivary gland at the end of embryonic stage 10, the first apparent change within the apical domain of placodal cells is apical constriction at the point that will form the first point of invagination or pit (Figure 1A,B; [Booth et al., 2014; Myat and Andrew, 2000b]). Apical constriction at this point in fact preceded actual tissue bending (Figure 1B). We have previously shown that apical constriction is clustered around the pit and is important for proper tissue invagination and tube formation (Booth et al., 2014). In order to discover if any further cell behaviours in addition to apical constriction contribute to tissue bending and tube invagination in the early placode, we employed quantitative morphometric tools to investigate the process in comparable wild-type time-lapse movies using strain rate analysis (Blanchard et al., 2009). We imaged embryos expressing a lateral plasma membrane label in all epidermal cells as well as a marker that allows identification of placodal cells (Figure 1C and Video 1; for genotypes see Table 1). Time-lapse movies were segmented and cells tracked using previously developed computational tools that allow for the curvature of the tissue to be taken into account (Figure 1C–F) (Blanchard et al., 2009; Booth et al., 2014). The cells of the salivary gland placode that will later form the secretory part of the gland are organised into a roughly circular patch of tissue prior to invagination and maintain this shape during the process (Figure 1C). Within this circular patch, the invagination pit is located at the dorsal-posterior edge rather than within the centre of the placode. We therefore employed a radial coordinate system, with the forming pit as its origin, in which to analyse and express any changes (Figure 1E). We locally projected 2D strain (deformation) rates and other oriented measures onto these radial (‘rad’) and circumferential (‘circ’) axes (Figure 1F).

Video 1

Download asset

We focused our analysis on the early stages of tissue bending and invagination, defining as t = 0 the time point before the first curvature change at the point of invagination at the tissue level could be observed. Dynamic analysis of the changes in apical area of placodal cells, in the time interval of 18 min prior to and 18 min after the first tissue bending, revealed distinct zones of apical cell behaviour. Grouping the placodal cells into ~ 2 cell-wide stripes concentric to the pit for apical area change analysis (Figure 1G; Figure 1—figure supplement 1 and Video 2) revealed a split into cells whose apical area over this time interval did not change or changed only slightly (Figure 1G’, red, yellow and green lines) and cells whose apical area progressively decreased (Figure 1G’, blue and purple lines). The clear split into two zones of differing cell behaviours, defined by radial distance from the invagination point at zero minutes (when the pit starts to invaginate), prompted us to analyse these regions independently in our strain rate analyses (Figure 1H).

Video 2

Download asset

Methods have been developed previously to calculate strain (deformation) rates for small patches of tissue, and to further decompose these into the additive contributions of the rates of cell shape change and of the continuous process of cell rearrangement (intercalation) (Blanchard, 2017; Blanchard et al., 2009) (Figure 2A,B). The levels of both contributions vary dramatically between different morphogenetic processes (Blanchard et al., 2010; Bosveld et al., 2012; Butler et al., 2009; Etournay et al., 2015; Guirao et al., 2015). Cell divisions have ceased in the placode around the time of specification, and there is also no cell death, therefore none of these processes contributed to the overall tissue deformation in this case. Upon segmentation of cell outlines, the rate of tissue deformation was calculated from the relative movement of cell centroids in small spatio-temporal domains of a central cell surrounded by its immediate neighbours and over three movie frames (~6 min). Independently, for the same domains, rates of cell shape change were calculated mapping best-fitted ellipses to actual cell shapes over time. The rate of cell intercalation could then be deduced as the difference between these two measures (Figure 2A,B and Video 3; see Materials and methods for a detailed description). The three types of strain rate measure were then projected onto our radial coordinate system.

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 2—source data 1

Cumulative strain and statistics for cells near to the pit and far from the pit.

Related to Figure 2D–E.

https://doi.org/10.7554/eLife.35717.012

Download elife-35717-fig2-data1-v4.xlsx

Video 3

Download asset

Strain rate analysis of changes within the apical domain of the epithelial cells in the early salivary gland placode clearly confirmed the existence of two distinct zones of cell behaviour. Spatial plots summarising ~ 36 mins of data centred around the first tissue bending event (combined from the analysis of 9 wild-type embryos; Figure 2—figure supplement 1A) revealed strong isotropic tissue contraction near the invagination pit (Figure 2C’ and Figure 2—figure supplement 1B, green) that was mainly contributed by apical cell constriction (Figure 2C’’ and Figure 2—figure supplement 1B’). At a distance from the pit, tissue elongation dominated in the radial direction (Figure 2C’, magenta) and was contributed mostly by cell intercalation (Figure 2C’’,C’’’). The split in behaviour and the strong contribution of cell intercalation to the early invagination and tube formation was even more apparent from time resolved strain rate analyses. At the whole tissue level, apical cell constriction began more than 10 min before any curvature change at the tissue level (Figure 2—figure supplement 1C,C’, green curve), and this change was most pronounced in the cells near the invagination pit (Figure 2D,D’ versus E, E’, green curve). In addition, cell intercalation also commenced about 10 min prior to tissue bending (Figure 2—figure supplement 1C,C’, orange curve), but in this case, the stronger contribution came from cells far from the pit (Figure 2E,E’ versus D,D’, orange curve).

Although constriction was isotropic near the pit, with equally large magnitudes contributing both radially and circumferentially (Figure 2D,D’, green curves), intercalation was clearly polarised towards the invagination pit, with expansion radially in the orientation of the pit (Figure 2D,E and Figure 2—figure supplement 1C, ‘rad’, orange curves), and contraction circumferentially (Figure 2D’,E’ and Figure 2—figure supplement 1C', ‘circ’, orange curves; see also Figure 2—figure supplement 1F).

Thus, in addition to apical constriction, directional cell intercalation constitutes a major second cell behaviour occurring during tissue bending and invagination of a tube. Furthermore, the amount of both cell behaviours occurring was radially patterned across the placode. However, it was not clear whether both these behaviours were entirely being driven by active apical mechanisms, or whether further basal events contributed, so we next investigated the 3D behaviours of placode cells.

