1. Cell Biology
  2. Developmental Biology
Download icon

Radially patterned cell behaviours during tube budding from an epithelium

  1. Yara E Sanchez-Corrales
  2. Guy B Blanchard  Is a corresponding author
  3. Katja Röper  Is a corresponding author
  1. MRC Laboratory of Molecular Biology, United Kingdom
  2. University of Cambridge, United Kingdom
Research Article
  • Cited 3
  • Views 2,140
  • Annotations
Cite this article as: eLife 2018;7:e35717 doi: 10.7554/eLife.35717

Abstract

The budding of tubular organs from flat epithelial sheets is a vital morphogenetic process. Cell behaviours that drive such processes are only starting to be unraveled. Using live-imaging and novel morphometric methods, we show that in addition to apical constriction, radially oriented directional intercalation of cells plays a major contribution to early stages of invagination of the salivary gland tube in the Drosophila embryo. Extending analyses in 3D, we find that near the pit of invagination, isotropic apical constriction leads to strong cell-wedging. Further from the pit cells interleave circumferentially, suggesting apically driven behaviours. Supporting this, junctional myosin is enriched in, and neighbour exchanges are biased towards the circumferential orientation. In a mutant failing pit specification, neither are biased due to an inactive pit. Thus, tube budding involves radially patterned pools of apical myosin, medial as well as junctional, and radially patterned 3D-cell behaviours, with a close mechanical interplay between invagination and intercalation.

https://doi.org/10.7554/eLife.35717.001

eLife digest

Tubes form many of the organs in the animal body, from lungs to kidneys to intestines; but how are these structures created during development? For example, the tube that composes the salivary gland of the fruit fly emerges from a flat patch of cells. First, a dimple develops in the cell layer and moves inwards to create the tube pit. Then, like water down a plughole, the rest of the cells flow towards this point and fold downwards into the tube. However, it is still unclear exactly which mechanisms drive this process.

Here, Sanchez-Corrales et al. combine microscopy and computational approaches to follow how cells behave in a fruit fly embryo as they build the future salivary gland. The results confirmed that the tube starts forming because of ‘apical constriction’: cells, which normally look like prisms, squeeze into a wedge shape. This helps the tissue to bend and create the tube pit.

Other mechanisms contribute to the extension of the tube by turning the flat surface of cells into a curved one. In particular, further away from the dimple, cells become tilted towards the cylinder as they move into it. Another process reshuffles how these cells are connected to each other – a mechanism known as neighbour exchange – which leads to an overall movement towards the dimple. As the tube develops, this creates an increasing number of smaller rings of cells around the pit, which helps the cells to form the walls of the cylinder.

Many developmental processes are similar across organs and even species, and a next step could be to explore whether the mechanisms described by Sanchez-Corrales et al. are also present outside of the fruit fly’s salivary glands. If so, this could shed light on what happens when tubes fail to form correctly in an embryo, and on how we could create these structures in the laboratory.

https://doi.org/10.7554/eLife.35717.002

Introduction

During early embryonic development, simple tissue structures are converted into complex organs through highly orchestrated morphogenetic movements. We use the formation of a simple tubular epithelial structure from a flat epithelial sheet as a model system to dissect the processes and forces that drive this change. Many important organ systems in both vertebrates and invertebrates are tubular in structure, such as lung, kidney, vasculature, digestive system and many glands. The formation of the salivary glands from an epithelial placode in the Drosophila embryo constitutes such a simple model of tubulogenesis (Girdler and Röper, 2014; Sidor and Röper, 2016). Each of the two placodes on the ventral side of the embryo (Figure 1A) consists of about 100 epithelial cells, and cells in the dorso-posterior corner of the placode begin the process of tube formation through constriction of their apical surfaces, leading to the formation of an invagination pit through which all cells eventually internalise (Girdler and Röper, 2014; Sidor and Röper, 2016).

Figure 1 with 1 supplement see all
Morphogenetic events in the salivary gland placode show a radial organisation.

(A, A’). Schematic of a stage 11 Drosophila embryo highlighting the position of the salivary gland placode (green) in lateral (A) and ventral (A’) views; A: anterior, P: posterior; D: dorsal; V: ventral. (B, B’). Surface and cross-section views of the salivary gland placode just prior to the first tissue bending (B) and once the initial pit of invagination has formed (B’). Lateral membranes are labelled by ScribbleGFP (green) and apical cell outlines by Crumbs (magenta). Scale bars are 5 µm. (C-F) Workflow of the morphometric analysis. Early salivary gland placode morphogenesis is recorded by time-lapse analysis using markers of cell outlines (C), cells are segmented and tracked over time (D; exemplary tracks of individual cells are indicated by coloured lines). Data are recorded and expressed in a radial coordinate system with the invaginating pit as the origin (E; ‘rad’ is the vectorial contribution radially towards the pit, ‘circ’ is the vectorial circumferential contribution). Various derived measures are projected onto the radial coordinate system. Here, example projected small domain deformation (strain) rates are shown (F; contraction in green, expansion in magenta; see Figure 2B). See also Videos 1 and 2. (G-H) Cumulative time-resolved analysis of apical area change for nine embryos (G’), the cumulative area strain is plotted, grouping the placodal cells into radial stripes (G) for the analysis. This reveals a clear split in behaviour between cells far from the pit (H, top; negligible area change) and cells near the pit (H, bottom; constriction, consistently negative area change). See also Figure 1—figure supplement 1.

https://doi.org/10.7554/eLife.35717.003

Apical constriction, a cell behaviour of epithelial cells that can transform columnar or cuboidal cells into wedge-shaped cells and can thereby induce and assist tissue bending, has emerged as a key morphogenetic module utilised in many different events ranging from mesoderm invagination in flies, Xenopus and zebrafish to lens formation in the mouse eye (Lee and Harland, 2007; Martin and Goldstein, 2014; Martin et al., 2009; Plageman et al., 2011). Apical constriction relies on the apical accumulation of actomyosin, that when tied to junctional complexes can exert pulling forces on the cell cortex and thereby reduce apical cell radius (Blanchard et al., 2010; Mason et al., 2013). Two pools of apical actomyosin have been identified: junctional actomyosin, closely associated with apical adherens junctions, as well as apical-medial actomyosin, a highly dynamic pool underlying the free apical domain (Levayer and Lecuit, 2012; Röper, 2015).

Another prominent cell behaviour during morphogenesis in all animals is cell intercalation, the directed exchange of neighbours, that is for instance the driving force behind events such as convergence and extension of tissues during gastrulation. Also during cell intercalation apical actomyosin activity is crucial to processes such as junction shrinkage and junction extension that underlie this cell behaviour (Bertet et al., 2004; Collinet et al., 2015; Rauzi et al., 2010; Zallen and Wieschaus, 2004). Importantly, all cell behaviours during morphogenesis require close coordination between neighbouring cells. This is achieved on the one hand through tight coupling of cells at adherens junctions, but also through coordination of actomyosin behaviour within groups of cells, often leading to seemingly supracellular actomyosin structures in the form of interlinked meshworks and cables (Blankenship et al., 2006; Röper, 2012, 2013).

The coordination between cells at the level of adherens junctions as well as actomyosin organisation and dynamics allows a further important aspect of morphogenesis to be implemented: the force propagation across cells and tissues. There is mounting evidence from different processes in Drosophila that force generated in one tissue can have profound effects on morphogenetic behaviour and cytoskeletal organisation in another tissue. For instance, during germband extension in the fly embryo, the pulling force exerted by the invagination of the posterior midgut leads to both anisotropic cell shape changes in the germband cells (Lye et al., 2015) and also assists the junction extension during neighbour exchanges (Collinet et al., 2015). During mesoderm invagination in the fly embryo, anisotropic tension due to the elongated geometry of the embryo leads to a clear anisotropic polarisation and activity of apical actomyosin within the mesodermal cells (Chanet et al., 2017).

We have previously shown that in the salivary gland placode during early tube formation when the cells just start to invaginate, the placodal cells contain prominent junctional and apical-medial actomyosin networks (Booth et al., 2014). The highly dynamic and pulsatile apical-medial pool is important for apical constriction of the placodal cells, and constriction starts in the position of the future pit and cells near the pit continue to constrict before they invaginate (Booth et al., 2014). The GPCR-ligand Fog is important for apical constriction and myosin activation in different contexts in the fly (Kerridge et al., 2016; Manning et al., 2013), and fog expression is also clearly upregulated in the salivary gland placode downstream of two transcription factors, Fkh and Hkb (Chung et al., 2017; Myat and Andrew, 2000b). Fkh is a key factor expressed in the placode directly downstream of the homeotic factor Sex combs reduced (Scr) that itself is necessary and sufficient to induce gland fate (Myat and Andrew, 2000a; Panzer et al., 1992). fkh mutants fail to invaginate cells from the placode, with only a central depression within the placode forming over time (Myat and Andrew, 2000a).

Here, we use morphometric methods, in particular strain rate analysis (Blanchard et al., 2009), to quantify the changes occurring during early tube formation in the salivary gland placode. Many morphogenetic processes are aligned with the major embryonic axes of anterio-posterior and dorso-ventral. An excellent example is germband extension in the fly embryo, where polarised placement of force-generating actomyosin networks is downstream of the early anterio-posterior patterning cascade (Blankenship et al., 2006; Simões et al., 2010). In the case of the salivary gland placode, the primordium of the secretory cells that invaginate first is roughly circular, with an off-center focus due to the invagination point being located in the dorsal-posterior corner, prompting us to assess changes within a radial coordinate framework. In addition to previously characterised apical constriction (Booth et al., 2014; Chung et al., 2017), we demonstrate that circumferentially oriented directional intercalation of placodal cells plays a major contribution to ordered invagination at early stages.

In addition, we compare quantitative planar strain rate analysis at different apical-basal depths of the tissue, and link cells between the planes to calculate rates of change of local geometry in depth, as a proxy for a full 3D analysis. We uncover that cell geometries and behaviours in 3D are also radially patterned: near the pit of invagination, where apical-medial myosin II is strong (Booth et al., 2014), cells are isotropically constricting apically leading to apical cell wedging, and with distance from the pit cells progressively tilt towards the pit. Cells also interleave apically towards each other in a circumferential direction, which can lead to different neighbour connectivity along their basal-to-apical length, equivalent to a T1 transition in depth. This strongly suggests apically driven active intercalation behaviours. We further show that several measures of ‘geometrical stress’ have signatures indicating that circumferential intercalation in the cells away from the pit is active. In addition, across the placode junctional myosin II is enriched in circumferential junctions, suggesting polarised initiation of cell intercalation through active junction shrinkage. This is followed by polarised resolution of exchanges towards the pit, thereby contributing to tissue invagination. forkhead (fkh) mutants, that fail to form an invagination, still show cell intercalations within the placode at a high rate, further supporting the active nature of the intercalations. Thus, tube budding depends on a radial organisation of 3D cell behaviours, that are themselves patterned by the radially patterned and polarised activity of apical myosin II pools, with apical-medial myosin dominating near the invagination point and polarised junctional myosin dominating further away from the pit. The continued initiation of cell intercalation but lack of polarised resolution in the fkh mutant, where the invagination is lost, could suggest that a tissue-intrinsic mechanical interplay also contributes to successful tube budding.

Results

Apical cell constriction is organised in a radial pattern in the salivary gland placode

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).

Table 1
Number of cells in time-lapse experiments.
https://doi.org/10.7554/eLife.35717.006
FigureCellsNumber of embryosnumber of cells per time point at [−18, −13.5, −9, −4.5, 0, 4.5, 9, 13.5, 18] min
Figure 2Apical cells near to pit9[123, 599, 724, 1132, 980, 737, 468, 219, 80]
Figure 2Apical cells far from pit9[156, 837, 1183, 1880, 1731, 1484, 1151, 744, 319]
Figure 3Basal cells near to pit6[260, 453, 726, 569, 546, 506, 360, 89]
Figure 3Basal cells far from pit6[382, 762, 1309, 1200, 1145, 1133, 770, 247]
Figure 43D proxy cells near to pit5[36, 307, 420, 479, 336, 410 260 148 58]
Figure 43D proxy cells far from pit5[37, 350, 515, 652, 479, 564, 394, 243, 131]
Figure 8Apical fkh6 cells near to pit5[27, 161, 248, 269, 215, 238, 233, 186, 85]
Figure 8Apical fkh6 cells far from pit5[52, 343, 626, 735, 615, 706, 655, 550, 239]
FigureCellsNumber of embryosnumber of cells per time point at [−13.5, −9, −4.5, 0, 4.5, 9, 13.5, 18] min
Figure 7Uni and Bi-polarity WT4[207, 522, 1115, 918, 1091, 1125, 1047, 484]
Figure 9Uni- and Bi-polarity fkh65[528, 783, 815, 712, 809, 786, 708, 307]
Video 1
Example movie of early salivary gland placode morphogenesis in 3D.

Embryo of the genotype Scribble-GFP/fkhGal4::UAS-palmYFP as shown if Figure 1C and Figure 3A. Time stamp indicates time before and after initiation of tissue bending at t = 0. Scale bar 20 µm.

https://doi.org/10.7554/eLife.35717.007

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
Example movie showing evolution of apical cell identities and apical area.

Movie ExpID0356 with 100% tracking, is shown. Cell identities are randomly colour-coded on the left, apical area on the right. Area colour codes are identical to Figure 8C, please refer to this scale. Time stamp indicates time before and after initiation of tissue bending at t = 0. Scale bar 20 µm.

https://doi.org/10.7554/eLife.35717.008

Both apical constriction and cell intercalation are prominent during the early stages of tube morphogenesis

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
Apical strain rate analysis of early events during salivary gland placode invagination.

(A–A’’) During early salivary gland placode development the change in tissue shape (A) can be accounted for by cell shape changes (A’), cell intercalation (A’’) or any combination of the two. During this phase, cell division and gain or loss of cells from the epithelium (other than cells invaginating into the pit) do not occur. (B) For small domains of a focal cell and its immediate neighbours, tissue shape change is the sum of two additive contributions, cell shape change and cell intercalation (Blanchard et al., 2009). Both tissue shape change and cell shape change can be measured directly from the segmented and tracked time-lapse movies, so the amount of cell intercalation can be deduced. Strain rates are depicted as orthogonal lines, with contraction in green and expansion in magenta, and the length of the line proportional to the rate of strain (grey circles mark 2 and 4% change per minute). The last panel shows the group of cells (central cell and first corona of neighbours) for which the example strain rates in (B) were calculated, with orthogonal lines indicating the lengths of the major and minor axes of the cells. The vertical contraction in the group of cells can be clearly seen as a combination of vertical cell shortening and continuous rearrangement/intercalation (evidenced by one topological change). (C–C’’’) Spatial maps summarising the strain rate analysis covering 18 min prior to 18 min post commencement of tissue bending. Mapped onto the shape of the placode (C) the strain rate contribution towards the pit (‘rad’) is shown, quantified from data from nine embryos (see Figure 2—figure supplement 1). (C’) Tissue contraction (green) dominates near the pit, with expansion (magenta) towards the periphery. The tissue contraction near the pit is mostly due to cell constriction near the pit (C’’), whereas the tissue expansion is contributed to by both cell expansion and cell intercalation far from the pit (magenta in C’’ and C’’’). Strain rates are given in proportion (pp) of change per minute. (D–E’’) Regional breakdown of time-resolved cumulative strain. In the area near the pit (D’’) tissue constriction dominates (grey curves in D and D’) and is due to isotropic constriction at the cell level (green curves in D and D’), whilst intercalation only plays a minor role in this region (orange curves in D, D’). Far from the pit (E’’), the tissue elongates towards the pit (E, grey curve), with a corresponding contraction circumferentially (E’, grey curve), and this is predominantly due to cell intercalation (orange curves in E and E’). Statistical significance based on a mixed-effects model (see Materials and methods) and a p<0.05 threshold (calculated for instantaneous strain rates [see Figure 2—figure supplement 1]), is indicated by shaded boxes at the top or bottom of each panel (tissue vs cell shape is green and tissue vs intercalation is orange). See also Video 3. Cumulative strains here and in all subsequent plots are calculated as the exponent of the cumulative instantaneous strain rates, which are shown in Figure supplements.

https://doi.org/10.7554/eLife.35717.009
Video 3
Example movies of Tissue strain rate tensor (SRT), Cell Shape SRT and Intercalation SRT.

