Cell dynamics underlying oriented growth of the Drosophila wing imaginal disc

The Drosophila larval wing imaginal disc is a powerful model system in which to study the integration of diverse types of regulatory cues for tissue growth. The wing disc is a relatively flat epithelial sac that grows during larval stages of development. As it grows, the rapidly proliferating cells on one side (the wing pouch) become pseudostratified, whereas cells on the other side become squamous. At pupariation, the wing pouch everts and flattens to assume the adult wing shape (Waddington, 1940).

The size and shape of the wing are largely, but not entirely, determined during the larval growth phase. At this stage, signaling centers located at the anterior-posterior (AP) and dorsal-ventral (DV) compartment boundaries not only establish patterns of gene expression but also promote growth (Beira and Paro, 2016; Hartl and Scott, 2014). Hedgehog produced in posterior cells signals across the AP boundary to stabilize the transcriptional activator Ci155 (Ci) and induce the expression of different target genes [i.e. engrailed and decapentaplegic (dpp)] at different distances (reviewed by Hartl and Scott, 2014). Dpp is itself a secreted signaling molecule that establishes a bidirectional gradient anteriorly and posteriorly, promoting patterned gene expression through graded phosphorylation of SMAD transcription factors (reviewed by Affolter and Basler, 2007). At the DV boundary, Wingless (Wg) expression and Notch signaling are maintained by a positive-feedback loop and together pattern the DV axis (Micchelli and Blair, 1999; Micchelli et al., 1997; Rulifson et al., 1996). Signaling from the AP and DV organizers is required for wing disc growth, even though cell proliferation is not especially concentrated near these regions (González-Gaitán et al., 1994; Milan et al., 1996; Schwank et al., 2011).

The orientation of tissue growth is strikingly non-uniform: marked clones strongly elongate along the proximo-distal (PD) axis of the adult wing (González-Gaitán et al., 1994; Resino et al., 2002; Baena-López et al., 2005; Mao et al., 2013; Worley et al., 2013; Heemskerk et al., 2014). In central regions of the larval wing pouch, the adult PD axis is generally orthogonal to the DV boundary. Signaling at organizer regions influences the growth orientation by establishing tissue-wide patterns of two planar cell polarity (PCP) systems, Fat and Core (Aw and Devenport, 2017). Both systems develop the same global pattern of planar polarity that presages the orientation of growth (Rogulja et al., 2008; Ambegaonkar et al., 2012; Sagner et al., 2012; Brittle et al., 2012). Perturbing Fat PCP or dominantly reversing the orientation of Core PCP both disrupt wing size and shorten it in the PD axis (Bryant et al., 1988; Clark et al., 1995; Mao et al., 2006; Merkel et al., 2014).

How the PCP patterns direct oriented tissue growth remains unclear. PD-oriented cell divisions clearly contribute to oriented tissue growth and are disturbed by genetic disruption of the Fat pathway (Baena-López et al., 2005; Mao et al., 2011). However, the extent to which oriented cell divisions or other dynamic behaviors (T1 rearrangements, cell shape changes or extrusions) quantitatively contribute to the tissue's growth pattern is unknown. Analytical tools now exist to describe tissue growth fully from cellular dynamics (Merkel et al., 2017; Etournay et al., 2016, 2015; Guirao et al., 2015), but such methods demand long-term time-lapse imaging at high spatial and temporal resolution. In vivo imaging has an insufficient time resolution (Heemskerk et al., 2014), necessitating the use of ex vivo culture. The achievement of prolonged ex vivo wing disc growth has proven difficult, even though the eversion process that imaginal tissues undergo at the onset of pupariation can be successfully replicated in culture (Fristrom et al., 1973; Milner, 1977; Aldaz et al., 2010). Importantly, in order to study properly the cellular dynamics underlying oriented growth, we require a culture condition that maintains the signaling from AP/DV organizers and global PCP patterns, and how in vitro culture affects these systems is completely unknown.