The apical constriction as observed in the placodal cells near the pit is indicative of a redistribution of cell volume more basally, which could result in a combination of cell wedging and/or cell elongation in depth. The 3D-cell behaviour of acquiring a wedge-like shape is of particular interest in the placode because it is capable of tissue bending. Apical constriction (coupled with corresponding basal expansion if cell volume is maintained, as in the case of the salivary gland placode [data not shown]) is known to deform previously columnar or cuboidal epithelial cells into wedge-shaped cells (Martin and Goldstein, 2014; Wen et al., 2017a). This is also true in the salivary gland placode, where cross-sections in xz of fixed samples and time-lapse movies confirm the change from columnar to wedged morphology (Figure 1B,B') (Girdler and Röper, 2014; Myat and Andrew, 2000b). Most analyses of morphogenetic processes are conducted with a focus on events within the apical domain given the prevalent apical accumulation of actomyosin and junctional components (Figure 3A,A’) (Bosveld et al., 2012; Butler et al., 2009; Martin et al., 2009; Rauzi et al., 2010; Simões et al., 2010). However, this apico-centric view neglects most of the volume of the cells. For instance, in the case of the salivary gland placode, cells extend up to 15 µm in depth (Figure 3B). We thus decided to analyse cell and tissue behaviour during early stages of tube formation from the placode in a 3D context. Automated cell segmentation and tracking of 3D behaviours is still unreliable in the case of tissues with a high amount of curvature, such as in the salivary gland placode once invagination begins. To circumvent this issue, we used strain rate analysis at different depth as a proxy for a full 3D analysis (Figure 3A’,C). After accounting for the overall curvature of the tissue, we segmented and tracked placodal cells not only within the apical region (as shown in Figures 1 and 2; Figure 3A’,C pink), but also at a more basal level, ~8 µm below the apical domain (Figure 3A’,C blue), and repeated the strain rate analysis at this depth (Figure 3D–E’’ and Figure 3—figure supplement 1 and Figure 3—figure supplement 2). We used the same radial coordinate system with the pit as the origin for both layers, so we were able to compare cell behaviours between depths. Segmentation and analysis at the most-basal level of placodal cells was not reliably possible due low signal-to-noise ratio of all analysed membrane reporters at this depth (data not shown).

Figure 3 with 2 supplements see all

Download asset Open asset

Figure 3—source data 1

Cumulative strain and statistics for cells near to the pit and far from the pit at apical and mid-basal depth.

Related to Figure 3D–E.

https://doi.org/10.7554/eLife.35717.019

Download elife-35717-fig3-data1-v4.xlsx

Overall, the spatial pattern of change at mid-basal level was similar to changes within the apical domain during 33 min centred around the start of tissue bending (Figure 3—figure supplement 1B,C). Isotropic tissue contraction at mid-basal depth was clustered around the position of the invagination pit (Figure 3—figure supplement 1B,C, green), and this was due to cell constriction at this level (Figure 3—figure supplement 1B’,C’), whereas the tissue expansion at a distance from the pit in the radial direction (Figure 3—figure supplement 1C) was due primarily to cell intercalation (Figure 3—figure supplement 1C’’). Similarly to events at the apical level, this radial expansion towards the pit was accompanied by a circumferential contraction, again due primarily to cell intercalation in the region away from the pit (Figure 3—figure supplement 1B,B’’).

Comparing apical and basal strain rates at the cell and tissue level with respect to their radial and circumferential contributions revealed an interesting picture. In temporally resolved plots, isotropic cell constriction dominated apically in cells near the pit (Figure 3D, pink), but with a slower rate of constriction at the mid-basal depth (Figure 3D, blue). This was confirmed by cross-section images that show that, near the pit and once tissue-bending had commenced, the basal surface of the cells was displaced even further basally than in the rest of the placode, and cells were expanded at this level, leading to an overall wedge-shape (Figure 1B’). In cells away from the pit, similar to the apical region, cell shapes did not change much (Figure 3E). In contrast to cell shape changes that diverged at depth at least in the cells near the pit, intercalation behaviour appeared to be highly coordinated between apical and basal levels with near identical contributions at both particularly in cells far from the pit (Figure 3D’, E’ and Figure 3—figure supplement 1D’’).

In cells near to the pit, the faster rate of apical constriction implies that cell wedging is occurring, but we have not confirmed this in the same dataset by measuring the relative sizes of apical and basal cell diameters. Similarly, although the rates of apical and basal intercalation are remarkably similar, this does not rule out that the actual arrangement of cells at one level is ‘tipped’ ahead of the other, through interleaving in depth (akin to a T1 transition along the apical-basal axis, see z-sections in Figure 4C and Figure 4—figure supplement 1Dd'’). For example, if cell rearrangement is being driven by an active apical mechanism (see below and Figure 6), we predict that apical cell contours would be intercalating ahead of basal cell contours, even while their rates remain the same. Both wedging (Figure 4B) and interleaving (Figure 4C) have implications for the tilt, or lean, of cells relative to epithelial surface normals (Figure 4D and Figure 4—figure supplement 1A–C), with a gradient of tilt expected for constant wedging or interleaving in a flat epithelium. We therefore set out to quantify 3D wedging, interleaving and cell tilt with new methods.

Figure 4 with 1 supplement see all

Download asset Open asset

Figure 4—source data 1

Wedging, interleaving and tilt mean ± confidence intervall (CI) and statistics.

Related to Figure 4B’, C’ and D’.

https://doi.org/10.7554/eLife.35717.023

Download elife-35717-fig4-data1-v4.xlsx

We used the placodes (n = 5) for which we have tracked cells at both apical and mid-basal levels. First, we developed a semi-automated method to accurately match cell identities correctly between depths (Figure 4A, Video 4, and see Materials and methods). We then borrowed ideas from recent methods developed to account for epithelial curvature in terms of the additive contributions of cell wedging and interleaving in depth (Deacon, 2012). In the early developing salivary gland placode, average tissue curvature is very slight, so we simplified the above methods for flat epithelia. We applied exactly the same methods that we have used in Figures 2 and 3 to calculate strain rates for small cell domains, but rather than quantifying rates of deformation over time, now we quantify rates of deformation in depth (see z-level illustrations in Figure 4B,C,D and Figure 4—figure supplement 1D). The cell shape strain rate becomes a wedging strain in depth, in units of proportional shape change per micron in z (Figure 4B,B’ and Figure 4—figure supplement 1Dc, c'), the intercalation strain rate becomes the interleaving strain in depth, in the same units (Figure 4C,C’ and Figure 4—figure supplement 1Dd, d'), and translation velocity becomes the cell tilt (Figure 4D,D’ and Figure 4—figure supplement 1A,Da,a’; see Materials and methods for details). Once again, we projected the z-strain rates and tilt onto our radial coordinate system.

Video 4

Download asset

Cells across the placode started out at −18 min before pit invagination unwedged and mostly untilted in radial and circumferential orientations (Figure 4B’,D’). Cells near the pit became progressively wedge-shaped over the next 30 min, with smaller apices (Figure 4B’, orange lines; Figure 4—figure supplement 1C). Cell wedging was reasonably isotropic, but with circumferential wedging always stronger than radial. Away from the pit, progressive wedging was less rapid, again with a strong circumferential contribution but nearly no radial contribution (Figure 4B’, purple lines). That cells were less wedged radially might be because this is the orientation in which cells move into the pit, releasing radial pressure due to apical constriction near the pit.