Local strain rates are extracted from segmented and tracked movies. Rates of tissue shape change (left), cell shape change (centre) and intercalation (right) are shown. Green vectors represent contraction and magenta vectors represent expansion. Time stamp indicates time before and after initiation of tissue bending at t = 0. See also Figure 2.

https://doi.org/10.7554/eLife.35717.013

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.

A quasi-3D tissue analysis at two depths shows coordination of cell behaviours in depth

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
Strain rate analysis of early tubulogenesis using a 3D proxy.

(A–B) Epithelial cells of the placode are about 2–5 µm in apical diameter but extend about 15 µm into the embryo along their apical-basal axis. To assess behaviour of the tissue and cells at depth, we analysed a mid-basal level, about 8 µm basal to the apical surface. Scale bars are 5 µm. (C) Illustration of the method used: we segmented and tracked cells at a depth of ~8 µm, and repeated the strain rate analysis at this mid-basal level (blue) to compare with apical (pink). (D–D’) Comparison of apical (pink, as shown in Figure 2) and basal (blue) cumulative strains, for cell shape changes (D, E) and cell intercalation (D’, E’), near the pit (D,D’) and far from the pit (E,E’). Note how cell shape changes apical versus basal near the pit suggest progressive cell wedging as apices contract isotropically more rapidly than basally (D), and how the rate of cell intercalation across the tissue is highly coordinated between apical and basal, especially in the cells far from the pit (D’,E’). Statistical significance based on a mixed-effects model and a p<0.05 threshold (calculated for instantaneous strain rates [see Figure 2—figure supplement 1 and Figure 3—figure supplement 2]), is indicated by shaded boxes at the top of each panel: apical ‘rad’ vs basal ‘rad’ (dark grey) and apical ‘circ’ vs basal ‘circ’ (light grey).

https://doi.org/10.7554/eLife.35717.014

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’’).

Patterns of cell wedging, interleaving and tilt in the placode

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
3D cell geometries during early tube budding.

(A) Illustration of the method used for calculating 3D domain geometries using tracked cells at apical and mid-basal levels throughout the salivary placode. Although for simplicity only two matched cells are shown here (green outlines in both z-levels), all cells were matched accurately between levels (see text and Video 4). Small domains of a focal cell and its immediate neighbours were used to calculate local rates of wedging, interleaving and tilt (from cell ‘in-lines’, pink dashed lines) using z-strain rate methods (see text and Figure 4—figure supplement 1A–D). Measures were projected onto the pit-centred radial (dotted black lines) and circumferential axes. In this tilted side-view cartoon, the z distance between apical and mid-basal levels has been exaggerated. (B) Schematic of cell wedging in a cross-section and as individual z-levels (z1–z3) in the marked red cell. (B’) Cell wedging is patterned across the placode, increasing most strongly in cells near to the pit, in accordance with the isotropic apical constriction observed here (orange lines). Away from the pit, wedging contributes mainly in the circumferential direction (purple dashed line). (C) Schematic of cell interleaving in a cross-section and as individual z-levels (z1–z3) in the group of marked red cells. (C’) Radial cell interleaving (solid line) is always more positive than circumferential interleaving (dashed lines), often significantly so. Interleaving contributes to a radial expansion apically and a concomitant circumferential contraction. (D) Schematic of cell tilt in a cross-section and as individual z-levels (z1–z3) in the marked red cell. (D’) Cell tilt increased continuously in the radial direction (solid lines) apically towards the pit over the period of our analysis. A stronger rate of tilt was observed for the cells further from the pit (purple solid line), which is expected from the radial wedging seen near the pit. (B’, C’, D’) Error bars show the mean of the within-embryo variances for five movies. Significance at p<0.05 using a mixed-effect model (see Materials and methods) is depicted as shaded boxes at the top of the panel: B’) wedging in radial orientation, near to vs far from pit (dark grey, ‘rad’) and wedging in circumferential orientation, near to vs far from pit (light grey, ‘circ). (C’) Interleaving near to the pit, radial vs circumferential (orange, ‘near’) and interleaving far from the pit, radial vs circumferential (purple, ‘far’). (D’) Tilt in radial orientation, near to vs far from pit (dark grey, ‘rad’) and tilt in circumferential orientation, near to vs far from pit (light grey, ‘circ).

https://doi.org/10.7554/eLife.35717.020

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
Example of matching cells through depth.

Apical cell identities matched to basal cell identities are shown within the placode. Stamp refers to the depth in the tissue. Scale bar 20 µm. See also Figure 4.

https://doi.org/10.7554/eLife.35717.024

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.

Radially polarised T1 and rosette formation and resolution underlie cell intercalation in the placode

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
Cell intercalation during tube formation combines T1 exchanges and rosette-formation/resolution.

(A,B) Depending on the number of cells involved, intercalation can proceed through formation and resolution of a four-cell vertex, a so-called T1-exchange (A), or through formation and resolution of a rosette-like structure than can involve 5–10 or more cells (B). (C–C’’) Quantification of neighbour gains as a measure of T1 and intercalation events, with an example of a circumferential neighbour gain (leading to radial tissue expansion) shown in (C). (C’) Circumferential neighbour gains dominate over radial neighbour gains. (C’’) Rate of productive gains, defined as the amount of circumferential neighbour gains leading to radial tissue elongation and expressed as a proportion (pp) of cell-cell interfaces tracked at each time point. Data are pooled from seven embryo movies. (D) Represented for a single exemplary movie analysed (ExpID0356), interfaces that will be lost (blue) or have been gained (red) are shown, mapped onto a segmented version of the placode at the start of the movie for future losses (t = −7.5 min) and at the end of the movie for past gains (t = +20.0 min). Note that lost interfaces are mostly oriented circumferentially, whereas gained interfaces are mostly radial (see alternative visualisation in Figure 5—figure supplement 1). (E) Stills of two time-lapse movies used for the strain rate analysis, illustrating the appearance of rosette structures prior to (t= −11:44 min) and after (t = +5:18 min) tissue bending. (E’) The majority of rosettes observed in the salivary gland placode consist of five to six cells. Data are pooled from three embryo movies. (F–H) Stills of a time-lapse movie of an example of rosette formation-resolution. (F) The cluster of cells contracts along the circumferential direction of the placode, and resolution from the six-cell-vertex is oriented towards the pit (arrow). Label is Ubi-RFP-CAAX. (G,H) The same rosette as in (F) in close-up, showing Ubi-RFP-CAAX to label membranes in magenta (G) and myosin II-GFP (sqhGFP) in green (H). Note the prominent but transient myosin accumulation at the centre of the rosette-forming group of cells. (I) is a schematic of the rosette formation-resolution analysed in (F–H). See also Video 5. Scale bars in (E, F, H) are 5 µm.

https://doi.org/10.7554/eLife.35717.025

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 (e0.09 = 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
Example movie of apical rosette formation/resolution.

Embryo of the genotype sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX, only the UbiRFP label is shown. A group of cells going through rosette formation/resolution is highlighted, still of the movie are shown in Figure 5F. Time stamp indicates time before and after initiation of tissue bending at t = 0. Scale bar 20 µm.

https://doi.org/10.7554/eLife.35717.028

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
Example movie of cell shape and myosin II localisation in a control embryo.

Embryo of the genotype sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX used for the myosin II uni- and bi-polarity quantifications as shown in Figure 7. Time stamp indicates time before and after initiation of tissue bending at t = 0. Scale bar 20 µm.

https://doi.org/10.7554/eLife.35717.029

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

Analysis of signatures of active versus passive cell intercalation

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
Analysis of signatures of active versus passive cell intercalation.

(A, A’). Cell intercalation in cells far from the pit could either be an active process, helping to draw cells together circumferentially and to extend tissue towards the pit (A), or the active pulling from the contractile pit could lead to a passive funnelling of cells towards the pit with associated passive intercalations (A’). We considered four measures should allow to distinguish between the above active and passive mechanisms. (BB’) The lengths of actively shrinking circumferential junctions we predicted would be shorter than equivalent junction lengths of a Voronoi tessellation seeded with real cell centroid locations. We used the Voronoi tessellation as a proxy for ‘relaxed’ (not actively or passively altered) junction lengths and angles. Junctions pulled radially by the pit would be longer than predicted by Voronoi tessellation. (C,C’) Angles at vertices opposite actively shrinking circumferential junctions we predicted to be more acute than angles derived from a relaxed Voronoi tessellation, with the opposite predicted for pulled radial junctions. (D, D’) Active intercalation should lead to circumferential elongation of cells connected at vertices to shrinking junctions, whereas pulling from the pit would lead to radial cell elongation. (E, E’) Junctional actomyosin is predicted to be enriched in circumferential junctions that are actively shrinking. A response to pulling from the pit could lead to no enrichment or radially enriched junctional myosin (see text). (F) The junction lengths of both circumferentially and radially oriented junctions undergoing a T1 exchange showed an active signature (shorter than expected) compared to similarly oriented non-intercalating junctions. N = 135, 523, 78 and 618 junctions for the four distributions, respectively, from seven embryos. (G) Circumferentially oriented junctions far from the pit had an active signature of their vertex angles being more acute compared to radially oriented junctions, though junctions undergoing a T1 did not have more acute angles than junctions not directly involved in a T1. N = 139, 571, 84 and 675 junctions for the four distributions, respectively, from seven embryos. (F) Cells far from the pit had an active signature of being more elongated in the circumferential orientation. Cells at the ends of junctions involved in a T1 did not behave differently to junctions not involved in a T1. N = 135, 523, 78 and 618 junctions for the four distributions, respectively, from seven embryos. Error bars in (F–H) are ±SD.

https://doi.org/10.7554/eLife.35717.030

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.

Myosin is enriched in circumferential junctions across the 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.

Junctional myosin II shows a strong circumferential polarisation during tube formation.

(A–B) Example of junctional myosin II intensity, visualised using sqh[AX3]; sqhGFP in fixed embryos; (A) shows a heat map of intensity. (B) Quantification of apical junctional myosin polarisation, depicted as the ratio of junctional intensity (circumferential or radial) over embryo average (5 placodes from 5 embryos, 83 circumferential junctions versus 83 radial junctions; significance calculated using unpaired t-test; embryo average is average apical myosin intensity in the epidermis outside the placode). (C, C’) Myosin II enrichment was quantified from segmented and tracked time-lapse movies. (C) depicts the sqhGFP signal from a single time point of a movie, (C’) shows the calculated junctional myosin II intensity. See also Videos 6 and 7. (D–F) Myosin enrichment at junctions can occur in two flavours: (D’, E) Myosin II bi-polarity is defined as myosin II enrichment at two parallel oriented junctions of a single cell, calculated as the magnitude of a vector pointing at the enrichment (D’). Circumferential myosin II bi-polar enrichment (i.e. the radial bi-polarity vector, green in (D’) pointing at myosin II enrichment), increases between −15 min and + 18 min (E), green curve). (D’’, F) Myosin II unipolarity is defined as myosin II enrichment selectively on side of a cell (D’’). Circumferential myosin II uni-polar enrichment (i.e. the radial uni-polarity vector, green, in (D’’) pointing at myosin II enrichment), increases between −15 min and +18 min (F), green curve), whereas radial uni-polar enrichment (i.e. the circumferential uni-polarity vector in magenta in D’’), does not increase (F), magenta curve). (E, F) Error bars represent the mean of within-embryo variances of four movies, and significance between radial and circumferential bi- and uni-polarity at p<0.05 using a mixed-effects model (see Materials and methods) is depicted as shaded boxes at the bottom of panels E and F. See also Video 7. (G–H’’’) Examples of myosin II and Bazooka/Par3 polarisation in fixed embryos, two different regions of two placodes are shown. Note the complementary localisation of myosin II (green) and Bazooka (magenta). Scale bars are 5 µm. (I) Quantification of myosin II and Bazooka/Par3 polarisation in areas of strong junctional myosin II enrichment at circumferential junctions as shown in (G’) and (H’). Both circumferential (yellow arrows in H’) and radial (turquoise arrows in H’) fluorescence intensity are shown as ratios over embryo average, both for myosin II and Bazooka in the same junctions (from five embryos; mean and SEM are shown; paired t-test for comparison in the same junctions, unpaired t-tests for comparison of circumferential myo vs radial myo and circumferential Baz vs radial Baz; N = 33 circumferential junctions and N = 38 radial junctions across five embryos; embryo average is average apical myosin II or bazooka intensity in the epidermis outside the placode). (J, K) In contrast to the well-documented A-P/D-V polarity of cells in the Drosophila germband during elongation in gastrulation (J), the salivary gland placode appears to show a radial tissue polarisation, with a radial-circumferential molecular pattern imprinted onto it that instructs the morphogenesis (K).

https://doi.org/10.7554/eLife.35717.034

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
Example movie of automatic junctional myosin II quantification.

Myosin II uni-polarity vectors (left) and bi-polarity vectors (right) are shown. Junctions are colour coded according to their myosin II intensity levels as shown in Figure 7C’ (middle). Time stamp indicates time before and after initiation of tissue bending at t = 0. Scale bar 20 µm. See also Figure 7.

https://doi.org/10.7554/eLife.35717.036

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).

Resolution but not initiation of cell intercalation is disrupted in the absence of Forkhead

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
Loss of radial patterning of cell behaviours in salivary gland placodes lacking Fkh.

(A–B’) Examples of fkh mutant and wild-type placodes illustrating the lack of pit formation. A control placode at early stage 11 shows clear constriction and pit formation in the dorsal-posterior corner (yellow dotted line in cross-sections in B, B’), whereas even at late stage 11 a fkh mutant placode (beyond the time frame of our quantitative analysis) does not show a pit in the dorsal-posterior corner, instead a shallow central depression forms with some constricted central apices (yellow dotted line in cross-sections in A), (A’). Myosin II (sqhGFP) is in green and DE-Cadherin in magenta, white dotted lines in main panels outline the placode boundary. (C, D) Stills of segmented and tracked time lapse movies for control (C) and fkh[6] mutant (D) placodes, apical cell outlines are shown and colour-coded by apical cell area. Note the lack of contraction at the tissue and cell level in the fkh[6] mutant. (E–F’’) Analysis of cumulative apical strains in fkh mutant placodes (from five movies) in comparison to the analysis of wild-type embryos (as shown in Figure 2). Over the first 36 min of tube budding centred around the first appearance of tissue-bending in the wild-type and an equivalent time point in the fkh mutants, the fkh mutant placodes show only a slight expansion at the tissue level in the cells far from the predicted pit position (F) due to cell shape changes (F’) with very little intercalation contributing to the change (E’’, F’’). Statistical significance based on a mixed-effects model and a p<0.05 threshold (calculated for instantaneous strain rates [see Figure 2—figure supplement 1 and Figure 8—figure supplement 2]), is indicated by shaded boxes at the top of each panel: wt ‘rad’ vs fkh ‘rad’ (dark grey) and wt ‘circ’ vs fkh ‘circ’ (light grey).

https://doi.org/10.7554/eLife.35717.037
Video 8
Example movie of cell shape and myosin II localisation in a fkh[6] mutant embryo.

Embryo of the genotype sqh::sqhGFP42, UbiRFP-CAAX; fkh[6] used for the myosin II uni- and bi-polarity quantifications as shown in Figure 9. Time stamp indicates time before and after initiation of tissue bending at t = 0. Scale bar 20 µm.

https://doi.org/10.7554/eLife.35717.042

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).

Analysis of myosin patterns and intercalation behaviour in fkh mutants.