Insulin signaling regulates animal size in response to nutrition, and bovine insulin has been widely used to stimulate growth of explanted discs (Zartman et al., 2013; Handke et al., 2014; Mao et al., 2011, 2013; Legoff et al., 2013; Heller et al., 2016; Tsao et al., 2016; Strassburger et al., 2017). However, proliferation in insulin-cultured discs arrests within a few hours (Handke et al., 2014; Tsao et al., 2016); thus, clearly other signals are required for long-term growth. One such missing signal could be ecdysteroids, a family of steroid hormones produced by the ring gland in arthropods with well-characterized functions in regulating major developmental transitions (Kozlova and Thummel, 2000). A peak of 20-hydroxyecdysone (20E) and related hormones at the larval-to-pupal transition induces pupariation and imaginal disc eversion (Fristrom et al., 1973; Milner, 1977). This peak is preceded by a smaller elevation in hormone levels that starts in the mid-third instar (Kozlova and Thummel, 2000; Lavrynenko et al., 2015). The levels of ecdysteroid at this stage are about 4% of the peak concentration. Several lines of evidence now suggest that these lower concentrations promote imaginal growth and development during larval stages (Bodenstein, 1943; Brennan et al., 1998, 2001; Mirth et al., 2009; Delanoue et al., 2010; Mitchell et al., 2013; Herboso et al., 2015).

Here, we show that physiologically low levels of 20E, in the absence of insulin, extend the proliferation of explanted wing discs compared with insulin alone. Transcriptome sequencing reveals that 20E is required in culture for the normal expression of numerous genes involved in wing patterning, whereas insulin promotes a strong but transient anabolic growth response. Genetic perturbations in vivo confirm that 20E has widespread effects on morphogen signaling and PCP in the larval wing, highlighting the importance of including 20E in culture media for replicating in vivo growth patterns. We exploit this improved system to perform live imaging and apply recently developed tools for decomposing tissue growth into cellular contributions (Merkel et al., 2017; Etournay et al., 2016, 2015). Our analysis indicates that divisions alone are insufficient to account quantitatively for the anisotropy of tissue; cell rearrangements and cell shape changes play an equally important role. This work inspires new directions for exploration into the regulation of tissue size and shape.

We directly tested the ability of low levels of 20E, with or without insulin, to maintain proliferation in explants from mid-third instar larvae [96 h after egg laying (AEL)]. In vivo, these discs would continue to proliferate for another 24 h (pupariation at ∼120 h AEL). We monitored the numbers of phospho-histone H3-positive proliferating cells in freshly explanted discs, and compared them with discs cultured in the absence of hormones, or in the presence of 20 nM 20E, 1 µM insulin, or both hormones (Fig. 1A,B). In the absence of any hormones, the number of mitotic cells per area (proliferation index) is dramatically reduced by 4 h and negligible by 9 h (Fig. 1Ai,Bi), consistent with previous work (Zartman et al., 2013; Handke et al., 2014). In the presence of insulin, the proliferation index drops to half that of freshly explanted discs after 4 h and continues to decline thereafter (Fig. 1Aii,Bii). These indices quantitatively agree with those of previous analyses of insulin-culture discs (Handke et al., 2014).

Strikingly, in the presence of 20E alone, the proliferation index is similar to that of insulin-cultured discs at 4 h, but instead of declining thereafter, the index actually increases (Fig. 1Aiii,Biii). From 9-16 h after explant, the same density of mitotic nuclei is found in 20E-cultured discs as in freshly explanted discs. Proliferation even extends to 24 h, albeit at a lower density than between 9-16 h. By 24 h, the overall morphology becomes somewhat abnormal (Fig. S1), perhaps reflecting the fact that components of the extracellular matrix come from the fat body in vivo (Pastor-Pareja and Xu, 2011).

The simultaneous addition of both hormones promotes proliferation mildly better than either alone for 4 h; but by 9 h, proliferation declines, as in the presence of insulin alone. Unlike in any other condition, however, the morphology of discs cultured for 9 h in insulin+20E becomes abnormal, as visualized by the cell area distribution in Fig. 1C. Proliferation is not significantly improved by adjusting 20E concentration (Fig. S1). Thus, under these explant conditions, the response triggered by the combined addition of hormones is incompatible with long-term proliferation with normal morphology.