The wedging anisotropy is also compatible with active circumferential contraction. Indeed, circumferential interleaving was always more negative than radial interleaving, often significantly so (Figure 4C’, solid vs dashed lines). Thus, interleaving contributes a circumferential tissue contraction apically, with a concomitant radial expansion. This pattern is thus also compatible with an apical circumferential contraction mechanism, possibly driving cell rearrangements.

Cell tilt, a measure of the divergence of a cell’s in-line from the surface normal (Figure 4D and Figure 4—figure supplement 1A,B,D), increased continuously in the radial direction towards the pit over the period of our analysis (Figure 4D’, solid lines). A stronger tilt was observed for the cells further from the pit (Figure 4D’, purple solid line), which is expected from the radial wedging seen near the pit (Figure 4B’, solid orange line).

Hence, the relatively isotropic rates of apical constriction near the pit and the very similar rates of intercalation apically versus basally were in fact grounded in anisotropic wedging near the pit and in an interleaving difference between apical and basal. 3D tissue information such as wedging, interleaving and tilt are therefore essential to fully understand planar cell behaviours such as cell shape change and intercalation. Overall, our combined analysis so far suggests that isotropic apical constriction near the pit combines with an apically led circumferential contraction mechanism. We now investigate the possible origins of the latter.

Our strain rate analysis has revealed that the intercalation strain rate, representing the continuous process of slippage of cells past each other, was a major contributor to early tube formation and was highly coordinated between apical and basal domains. 3D domain interleaving further revealed that cell rearrangement convergence is more advanced apically in the circumferential orientation. Both these findings are measured from small groups of cells across the placode but are agnostic about neighbour exchange events or more complicated multi-neighbour exchanges. Neighbour exchanges or cell intercalations are usually thought to occur through one of two mechanisms: groups of four cells can exchange contacts through the formation of a transient four-cell vertex structure, in a typical T1 exchange (Figure 5A), whereas groups of more than four cells can form an intermediary structure termed a rosette, followed by resolution of the rosette to create new neighbour contacts (Figure 5B). During convergence and extension of tissues in both vertebrates and invertebrates, the formation and resolution of these intermediate structures tend to be oriented along embryonic axis, with the resolution occurring perpendicular to the formation (Blankenship et al., 2006; Lienkamp et al., 2012). We therefore decided to also analyse intercalation in the placode by following individual events to identify the underlying mechanistic basis.

Figure 5 with 1 supplement see all

Download asset Open asset

Figure 5—source data 1

Cumulative neighbour gains and cumulative proportions of circumferential productive neighbour gains.

Related to Figure 5C.

https://doi.org/10.7554/eLife.35717.027

Download elife-35717-fig5-data1-v4.xlsx

From our database of apical cell tracks and their connectivity, we identified all T1 transitions, classifying the time point when pairs of cells become new neighbours as ‘neighbour gains’. We further sub-classified neighbour gains as being either radially or circumferentially oriented, depending on which orientation was closest to the centroid-centroid line of two cells involved in a neighbour gain (Figure 5C). T1s occurred at a constant rate over our study period, and we observed T1s in both orientations, revealing that neighbour connectivity was quite dynamic (Figure 5C’,D). Nevertheless, over two-thirds of neighbour gains were oriented circumferentially, a bias that correlates with the intercalation strain rate contraction circumferentially (see for example Figure 2E’). This bias was also evident when visualising interface losses and gains over time for an individual placode, with losses preferentially occurring for circumferential interfaces and gains for radial interfaces (Figure 5D and Figure 5—figure supplement 1). We defined the number of productive neighbour gains as the difference between circumferential and radial gains, since equal numbers of both would cancel each other out. In order to control for any variability between placodes or between the number of cells tracked per placode, we expressed the number of productive gains as a proportion of the number of cell-cell contacts that were available to perform a circumferential T1 per time step (see Materials and methods). The proportion of productive circumferential gains was approximately constant, which lead to a steady net gain over time (Figure 5C’’). Furthermore, the cumulative strain attributable to discrete T1s calculated here (e^0.09 = 1.09) is in good agreement with the cumulative strain over the same time window that is attributed to the continuous process of intercalation (1.1 in Figure 2E).

In addition to typical T1 exchanges, multicellular rosette structures could easily be identified amongst the placodal cells (Figure 5E–F). Rosette formation began prior to the first sign of tissue-bending, but the number of rosettes per placode increased afterwards (Figure 5E,E’). Rosettes were usually formed of five to seven cells, with most involving only five cells (Figure 5E’, F). By contrast, rosettes observed during Drosophila germband extension can be formed of up to 12 cells (Blankenship et al., 2006). The strain rate analysis already indicated that, overall, intercalation events should be polarised to produce a contraction in the circumferential orientation with the corresponding expansion polarised towards the pit (Figure 2C–E’). Analysis of rosette formation and resolution in our time-lapse datasets of embryos expressing a membrane marker demonstrated that groups of cells contracted in a circumferential orientation to form a rosette, the resolution of which then moved individual cells towards the invaginating pit (Figure 5F and Video 5), thereby leading to the expansion observed in the strain rate analysis.

Video 5

Download asset

Non-muscle myosin II is the major driver of cell shape changes in many different contexts (Levayer and Lecuit, 2012; Röper, 2013, 2015), and it has been shown to play an important role in T1 transitions and in rosette formation driving the convergence and extension events during germband extension in the Drosophila embryo (Blankenship et al., 2006; Fernandez-Gonzalez et al., 2009; Rauzi et al., 2010). We imaged embryos expressing palmitoylated RFP (Ubi-TagRFP-CAAX) as a membrane label and a GFP-tagged version of non-muscle myosin II regulatory light chain (called spaghetti squash, sqh, in Drosophila) under control of its own promoter in the null mutant background (sqhAX[3]; sqhGFP42) to assess myosin II distribution and intensity as a proxy for myosin activity (Video 6). In all individual rosette formation-resolution examples analysed (n = 29), junctional myosin II appeared particularly enriched in the form of short cable-like structures at the central contact sites of the rosette (Figure 5G,H). These cables initially spanned several cell diameters and shortened concomitant with the cells being drawn into a central vertex (Figure 5I). The orientation of the short myosin cables correlated with the direction of rosette-formation, but in contrast to germband extension was not always oriented parallel to the dorsal-ventral axis, but rather following the circumferential coordinates of the placode.

Video 6

Download asset

We now investigated whether these circumferential intercalations were actively driven within the apical domain to promote invagination.