(A,B) Analysis of myosin II unipolarity (A) and bi-polarity (B) in fkh[6] mutant placodes compared to wild-type (as shown in Figure 7, see also Videos 7 and 8). Overall myosin II levels in fkh[6] mutant placodes are lower (not shown) and both unipolarity and bi-polarity are decreased, with bi-polarity near zero (B, solid curves). Error bars show intra-embryo variation of five embryo movies for fkh[6] and four embryo movies for wt. Statistical significance at p<0.05 using a mixed-effect model is indicated as shaded boxed at the bottom of the panels: wt vs fkh[6] for the radial vector/circumferential enrichment (green, ‘rad’) and wt vs fkh for the circumferential vector/radial enrichment (purple, ‘circ’). (C–F) Neighbour exchanges still occur in fkh[6] mutant placodes. (C) Sill pictures from a tracked and segmented movie of a fkh[6] mutant placode (labelled with membrane-RFP), frames (labelled a–j) are 1:55 min apart; the full placode view in (D) is from time point (a, d’) shows the myosin accumulation at the centre of the rosette structure observed in (d). (E) In comparison to wt where circumferential neighbour gains dominate over radial ones (dashed lines), in the fkh[6] mutant placodes both occur with equal frequency (solid lines). This leads to nearly no productive circumferential neighbour gains in the mutant, significantly fewer than in the wt (F); Kolmogorov-Smirnov two sample test D = 0.5833, p=0.0191).

https://doi.org/10.7554/eLife.35717.043

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.

Discussion

Morphogenesis sculpts many differently shaped tissues and structures during embryogenesis. A core set of molecular factors that are the actual morphogenetic effectors, such as actomyosin allowing contractility or cell-cell adhesion components allowing coordination and mechanical propagation of cell behaviours across tissues, are used iteratively in different tissues and at different times. By contrast, the activity of upstream activating gene regulatory networks leading to tissue identity, but also initial tissue geometry and mechanical constraints, are highly tissue-specific.

During tube formation of the salivary glands in the fly embryo, we observe a clear tissue-level radial organisation of cell behaviours, with the off-centre located invaginating pit as the organising focal point (Figure 10A). When analysed in 2D within the apical domain of the placodal epithelial cells, apical constriction dominates at the future pit of invagination and directional cell intercalation dominates further away from the pit. This intercalation achieves a circumferential convergence and radial extension of the tissue towards the invagination point (Figure 10B). Interestingly, these cell behaviours (cell shape change and cell intercalation) have previously been shown to drive other morphogenetic processes (Butler et al., 2009; Collinet et al., 2015; Lee and Harland, 2007; Lye et al., 2015; Martin and Goldstein, 2014; Martin et al., 2009; Plageman et al., 2011; Rauzi et al., 2010), but in our system they are utilised within a radial coordinate system, with the morphogenetic outcome being the formation of a narrow tube of epithelial cells from a round and flat placode primordium.

Summary of the radial patterning of 2D and 3D cell behaviours and actomyosin pools across the salivary gland placode during early tube formation.

(A) Circumferential tissue convergence through intercalation and apical constriction at the pit combine to result in radial tissue expansion towards the invagination point. (B) These 2D behaviours are associated with different actomyosin pools: far from the pit, circumferential junctional actomyosin underlies active intercalation through junction shrinkage, and near to the pit a pulsatile apical-medial actomyosin underlies apical constriction (Booth et al., 2014). (C) Quasi-3D analyses revealed that active apical intercalation (thick brown arrows) and isotropic constriction (thin brown arrows) lead in 3D to strong wedging of cells near the pit, strong tilting of cells far from the pit always towards the pit, as well as interleaving (i.e. change of neighbour connectivity along the apical-basal axis) across the tissue, aiding circumferential convergence and radial extension (cell and tissue movement indicated by blue arrows). (D) Once tissue bending has commenced at the pit, active apical constriction in and around the pit (thick and thin brown arrows) cooperates with active circumferential intercalations (medium brown arrows) to feed the elongation of the pit tube (blue arrows).

https://doi.org/10.7554/eLife.35717.045

In order to understand complex organ formation, it is important to understand cause and effect during the process. Cell behaviours can either be actively driven through for instance patterned actomyosin activity or they can be a mechanical response to events, or a combination of the two. An excellent example for active behaviours intersecting and being influenced by nearby events is the extension of the germband during Drosophila embryogenesis. In this tissue, an active mechanism of polarised intercalation combines with an extrinsic pulling force from the invaginating posterior midgut that helps the directional resolution of T1s (Butler et al., 2009; Collinet et al., 2015; Lye et al., 2015). During tube formation of the salivary gland placode, a similar intersection of active and passive mechanisms could be taking place: actively initiated intercalations combine with an active apical constriction at the pit that polarises the resolution of the exchanges (Figure 10C,D). The ongoing but unproductive intercalations in the fkh mutant, that lack directional formation and resolution, support the notion that the pulling of the constricting and invaginating pit reinforces the directionality of resolution in the wild type. The setup of active constriction and active intercalation in adjacent regions combined with some aspect of mechanical coupling between the two would thus be similar to the germband and posterior midgut, although with the placode being radially organised and the germband axially patterned. Thus, our work suggests the existence of iteratively used morphogenetic mechanisms that are highly adaptable to a particular tissue geometry and size. Compared to germband extension (Tetley et al., 2016), the signatures of active apical intercalation behaviour analysed here are reduced in magnitude. In particular, interfaces contract much more strongly away from the neutral Voronoi geometry in the dorso-ventral axis in the germband, compared to the circumferential axis in the placode. Furthermore, individual interfaces approaching a T1 are not significantly different from similarly oriented interfaces in the placode, whereas the strongly increasing fluorescence density of myosin associated with junction shortening in the germband distinguishes the geometry of these interfaces from non-T1 interfaces. This could be due to the overall much smaller size of the tissue, with changes accumulating at a tissue level that restrict further drastic deformation prior to a T1. This aspect can be addressed in future studies comparing even more active intercalation events in other tissues or through modelling approaches. It will also be crucial to determine what underlies the patterning of accumulation of the different myosin pools across the placode that likely allows the mechanical coupling of behaviours (Figure 10B).

The quasi-3D analysis comparing apical to basal sections also revealed a radial organisation of cell behaviours, specifically cell wedging, tilting and interleaving (Figure 10C,D). The constriction near the pit leads to strong cell wedging, with less wedging at a distance from the pit. Also in 3D, cells are tilting towards the point of invagination, with more tilt at a distance. In addition, interleaving of cells (the continuous change in arrangement of cells along their basal to apical length that can be but is not necessarily associated with discrete changes in neighbour connectivity in depth), occurs across the placode and is compatible with active apical circumferential convergence and radial extension of the tissue overall. Thus, our work revealed novel patterns of cell behaviours that could only be uncovered by considering the 3D context of a developing tissue.

Non-equilibrium 3D cell geometries in a flat epithelium, such as those caused by the wedging, interleaving and tilting analysed here, evolve during early placode morphogenesis in revealing patterns (Figure 10C,D). The appearance and progressive increase of these out-of-equilibrium geometries in the placode precede the start of 3D pit invagination by more than 10 min. This could suggest that during this phase a pre-pattern of tension could build up across the placode that would make the pit invagination more efficient, once initiated. Some of the cell behaviours we observe and the resulting complex 3D shapes might then be the integrated results of the balance of forces in the changing mechanical context of the placode. This will be possible to test experimentally by interference with certain behaviours or through ectopic induction of others, and can also be tested in silico in the future.

Our strain rate analysis in depth has revealed many interesting features of epithelial geometry and behaviour. What is still quite unclear during epithelial morphogenesis is whether morphogenetic behaviours are always initiated apically and propagated basally, or whether there is in fact an active contribution through events initiated at lateral or basal sides. A few recent reports suggest that not all is apically initiated (Monier et al., 2015; Sun et al., 2017), but whether this is a general principle or highly tissue-specific is unclear. It is also unclear how any cell behaviour, whether initiated apically or basally, is communicated and propagated across the length of the cell. In many morphogenetic processes including the tube budding from the salivary gland placode, actomyosin and other morphogenetic effectors are concentrated within the apical junctional domain. Our data at depth reveal that there is close coordination of intercalation rates between apical and mid-basal levels, and that the apically led interleaving and progressive wedging strongly support an active apical mechanism that is followed further basally.

In summary, our work uncovers a dynamic interplay of highly patterned cell behaviours within a radially organised tissue. Future research will show whether such radial patterning of myosin and cell behaviours is conserved across other tube-forming tissue primordia.

Materials and methods

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional information
Gene
(D. melanogaster)
Fork HeadNAFLYB:FBgn0000659
Gene
(D. melanogaster)
non-muscle myosin II/sqhNAFLYB:FBgn0003514
Gene
(D. melanogaster)
Bazooka/Par3NAFLYB:FBgn0000163
Genetic reagent
(D. melanogaster)
Scribble-GFPKyoto Drosophila
Genomic Research
Centre
Genetic reagent
(D. melanogaster)
fkh-Gal4PMID:10625560
Genetic reagent
(D. melanogaster)
Fkh-Gal4::UAS-palmYFPPMID:10625560,
PMID:21297621
and this study
stock generated upon
recombination of
Brainbow
Cassette stock
Genetic reagent
(D. melanogaster)
sqh[AX3]; sqh::sqhGFP42PMID:14657248
Genetic reagent
(D. melanogaster)
sqh::sqhGFP42,
UbiRFP-CAAX
Kyoto Drosophila Genomic
Research Centre Number
109822
Genetic reagent
(D. melanogaster)
fkh[6]Bloomington Drosophila
Stock Center, PMID:2566386
FLYB:FBal0004012
Antibodyanti-DE-Cadherin
(rat monoclonal)
Developmental Studies
Hybridoma Bank at the
University of Iowa
DSHB:DCAD2(1:10)
Antibodyanti-Crumbs
(mouse monoclonal)
Developmental Studies
Hybridoma Bank at the
University of Iowa
DSHB:Cq4(1:10)
Antibodyanti-Bazooka
(rabbit polyclonal)
PMID:10591216(1:500)
Antibodyanti-Forkhead
(guinea pig plyclonal)
PMID:2566386(1:2000)
AntibodyAlexa Fluor 488/549/649-
coupled secondary
antibodies
Molecular Probes
AntibodyCy3-, Cy5- coupled
secondary antibodies
Jackson Immuno
Research
SoftwareotracksPMID:19412170,
PMID:24914560
Software file (custom
software written in IDL)
Softwarend-safirPMID:19900849Denoising algorithm.
Available at
http://serpico.rennes.inria.fr/doku.php?id=software:nd-safir:index

Fly stocks and husbandry

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 sqhAX3; P{w+mC = sqh-GFP.RLC}C-42 M{w+mC = Ubi-TagRFP-T-CAAX}ZH-22A (Kyoto DGRC Number 109822, referred to as sqhAX3;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.

Embryo immunofluorescence labelling, confocal, and time-lapse

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], [sqhAX3; 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 segmentation and tracking

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)).

Table 2
Genotypes used for Figure Panels.
https://doi.org/10.7554/eLife.35717.046
FigurePanelEmbryo genotypesNumber of embryos used for quantitative analysis
Figure 1B, B'Scribble-GFPpanel representative of genotype (>30 embryos inspected)fixed
Figure 1CfkhGal4,UASpalm-YFP/Scribble-GFPpanel representative of genotype (>30 embryos inspected)live
Figure 1G,G’fkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1 and 8 (i.e. 9 in total)live
Figure 2C,D,EfkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1 and 8 (i.e. 9 in total)live
Figure 3A,BfkhGal4,UASpalm-YFP/Scribble-GFPpanels representative of genotype (>30 embryos inspected)live
Figure 3D,E,Fsqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX5live
Figure 4C,E,GfkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1 and 4 (i.e. 5 in total)live
Figure 5C',C'', C'''fkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP1 and 6 (i.e. 7 in total)live
Figure 5E,E'sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX3live
Figure 5F,G,Hsqh[AX3];sqh::sqhGFP42, UbiRFP-CAAXpanels representative of genotype (>30 embryos inspected)live
Figure 6F,G,HfkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP1 and 6 (i.e. 7 in total)live
Figure 7Asqh[AX3]; sqh::sqhGFP42panel representative of genotype (>30 embryos inspected)fixed
Figure 7Bsqh[AX3]; sqh::sqhGFP425fixed
Figure 7Csqh[AX3]; sqh::sqhGFP42 UbiRFP-CAAXpanels representative of genotype (>30 embryos inspected)live
Figure 7E,Fsqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX4live
Figure 7Gsqh[AX3]; sqh::sqhGFP42panels representative of genotype (>30 embryos inspected)fixed
Figure 7Hsqh[AX3]; sqh::sqhGFP42panels representative of genotype (>30 embryos inspected)fixed
Figure 7Isqh[AX3]; sqh::sqhGFP425fixed
Figure 8Asqh::sqhGFP42, UbiRFP-CAAX; fkh[6]panels representative of genotype (>30 embryos inspected)fixed
Figure 8Bsqh::sqhGFP42, UbiRFP-CAAX; fkh[6]/TM3panels representative of genotype (>30 embryos inspected)fixed
Figure 8Csqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAXpanels representative of genotype (>30 embryos inspected)live
Figure 8Dsqh::sqhGFP42, UbiRFP-CAAX; fkh[6]panels representative of genotype (>30 embryos inspected)live
Figure 8E,F (as in Figure 2C,D,E)fkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1 and 8 (i.e. 9 in total)live
Figure 8E,Fsqh::sqhGFP42, UbiRFP; fkh[6]5live
Figure 9A,Bsqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX4live
Figure 9A,Bsqh::sqhGFP42, UbiRFP-CAAX; fkh[6]5live
Figure 9C,Dsqh::sqhGFP42, UbiRFP-CAAX; fkh[6]3live
Figure 9Esqh[AX3];sqh::sqhGFP42, UbiRFP-CAAX6live
Figure 9Esqh::sqhGFP42, UbiRFP-CAAX; fkh[6]5live
Figure 9FfkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP1 and 6 (i.e. 7 in total)live
Figure 9Fsqh::sqhGFP42, UbiRFP-CAAX; fkh[6]5live
Figure SupplementPanelEmbryo genotypeNumber of embryos used for quantitative analysis
Figure 2—figure supplement 1B-E'fkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1 and 8 (i.e. 9 in total)live
Figure 2—figure supplement 1Fsqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1live
Figure 3—figure supplement 1B-Fsqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX5live
Figure 3—figure supplement 2A-B’sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX5live
Figure 4—figure supplement 1B,Csqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1live
Figure 4—figure supplement 1EfkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1 and 4 (i.e. 5 in total)live
Figure 5—figure supplement 1A,Bsqh[AX3]; sqh::sqhGFP42, UbiRFP-CAAX1live
Figure 6—figure supplement 1fkhGal4,UASpalm-YFP/Scribble-GFP and sqh[AX3]; sqh::sqhGFP42, UbiRFP1 and 6 (i.e. 7 in total)live
Figure 8—figure supplement 1A-B'sqh::sqhGFP42panels representative of genotype (>30 embryos inspected)fixed
Figure 8—figure supplement 1D-F’’’sqh::sqhGFP42, UbiRFP-CAAX; fkh[6]panels representative of genotype (>30 embryos inspected)fixed
Figure 8—figure supplement 2A-E'sqh::sqhGFP42, UbiRFP-CAAX fkh[6]5live

Mobile radial coordinate system for the salivary placode

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.

Morphogenetic strain rate analysis

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.

Matching cells between apical and mid-basal layers

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).

z-strain rates: 3D domain geometries

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.

Neighbour exchange analysis

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.

Calculation of signatures of active versus passive intercalation

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).

Fluorescence intensity quantifications of Myosin and Bazooka

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.

Automated myosin II quantification and polarity

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 analysis of time-lapse data

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.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
    Quantification of cell and tissue behaviours during morphogenesis
    1. W Deacon
    (2012)
    Department of Physiology, Development and Neuroscience, University of Cambridge, Cambridge, United Kingdom.
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
    Fork head prevents apoptosis and promotes cell shape change during formation of the Drosophila salivary glands
    1. MM Myat
    2. DJ Andrew
    (2000a)
    Development 127:4217–4226.
  34. 34
    Organ shape in the Drosophila salivary gland is controlled by regulated, sequential internalization of the primordia
    1. MM Myat
    2. DJ Andrew
    (2000b)
    Development 127:679–691.
  35. 35
  36. 36
    Organogenesis in Drosophila melanogaster: embryonic salivary gland determination is controlled by homeotic and dorsoventral patterning genes
    1. S Panzer
    2. D Weigel
    3. SK Beckendorf
    (1992)
    Development 114:49–57.
  37. 37
  38. 38
    Mixed-Effects Models in S and S-PLUS
    1. J Pinheiro
    2. D Bates
    (2000)
    Springer.
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
    Genetic Control of Salivary Gland Tubulogenesis in Drosophila
    1. C Sidor
    2. K Röper
    (2016)
    In: J Castelli-Gair Hombría, P Bovolenta, editors. Organogenetic Gene Networks. Switzerland: Srpinger International Publishing. pp. 125–149.
    https://doi.org/10.1007/978-3-319-42767-6_5
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55

Decision letter

  1. Didier YR Stainier
    Senior Editor; Max Planck Institute for Heart and Lung Research, Germany
  2. Stefan Luschnig
    Reviewing Editor; University of Münster, Germany

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: the authors were asked to provide a plan for revisions before the editors issued a final decision. What follows is the editors’ letter requesting such plan.]