To assess the consequences for gene expression of culturing under different hormonal conditions, we sequenced the transcriptomes of freshly explanted discs (96 h AEL), discs cultured without hormones for 4 h, and discs cultured for 4 or 9 h in either 20E or insulin alone. Relative to freshly explanted discs, the expression of 1604 genes goes down and 1409 genes goes up after 4 h of culture without hormones. Changed expression of 1911 genes (63%) could be prevented, at least transiently, by the addition of either 20E or insulin. Some genes responded to both 20E and insulin, whereas others were hormone specific.

Fig. 2 shows the size and overlap of the hormone-responsive gene sets. Comparing the transcriptional response to insulin at 4 and 9 h reveals that it is largely transient (Fig. 2A), consistent with the decreased proliferative response to insulin by 9 h (Fig. 1Aii,Bii). In contrast, although 20E regulates a smaller subset of genes, its effects are longer lasting (Fig. 2A). A subset of genes is regulated by either hormone (Fig. 2B). Maximum overlap (353 genes) occurs at time points when each hormone supports maximum proliferation (insulin at 4 h, 20E at 9 h). This overlap represents 22% of all insulin-regulated genes and 60% of 20E-regulated genes.

The transcriptional response to insulin that we observe in the wing disc is consistent with the results of many previous studies characterizing nutrient-dependent gene expression in other contexts (Fig. 3, Figs S2, S3) (Gershman et al., 2007; Teleman et al., 2008; Li et al., 2010). Genes induced by insulin are significantly enriched in gene ontology (GO) terms describing translation and ribosome biogenesis, cell cycle/DNA replication, and mitochondrial biogenesis (Fig. 3, Fig. S2). Conversely, insulin reduces the expression of genes involved in autophagy and cell death (Fig. 3, Fig. S6C). By 9 h of culture, however, the response of >70% of all insulin-regulated genes either weakens or disappears. This transient-responding group (Fig. 3Aii,B) includes many previously identified targets of the TOR pathway and the transcription factor FoxO, which lie downstream of insulin signaling (Fig. S3). The only enriched GO categories in the stably regulated genes (∼30% of total) describe oxoacid metabolic reactions (Fig. 3Ai).

To determine whether the decline in the transcriptional response to insulin reflects a loss of sensitivity to the hormone during culture, we examined two readouts of pathway activity. Binding of insulin to its receptor activates phosphoinositide 3-kinase (PI3K), activity of which can be monitored with a reporter that binds to the enzyme's product, phosphatidylinositol (3,4,5)-trisphosphate [PtdIns(3,4,5)P3 or PIP3]. Even after 9 h of culture, we observe strong membrane localization of the reporter (Fig. S4), suggesting that PI3K activity is stable. We also examined the intracellular localization of FoxO, a transcription factor that is phosphorylated and retained in the cytoplasm during insulin signaling. Even after 9 h, insulin maintains robust nuclear exclusion of FoxO (Fig. S5). This is surprising because we found that ostensibly FoxO-induced targets that repress growth (i.e. the translational repressor 4E-BP; also known as Thor) are reactivated after 9 h of insulin culture (Fig. S3B). Thus, the retention of FoxO in the cytoplasm is not sufficient for maintaining the insulin-dependent transcriptional state or growth in the wing disc. Additional insulin-independent transcriptional regulators must affect these growth-inhibitory FoxO target genes.

About 22% of insulin-responsive genes are also regulated by 20E, at least after 9 h of culture. Both hormones repress genes involved in DNA-damage response and cell death and induce genes involved in DNA replication (Fig. S6). Amongst the shared targets, we found a subset involved in protein synthesis and nutrient transport, as well as some previously identified TOR- and FoxO-responsive genes (Figs S6 and S3). For example, 4E-BP is ectopically activated when explanted discs are cultured without hormone, but can be repressed either by insulin (at 4 h) or 20E. These data suggest that there could be crosstalk between the 20E- and insulin-signaling pathways and are consistent with a previous report that 20E can affect 4E-BP levels in wing discs in vivo (Herboso et al., 2015).