Even though the quasi-3D analysis detailed above strongly indicates that the intercalation of cells far from the pit initiates from the apical domain, the process of initiation itself could be either actively driven or a passive response. Intercalation far from the pit would be active if it arose as an intrinsic property of this part of the tissue and actively drove circumferential contraction of the tissue (Figure 6A; red curved arrows). The capacity of these cells for such active behaviour would likely be achieved through genetic patterning. Intercalation would be passive if the active apical constriction near the pit drove a passive ‘funnelling’ of far cells towards the pit, the radial pull leading to passive polarised intercalations (Figure 6A’; blue arrows).

Figure 6 with 1 supplement see all

Download asset Open asset

Figure 6—source data 1

Predictions (junction length, angles opposite shrinking junctions and cell elongation) for circumferential T1, circumferential baseline, radial T1 and radial baseline and their statistics.

Related to Figure 6F–H.

https://doi.org/10.7554/eLife.35717.033

Download elife-35717-fig6-data1-v4.xlsx

In order to distinguish between an active mechanism and a passive mechanism, we made four predictions (Figure 6B–E’), comparing metrics of junction length, angles at vertices, cell elongation and junctional myosin II accumulation. For each metric, we compared circumferentially oriented junctions with radial junctions. We also distinguished whether junctions were or were not shrinking junctions, about to be involved in a T1 event (see Materials and methods). We compared T1 data with similarly oriented non-T1 data to ask if there was a more active signature to shrinking junctions. We also compared radial versus circumferential non-T1 data to ask whether junctions as a whole in one orientation had a more active signature.

For the first two predictions (Figure 6B–C’), we compared real interface lengths and vertex angles with interface lengths and angles predicted by a Voronoi tessellation (see Materials and methods). We considered that a Voronoi tessellation generated from cell centroid seeds represents a mechanically neutral configuration for the cell-cell junctions and angles, and controlled for variation in local geometry around focal junctions. The direction in which real junction lengths or angles deviated from neutral Voronoi geometries was indicative of an active or a passive mechanism.

The deviation of real junction length from predicted Voronoi junction length (Figure 6B,B’) has previously been used as a geometric proxy for junction stress in the germband (Tetley et al., 2016). In the salivary gland placode, junction lengths were shorter than Voronoi predicted lengths for shortening T1 junctions compared to non-T1 junctions, most strongly for circumferentially oriented junctions (Figure 6F and Figure 6—figure supplement 1B). Circumferential non-T1 interfaces also deviated from neutral geometries by being shorter on average than expected, with radial junctions longer (Figure 6F). This suggests that circumferential junctions, and circumferential T1s in particular, were contracted, possibly by an intrinsic contractile mechanism, rather than the cells being pulled away from each other radially by the pit.

We predicted that active circumferential junction contraction would impose more acute angles at their associated vertices (Figure 6C, [Rauzi et al., 2008]), whereas a radial pull would lead to more obtuse angles (Figure 6C’). On average, angles linked to circumferential junctions were indeed more acute, relative to neutral Voronoi geometry, than those linked to radial junctions (Figure 6G and Figure 6—figure supplement 1C). Similarly, cells were on average more elongated circumferentially (Figure 6H and Figure 6—figure supplement 1D) as predicted by actively contracting circumferential junctions and incompatible with a radial pull.

These three geometrical measures for cells far from the pit, where intercalation dominates, are therefore consistent with an active intercalation mechanism, leading to circumferential neighbour gains and thus circumferential convergence.

We also predicted that active circumferential junction contraction would be driven by myosin II accumulation at circumferential junctions (Figure 6E). Recent studies have indicated that in some tissues, as response to mechanical pulling, myosin II accumulates and becomes polarised at junctions parallel to the pulling force (Duda et al., 2018; Fernandez-Gonzalez et al., 2009). Hence if intercalation occurred as a passive response, we would expect either no myosin II polarisation or mechanically induced localisation to radially oriented junctions (Figure 6E’). The localisation and increase in myosin II observed at the central junctions during rosette formation/resolution in the placode, as shown above (Figure 5G–I), indicated that some junctional myosin II was circumferentially enriched. We now set out to analyse junctional myosin II distribution systematically across the whole placode.

When apical junctional myosin intensity was quantified in fixed samples of sqhGFP embryos across the early placode, a clear enrichment of myosin II was apparent in circumferentially compared to radially oriented junctions (Figure 7A,B). Such tissue-wide circumferential versus radial polarisation of myosin clearly supported an active mechanism of cell intercalation, with circumferential myosin II likely assisting both rosette and T1 vertex formation during the polarised cell intercalation events.

Figure 7

Download asset Open asset

Figure 7—source data 1

Myosin and Bazooka fluoresce intensity in radial and circumferential junctions and Myosin unipolary and bipolarity over time.

Related to Figure 7B,E,F and I.

https://doi.org/10.7554/eLife.35717.035

Download elife-35717-fig7-data1-v4.xlsx

To compare and correlate junctional myosin II dynamics with the above strain rate analysis, we analysed myosin II polarisation dynamically across the whole tissue (Figure 7C,C’). Quantitative tools that allow junctional polarisation of myosin and other players to be quantified have previously been established (Figure 7D,D’’; [Tetley et al., 2016]). Myosin II polarisation at circumferential junctions could either occur through enrichment at two opposite junctions or sides, termed bi-polarity (Figure 7D’ and Video 7), or through enrichment at a single junction or side within a cell, termed unipolarity (Figure 7D’’ and Video 7). Starting about 10 min prior to tissue bending, bi-polar enrichment of myosin II at circumferential junctions was significantly stronger than at radial junctions when measured across the whole placode (Figure 7E). Similarly, starting ~ 5 min later, unipolar enrichment of myosin II at circumferential junctions dominated over radial enrichment when measured across the whole placode (Figure 7F). This clear and increasing circumferential junctional polarisation of myosin II together with the geometrical signatures discussed above strongly supported an active mechanism of T1 vertex and rosette formation (Figure 6A). Less clear is whether the resolution of rosettes and vertices is equally actively driven. Instead, this part of the intercalation events could respond to cell-extrinsic cues, such as the pulling of the invaginating pit, akin to the role of the posterior midgut invagination during germ band extension (Collinet et al., 2015; Lye et al., 2015) or pulling forces generated through micro-aspiration re-orienting intercalations in embryonic mouse tissues (Wen et al., 2017b).