Thank you for sending your article entitled "Radially-polarised cell behaviours drive tube budding from an epithelium" for peer review at eLife. Your article is being evaluated by four peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation is being overseen by Didier Stainier as the Senior Editor.

Given the list of essential revisions, including new experiments, the editors and reviewers invite you to respond within the next two weeks with an action plan and timetable for the completion of the additional work. We plan to share your responses with the reviewers and then issue a binding recommendation.

The full reviews are also included for your reference, as they contain detailed and useful suggestions. As you will see, the reviewers clearly find interest in the description of radially polarized cell behaviors during tube invagination. However, a number of issued and concerns are raised, a major one regarding the causal relationship between radially organized cell behavior and pit invagination.

The reviewers raised the following main points:

1) Causality. Cause and consequence are not clear with respect to the role of polarized cell behaviors in tube budding. Radially polarized intercalation of cells surrounding the pit may be, at least in part, a consequence of forces generated by the invaginating pit, rather than vice versa. Additional experiments would be required to dissect pit formation on the one hand and peripheral intercalation and myosin accumulation on the other hand. For instance, optogenetic tools might be used to trigger apical constriction and an ectopic invagination pit. This may allow to test whether intercalation behavior and myosin pattern respond to these changes, supporting a mechanical model, or if they do not respond, supporting intrinsic genetic control. If experimental evidence supporting such a causal role cannot be provided, claims about causality (including in the title) need to be diminished accordingly.

2) The analysis of forkhead mutants is interesting, as it reveals cell behavior in the absence of an invagination pit. However, it contributes little, if anything, to the resolution of issues related to causality. This should be discussed and made clearer in the text.

3) The analysis of cell behaviors is not truly "3D", as stated in the manuscript, but in fact cell shape is analyzed in 2D at two different levels along the apical-basal axis of the cells. An analysis of shape changes in 3D, and in particular of cell volume, would be more informative and would qualify as a "3D" analysis.

4) In the first part of the work, cell intercalation is inferred only indirectly from tissue behavior across a field of cells. Even though such measures are interesting to assess the relative contribution of cell shape and intercalation, the data would be more convincing if data are visualized per cell and if cell intercalation was assessed directly by quantifying the number and orientation of intercalation events, as in Figure 5. Reviewer 4 gives detailed suggestions on how this could be done.

5) There are a number of questions related to statistical analysis and the description of statistical methods in the manuscript. Mixed-model statistics needs to be explained. Experimental variance (e.g., confidence intervals) needs to be documented.

6) The description of experimental results should be removed from the Discussion section.

Reviewer #1:

This manuscript provides a comprehensive analysis of cellular behavior during the formation of a tubular epithelial organ, using the embryonic salivary gland of Drosophila as a model. Budding of tubular structures from flat epithelial sheets has previously been thought to be mainly driven by apical constriction, leading to the formation of an invagination pit. Here the authors used live imaging and extensive quantitative analyses to demonstrate that, in addition to apical constriction, salivary gland invagination is associated with radially oriented intercalation of cells surrounding the invagination pit. Importantly, the authors extended their analysis to 3D, which revealed additional cellular behaviors (wedging, tilting, interleaving) that are oriented either circumferentially or radially with respect to the invagination center. Interestingly, polarized cell intercalation is associated with accumulation of myosin II along circumferential junctions, suggesting that myosin activity is polarized radially across the placode and contributes to tube invagination. Finally, the authors show that in fkh mutants, which fail to form an invagination, intercalation still takes place, but intercalating cell clusters fails to resolve in a directional manner. Here, cause and consequence are not clear – do polarized cell behaviors 'drive' tube budding, as stated in the title, or are they, at least in part, a consequence of forces generated by the invaginating pit that are transmitted across the field of cells?

This work reveals important new insights into the cellular behaviors underlying tube formation and contributes significantly to a better mechanistic understanding of this fundamental process. The findings are original and are very likely to be of general relevance, as the cellular behaviors and topologies described here in Drosophila closely resemble those found in developing organs of other animals, including vertebrates.

The work is substantial, of very high standard and thoroughly described, although the description of image analysis methodology and statistics is not readily accessible to the non-specialist reader and makes some sections of the text difficult to read.

The manuscript is likely to be of significant interest to a broad audience of cell and developmental biologists. Overall, this is a strong piece of work that is appropriate for publication in eLife. However, the authors need to address the following points.

It is not easily conceivable how a symmetric tube can result from the invagination of cells at an asymmetrically (peripherally) placed invagination pit. What happens with the cells located between the pit and the margin of the placode (upper-right part of placode; Figure 1E)? Do radially polarized cell behaviors spread beyond the boundary of the placode, and do cells outside the placode participate in invagination?

Introduction section: "forkhead mutants, that fail to form an invagination, only show unproductive intercalations that fail to resolve directionally, likely due to the lack of an active pit."

This sentence seems to contradict the statement in the title that radially-polarised cell behaviours "drive" tube budding. Does the invagination pit drive the polarized intercalation of nearby cells, or vice versa? The issue of cause and consequence needs to be clarified and discussed in the text. The wording of the title may need to be adjusted accordingly.

Results, subsection “3D tissue analysis at two depth shows coordination of cell behaviours in depth”: "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 3E', magenta), but with a slower rate of constriction at mid-basal depth."

Isn't this precisely what is to be expected if cells near the pit undergo apical constriction (which was known before)?

Figure 6 E, F, and the accompanying text, appear unnecessarily complicated. The text refers to circumferential myosin enrichment, whereas Figure 6E, F and the legend refer to radial bi- or unipolarity of myosin distribution, respectively. Please simplify.

Is the distribution of myosin shown in Figure 6D a schematic drawing or representative of a real image?

Figure legends often lack information that is necessary to understand the data. Axis labels (e.g., "pp per min", "fluorescence increase over embryo average") need to be explained in the figure legends.

Reviewer #2:

This is an interesting manuscript by Sanchez-Corrales et al. that explores the cell shape changes that underlie the invagination of the salivary gland placode. The authors show that a combination of two common tissue shaping events, apical constriction and cell intercalation, function during early invagination. They also interpret their results in a radial coordinate system, which is nicely appropriate to the tissue context, and observe different topology-altering regimens dominating in different proximal-distal regions of the coordinate axis. Finally, they examine these cell behaviors at two different levels along the apical-basal axis as well as in a mutant that affects specification of the salivary placode. In general, these are interesting measurements and provide insight into tube and salivary placode invagination, although the impact of it is slightly hurt by an inability to tease apart the relative functional contributions of each regimen to invagination (i.e., it is not possible to alternately remove cell intercalation or apical constriction from the tissue) and the fact that the role of apical constriction in placode invagination is well-established. The figures are very nicely laid out, and, in some cases, quite beautiful. Their data does broadly support the interpretations presented in the paper. The attention paid to orienting schematics is appreciated and nicely done. The authors are very well-read and knowledgeable about morphogenesis, and carefully build and extend on the work from other systems, while the observation of circumferential intercalation is very nice. I would give a mild recommendation for publication, but will be interested to see the evaluations of other reviewers (which also partly goes to a level of confusion of what level of impact eLife requires). Apologies in advance if data was missed that explains any of the below.

Some issues to be addressed:

I'm not sure that "3D" analysis is accurate – this is 2D analysis done at a fairly limited two different planes (0µm and -8µm) in the tissue. It would be more accurate to say that this is an examination of apical and more basal contributions to cell shape and intercalation. Also, can the authors say whether cell volumes are maintained during invagination? It is a bit strange that similar constrictive behaviors are observed at -8µm; at some point there should be a cellular accommodation of the volume shifted due to apical constriction – either through a widening or deepening of the cell. There is a passing reference to cell deepening, but no hard data is presented. The authors should elaborate on this in the Results and Discussion. The analysis of cell wedging and the relationship to apical constriction and cell intercalation is quite interesting, although it takes significant effort to follow. Any editing and addition to the Discussion on these points will be helpful.

N numbers for the number of cells quantified in each category (for example, "near pit" and "far pit" cells) should be reported in each figure. The current reporting of the number of embryos should be kept.

Much of the statistical analysis is calculated through a mixed-effects model. More information on how mixed-effects models were applied to the data is needed to be able to evaluate the appropriateness of this statistical measurement. There should be at least a brief methods section on this.

Was there a statistical reason for splitting cells into "near" and "far" bins? I didn't see a clear statistical justification for this.

I was somewhat troubled by the inferred "intercalation strain rate" in the first sections of the manuscript, as this is a rather indirect measurement of topological changes that can be concretely measured, but they subsequently do exactly this in later portions of the manuscript.

There is an unofficial "results" section embedded into the Discussion. These should be moved to the results, or alternately, saved for a future publication. This section is a bit incongruous, given the Results sections, as it jumps into a cell fate discussion in the Discussion. The cell fate results are weakened by being correlative, without a functional component to test the hypothesis. An advantage of saving these results for another publication would be the ability to analyze such functional disruptions.

Discussion section penultimate paragraph – I am curious which "non-intuitive" results the authors are referring to?

Reviewer #3:

The study analyzes cellular mechanisms underlying initial events in salivary gland invagination, when a flat patch of cells invaginates to form a tube. Invagination is initiated by apical constriction of a few pit cells. The surrounding cells follow suit. Although the role of apical constriction has been reported previously, the cellular behaviour of the surrounding cells and their function for tube formation has not been dissected.

1) The authors report cell intercalation specifically towards the invagination site and propose that this may contribute to tube budding.

2) The authors aim to establish a three-dimensional view of the process by introducing a second focal layer (mid-basal) in addition to the apical view of the cells.

3) The authors report a polarized distribution of myosin II towards the site of invagination.

4) A mutant (forkhead) in which the cells are not specified displays only isotropic cell behaviour and myosin II distribution.

The authors conclude that a radially polarised pattern of cell behaviour, including apical constriction (as previously reported) and cell intercalation drives tube formation. The authors compare this pattern of cell behaviour with a radial pattern of myosin II accumulation and conclude that the radial pattern of myosin II would lead to radially polarised cell behaviour which would drive tube budding.

In my view the data do not allow to derive a causality. I accept that in wild type embryos the polarized myosin II pattern matches globally the preference of cell intercalation events. To address causality, the authors analyse forkhead mutants which completely lack tube invagination because cell specification is disrupted. It does not come as a surprise that cell and myosin II behave isotropically. The authors do not address the possibility that the radially polarized pattern of myosin II and cell behaviour is a mere consequence of mechanical pulling by the constricting and invaginating pit cells. forkhead mutants to not allow to rule out this option, as no invagination is observed in these mutants. The headline is thus an unjustified overstatement, as the radially polarized cell behaviour is likely to be a mere consequence of the activity of the small group of pit cells and certainly do not drive tube budding.

Specific issues:

In many figures the experimental variance is not visible. I understand that the authors tested the statistical significance. Yet, any sort of representation of the variance (e. g. confidence intervals) would help to assess the quality and the degree of uniformity of the data.

In the first part of the study, the degree of cell intercalation is only indirectly derived from the data as difference of total tissue behaviour and fitted cell shape changes. In Figure 5, the number and orientation of intercalation events is directly counted. The data would be more convincing, if also in the first part the contribution of intercalation would be directly measured. Given that the contribution of intercalation is the central conclusion of the study, it seems to be important that this parameter is directly measured.

3D: This part of the study is not convincing. I am fully aware of the difficulties and importance of a three-dimensional view of cell behaviour. Adding a second layer of recording does not contribute much to the issue however. In my view the presented data are not convincing and, in the end, even weaken the central conclusions. The authors conduct a sort of strain analysis along the axial direction, however with only two data points along the axis (apical and mid-basal).

Time axis: I find it confusing that the specific cell behaviour starts already in negative time. I understand that T=0 is set to the first visible invagination. As the polarized intercalation starts already at t=-15min, it would help to correlate the time axis to events at the time. Is this the first time when apical constrictions are observed?

Reviewer #4:

The paper offers interesting insights into tissue morphogenesis. Even though the single mechanical processes underlying this morphogenesis are not new, it is interesting to see this combination for a radial system. For the image analysis elegant and powerful methods were used. However, the principles underlying these analyses are not common knowledge among biologists. Therefore, extra care should be taken for the presentation of the results in order to make sure that biologists can still understand them. The current presentation is insufficiently intuitive and should be clearly improved. This should also allow readers to better judge the quality of the tracking. In addition, because of the simplicity of the system, the analysis method should be slightly simplified as control and also to contribute to better understanding.

Major points (better presentation and minor simplification) in detail:

1) When reading the manuscript, it becomes immediately clear that cell areas decrease at the pit. However, it is much less clear what happens to the regions further away. In addition, currently, it is not easy to see what happens with single cells. This hinders getting an intuitive picture of what happens to shapes of single cells and it does not become clear what the quality of the tracking was. Therefore, movies should be added that look similar to Figure 7C and D but with different coloring. In one movie, it should become visible what happens to the shapes of the radial stripes (Figure 1H’) and, in another one, what happens to individual cells.

2) The relative simplicity of the system should be exploited more to generate additional controls of the method, to simplify the understanding of the results, and possibly to obtain more precise outcomes. Currently, there is a focus on average strain rates and their accumulation. However, since the system has no apoptosis or division and relatively small strain changes, it should, at least for cellular strain rates, be possible to calculate total changes in strain directly for each cell, by comparing shapes in a certain time step directly with those in the first time step. For calculating the change in area, this would be very straightforward. It seems as if the cumulative area change is now calculated indirectly (but it is not clearly described whether this is indeed the case): ellipses are fitted to cells in such a way that the ellipses have the same areas as the cells. Then for each cell a strain rate tensor is obtained that is constrained to conserve the cell's area change, as assessed by the tensor's trace, which is a first order approximation of area change. Then the cumulative area change is obtained by adding up the traces, thus approximating total relative area change (1-0.1-0.1 is not the same as 1*0.9*0.9). Relative area change can also be obtained by only comparing areas at the first time step with those at a later time step. The authors should compare these values as a control of the method. In addition, they should use the direct method for area change, since it is easier to understand and a reader doesn't have to wonder for example why the cumulative area change is in pp per minute and not just in pp in Figure 1H. In addition, the total relative area change is actually the biologically relevant value in my opinion. Now it should not be a problem to do something similar for the strain change of the cells: here the fitted ellipse of a cell at a certain time step can be directly compared with the fitted ellipse of the same cell at the beginning and this should be used as a control. For calculating tissue shape change, the situation is a bit more complicated, since neighbors, and thus calculations, change during the time sequence. In order to do a direct comparison, the same cells should be compared at the beginning and the end. This should in principle be possible though, as long as the neighboring cells in the beginning (mostly) stay together. If other cells mix too much with the initial cells, definitions of strain do not really make sense any more. The authors should judge whether the direct approach is useful for tissue shape change. If it is, it would be useful to replace cumulative strain rates by these values, since they are a bit more straightforward.

3) The presentation of the figures should be clearly improved. Generally speaking, figures are currently missing that give a clear and intuitive overview of what is happening where in the tissue. In addition, tissue shape change, cell shape change, and cell intercalation are currently visualized by coloring according to strain rates, which is not a very intuitive quantity for many biologists. Instead, the figures should be such that readers can make a direct connection between quantitative values and shapes.