Unlike in the insulin targets, significantly enriched GO terms in the 20E-induced genes also describe a variety of developmental processes (Fig. 4). Genes with these GO terms include components and targets of the Notch, Wg, EGF and JAK/STAT pathways (Fig. 5). Also significantly affected was sugarless, which is required to synthesize the heparan sulfate moieties of proteoglycans needed for Dpp, Hh, FGF and Wg signaling (Hacker et al., 1997; Bornemann et al., 2004; Yan and Lin, 2009). 20E, and not insulin, also supports the expression of a wide variety of developmentally important transcription factors, chromatin-modifying proteins, and many genes that regulate tissue morphogenesis, including junctional, cytoskeletal and ECM components (Fig. 5). A smaller set of developmentally important genes seems to be regulated by either hormone; however, 20E generally causes a more potent and long-lasting response in this group (Fig. S6). Of the genes that only 20E induces, roughly 30% are already known to perturb wing development when mutated.

Although the expression of many of these patterning genes is stably maintained by 20E in culture, a subset of 20E genes are only normalized in culture after 9 h but less well at 4 h (Fig. 4B). This subset is enriched for genes involved in DNA replication, consistent with the higher proliferation index at 9 h than at 4 h in culture with 20E (Fig. 1Aiii,Biii). The molecular mechanisms responsible for this delay in full response remain unclear. We also identified 102 genes that were only responsive to 20E at 4 h and not 9 h, but only one GO term, cuticle production, was enriched in this subset (data not shown).

On the whole, these transcriptional data suggest that insulin transiently promotes proliferation through its well-understood effects on anabolic metabolism. 20E promotes proliferation over a longer time scale, but influences only a subset of the direct growth-promoting genes activated by insulin. Instead, 20E supports the expression of genes involved in wing patterning and morphogenesis.

To determine whether our transcriptomic data from explants reflect an in vivo requirement for 20E in maintaining patterning systems during growth, we genetically perturbed 20E signaling in vivo by transiently overexpressing a dominant-negative Ecdysone receptor (EcR-DN) during the third instar and then assessing its effect on a broad range of morphogen and PCP pathways (Fig. 6). EcR-DN expression autonomously reduces tissue size and mitotic cell number (Fig. S7A) (Herboso et al., 2015). However, the tissue appears otherwise healthy: apoptosis is not affected (Herboso et al., 2015; data not shown), and neither the localization nor protein levels of Discs large is perturbed (Fig. 6A).

Expressing EcR-DN in the dorsal compartment decreases Wg expression, both in the hinge and in the stripe at the DV boundary (Fig. 6Bi). The autonomous effect of EcR-DN on the DV boundary Wg stripe is more obvious when it is overexpressed in the posterior compartment (Fig. S8). In addition, wing discs deprived of 20E in vivo have severely reduced levels of Wg and also grow less (Fig. S9A, Fig. S7B). These results are consistent with previously published data using similar perturbations (Mirth et al., 2009; Herboso et al., 2015).

In wild-type discs, both E-Cadherin (Shotgun) and Delta are upregulated by Wg signaling on either side of the DV boundary (Micchelli et al., 1997; Jaiswal et al., 2006; de Celis and Bray, 1997). Dorsal EcR-DN expression prevents both from accumulating on the dorsal side of the DV boundary, consistent with reduced Wg signaling in this region (Fig. 6Bii,Biii). Like at the DV boundary, EcR-DN expression also affects elevated Delta expression on either side of a region of high Hedgehog signaling near the AP boundary (Fig. 6Bii). This effect on Delta is consistent with our transcriptomic data showing that mRNA levels of Delta itself, as well as those of effectors of the EGF and Notch pathways that regulate the AP Delta pattern, drop in culture without 20E (Fig. 5).