Video 7

Download asset

In cell intercalation events during germband extension, myosin polarisation is complementary to enrichment of Par3/Bazooka (Baz) as well as Armadillo/β-catenin, with both enrichments controlled by the upstream patterning and positioning of transmembrane Toll receptors and Rho-kinase (Rok) (Blankenship et al., 2006; Paré et al., 2014; Simões et al., 2010). We therefore analysed whether such complementarity was also present in the early salivary gland placode. Antibody labelling of Baz often showed a complementarity in membrane enrichment to myosin II (Figure 7G–I), most pronounced where myosin II was organised into circumferential mini cables during intercalation events (Figure 7G,H). Baz was enriched at radially oriented junctions where myosin II was low, and vice versa at circumferential junctions, though in contrast to germband extension, the Baz polarisation did not extend uniformly across the tissue and was overall less strong than the myosin II polarisation (Figure 7I).

Thus, in order to adapt to a circular tissue geometry and to the need for ordered invagination through a focal point during the process of tube budding, a conserved molecular pattern of myosin-Baz complementarity is apparently imprinted onto the salivary gland placode in a radial coordinate pattern, rather than the prevailing A-P/D-V pattern of the earlier embryo during germband extension (Figure 7J,K).

In order to address how the radial pattern of behaviours and molecular factors across the placode is established, we analysed mutants in a key factor of salivary gland tube invagination, the transcription factor Fork head (Fkh). Fkh is expressed just upon specification of the placodal cells in a dynamic pattern spreading across the whole placode (Figure 8—figure supplement 1A–C), directly downstream of the homeotic factor Scr (Zhou et al., 2001). In fkh mutants, invagination of the placode fails, and towards the end of embryogenesis salivary gland-fated cells undergo apoptosis as Fkh appears to prevent activation of pro-apoptotic factors (Jürgens and Weigel, 1988; Myat and Andrew, 2000a). Previous studies have concluded that Fkh promotes cell shape changes important for invagination, in particular apical constriction (Chung et al., 2017; Myat and Andrew, 2000a). Fkh is not the only transcription factor important for correct changes during invagination, but works in parallel to for example Huckebein (Hkb), the lack of which also confers significant problems with apical cell shape changes and invagination (Myat and Andrew, 2000b, 2002).

We combined a fkh null mutant (fkh[6]) with markers allowing membrane labeling and cell segmentation (Ubi-TagRFP-CAAX) as well as myosin II quantification (sqhGFP; see Video 8). When fkh mutant placodes were compared to wild-type ones at late stage 11 (beyond the time window analysed here) then fkh mutant placodes showed no sign of a dorsal-posterior invagination point (Figure 8A–B’; Figure 8—figure supplement 1D–E’’). In fact, many placodes beyond late stage 11 showed a centrally located shallow depression (Figure 8A,A’, yellow dotted lines). Strain rate analysis of five segmented movies of fkh[6] mutant placodes, spanning the equivalent time period to the wild-type movies analysed above (Figure 8—figure supplement 2C), showed that there was no constriction at the tissue level near the pit (Figure 8C–E and Figure 8—figure supplement 2D,D’). In fact, if anything, there was a slight tissue expansion (Figure 8E) caused by a slight expansion at the cell level (Figure 8E’), with zero intercalation (Figure 8E’’). Away from the pit fkh[6] mutant placodes expanded slightly (Figure 8F and Figure 8—figure supplement 2E,E’), again mostly due to cell shape changes (Figure 8F’, F’’).

Figure 8 with 2 supplements see all

Download asset Open asset

Figure 8—source data 1

Cumulative strain and statistics for cells near to the pit and far from the pit for wild type and fkh[6] mutant.

Related to Figure 8E and F.

https://doi.org/10.7554/eLife.35717.041

Download elife-35717-fig8-data1-v4.xlsx

Video 8

Download asset

Is this strong reduction of cell behaviours due to a complete ‘freezing’ of the placode in the fkh[6] mutants? We analysed whether junctional myosin II was still polarised in the fkh[6] mutant placodes. In fixed and live samples, fkh[6] mutant placodes still showed the circumferential actomyosin cable surrounding the placode (Figure 8—figure supplement 1E’’, compare to D’’; [Röper, 2012]). When myosin II unipolarity and bi-polarity were quantified from five segmented and tracked movies, at the tissue-level there was a strong and significant reduction in myosin II polarisation, with no difference between radial and circumferential orientations in either measure and myosin II bi-polarity indistinguishable from zero (Figure 9A,B). Nonetheless, cells were not static but in fact neighbour exchanges were present, both in form of T1 exchanges and rosettes (Figure 9C–F). The quantitative analysis of neighbour gains from segmented and tracked movies in the fkh[6] mutant revealed that gains accumulated both circumferentially as well as radially, at a rate comparable to that observed in the control (Figure 9E). However, in contrast to the control, where circumferential gains significantly outweighed radial neighbour gains leading to a positive net rate of productive circumferential gains (Figure 9F, green line), in the fkh[6] mutant circumferential and radial gains occurred in equal amounts (Figure 9E) leading to a near zero net gain (Figure 9F, pink line). Interestingly, when focusing on individual events such as rosettes, despite a loss of tissue-wide myosin II polarisation, myosin II was still enriched at the central constricting junctions (Figure 9Cd’), and in fixed embryos smaller regions of myosin II-Baz complementarity could be identified (Figure 8—figure supplement 1F–G).

Figure 9

Download asset Open asset

Figure 9—source data 1

Unipolarity and bipolarity and dynamics of neighbour gains in fkh[6] mutant.

Related to Figure 9A–B and E–F.

https://doi.org/10.7554/eLife.35717.044

Download elife-35717-fig9-data1-v4.xlsx

Therefore, although at the tissue level fkh[6] mutant placodes appear near static, the close analysis of individual events revealed a highly dynamic but unpolarised intercalation behaviour across the mutant placodes. Without an actively invaginating pit and focussed apical constriction taking place in the fkh[6] mutant, the unpolarised intercalation events are not being resolved radially because of the absence of the pull from the pit. This finding therefore also suggests an active intercalation mechanism, where the remaining local increases in junctional myosin II still support formation of T1 vertices and rosettes, but without any overall directionality to their resolution.

The following transgenic fly lines were used: sqhAX3; sqh::sqhGFP42 (Royou et al., 2004) and fkhGal4 (Henderson and Andrew, 2000; Zhou et al., 2001) [kind gift of Debbie Andrew]; Scribble-GFP (DGRC Kyoto), UAS-palmYFP (generated from membrane Brainbow; [Hampel et al., 2011]), y¹ w^* cv¹ sqh^AX3; P{w^+mC = sqh-GFP.RLC}C-42 M{w^+mC = Ubi-TagRFP-T-CAAX}ZH-22A (Kyoto DGRC Number 109822, referred to as sqh^AX3;sqhGFP; UbiRFP); P{w + mC = sqh-GFP.RLC}C-42 M{w + mC = Ubi-TagRFP-T-CAAX}ZH-22A; fkh[6]/TM3 Sb Twi-Gal4::UAS-GFP (fkh[6] allele from Bloomington). See Table 1 for details of genotypes used for individual figure panels.