List of detailed suggestions to improve figures:

– Figures 1F and 3C are not very intuitive. First, a longer line is usually not associated with contraction. This should be changed. For example, arrows could be used, or the line length could be 1 as a standard and then be shortened or lengthened based on contraction and expansion, respectively. Secondly, the figure shows cell shapes, even though the strain rates are calculated based on a cell and its neighbors. It would therefore be better to make the cell outlines grey or something. Third, as far as I understand it, instantaneous strain rates are shown, so that there is much noise in the results and looking at the single figures may therefore not give much information. This can be improved by taking the total strain change (see point 2 of main points) or otherwise the cumulative strain rate, which should thus be calculated per cell.

– At the moment, Figure 1 does not clearly show what happens to the shapes of the radial stripes, even though this would be useful to better understand the movements in the tissue. This should be improved. One way could be to replace Figure 1G. This figure gives a good impression about time and position dependent cell shape changes, but may be replaced by another one that contains this information together with information at tissue level: one could for example replace it by a figure that contains a segmented image at the beginning, at t=0 and at the end. The boundaries between the radial stripes are thicker, so that it can be seen directly how their stripe shapes change. Cell shapes are then directly visible and can thus be assessed directly. However, in order to stress differences in area and compensate for interpretation difficulties due to the 2D projection, it would then still be useful to color individual cells according to area change. If desirable, cell aspect ratios could be color coded in additional images.

– The interpretation of Figure 2C, Figure 2—figure supplement 1B, Figure 3D, Figure 3—figure supplement 1B, and Figure 6—figure supplement 1A and B is not straightforward. The reader has to look up what the different colors mean and then couple that to the small inlet, which indicates whether the radial or circumferential direction is at play and then read the type of change that the figure shows. Even though the colors are nice to recognize patterns quickly, they don't confer any intuition on the extent of strain. In order to get such an idea, the data in the color bar need to be combined with reading the text to find out what kind of strain is visualized exactly and what the squares mean exactly. The authors should make these figures more intuitive. For example, they could use a segmented image of the end of a video. Each cell could then have arrows indicating strain changed in the radial (or circumferential) direction. Again, the total strain change (or the cumulative strain rate) can be used. In this way, it is immediately clear whether radial or circumferential strains are visualized and they allow for a direct more quantitative comparison between cells. In order to see quickly what is happening where, cells can be colored according to strain changes similarly to how the squares are colored now. Because of the presence of the arrows, it is then also clearer which color codes what. Depending on what the figure looks like exactly in the end, it may be useful to add a segmented image of the start as well and add radial stripe boundaries, so that the shape changes of single cells and their environment can also be looked at. In this way, shape changes of a cell and its neighbors could be directly compared to the length of the arrows and thus create an intuition for strain change. This would of course only show data of one embryo, but the quantification is in other figures anyway.

– Figure 4 does not give an intuitive overview of where which shape changes are present between the two layers. This should be improved. For example, two segmented z-projections could be shown in different colors in one figure. Then the reader could directly look at shape changes between the levels and see whether he can distinguish the wedging, interleaving and tilt.

– Figure 5 does not give any intuitive overview of where which neighbor exchanges occur. Such an overview should be added. For example, segmented images of the beginning and the end of a movie can be shown. Cell boundaries that will disappear may for example be drawn grey in the first image. New cell boundaries may for example be drawn blue in the image of the end of the movie. To get an idea of the number of reversals, each boundary that disappeared and then appeared again, could get another color.

[Editors’ note: formal revisions were requested, following approval of the authors’ plan of action.]

Thank you very much for your response, which was considered by the handling Senior Editor, the Reviewing Editor and the original reviewers. We are in general satisfied with your response and therefore would like to invite you to proceed with revisions. Please note that a major point that still remains to be resolved is the issue of causality – in the revised manuscript, any statements, including the title, implying that polarised cell behaviours cause ("drive") tube invagination will need to be carefully rephrased and toned down accordingly.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Radially-patterned cell behaviours during tube budding from an epithelium" for further consideration at eLife. Your revised article has been favorably evaluated by Didier Stainier (Senior Editor), a Reviewing Editor, and three reviewers.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

In the Abstract and in the manuscript text, there are several remaining instances of misleading statements (italicized) inferring a causative role of polarised cell behaviours in tube budding, where no such causative role is supported by experimental evidence:

Last sentence of the Abstract: "Thus, tube budding involves radially-patterned pools of apical myosin, medial as well as junctional, leading to radially-patterned 3D-cell behaviours."

Introduction section: "In addition, across the placode junctional myosin II is enriched in circumferential junctions leading to polarised initiation of cell intercalation through junction shrinkage."

Subsection “Analysis of signatures of active versus passive cell intercalation”: "This shows that circumferential junctions, and T1s in particular, are likely to be driven by an intrinsic contractile mechanism, and are not the result of cells being pulled away from each other radially."

Discussion section: "in our system they are utilised within a radial coordinate system, thereby leading to the formation of a narrow tube of epithelial cells from a round and flat placode primordium."

These and corresponding claims in the text need to be carefully adjusted and toned down accordingly, so as to avoid misleading implications that polarised cell behaviours are a main driving force of invagination.

https://doi.org/10.7554/eLife.35717.052

Author response

[Editors’ note: what follows is the authors’ plan to address the revisions.]

The full reviews are also included for your reference, as they contain detailed and useful suggestions. As you will see, the reviewers clearly find interest in the description of radially polarized cell behaviors during tube invagination. However, a number of issued and concerns are raised, a major one regarding the causal relationship between radially organized cell behavior and pit invagination.

The reviewers raised the following main points:

1) Causality. Cause and consequence are not clear with respect to the role of polarized cell behaviors in tube budding. Radially polarized intercalation of cells surrounding the pit may be, at least in part, a consequence of forces generated by the invaginating pit, rather than vice versa. Additional experiments would be required to dissect pit formation on the one hand and peripheral intercalation and myosin accumulation on the other hand. For instance, optogenetic tools might be used to trigger apical constriction and an ectopic invagination pit. This may allow to test whether intercalation behavior and myosin pattern respond to these changes, supporting a mechanical model, or if they do not respond, supporting intrinsic genetic control. If experimental evidence supporting such a causal role cannot be provided, claims about causality (including in the title) need to be diminished accordingly.

We would firstly like to apologise as apparently our phrasing of the title led to an interpretation or focus on an aspect of our findings that we did not intend. Our title was meant to refer to a radial patterning of [levels of] different cell behaviours across the placode, with isotropic constriction dominating in the domain near the pit, and oriented cell intercalation dominating in the region further away from the pit, as we illustrate in the schematic in Figure 8G. We did not mean to imply that the polarised cell intercalations that we describe are the main driving force of the invagination nor that the forces generated by the invagination pit do not play a role.

Second, to address directly the point raised, that the observed intercalation could be either a passive consequence of pit contraction or be intrinsically active, we propose to compare these two mechanisms comprehensively. The mechanisms lead to different predictions about cell behaviours, cell geometries and myosin behaviour (see Author response image 1). As we detail in the ‘action plan’ below, we have already tested one prediction, but propose to test a further three, that we anticipate will considerably improve this aspect of the paper.

In the ‘Active’ mechanism, polarised intercalation far from the pit arises from ‘intrinsic genetic control’ and actively drives circumferential contraction of the tissue (Author response image 1A). In the ‘Passive’ mechanism, the active apical constriction near the pit drives a passive ‘funneling’ of far cells towards the pit, the radial pull and circumferential crowding compression leading to passive polarised intercalations (Author response image 1B). This latter mechanism is what Reviewer 3 suggests with “the radially polarized cell behaviour is likely to be a mere consequence of the activity of the small group of pit cells”.

Author response image 1
Active versus Passive mechanisms driving intercalation.
https://doi.org/10.7554/eLife.35717.049

Before detailing specific predictions that allow us to distinguish between these mechanisms, we need to consider what we expect to see if placode morphogenesis is an overlay of both rather than one single mechanism. The Active mechanism could interact with apical constriction at the pit, the latter helping to pull out and thereby orient the resolution of T1s. Indeed, we believe this is likely to be the case, pending further confirmation from analyses we propose below. If both Active and Passive mechanisms are equally strong they could cancel, leaving no detectable signature of either. However, we think it likely that one will predominate, in which case we will detect its signature, even if this has been attenuated by the other mechanism.

We think it instructive to consider overlapping forces in another tissue earlier in Drosophila embryogenesis. Germband extension results from a combination of active polarized intercalation (set up by the AP-patterning system, work from Lecuit & Zallen labs) and an extrinsic pulling force from the invaginating posterior midgut (Lye et al., 2015; Collinet et al., 2015). These forces are analogous to the Active and Passive mechanisms in the placode, respectively, and in the same relative orientations. In the germband, the signature of both of these forces is detectable in cell geometries, myosin organization and strain rate gradients. We will further develop the discussion and Figure 8H along the above lines, pending what we find in proposed analyses below, and we will discuss in relation to the fkh mutant in which there is no invagination and cell neighbour exchange still occurs but it is not directional (see next section).

We believe four lines of evidence will allow us to distinguish which mechanism predominates:

Prediction 1

Orientation of junctional myosin II. According to the Active mechanism, we expect junctional myosin to be polarized, located along circumferentially oriented junctions, whereas if intercalation occurs as a passive response, we would expect either no myosin polarisation or mechanically induced localization in the perpendicular orientation (Author response image 1C). Therefore, the strong polarisation of junctional myosin along circumferential junctions that we have documented in the paper (Figure 6) strongly supports the Active mechanism.

In response to a point raised by Reviewer 3 (“the possibility that the radially polarized pattern of myosin II […] is a mere consequence of mechanical pulling”), to our knowledge only two studies have analysed myosin subcellular localisation in response to mechanical pulling. These are a very recent study using the Drosophila wing disc epithelium subjected to mechanical pulling (Duda et al., 2018) and an earlier study of supracellular actomyosin cables during germband extension (Fernandes-Gonzales et al., 2009). In both cases, myosin accumulates and becomes polarised at junctions parallel to the pulling force. Extrapolating to our placode system, we would therefore expect an accumulation of myosin at radially oriented junctions if this is due to mechanical pulling, when we observe the opposite. Circumferentially polarized junctional myosin thus supports the Active mechanism.

Prediction 2

Cell elongation. The Active mechanism predicts cells will have a preferential elongation in the circumferential orientation, drawn out by myosin-driven junction contraction, while the Passive mechanism predicts radial elongation, pulled from the pit (Author response image 1D). To establish the orientation of cell elongation, we first calculate best-fit ellipses to cell shapes, constraining the area of the ellipse to be the same as the cell area. Then we project the ellipse shape onto the radial axes to get radial and circumferential cell lengths. The strength of radial elongation is then the log-ratio of the radial to circumferential cell lengths, which can be positive (radially) or negative (circumferentially elongated). We will test whether on average cell elongation is significantly different from zero and in which orientation.

Prediction 3

Angles at vertices (following Rauzi et al., 2008). The Active mechanism predicts that angles opposite circumferentially oriented junctions will on average be more acute than 90 degrees, stretched by the contracting junctions in between (Author response image 1E). Conversely, the Passive mechanism predicts less acute angles as cells are stretched towards the pit.

Prediction 4

Junction lengths compared to ‘relaxed’ Voronoi lengths. To distinguish between the junctions that are shortening actively from those that may shorten passively, we will use an inference method to probe geometric stress. We assume that a Voronoi tessellation based on cell centroid locations represents a mechanically neutral configuration for the cell-cell junctions. We will compare real interface lengths with interface lengths predicted by a Voronoi tessellation to extract a length deviation from the Voronoi tessellation, a geometric proxy for local stress (Tetley et al., 2015). Circumferential junctions that are shorter than expected are likely to have been actively shortened by local myosin activity whereas longer junctions are a likely result of cells being stretched towards the pit and subsequently rearranging passively (Author response image 1F).

Finally, you suggest using optogenic methods to induce an ectopic pit. We have performed experiments along those lines and have tried the optogenetic tool developed by the de Renzis lab at EMBL to modify myosin activity (Guglielmi et al., 2015). However, this system has the inherent problem that two imaging channels are required for the optogenetic tool, thus we cannot simultaneously image a membrane channel and myosin channel to make our morphometric measurements. We also have tried several other experimental approaches with the purpose of inducing an ectopic constriction point. We have expressed a number of factors that should lead to increased myosin activity (such as an activated form of myosin or the catalytic domain of the myosin-activator Rho-kinase) in a stripe that covers the anterior part of the placode (i.e. opposite the forming pit, the domain where engrailed is expressed, using enGal4 for transgene overexpression). So far, none of these conditions has allowed us to genetically induce strong constriction. Therefore, although the idea of induction of an ectopic pit is tempting, technically this seems to still be impossible with current tools.

Moreover, in the scenario where we would be able to induce a second focus of invagination, and we could not detect a shift in the direction of intercalation, we have no means to rule out the possibility that the ectopic pit did not constrict as strongly as the original pit or that the tissue reached a new equilibrium to buffer an ectopic force. In our view, a quantitative assessment of the relative forces in our tissue deserves a full analysis, for example with various physical manipulations and force inference, which are beyond the scope of this work.

Action plan:

1) Change the title to make our interpretation clearer, using for instance “Radially-patterned cell behaviours drive tube budding from an epithelium”.

2) Explicit introduction of the Active and Passive mechanisms and their predictions (as shown here in Author response image 1) to current Figure 4.

3) Extend our analysis to test Predictions 2-4 listed above.

4) Re-write the manuscript to explain the mechanisms, predictions and new analysis results.

5) Elaborate more in the discussion about the interaction between the Active and Passive mechanisms, how they are synergistic but with distinct signatures, and allude to a combination in germband extension that has some similarities

2) The analysis of forkhead mutants is interesting, as it reveals cell behavior in the absence of an invagination pit. However, it contributes little, if anything, to the resolution of issues related to causality. This should be discussed and made clearer in the text.

We will adjust the results and discussion of fkh mutant in light of the two mechanisms proposed above (Author response image 1).

As the reviewers appreciated, the fkh mutant still shows intercalations occurring. Although the myosin polarisation is dramatically reduced in the fkh mutant, we could detect that the formation of rosettes still correlates with increased junctional myosin at the central junction of the forming rosette, as in the wild-type (example shown in Figure 8C). This finding thus supports an ‘Active’ mechanism in which cell intercalation is an active behaviour, rather than a passive response to activity elsewhere in the tissue.

Moreover, after testing our predictions (Author response image 1), we will contrast our findings in WT and fkh mutant within a radially organised tissue to an axially patterned tissue. In germband extension, an Active mechanism of polarised intercalation combines with an extrinsic pulling force from the invaginating posterior midgut that helps the directional resolution of T1s (Lye et al., 2015; Collinet et al., 2015). Our analysis in the fkh mutant shows that the productive directional neighbour exchange is dramatically decreased when an invaginating pit is not present (Figure 8E-F). Thus, an active intercalation in our tissue could combine with an apical constriction at the pit, the latter helping to pull out the resolution of T1s.

Another recent example of the role of forces to bias rosette resolution comes from the mouse embryo (Wen et al., 2017). Using micropipette aspiration to create an anisotropic force, the authors show that this ectopic stress is sufficient to rescue the resolution of T1s and rosettes in a mutant defective in intercalation (Fgfr2). Thus, the discussion of fkh mutant in this context will point towards a more generic mechanism of active intercalation combined with forces within the developing embryo to bias the direction of neighbour exchange resolutions.

Action plan:

1) Discuss our findings in fkh mutant in relation to possible Active and Passive mechanisms, comparing our findings to germband extension that has an analogous overlay of forces.

3) The analysis of cell behaviors is not truly "3D", as stated in the manuscript, but in fact cell shape is analyzed in 2D at two different levels along the apical-basal axis of the cells. An analysis of shape changes in 3D, and in particular of cell volume, would be more informative and would qualify as a "3D" analysis.

We completely agree that a true 3D analysis would be preferable. We therefore stated in the manuscript: To circumvent this issue, we used strain rate analysis at different depth as a proxy for a full 3D analysis”. As we hoped to explain in the manuscript, obtaining a full 3D analysis is still very unreliable (apart from hand-segmentation, which is not feasible – one movie tracing 80 cells over 20 sections in depth and 40 time points would require drawing and linking 64,000 cell outlines by hand). We have tested various available segmentation methods, some of which allow 3D segmentation (such as EDGE4D [Khan et al., 2014, Development]; and RACE [Stegmaier et al., 2016, Dev. Cell]), but these do not track cells in 3D reliably. If unreliable segmentation leads to too much loss of cell identities over time, then our computational methods break down for the strain rate analysis and the data analysed will be too sparse. Therefore, we decided that a 3D proxy as used in our study currently allows much better segmentation and tracking coverage and thus far better strain rate analysis (both in time and in z). We are happy to emphasise even further that our methods are a 3D proxy, or any other term that reflects this approximation better.