Embryos were collected on apple juice-agar plates and processed for immunofluorescence using standard procedures. Briefly, embryos were dechorionated in 50% bleach, fixed in 4% formaldehyde, and stained with primary and secondary antibodies in PBT (PBS plus 0.5% bovine serum albumin and 0.3% Triton X-100). anti-Crumbs and anti-E-Cadherin antibodies were obtained from the Developmental Studies Hybridoma Bank at the University of Iowa; anti-Baz was a gift from Andreas Wodarz (Wodarz et al., 1999); anti-Fkh was a gift from Herbert Jäckle (Weigel et al., 1989). Secondary antibodies used were Alexa Fluor 488/Fluor 549/Fluor 649 coupled (Molecular Probes) and Cy3 and Cy5 coupled (Jackson ImmunoResearch Laboratories). Samples were embedded in Vectashield (Vectorlabs).

Images of fixed samples were acquired on an Olympus FluoView 1200 or a Zeiss 780 Confocal Laser scanning system as z-stacks to cover the whole apical surface of cells in the placode. Z-stack projections were assembled in ImageJ or Imaris (Bitplane), 3D rendering was performed in Imaris.

For live time-lapse experiments embryos from [Scribble-GFP,UAS-palmYFP fkhGal4], [sqh^AX3; sqhGFP,UbiRFP] or [sqhGFP,UbiRFP; fkh[6]] were dechorionated in 50% bleach and extensively rinsed in water. Embryos were manually aligned and attached to heptane-glue coated coverslips and mounted on custom-made metal slides; embryos were covered using halocarbon oil 27 (Sigma) and viability after imaging after 24 h was controlled prior to further data analysis. Time-lapse sequences were imaged under a 40x/1.3NA oil objective on an inverted Zeiss 780 Laser scanning system, acquiring z-stacks every 0.8–2.6 min with a typical voxel xyz size of 0.22 × 0.22 × 1 μm. Z-stack projections to generate movies in Supplementary Material were assembled in ImageJ or Imaris. The absence of fluorescent Twi-Gal4::UAS-GFP was used to identify homozygous fkh[6] mutant embryos. During the early stages of salivary gland placode morphogenesis analysed here, fkh[6] mutants showed no reduction in cell number or initiation of apoptosis (data not shown). The membrane channel images from time-lapse experiments were denoised using nd-safir software (Boulanger et al., 2010).

Cell tracking was performed using custom software written in IDL (code provided in (Blanchard et al., 2009) or by email from G.B.B.). First, the curved surface of the embryonic epithelium was located by draping a ‘blanket’ down onto all image volumes over time, where the pixel-detailed blanket was caught by, and remained on top of binarised cortical fluorescence signal. Different quasi-2D image layers were then extracted from image volumes at specified depths from the surface blanket. We took image layers at 1–3 and 7–8 μm for the apical and mid-basal depths, respectively. Image layers were local projections of 1–3 z-depths, with median, top-hat or high/low frequency filters applied as necessary to optimise subsequent cell tracking.

Cells in image layers at these two depths were segmented using an adaptive watershedding algorithm as they were simultaneously linked in time. Manual correction of segmented cell outlines was performed for all fixed and time-lapse data. The segmentation of all the movies used in this study was manually corrected to ensure at least 90% tracking coverage of the placode at all times. Tracked cells were subjected to various quality filters (lineage length, area, aspect ratio, relative velocity) so that incorrectly tracked cells were eliminated prior to further analysis. The number of embryos analysed and number of cells can be found in Tables 1 and 2 (also see Figure 2—figure supplement 1 (WT apical), Figure 3—figure supplement 1 (WT basal) and Figure 8—figure supplement 1(f, K, h)).

WT movies were aligned in time using as t = 0 min the frame just before the first sign of invagination of cell apices at the future tube pit was evident. fkh[6] mutants were aligned using as a reference of embryo development the level of invagination of the tracheal pits that are not affected in the fkh[6] mutant as well as other morphological markers such as appearance and depth of segmental grooves in the embryo. Cells belonging to the salivary placode (without the future duct cells that comprise the two most ventral rows of cells in the primordium) were then manually outlined at t = 0 mins using the surrounding myosin II cable as a guide and ramified forwards and backwards in time. Only cells of the salivary placode were included in subsequent analyses.

At t = 0 min, the centre of the future tube pit was specified manually as the origin of a radial coordinate system, with radial distance (in µm) increasing away from the pit (e.g. Figure 1G). Circumferential angle was set to zero towards Posterior, proceeding anti-clockwise for the placode on the left-hand side of the embryo, and clockwise for the placode on the right so that data collected from different sides could be overlaid.

The radial coordinate system was ‘mobile’, in the sense that its origin tracked the centre of the pit, forwards and backwards in time, as the placode translated within the field of view due to embryo movement or to on-going morphogenesis.

Detailed spatial patterns of the rates of deformation across the placode and over time quantify the outcome of active stresses, viscoelastic material properties and frictions both from within and outside the placode. We quantified strain (deformation) rates over small spatio-temporal domains composed of a focal cell and one corona of immediate neighbours over a ~5 min interval ([Blanchard et al., 2009] and reviewed in [Blanchard, 2017]). On such 2D domains, strain rates are captured elliptically, as the strain rate in the orientation of greatest absolute strain rate, with a second strain rate perpendicular to this (Figure 2B).

For the early morphogenesis of the salivary gland placode, in which there is no cell division or gain/loss of cells from the epithelium, three types of strain rate can be calculated. First, total tissue strain rates are calculated for all local domains using the relative movements of cell centroids, extracted from automated cell tracking. This captures the net effect of cell shape changes and cell rearrangements within the tissue, but these can also be separated out. Second, domain cell shape strain rates are calculated by approximating each cell with its best-fit ellipse and then finding the best mapping of a cell’s elliptical shape to its shape in the subsequent time point, and averaging over the cells of the domain. Third, intercalation strain rates that capture the continuous process of cells in a domain sliding past each other in a particular orientation, is calculated as the difference between the total tissue strain rates and the cell shape strain rates of cells. Strain rates were calculated using custom software written in IDL (code provided in (Blanchard et al., 2009) or by email from G.B.B.).