It is important to note that our work would thus far be the first published method allowing quantification of changes in cell wedging, cell interleaving or cell tilting as 3D cell behaviours. Adding a second layer (with both layers parallel to the curved epithelial surface) and especially in combination with linking identities of cells between these layers as done in Figure 4 allowed us to approximate changes in 3D cell shape such as cell wedging, cell interleaving and cell tilting. None of these behaviours are obvious or could be deduced from an apical analysis only: apical area shrinkage could well be compensated for by cell elongation in depth, requiring analysis in a different plane; cell interleaving (a change in neighbour contacts along the z-length of the cell) cannot be observed from a single plane, neither can cell tilting. Importantly, we show in Figure 4 that these three changes to 3D cell shape (even if only approximated by the analysis at two z-levels) is clearly radially patterned across the tissue. We will further integrate these results within the framework described here in Author response image 1.

Action plan:

1) Modify the text to emphasise that our method is a 3D approximation.

2) Include in the summary Figure 8G the 3D signatures revealed in our analysis.

4) In the first part of the work, cell intercalation is inferred only indirectly from tissue behavior across a field of cells. Even though such measures are interesting to assess the relative contribution of cell shape and intercalation, the data would be more convincing if data are visualized per cell and if cell intercalation was assessed directly by quantifying the number and orientation of intercalation events, as in Figure 5. Reviewer 4 gives detailed suggestions on how this could be done.

We would first like to point out that the term intercalation is not restricted to one particular measure. The process of intercalation can be measured in two ways, either as the continuous process of slippage of cells past each other (Blanchard et al., 2009; Kabla et al., 2010; Blanchard, 2017) that is captured by our strain rate analysis, or as discrete T1 topological change events. These are equally valid, though can be interestingly different. For example, looking at Figure 4E and 4F from Blanchard (2017) (PMC5379022 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5379022/figure/RSTB20150513F4/), though the cumulative amount of intercalation measured by both methods during germband extension in Drosophila is the same, the time-courses are quite different, with the continuous process starting some 5-10 minutes before T1 events occur, as would be expected from starting from an approximately hexagonal arrangement of cells. One could argue that the continuous process of slippage is a more faithful measure of the underlying myosin-based contractility that drives intercalation than the counting of topological T1 events.

Since in our tissue there is no contribution to tissue strain due to either cell divisions or cell death/cell delaminations, the simple equation of cell shape change plus intercalation equals tissue shape change holds true (Blanchard et al., 2009). So we would not agree that the intercalation strain rate is somehow less real for being ‘inferred’, rather it is precisely calculated as a difference between two directly measurable quantities.

Importantly, our direct measurement and analysis of T1 events in the placode as presented in Figures 5 and 6 closely matches the strain rate analysis. Specifically, the cumulative strain over our observed time window that is attributed to the continuous process of intercalation is 0.1 (Figure 2E) while it is 0.09 for discrete T1s (Figure 8F). This strongly supports the validity and complementarity of both approaches.

Action plan:

1) Highlight in the text that the intercalation strain rates, calculated either as a difference of two directly measured quantities (Figure 2 in the main text) and as amount of neighbour exchange (Figure 5 and 6), are comparable.

In addition:

We thank Reviewer 4 for very helpful suggestions on the presentation of the strain rate data. We propose making the following changes:

1) We will change all cumulative plots to strains, which means they will all start at 1.0 and deviate over time from 1, at the end of the study period showing the proportion of its original size the tissue has reached (due to a particular behaviour). This could be more intuitive, as the reviewer suggests. In Author response image 2 is a version of Figure 2D’ (near pit cells, circumferential strain) in this new representation so that you can see how cumulative panels will look. Note that one aspect of this new presentation that will be different is that a halving of size (to 0.5) and a doubling of size (to 2.0) will not look symmetrical on these new graphs, whereas they do so in our current presentation.

Author response image 2
New cumulative strain graph to be used.
https://doi.org/10.7554/eLife.35717.050

2) A number of the reviewer’s comments refer to visualising and comparing cells or rings of cells at the start versus at the end of our study period. We agree that this will indeed help to show the changes that occur more intuitively so we will present figure panels for a representative embryo in the various ways suggested (showing rings of near and far from pit cells related to Figure 1H, comparing cell areas and shapes, cumulative strains in the different cell behavioural flavours, overlaying apical and mid-basal cell outlines, colour-coding interfaces that have been involved in a T1). Note however that we are presenting data from multiple placodes and our analysis and conclusions rely on the robustness we get from including as much data as we can. So we will keep the existing multi-placode summaries, though we will now add a detailed case study of one placode.

3) For this case study, we will show that the cumulative strain rates do indeed agree with a comparison of beginning and end sizes. We can do this for cell shape, as the reviewer suggests, but note that the strain rate calculation methods we use rely on small changes per time step of less than 20% strain (Blanchard et al., 2009) and we chose the imaging frame rate to ensure this was the case. It might therefore be difficult to do the same start to end measurement for tissue and intercalation strains.

4) We will try for Figures 1F and 3C to present these as cumulative strains across a movie and see how they look. If differences and patterns are sufficiently clear we will present these as suggested, though in their current form they are useful in showing a snapshot of the instantaneous dynamics, which would otherwise not be shown.

5) There are a number of questions related to statistical analysis and the description of statistical methods in the manuscript. Mixed-model statistics needs to be explained. Experimental variance (e.g., confidence intervals) needs to be documented.

We constructed mixed-effects models to test for differences between regions (far and near pit), directions (radial and circumferential) and genotypes (wild type and forkhead mutant) from multi-embryo time-lapse movies.

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 JC and Bates D. (2000) Mixed effects models in S and SPLUS. Springer New York). In our work, to test for differences in instantaneous strain rates we used as fixed effect the genotypes (wild type and forkhead) while the variation within-embryo (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. A simpler alternative would have been to pool all data from different embryos of the same genotype, then testing for genotype differences. However, as some of the biologist reviewers will know well, inter-embryo variability in morphogenetic movements cannot be ignored, such as variation in the absolute and relative timings of the onset of mechanical processes and their subsequent morphological effects.

In the main figures we generally present cumulative strains, that are generated from instantaneous strain rate data shown in Supplemental Figures. 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 do 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; Tetley et al., 2015; Lye et al., 2014).

Action plan:

1) Extend the section in the Materials and methods to explain mixed-model methods, and refer to this more frequently in the text and figure legends.

Reference not cited in the paper:

Kabla AJ, Blanchard GB, Adams RJ, Mahadevan L: Bridging cell and tissue behaviour in embryo development. In Cell mechanics: from single scale-based models to multiscale modeling. Edited by Chauviere A, Preziosi L, Verdier C: Chapman & Hall/CRC Mathematical & Computational Biology; 2010. DOI: https://dx.doi.org/10.1098/rstb.2015.0513

[Editors’ notes: the authors’ response after being formally invited to submit a revised submission follows.]

Reviewer #1:

This manuscript provides a comprehensive analysis of cellular behavior during the formation of a tubular epithelial organ, using the embryonic salivary gland of Drosophila as a model. Budding of tubular structures from flat epithelial sheets has previously been thought to be mainly driven by apical constriction, leading to the formation of an invagination pit. Here the authors used live imaging and extensive quantitative analyses to demonstrate that, in addition to apical constriction, salivary gland invagination is associated with radially oriented intercalation of cells surrounding the invagination pit. Importantly, the authors extended their analysis to 3D, which revealed additional cellular behaviors (wedging, tilting, interleaving) that are oriented either circumferentially or radially with respect to the invagination center. Interestingly, polarized cell intercalation is associated with accumulation of myosin II along circumferential junctions, suggesting that myosin activity is polarized radially across the placode and contributes to tube invagination. Finally, the authors show that in fkh mutants, which fail to form an invagination, intercalation still takes place, but intercalating cell clusters fails to resolve in a directional manner. Here, cause and consequence are not clear – do polarized cell behaviors 'drive' tube budding, as stated in the title, or are they, at least in part, a consequence of forces generated by the invaginating pit that are transmitted across the field of cells?

As discussed in our initial response, the phrasing of our original title led to an interpretation or focus on an aspect of our findings that we did not intend. Our title was meant to refer to a radial patterning of [levels of] different cell behaviours across the placode, with isotropic constriction dominating in the domain near the pit, and oriented cell intercalation dominating in the region further away from the pit, now illustrated in the new summary Figure 10. We did not mean to imply that the polarised cell intercalations that we describe are the main driving force of the invagination nor that the forces generated by the invagination pit do not play a role (see below).

We have changed the title to now say:

‘Radially-patterned cell behaviours during tube budding from an epithelium’

and hope that this describes the findings more accurately.

As suggested in our initial response, we have now added a description and assessment of an active versus a passive mechanism responsible for the cell intercalation observed in the cells far from the pit. The detailed introduction of these mechanisms and associated predictions can now be found in subsection “Analysis of signatures of active versus passive cell intercalation” of the manuscript and illustrated in the new Figure 6. This figure also contains the data of the analysis of active versus passive signatures (together with Figure 7 (previous Figure 6)). This analysis of geometrical indicators of tissue stress as well as the analysis of junctional myosin polarisation clearly support that intercalations leading to circumferential neighbour gains in the cells far from the pit are actively initiated and actively lead to vertex formation. We cannot at this moment distinguish whether the resolution of exchanges is also intrinsic and actively driven or responds to the pulling from the pit. The continuing intercalation without directed resolution in the fkh mutant hint at an extrinsic (passive) component to the resolution, and the analogy to the process of germband extension would predict the same: active initiation of intercalation combines with the pulling from the pit to lead to an overall polarised intercalation event that underlies the circumferential convergence and radial extension at the tissue level. We have also added a short section in the Discussion about why the active signatures in the placode are less strong than in the germband.

These findings and conclusions are now clearly stated in the manuscript and illustrated in Figures 6, 7, and 10.

It is not easily conceivable how a symmetric tube can result from the invagination of cells at an asymmetrically (peripherally) placed invagination pit.

We would suggest that the cell intercalations as well as 3D behaviours we describe are important to achieve this. Although this question is not specifically addressed in this current study, it is of course an important one: why does the placode have an acentral invagination point and is this important for wild-type morphogenesis? The answer to the latter is clearly yes, as previous studies have shown that for instance mutants that commence invagination in the centre of the placode show aberrant tube morphologies and diametres at later stages (Myat and Andrew, 2000a; Myat and Andrew, 2002).

The circumferential convergence far from the pit combined with radial expansion towards the pit is one aspect of how cells move towards the acentral invagination point. Once internalised, previous studies have inferred that further intercalation/convergence and extension must take place within the formed tube as the number of cells around the tube circumference decreases from 12 to about 8 during the later stages of morphogenesis, way beyond the time frame analysed here.

What happens with the cells located between the pit and the margin of the placode (upper-right part of placode; Figure 1E)?

Again, this together with the above question about the acentral invagination point probably warrants a complete additional study. Our preliminary observations, however, suggest that these dorsal posterior cells at the boundary of the placode and right next to the invaginating pit initially remain static and in their original position, but over time these cells also invaginate and move away from the placode boundary. It would be interesting to analyse the behaviour of these cells in the future.

Do radially polarized cell behaviors spread beyond the boundary of the placode, and do cells outside the placode participate in invagination?

No, the whole placode is surrounded by a supracellular actomyosin cable (Röper, 2012, Dev. Cell) that decreases in diameter as the placodal cells disappear from the surface of the embryo.

Introduction section: "forkhead mutants, that fail to form an invagination, only show unproductive intercalations that fail to resolve directionally, likely due to the lack of an active pit."

This sentence seems to contradict the statement in the title that radially-polarised cell behaviours "drive" tube budding. Does the invagination pit drive the polarized intercalation of nearby cells, or vice versa? The issue of cause and consequence needs to be clarified and discussed in the text. The wording of the title may need to be adjusted accordingly.

Please see the answer to Reviewer 1 above.

Subsection “3D tissue analysis at two depth shows coordination of cell behaviours in depth”: "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 3E', magenta), but with a slower rate of constriction at mid-basal depth."

Isn't this precisely what is to be expected if cells near the pit undergo apical constriction (which was known before)?

Yes, but so far this had never been analysed and quantified dynamically from time-lapse movies and in a 3D proxy analysis. But we agree, this just confirmed an aspect we were expecting.

Figure 6 E, F, and the accompanying text, appear unnecessarily complicated. The text refers to circumferential myosin enrichment, whereas Figure 6E, F and the legend refer to radial bi- or unipolarity of myosin distribution, respectively. Please simplify.

What we are plotting in panels D-F of revised Figure 7 (previously 6) are the vectors that are pointing towards to myosin II enrichment, and by definition these will be orthogonal to the orientation of the junction the myosin is enriched in. The paper describing this method of quantifying and representing myosin junctional polarisation was published in eLife previously (Tetley et al., 2016, eLife) and we have just kept the conventions as in this paper. We have tried to illustrate clearly what is plotted with the diagrams in Figure 7D’ and D’’.

To simplify or rather clarify, we have now changed the label in panels E, F to say ‘circumferential enrichment (rad. polarity vector)’ as well as ‘radial enrichment (circ. polarity vector)’ to make the plots more intuitive but also retain the correct nomenclature for what is measured and plotted in the method, and have also updated the figure legend accordingly.

Is the distribution of myosin shown in Figure 6D a schematic drawing or representative of a real image?

It is a schematic (drawn from real cell shapes), but representative of real images such as Fgure 7A, G, and H.

Figure legends often lack information that is necessary to understand the data. Axis labels (e.g., "pp per min", "fluorescence increase over embryo average") need to be explained in the figure legends.

The labels were changed and/or explained more clearly in the legend.

Reviewer #2:

This is an interesting manuscript by Sanchez-Corrales et al. that explores the cell shape changes that underlie the invagination of the salivary gland placode. The authors show that a combination of two common tissue shaping events, apical constriction and cell intercalation, function during early invagination. They also interpret their results in a radial coordinate system, which is nicely appropriate to the tissue context, and observe different topology-altering regimens dominating in different proximal-distal regions of the coordinate axis. Finally, they examine these cell behaviors at two different levels along the apical-basal axis as well as in a mutant that affects specification of the salivary placode. In general, these are interesting measurements and provide insight into tube and salivary placode invagination, although the impact of it is slightly hurt by an inability to tease apart the relative functional contributions of each regimen to invagination (i.e., it is not possible to alternately remove cell intercalation or apical constriction from the tissue) and the fact that the role of apical constriction in placode invagination is well-established.

Yes, we agree with the reviewer that apical constriction was previously well documented in the salivary gland placode, including by our own work (Booth et al., 2014). But thus far, the analysis of apical constriction was always only focused on the apical domain exclusively, or on a single 2D cross section of tissue in histological sections. Indeed, the Sun et al., (2017) paper revealing basal protrusions in the germband were something of a surprise for those who thought the germband was apically driven. Therefore, we would posit that our dynamic and 3D-proxy analysis, clearly showing strong wedging behaviour near the pit that is restricted to this region and increases over time (even prior to actual tissue-bending) adds to the previous descriptions. Furthermore, the novelty of our data lies in the identification of radially patterned behaviours 2D and 3D, the contribution of directed intercalations in 2D, interleaving, wedging and tilting in 3D and the analysis of active/passive signatures. For this much more in depth analysis of placode tube morphogenesis the apical constriction near the pit, albeit known before, was obviously an important factor that needed to be part of a more complete analysis.

As the reviewer will be well aware, in most morphogenetic processes, it is difficult to selectively affect only part of all actomyosin-dependent processes (i.e. constriction or intercalation), where this has been achieved it is through use of fortuitous mutants that mostly impair one but not the other process. We would argue that in our case the use of the fkh mutant achieved this in part: in the mutant, apical constriction is basically absent at early stages, whereas we show that intercalations occur at near wild-type levels but lack the polarised resolution.

Some issues to be addressed:

I'm not sure that "3D" analysis is accurate – this is 2D analysis done at a fairly limited two different planes (0µm and -8µm) in the tissue. It would be more accurate to say that this is an examination of apical and more basal contributions to cell shape and intercalation. Also, can the authors say whether cell volumes are maintained during invagination?

As explained in our initial response, we completely agree that a true 3D analysis would be preferable. We already stated in the original manuscript To circumvent this issue, we used strain rate analysis at different depth as a proxy for a full 3D analysis”, and have now made sure that this phrasing is used wherever we refer to the method.As we hoped to explain in the manuscript, obtaining a full 3D analysis is still very unreliable (apart from hand-segmentation, which is not feasible – one movie tracing 80 cells over 20 sections in depth and 40 time points would require drawing and linking 64,000 cell outlines by hand). We have tested various available segmentation methods, some of which allow 3D segmentation (such as EDGE4D [Khan et al., 2014, Development]; and RACE [Stegmaier et al., 2016, Dev. Cell]), but these do not track cells in 3D reliably. If unreliable segmentation leads to too much loss of cell identities over time, then our computational methods break down for the strain rate analysis and the data analysed will be too sparse. Therefore, we decided that a 3D proxy as used in our study currently allows much better segmentation and tracking coverage and thus far better strain rate analysis (both in time and in z).

It is important to stress again that our work would thus far be the first published method allowing quantification of changes in cell wedging, cell interleaving or cell tilting as 3D cell behaviours. Adding a second layer (with both layers parallel to the curved epithelial surface) and especially in combination with linking identities of cells between these layers as done in Figure 4 allowed us to quantify approximate changes in 3D cell shape such as cell wedging, cell interleaving and cell tilting. Indeed, even if full 3D shapes of all cells were known, the entry-level analysis one would do would likely start with simplifying cell shapes by fitting straight lateral sides to all cells (through regression or similar). The elegance of our approach is that we have effectively constructed this same simplified description of cells in a very efficient manner. Because we only sample at two depths, apically and mid-basally, our measures are likely to be more variable than if detailed 3D cells shapes were known. However, we would like to point out that changes in topology in depth seem remarkably smooth, as can be seen in Video 4 which progressively steps through z, so that sparse sampling in z will in fact capture much of the important 3D information.

None of the 3D geometries are obvious or could be deduced from an apical analysis only: apical area shrinkage could well be compensated for by cell elongation in depth, requiring analysis in a different plane; cell interleaving (a change in neighbour contacts along the z-length of the cell) cannot be observed from a single plane, neither can cell tilting. Importantly, we show in Figure 4 that these three changes to 3D cell shape (even if only approximated by the analysis at two z-levels) is clearly radially patterned across the tissue.

With regards to cell volume, we have analysed this from a single fully hand-traced movie and indeed cell volume appears to be conserved, and we now mention this as ‘data not shown’.

It is a bit strange that similar constrictive behaviors are observed at -8µm; at some point there should be a cellular accommodation of the volume shifted due to apical constriction – either through a widening or deepening of the cell.

In fact, the cell shape strain rate basally is significantly reduced compared to apically (Figure 3D), already indicating wedging behaviour, and the direct analysis of wedging in the 3D proxy analysis using z-strain rates confirms this (Figure 4B’). But the reviewer is correct in that cells are expanded even further more basally, at a depth that we cannot reliably segment and track. This was commented on and explained in the manuscript (Results, subsection “A quasi-3D tissue analysis at two depths shows coordination of cell behaviours in depth”).

The fact that intercalation strain rates are very similar apically and basally (Figure 3E’), especially in the cells away from the pit, prompted the 3D analysis in order to dissect whether apical or basal levels tipped ahead of each other (i.e. were interleaved) despite a nearly identical rate.

There is a passing reference to cell deepening, but no hard data is presented. The authors should elaborate on this in the Results and Discussion.

The slight deepening of the whole placode compared to the surrounding epidermis and even more so in the region of the forming pit is visible in the placodal cross-section in Figure 1B, and we point this out in the section of the manuscript commented on above under 3). This deepening has been described from fixed samples and histological sections by the Andrew lab in already in 2000 (Myat and Andrew, 2000a).

The role of the initial slight columnarisation of the placodal cells compared to the surrounding epidermis is not clear and deserves to be analysed in a future study.

The analysis of cell wedging and the relationship to apical constriction and cell intercalation is quite interesting, although it takes significant effort to follow. Any editing and addition to the Discussion on these points will be helpful.

Together with our introduction of possible active or passive mechanisms responsible for the intercalation away from the pit, we have significantly changed and restructured parts of the Result section and Discussion concerned with these two behaviours and their potential interplay. We hope this has made these sections clearer.

N numbers for the number of cells quantified in each category (for example, "near pit" and "far pit" cells) should be reported in each figure. The current reporting of the number of embryos should be kept.

We have added the total number of cells analysed per time-step of the analysis for the different regions, apical/basal, wt/fkh[6] to the Materials and methods section. We felt that adding these lists to the Figure legends would make these confusing.

Much of the statistical analysis is calculated through a mixed-effects model. More information on how mixed-effects models were applied to the data is needed to be able to evaluate the appropriateness of this statistical measurement. There should be at least a brief methods section on this.

Mixed effect models were previously discussed in the Materials and methods section, but we have now expanded the section, which describes the rationale for the use of mixed effect models. We have also clarified legend text where statistics are presented.

Was there a statistical reason for splitting cells into "near" and "far" bins? I didn't see a clear statistical justification for this.

We were prompted to use this split into near and far domains by our early analysis, as for instance shown in Figure 1G, G’, where cumulative apical area change of concentric rings around the pit clearly shows that only the cells near the pit (magenta and blue) are actively constricting at a steady rate, whereas cells far from the pit (green, red, yellow) change little or even slightly expand.

The apical strain rate analysis reported in Figure 2 and Figure 2—figure supplement 1also shows that only when cells are split into these two region can specific cell behaviours be identified as dominant in a region, whereas binning all placodal cells (as in Figure 2—figure supplement 1C, C’) has effects across the tissue cancel each other.

These regions thus distinguish dynamics that were otherwise masked when the tissue was considered as a whole. We cannot rule out the possibility that there are more than two regions within the placode (defined for instance by differential gene expression) and future studies will need to address this.

I was somewhat troubled by the inferred "intercalation strain rate" in the first sections of the manuscript, as this is a rather indirect measurement of topological changes that can be concretely measured, but they subsequently do exactly this in later portions of the manuscript.

As already mentioned in our initial response, we would like to point out that the term intercalation is not restricted to one particular measure. The process of intercalation can be measured in two ways, either as the continuous process of slippage of cells past each other (Blanchard et al., 2009; Kabla et al., 2010; Blanchard, 2017) that is captured by our strain rate analysis, or as discrete T1 topological change events. These are equally valid, though can be interestingly different. For example, looking at Figure 4E, F from Blanchard (2017) (PMC5379022 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5379022/figure/RSTB20150513F4/), though the cumulative amount of intercalation measured by both methods during germband extension in Drosophila is the very similar, the time-courses are quite different, with the continuous process starting some 5-10 minutes before T1 events occur, as would be expected from starting from an approximately hexagonal arrangement of cells. We would argue that the continuous process of slippage is a more faithful measure of the underlying myosin-based contractility that drives intercalation than the counting of topological T1 events.

Since in our tissue there is no contribution to tissue strain due to either cell divisions or cell death/cell delaminations, the simple equation of cell shape change plus intercalation equals tissue shape change holds true (Blanchard et al., 2009). So we would not agree that the intercalation strain rate is somehow less real for being ‘inferred’, rather it is precisely calculated as a difference between two directly measurable quantities.

Importantly, our direct measurement and analysis of T1 events in the placode as presented in Figures 5 and 6 closely matches the strain rate analysis. Specifically, the cumulative strain over our observed time window that is attributed to the continuous process of intercalation is 1.1 (Figure 2E) while it is e0.09 = 1.094 for discrete T1s (Figure 9F). This strongly supports the validity and complementarity of both approaches.

There is an unofficial "results" section embedded into the Discussion. These should be moved to the Results, or alternately, saved for a future publication. This section is a bit incongruous, given the Results sections, as it jumps into a cell fate discussion in the Discussion. The cell fate results are weakened by being correlative, without a functional component to test the hypothesis. An advantage of saving these results for another publication would be the ability to analyze such functional disruptions.

Thanks, yes, we have removed this section and part of the figure for future use.

Discussion section penultimate paragraph – I am curious which "non-intuitive" results the authors are referring to?

We have found cell interleaving to be a non-intuitive 3D geometry of multi-cellular domains when explained to people for the first time, particularly in comparison to wedging and tilting which can be seen directly for individual cells. Once understood, of course, it becomes intuitive, but we agree it is not a helpful term, so we have removed it.

Reviewer #3:

[...]

The authors conclude that a radially polarised pattern of cell behaviour, including apical constriction (as previously reported) and cell intercalation drives tube formation. The authors compare this pattern of cell behaviour with a radial pattern of myosin II accumulation and conclude that the radial pattern of myosin II would lead to radially polarised cell behaviour which would drive tube budding.

In my view the data do not allow to derive a causality. I accept that in wild type embryos the polarized myosin II pattern matches globally the preference of cell intercalation events. To address causality, the authors analyse forkhead mutants which completely lack tube invagination because cell specification is disrupted.

In the fkh mutant, not all cell specification is disrupted. The most upstream transcription factor driving salivary gland placode fate is Scr (in conjunction with Hth and Ext). In the fkh mutant, expression of CrebA (as a master activator of all aspects of increased secretory capacity of the glands) is normal, various other factors usually expressed within the placode are still expressed, though what is changed is the spatial pattern of expression of a number of downstream factors (i.e. being restricted to the dorsal-posterior corner at first and then expanding across the tissue).

It does not come as a surprise that cell and myosin II behave isotropically.

This is in fact only true over the time interval analysed here, at later stages the central part of the placode begins to buckle and form a very shallow indented invagination. Also, as visible in the myosin images and movie in the fkh mutant, the circumferential actomyosin cable is still specified and assembled, as is junctional myosin accumulation at the centre of groups of cells that still undergo intercalations (see the images of rosette formation in Figure 9). Thus, we would argue that what is lost in the fkh mutant is the sub-patterning of the placode with an acentrally positioned point of invagination.

As we discuss in detail now in the revised version, the fact that in the fkh mutant T1s still occur, at levels comparable to wild-type (though without directional coordination), strongly supports that the shortening of junctions to a T1 in the placode is indeed an active cell behaviour and not a passive response to other events within the placode.

The authors do not address the possibility that the radially polarized pattern of myosin II and cell behaviour is a mere consequence of mechanical pulling by the constricting and invaginating pit cells.

Please refer to our detailed response to Reviewer 1’s first point. We now introduce, and explicitly test active versus passive contributions.

With regards to myosin polarisation as a passive response, we discuss now in the manuscript that to our knowledge only two studies have analysed myosin subcellular localisation in response to mechanical pulling. These are a very recent study using the Drosophila wing disc epithelium subjected to mechanical pulling (Duda et al., 2018) and an earlier study of supracellular actomyosin cables during germband extension (Fernandes-Gonzales et al., 2009). In both cases, myosin accumulates and becomes polarised at junctions parallel to the pulling force. Extrapolating to our placode system, we would therefore expect an accumulation of myosin at radially oriented junctions if this is due to mechanical pulling, when we observe the opposite. As discussed in the revised manuscript, circumferentially polarized junctional myosin thus supports that intercalation is driven by an Active mechanism.

However, we also discuss that the mechanical pulling (due to apical constriction near the invagination point) and intercalation are integrated: polarised junctional actomyosin leads to polarised formation of a vertex or rosette, and the mechanical pulling helps to polarise the resolution of these events in a directional manner, similar to what has been found to occur during germband extension and posterior midgut invagination (Collinet et al., 2015).

forkhead mutants to not allow to rule out this option, as no invagination is observed in these mutants.

As rosette formation and T1 exchanges continue in the fkh mutant (albeit with non-directional resolutions) despite the lack of a pulling pit, we would argue that this supports cell intercalation as an active cell behaviour rather than a passive response, and we discuss this in detail in the revised manuscript.

The headline is thus an unjustified overstatement, as the radially polarized cell behaviour is likely to be a mere consequence of the activity of the small group of pit cells and certainly do not drive tube budding.

As we already pointed out in our initial response, we are sorry that the original title was misinterpreted and we have changed it accordingly.

Please see the detailed response to Reviewer 1 above.

Specific issues:

In many figures the experimental variance is not visible. I understand that the authors tested the statistical significance. Yet, any sort of representation of the variance (e. g. confidence intervals) would help to assess the quality and the degree of uniformity of the data.

In the main figures we generally present cumulative strains, that 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 do all statistics on the instantaneous data, using the mixed-effects model. 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; Tetley et al., 2015; Lye et al., 2014).

The variance is therefore visible in the instantaneous strain rates in the figure supplements.

In the first part of the study, the degree of cell intercalation is only indirectly derived from the data as difference of total tissue behaviour and fitted cell shape changes. In Figure 5, the number and orientation of intercalation events is directly counted. The data would be more convincing, if also in the first part the contribution of intercalation would be directly measured. Given that the contribution of intercalation is the central conclusion of the study, it seems to be important that this parameter is directly measured.

Please see the detailed response to Reviewer 2’s point about intercalation strain rate above.

3D: This part of the study is not convincing. I am fully aware of the difficulties and importance of a three-dimensional view of cell behaviour. Adding a second layer of recording does not contribute much to the issue however. In my view the presented data are not convincing and, in the end, even weaken the central conclusions. The authors conduct a sort of strain analysis along the axial direction, however with only two data points along the axis (apical and mid-basal).

Please see our detailed response to Reviewer 2 about the practicalities and what we see as the strengths of our method. We are glad that the other reviewers found the 3D proxy analyses and results interesting. We hope that we have addressed the concerns of this reviewer in the revision and explained the advantages of our methods more succinctly.

Time axis: I find it confusing that the specific cell behaviour starts already in negative time. I understand that T=0 is set to the first visible invagination. As the polarized intercalation starts already at t=-15min, it would help to correlate the time axis to events at the time. Is this the first time when apical constrictions are observed?

As explained in the manuscript, all time lapse movies are aligned to the point in time where the first tissue bending (curvature at the tissue level) can be observed as this emerged as the most reproducible point for such alignment. We thus designated this point as t= 0, which seemed intuitive to us as it represents the first time point of actual invagination.

We agree that it is interesting that changes to cell shapes in the placode commence way before this point, though this was noted and commented on by us already previously (Girdler and Röper, 2014). It is though not completely surprising that changes in the planar epithelium star to occur and build up stresses prior to an out of plane bending of the epithelium. But to dissect the contributions of these pre-bending changes to the efficiency and robustness of the invagination process will be analysed in a further study.

With regards to the first time point analysed and what this relates to: as we state in the results for Figure 4, cells across the placode started out at -18 mins before pit invagination unwedged and mostly untilted in radial and circumferential orientations (Figure 4B’, F’). Cells near the pit became progressively wedge-shaped over the next 30 minutes. Therefore, the pooled beginning of our time lapse analysis appears to capture the earliest changes to placodal cells.

Reviewer #4:

The paper offers interesting insights into tissue morphogenesis. Even though the single mechanical processes underlying this morphogenesis are not new, it is interesting to see this combination for a radial system.

To our knowledge, quantitative changes in cell wedging, cell interleaving and cell tilting as measure of 3D cell behaviour have not been reported previously. Certainly, our newly developed methods to asses these changes quantitatively from two or more z levels are novel and will be of general use in the epithelial morphogenesis field.

Furthermore, even within a 2D context, directed cell intercalation within the plane of the epithelial primordium had not been appreciated as a contributor to tube budding from an epithelium previously, to the best of our knowledge.