The three types of elliptical strain rate were projected onto our radial coordinate system (see Figure 1F), so that we could analyse radial and circumferential contributions. Strain rates in units of proportional size change per minute can easily be averaged across space or accumulated over time. We present instantaneous strain rates over time for spatial subsets of cells in the placode, and cumulative strain ratios for the same regions over time. These plots were made from exported data using MATLAB R2014b.

We calculated strain rates in layers at two depths for WT placodes. Because we applied a radial coordinate system originating in the centre of the future tube pit to both depths, and used the same reference t = 0 min, we were able to compare strain rates at the same spatio-temporal locations on the placode between the depths.

We also wanted to characterise the 3D geometries of cells within small domains, using the cell shapes and arrangements in the two depths we had tracked. To do this, we needed to correctly match cells between apical and mid-basal depths. We manually seeded 3–5 apico-basal cell matches per placode, then used an automated method to fill out the remaining cell matches across the placode and over time.

We did this by sequentially looking at all unmatched apical cells that were next to one or more matched cell. The location of the unmatched basal centroid was predicted by adding the vector between matched and unmatched apical cell centroids to the matched basal centroid. The nearest actual basal centroid to this predicted basal centroid location was chosen as the match, on condition that the prediction accuracy was within 0.25 of the apical centroid distance. Information from multiple matched neighbours was used, if available, improving the basal centroid prediction. Progressively matching cells out from known matched cells filled out placodes for all embryos. We visually checked apical-basal matches for each embryo in movies of an overlay of apical and basal cell shapes with apico-basal centroid connections drawn (Video 3).

Having matched cells between apical and mid-basal layers, we have access to information about approximate 3D cell shapes. For single cells, we can measure how wedged the cell shape is in any orientation and how tilted is the apical centroid to basal centroid ‘in-line’ relative to the surface normal (Figure 4A). In particular, we are interested in the amount of wedging and tilt in our radial and circumferential placode orientations (Figure 4A). However, an important aspect is missing which is that cells can be arranged differently apically versus basally, independently of any cell wedging. That is they can be to a greater or less extent interleaved. Interleaving in depth is completely analogous to cell intercalation in time (Figure 4—figure supplement 1D). The degree of interleaving is a multi-cell phenomenon, so we return to our small domains of a cell and its immediate neighbours. Within these small domains we want to quantify the amount of cell wedging, interleaving and cell tilt.

To correctly separate these quantities, we borrow from methods developed to separate out the additive contributions of wedging, interleaving and tilt that account for epithelial curvature (Deacon, 2012). Deacon shows that for small domains of epithelial cells, curvature across the domain is the sum of their wedging and interleaving, while cell tilt has no direct implication for curvature.

During our study period, salivary placodes have minimal curvature (average curvature is less than 0.05 µm⁻¹ in both AP and DV axes). We therefore simplify the problem to flat (uncurved) domains, uncurving any local curvature so that apical and basal cell outlines are flat and parallel. Usefully, we can then treat small domains in exactly the same way as we have done above to calculate strain rates above, but instead of quantifying the rate of deformation over time, here we calculate rate of deformation in depth, or ‘z-strain rates’ (Figure 4—figure supplement 1D).

The cell shape strain rate is equivalent to the wedging z-strain rate (Figure 4B) and the intercalation strain rate is the interleaving z-strain rate (Figure 4D). Both of these add up to a total or tissue strain rate and a total or tissue z-strain rate (Figure 4—figure supplement 1E), respectively. Note that the total z-strain rate of a domain can be the result exclusively of wedging or of interleaving (as in Figure 4B,D; as with temporal strain rates, Figure 2A), but some combination of the two is more likely. Domain translation and rotation for temporal domains become domain tilt (Figure 4F) and twist in the 3D domain geometries, respectively. Domain twist is very weak in spatial or temporal averages of the placode (data not shown).

The units of wedging and interleaving are proportional size change per µm in z, and tilt is simply a rate (xy µm/z µm). We use the convention that change is from basal to apical, so a bottle-shape has negative cell wedging. The number of embryos and cells analysed for 3D cell geometries can be found in Tables 1 and 2.

We used changes in neighbour connectivity in our tracked cell data to identify neighbour exchange events (T1 processes). Neighbour exchange events were defined by the identity of the pair of cells that lost connectivity in t and the pair that gained connectivity at t + 1. The orientation of gain we defined as the orientation of the centroid-centroid line of the gaining pair at t + 1. We further classified gains as either radially or circumferentially oriented, depending on which the gain axis was most closely aligned to locally. We did not distinguish between solitary T1s and T1s involved in rosette-like structures.

From visual inspection, we knew that some T1s were subsequently reversed, so we characterised not only the total number of gains in each orientation but also the net gain in the circumferential axis, by subtracting the number of radial gains. Furthermore, when comparing embryos and genotypes, we controlled for differences in numbers of tracked cells by expressing the net circumferential gain per time step as a proportion of half of the total number of tracked cell-cell interfaces in that time step. We accumulated numbers of gains, net gains, and proportional rate of gain over time for WT (Figure 5C) and fkh (Figure 9E,F) embryos. Two sample Kolmogorov-Smirnov test was used to determine significance at p<0.05 for data in Figure 9F.

In the region of the placode away from the pit, we considered that active circumferential intercalation driven by junctional shortening would produce a different signature in the geometries of cells from a pull from the actively constricting cells at the pit (as depicted in Figure 4A,A’). We took an approach that avoided using simplistic raw measures of junction lengths and angles as these do not take into account differences in the geometry of cells immediately surrounding a junction.

First, we considered that junctions approaching a T1-transition that are being actively shortened by myosin II motors would be shorter than expected according to some neutral geometry. We used a Voronoi tessellation based on actual cell centroids as our reference neutral geometry, allowing us to compare actual junction lengths with predicted neutral lengths (as used in [Tetley et al., 2016]). The Voronoi tessellation provides cell junctions as the set of points equidistant between neighbouring cell centroids (see magenta dashed lines in Figure 6B,B’). We calculated the deviation from Voronoi in µm, with a negative value indicating that junctions were likely to be actively shortened. A positive value would indicate a passively elongated junction. In practice, placode cells (and germ-band cells, see Figure 5D–F in [Tetley et al., 2016]) are less ordered than a Voronoi tessellation, with the result that cell-cell interfaces are on average longer than tessellated interfaces. The baseline average value is therefore positive and not zero (see dashed line in Figure 6F). In principle, it would be better to generate neutral geometries using a mechanical model but that would require a further set of untested assumptions about the rheology of the placode tissue. We chose the Voronoi tessellation because it offers a very simple way of generating a plausible neutral geometry.

Second, we measured the vertex angles in the cells at opposite ends of focal junctions (see Figure 6C,C’). Raw angles would be expected to reduce from 120° at three-way junctions to 90° at a four-way T1-transition. However, again to control for variation in local cell geometry, we compared angles relative to those predicted by a Voronoi tessellation constructed from cell centroid seeds. Our measure is therefore the angular deviation from the Voronoi tessellation, in degrees, with a negative value indicating a narrower angle, which we would interpret as being drawn out by an actively contracting junction. Our angular deviation measure is measured as the average for the two vertices at either end of a focal junction.

Third, we calculated the elongation of the pair of cells at either end of a focal junction (Figure 6D,D’). Again, we considered that an actively shrinking focal junction will tend to elongate these cells, whereas a contractile pit would elongate cells in the radial direction towards the pit.

We classified junctions as being in the run up to a T1 (T1 data) or not involved in T1 (baseline data), and according to their (radial or circumferential) orientation and asked two types of question. We compared circumferential baseline to radial baseline data, asking if there was an overall active signature bias in the tissue. We also compared T1 data to baseline data for similarly oriented junctions, asking whether junctions involved in a T1 have a specific effect on local geometries.

For statistical tests, we again employed the mixed-effects models (see Statistics section below). In order to avoid using temporally correlated data in the comparisons we selected one time point for each data set we compared. For baseline data, we could choose any developmental time point because the baseline data for all three measures did not vary with developmental time (Figure 6—figure supplement 1B–D). We therefore chose the time point −2–0 min from the start of pit invagination, which maximised the number of placodes (see Figure 2—figure supplement 1A) and hence N cells. For T1 data we chose −3 min from neighbour exchange, near the middle of the time window before exchange where junction length proceeds consistently to extinction (Figure 6—figure supplement 1A).

Embryos of the genotype sqhAX3; sqh::sqhGFP42 (Royou et al., 2004) were labelled with either with anti-Bazooka and anti-DE-Cadherin or anti-DE-Cadherin and phalloidin to highlight cell membranes. For Figure 7B, myosin II signal intensity of junctions oriented either circumferentially or radially from five placodes was quantified using Image J. For Figure 7I fluoresce intensity of myosin II and Bazooka of junctions such as the ones indicated by arrows in Figure 7H’, that were oriented either circumferentially or radially from 10 placodes was quantified using Image J. 3-pixel wide lines were manually drawn along each junction. Intensity values were normalised to average fluorescence outside the placodes. We used a paired t-test for comparison of intensities within in the same junction, unpaired t-tests for comparison of circumferential myosin II vs radial myosin II and circumferential Baz vs radial Baz.

Whereas previously we quantified apicomedial Myosin II (Booth et al., 2014), here we focused on junctional Myosin II. We extracted a quasi-2D layer image from the Myosin II channel at a depth that maximised the capture of the junctional Myosin. We background-subtracted the Myosin images and quantified the average intensity of Myosin along each cell-cell interface within the placode. To calculate the average intensity, we set the width of cell-cell interfaces as the cell edge pixels plus two pixels in a perpendicular direction either side. This captured the variable width of Myosin signal at interfaces.

We further summarised the uni- and bipolarity of Myosin for each cell. Methods to calculate Myosin II unipolarity and bipolarity are described in detail in Tetley et al. (2016). Briefly, the average interface fluorescence intensity around each cell perimeter as a function of angle is treated as a periodic signal and decomposed using Fourier analysis. The amplitude component of period two corresponds to the strength of Myosin II bipolarity (equivalent to planar cell polarity) (Figure 7D’). Similarly, the amplitude of period one is attributed to myosin unipolarity (junctional enrichment in a particular interface) (Figure 7D”). The extracted phase of periods 1 and 2 (bi/uni-polarity) represent the orientation of cell polarity. We projected both polarities onto our radial coordinate system. Both polarity amplitudes are expressed as a proportion of the mean cell perimeter fluorescence. Cells at the border of the placode neighbouring the supra-cellular actomyosin cable were excluded from the analysis. The number of embryos and number of cells analysed can be found in Tables 1 and 2.

Statistical tests to determine significance of data shown are indicated in the figure legends. Significance in time-lapse movies was calculated for bins of 4.5 min using a mixed-effects model implemented in R (‘lmer4’ package as in [Butler et al., 2009; Lye et al., 2015]) with a significance threshold of p<0.05.

A mixed-effect model has fixed effects and random effects. The former are variables associated with an entire population and are expected to have an effect on the dependent variable. Random effects are factors which are associated with individual experimental units drawn at random from a population and introduce variation that is desirable to account for (Pinheiro and Bates, 2000). Here, to test for differences in instantaneous strain rates we used as fixed effect the genotypes (wild type and forkhead) while the variation between embryos (from the same genotype) was considered a random effect. The advantage of using a linear mixed-effect model is that this framework allows testing for differences on certain variables while accounting for sources of variation that are present in the full data set.

In the main figures we generally present cumulative strains, which are generated from instantaneous strain rate data shown in Figure supplements. The cumulative strains are calculated only from instantaneous data averages and so do not carry any of the distributions that were used to calculate instantaneous means and associated confidence intervals. We therefore perform all statistics on the instantaneous data, using the mixed-effects model above. For mixed-effects models, one cannot portray a single overall confidence interval. Instead, we have a choice to show one of within- or between-genotype confidence intervals, and we have chosen the former, as has been done previously (for example in [Butler et al., 2009; Gorfinkiel et al., 2009; Lye et al., 2015; Tetley et al., 2016]).

Therefore, error bars in time-lapse plots show an indicative confidence interval of the mean, calculated as the mean of within-embryo variances. The between-embryo variation is not depicted, even though both are accounted for in the mixed effects tests.

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

The authors thank the following people; for reagents and fly stocks: Debbie Andrew, Andreas Wodarz, Herbert Jäckle; for help with image analysis: Jerôme Boulanger. Work in the Röper lab is supported by the Medical Research Council (file reference number U105178780). GBB was supported by grant no. 15.23(k) from the Isaac Newton Trust, by Wellcome Trust grant no. 100329/Z/12/Z awarded to William Harris and Biotechnology and Biological Sciences Research Council Standard Grant BB/J010278/1 to Richard Adams and Bénédicte Sanson. GBB also thanks Wellcome Trust Investigator Award [099234/Z/12/Z] to Bénédicte Sanson and the Department of Physiology, Development and Neuroscience, University of Cambridge, UK.

1. Received: February 6, 2018
2. Accepted: July 16, 2018
3. Accepted Manuscript published: July 17, 2018 (version 1)
4. Accepted Manuscript updated: July 23, 2018 (version 2)
5. Accepted Manuscript updated: July 24, 2018 (version 3)
6. Version of Record published: August 13, 2018 (version 4)

© 2018, Sanchez-Corrales 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.