[…]

Major points (better presentation and minor simplification) in detail:

1) When reading the manuscript, it becomes immediately clear that cell areas decrease at the pit. However, it is much less clear what happens to the regions further away. In addition, currently, it is not easy to see what happens with single cells. This hinders getting an intuitive picture of what happens to shapes of single cells

To aid this understanding, we had provided various means in the original manuscript, such as videos showing cell outlines, the still time points of apical cell area for wild-type and fkh mutant in Figure 8C, D as well as the marked examples of rosette formation/resolutions in Figures 7 and 9.

We agree, though, that even more could be done or at least tested, and so we have followed the reviewer’s suggestions and detail what we have done below.

and it does not become clear what the quality of the tracking was.

The segmentation of all movies used in this study was manually corrected to ensure at least 90% of coverage of each placode at each time point. This statement was now added to the Materials and methods section on segmentation and tracking.

We also added an example movie for one segmented and tracked placode in Video 2 that shows cell identities colour-coded for the length of the movie.

Therefore, movies should be added that look similar to Figure 7C and D but with different coloring. In one movie, it should become visible what happens to the shapes of the radial stripes (Figure 1H’) and, in another one, what happens to individual cells.

We agree that this will help to show changes that occur more intuitively, and so we have generated panels and movies for better illustration.

1) We have added a new Figure 1—figure supplement 1 that shows cell shapes and areas of cells within radial stripes (as in Figure 1G, G’) for an example movie (ExpID0356)

2) We also added a new Video 2 in which each cell identity in the same example experiment (ExpID0356) is colour-coded for identity as well as for apical cell area.

3) We include a new panel in Figure 2—figure supplement 1F that shows the total intercalation strain rate over a whole example experiment (ExpID0356). We did this by calculating strain rates at the middle time point of the movie, over a much larger time window than previously (+/- 11 frames rather than +/-1 frame), covering the whole of the movie. This very useful panel addition should be considered as an indicative example only, since some of the strain rates calculated over this longer time window were above the ~20% strain limit recommended for the methods that we use (explained in Blanchard et al., 2009). Nevertheless, the radial organization of intercalation (convergence is biased strongly to circumferential) in the panel is striking.

4) See also response below to Reviewer 4 for visual representations of where T1s occur in the placode, and their orientation.

2) The relative simplicity of the system should be exploited more to generate additional controls of the method, to simplify the understanding of the results, and possibly to obtain more precise outcomes. Currently, there is a focus on average strain rates and their accumulation. However, since the system has no apoptosis or division and relatively small strain changes, it should, at least for cellular strain rates, be possible to calculate total changes in strain directly for each cell, by comparing shapes in a certain time step directly with those in the first time step. For calculating the change in area, this would be very straightforward. It seems as if the cumulative area change is now calculated indirectly (but it is not clearly described whether this is indeed the case): ellipses are fitted to cells in such a way that the ellipses have the same areas as the cells. Then for each cell a strain rate tensor is obtained that is constrained to conserve the cell's area change, as assessed by the tensor's trace, which is a first order approximation of area change. Then the cumulative area change is obtained by adding up the traces, thus approximating total relative area change (1-0.1-0.1 is not the same as 1*0.9*0.9). Relative area change can also be obtained by only comparing areas at the first time step with those at a later time step.

We thank the reviewer for very helpful suggestions on the presentation of the strain rate data. We have made the following changes:

We have changed all cumulative plots from proportional strains to strains (or stretch ratios), which means they will all start at 1.0 and deviate over time from 1, at the end of the study period showing the proportion of its original size the tissue has reached (due to a particular behaviour). This could be more intuitive, as the reviewer suggests.

With regards to calculating total strain for individual cell versus our use of average cumulative strains:

For the embryo case study (ExpID0356) we introduced in response the above point, we compare firstly the radial cell lengths of cells at the beginning and at the end of the movie (Author response image 3A) to show the range of changes in size of cells over the whole movie (from t= -7.5 min to t= +20.0 min).

Author response image 3B and C show predicted (y-axis) versus actual (x-axis) radial cell lengths of cells, using two different cell shape strain rates to calculate the predicted length. For Author response image 3B, the cell shape strain rate is the radial projection of the tensorial strain rate that mapped best-fit cell shape ellipses at the start and end of the movie, over a +/- time window of 13.75 mins (half the movie). For Author response image 3C, the strain rate used for each cell is the average of the cell shape strain rates calculated for each of the 22 frames of the movie, where each strain is calculated over a smaller time window of +/- 1.25 mins (neighbouring movie frames).

The correlation in both cases is very strong. Note that, as explained for the above point, the strain rate calculation methods we use rely on small changes per time step of less than 20% strain (Blanchard et al., 2009) and we chose the imaging frame rate to ensure this was the case. Calculating start to end strain rate measurements as in Author response image 3B will therefore introduce small errors due to the larger strain rates in the larger time window, and this explains the minor discrepancies. There are also small imprecisions because the mapping of two ellipses onto each other is not perfect if they are elongated and their ellipse orientations are not aligned. The error though is expected to be and looks here unstructured (Gaussian) and so data averages will be accurate.

Thus, we strongly feel that our methodology correctly captures individual cell events in average cumulative strains. We hope that the change in depicting strains as well as the addition of further panels illustrating changes for an example placode will make the changes more intuitive to grasp.

The authors should compare these values as a control of the method. In addition, they should use the direct method for area change, since it is easier to understand and a reader doesn't have to wonder for example why the cumulative area change is in pp per minute and not just in pp in Figure 1H. In addition, the total relative area change is actually the biologically relevant value in my opinion. Now it should not be a problem to do something similar for the strain change of the cells: here the fitted ellipse of a cell at a certain time step can be directly compared with the fitted ellipse of the same cell at the beginning and this should be used as a control. For calculating tissue shape change, the situation is a bit more complicated, since neighbors, and thus calculations, change during the time sequence. In order to do a direct comparison, the same cells should be compared at the beginning and the end. This should in principle be possible though, as long as the neighboring cells in the beginning (mostly) stay together. If other cells mix too much with the initial cells, definitions of strain do not really make sense any more. The authors should judge whether the direct approach is useful for tissue shape change. If it is, it would be useful to replace cumulative strain rates by these values, since they are a bit more straightforward.

We thank the reviewer for these suggestions and have changed our cumulative strain rate plots to cumulative strains (or cumulative stretch ratios), which we agree are on balance more intuitive. (One downside is that the y-axis scale is now no longer symmetrical about the starting value. For example, for intercalation, which involves no area change, a doubling in length in one orientation (from 1 to 2) and a halving of length in the perpendicular orientation (from 1 to 0.5) are no longer symmetrical about the line y=1.)

3) The presentation of the figures should be clearly improved. Generally speaking, figures are currently missing that give a clear and intuitive overview of what is happening where in the tissue.

We have added a new summary Figure 10 that sums up the changes both for 2D cell behaviours observed within the apical domain only as well as the 3D cell behaviours that our 3D proxy analysis revealed. We hope the reviewer finds this a helpful addition.

In addition, tissue shape change, cell shape change, and cell intercalation are currently visualized by coloring according to strain rates, which is not a very intuitive quantity for many biologists. Instead, the figures should be such that readers can make a direct connection between quantitative values and shapes.

As for Reviewer 4’s point (2) above, the new cumulative strains on the y-axes are more intuitive in that one can immediately see the fractional change in size of the tissue attributable to each cell behavior over the course of developmental time.

List of detailed suggestions to improve figures:

– Figures 1F and 3C are not very intuitive. First, a longer line is usually not associated with contraction. This should be changed.

These are only illustrative panels to explain the methodology used and use the convention to depict strain rates that is also used and explained in Figure 2B. The colour-coding of the original Figures 1F and 3C was not the same as for Figure 2B, but we have now changed this and they all use magenta to indicate expansion and green to indicate contraction. The length of the lines in these illustrative panels is proportional to the amount of strain, as explained in Figure 2B. We have also added the original small example domain of cells from which the illustrative strain rates in Figure 2B were calculated, in which vertical cell shortening and cell rearrangement leading to further convergence vertically, as expressed in the strain rate motifs, can clearly be seen.

For example, arrows could be used, or the line length could be 1 as a standard and then be shortened or lengthened based on contraction and expansion, respectively. Secondly, the figure shows cell shapes, even though the strain rates are calculated based on a cell and its neighbors. It would therefore be better to make the cell outlines grey or something.

We agree with the general concern here, but our preference, based on a long experience of displaying strain rates going back to 2009 and before is that the fewer the number of lines the better, for ease of digest while also showing detail. Other publications with strain rate nematics showing deviations from 1 or deviations from a circle of standard radius (e.g. by Johanns Bellaîche/Francois Graner & Suzanne Eaton/Frank Jülicher) need quite strong strains to be obviously different and are better suited for data that have been either heavily averaged over space or accumulated over long times.

As the reviewer states, strains are calculated for small domain, i.e. a central cell and its first corona of neighbours, but this strain is then associated with the identity of the central cell, and is drawn for that cell. We think it should be clear from the cartoons representing the strain rate methods in Figure 2A, B and Figure 4—figure supplement 1A-D that this is what we are doing throughout.

Third, as far as I understand it, instantaneous strain rates are shown, so that there is much noise in the results and looking at the single figures may therefore not give much information. This can be improved by taking the total strain change (see point 2 of main points) or otherwise the cumulative strain rate, which should thus be calculated per cell.

We have changed plots in main figures to cumulative strains as suggested.

– At the moment, Figure 1 does not clearly show what happens to the shapes of the radial stripes, even though this would be useful to better understand the movements in the tissue. This should be improved. One way could be to replace Figure 1G. This figure gives a good impression about time and position dependent cell shape changes, but may be replaced by another one that contains this information together with information at tissue level: one could for example replace it by a figure that contains a segmented image at the beginning, at t=0 and at the end. The boundaries between the radial stripes are thicker, so that it can be seen directly how their stripe shapes change. Cell shapes are then directly visible and can thus be assessed directly. However, in order to stress differences in area and compensate for interpretation difficulties due to the 2D projection, it would then still be useful to color individual cells according to area change. If desirable, cell aspect ratios could be color coded in additional images.

We hope that our new panel of the placode separated into 5 radial stripes in Figure 1—figure supplement 1, colour coded by cell area at the start and end of an example movie, addresses this point satisfactorily.

– The interpretation of Figure 2C, Figure 2—figure supplement 1B, Figure 3D, Figure 3—figure supplement 1B, and Figure 6—figure supplement 1A, B is not straightforward. The reader has to look up what the different colors mean and then couple that to the small inlet, which indicates whether the radial or circumferential direction is at play and then read the type of change that the figure shows. Even though the colors are nice to recognize patterns quickly, they don't confer any intuition on the extent of strain. In order to get such an idea, the data in the color bar need to be combined with reading the text to find out what kind of strain is visualized exactly and what the squares mean exactly. The authors should make these figures more intuitive. For example, they could use a segmented image of the end of a video. Each cell could then have arrows indicating strain changed in the radial (or circumferential) direction. Again, the total strain change (or the cumulative strain rate) can be used. In this way, it is immediately clear whether radial or circumferential strains are visualized and they allow for a direct more quantitative comparison between cells. In order to see quickly what is happening where, cells can be colored according to strain changes similarly to how the squares are colored now. Because of the presence of the arrows, it is then also clearer which color codes what. Depending on what the figure looks like exactly in the end, it may be useful to add a segmented image of the start as well and add radial stripe boundaries, so that the shape changes of single cells and their environment can also be looked at. In this way, shape changes of a cell and its neighbors could be directly compared to the length of the arrows and thus create an intuition for strain change. This would of course only show data of one embryo, but the quantification is in other figures anyway.

We found it difficult to improve these plots as suggested. We hope that a combination of more intuitive cumulative strains in the main figures, the addition of panels of snapshots from a single movie in various figures, and better clarity in the text in response to various helpful points made here and by other reviewers, will make it easier for the reader to interpret these plots.

– Figure 4 does not give an intuitive overview of where which shape changes are present between the two layers. This should be improved. For example, two segmented z-projections could be shown in different colors in one figure. Then the reader could directly look at shape changes between the levels and see whether he can distinguish the wedging, interleaving and tilt.

To explain better what type of changes the z strain analysis is measuring and how the methods works in comparison to single plane time-strain analysis, we have added cross section panels to the schematics explaining the 3D behaviours, and have also added a whole section to the Figure 4—figure supplement 1that explains the z-strain rate method in comparison to t-strain rate analysis (Figure 4—figure supplement 1A-D).

– Figure 5 does not give any intuitive overview of where which neighbor exchanges occur. Such an overview should be added. For example, segmented images of the beginning and the end of a movie can be shown. Cell boundaries that will disappear may for example be drawn grey in the first image. New cell boundaries may for example be drawn blue in the image of the end of the movie. To get an idea of the number of reversals, each boundary that disappeared and then appeared again, could get another color.

We have added two panels (Figure 5D) for an example movie that show the interfaces that will be lost over the duration of the movie coloured in blue and labeled on a segmented image of the first time frame, and the interfaces that were gained over the course of the movie labeled on a segmented image of the last time point coloured in red. In addition, a variant of this is shown in Figure 5—figure supplement 1, where future neighbour losses and past neighbour gains are plotted, this time colour-coded for radial and circumferential orientation.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

In the Abstract and in the manuscript text, there are several remaining instances of misleading statements (italicised) inferring a causative role of polarised cell behaviours in tube budding, where no such causative role is supported by experimental evidence:

Last sentence of the Abstract: "Thus, tube budding involves radially-patterned pools of apical myosin, medial as well as junctional, leading to radially-patterned 3D-cell behaviours."

We changed the sentence to:

“Thus, tube budding involves radially-patterned pools of apical myosin, medial as well as junctional, and radially-patterned 3D-cell behaviours, with a close mechanical interplay between invagination and intercalation.”

Introduction section: "In addition, across the placode junctional myosin II is enriched in circumferential junctions leading to polarised initiation of cell intercalation through junction shrinkage."

We changed this sentence to:

“In addition, across the placode junctional myosin II is enriched in circumferential junctions, suggesting polarised initiation of cell intercalation through active junction shrinkage.”

Subsection “Analysis of signatures of active versus passive cell intercalation”: "This shows that circumferential junctions, and T1s in particular, are likely to be driven by an intrinsic contractile mechanism, and are not the result of cells being pulled away from each other radially."

We changed this sentence to:

“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.“

Discussion section: "in our system they are utilised within a radial coordinate system, thereby leading to the formation of a narrow tube of epithelial cells from a round and flat placode primordium."

We changed the sentence to:

“…but in our system they are utilised within a radial coordinate system, with the morphogenetic outcome being the formation of a narrow tube of epithelial cells from a round and flat placode primordium.”

and, former sentence:

“This finding therefore also supports an active intercalation mechanism, where the remaining local increases in junctional myosin II still drive formation of T1 vertices and rosettes, but without any overall directionality to their resolution.”

changed to:

“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.”

https://doi.org/10.7554/eLife.35717.053

Article and author information

Author details

  1. Yara E Sanchez-Corrales

    MRC Laboratory of Molecular Biology, Cambridge, United Kingdom
    Contribution
    Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1438-1994
  2. Guy B Blanchard

    Department of Physiology, Development and Neuroscience, University of Cambridge, Cambridge, United Kingdom
    Contribution
    Conceptualization, Software, Methodology, Writing—original draft
    For correspondence
    gb288@cam.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3689-0522
  3. Katja Röper

    MRC Laboratory of Molecular Biology, Cambridge, United Kingdom
    Contribution
    Conceptualization, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft
    For correspondence
    kroeper@mrc-lmb.cam.ac.uk
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3361-766X

Funding

Medical Research Council (U105178780)

  • Yara E Sanchez-Corrales
  • Guy B Blanchard
  • Katja Röper

Biotechnology and Biological Sciences Research Council (BB/J010278/1)

  • Guy B Blanchard

Wellcome (100329/Z/12/Z)

  • Guy B Blanchard

Isaac Newton Trust (15.23(k))

  • Guy B Blanchard

Wellcome (099234/Z/12/Z)

  • Guy B Blanchard

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

Acknowledgements

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.

Senior Editor

  1. Didier YR Stainier, Max Planck Institute for Heart and Lung Research, Germany

Reviewing Editor

  1. Stefan Luschnig, University of Münster, Germany

Publication history

  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)

Copyright

© 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.

Metrics

  • 2,140
    Page views
  • 274
    Downloads
  • 3
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)