Also consistent with our transcriptomic data, we found that Dally-like protein levels are considerably reduced in the EcR-DN-expressing compartment (Fig. 6Biv). Given the broad effects of heparan sulfate proteoglycans on morphogen signaling, we therefore also looked at Hh and Dpp pathways. Indeed, EcR-DN decreases levels of pMad (a readout of Dpp signaling) in the posterior compartment (Fig. 6Ci). Furthermore, Hh protein levels and signaling are strongly affected (Fig. 6Cii-Civ). Hh staining intensity decreases in the posterior producing cells, consistent with roles for Dally and Dally-like in controlling its trafficking there (Eugster et al., 2007; Ayers et al., 2010, 2012). Furthermore, the gradient in the anterior receiving cells is diminished and the amplitude and range of Hedgehog signaling outputs are reduced. Ci is stabilized over a shorter distance from the AP boundary (Fig. 6Ciii). Engrailed expression is also reduced, both anterior to the AP boundary, where expression is activated by Hh signaling, and posteriorly where its expression is maintained independently of Hh by chromatin modifications (DeVido et al., 2008) (Fig. 6Civ). The Hh gradient and signaling output were severely reduced in wing discs from larvae deprived of 20E in vivo, confirming that the effects of EcR-DN expression reflect a real requirement for 20E (Fig. S9).

Our transcriptomic analysis also revealed that 20E affects the expression of the Core PCP component Prickle and the Fat PCP component Four-jointed (Fig. 5, Fig. S6). To examine global patterns of PCP, we stained for Dachsous (Ds) and Flamingo (Fmi; Starry night, Stan), which were not transcriptionally affected by 20E in culture but should nonetheless reveal any defects in the global patterns of PCP. Interestingly, EcR-DN expression was found to dramatically lower expression of both Ds and Fmi, presumably post-transcriptionally, such that any residual planar polarity was unquantifiable (Fig. 6D).

In summary, both in vivo experiments and transcriptomic data from explanted discs suggest that 20E is required to maintain the signal transduction and PCP pathways that drive growth and patterning in the wing disc. Given its key role in regulating patterning in vivo and in vitro, addition of 20E to explant culture could be crucial for measuring the normal patterning of growth.

Having established that 20E supports longer proliferation and is better than insulin at maintaining the gene expression patterns underlying oriented growth, we used the new 20E-containing media to investigate the cell dynamics underlying growth and its anisotropy with live imaging. We analyzed the central wing pouch (shown in Fig. S10) of three discs in 13 h time-lapse experiments (shown together in Movie 1).

We first examined area change in the wing pouch and its cellular contributions (Fig. 7Bi). We note that tissue area (Fig. 7Bi, blue line) decreases during the first 2 h, almost entirely as a consequence of cell area changes (Fig. 7Bi, green line), suggesting that some time is required for discs to adapt to culture or recover from mounting. However, after the adaptation phase, tissue area grows. Cell divisions (Fig. 7Bi, orange line) contribute positively to tissue area growth and are only partly counteracted by modest levels of cell extrusion (Fig. 7Bi, cyan line) and slight cell area reduction (Fig. 7Bi, green line). This slight reduction in cell area is consistent with in vivo observations of growth over development time (Aegerter-Wilmsen et al., 2012; Mao et al., 2013).

To analyze the anisotropy of tissue growth in 20E, we measured the patterns of average pure tissue shear and its cellular contributions accumulated over 11 h after the 2 h adaptation phase using the Tissue Miner computational framework (Etournay et al., 2016) and Triangle Method (Merkel et al., 2017) (Fig. 7A, Fig. S11). Pure shear describes a change in aspect ratio that is independent of area change. In Fig. 7A, total tissue shear, determined on a grid of squares and accumulated over time, is indicated by lines, the length of which is proportional to the magnitude of pure shear and the orientation indicates its axis. We observe that tissue shear is oriented perpendicular to an axis defined by the DV boundary throughout the tracked region of the wing pouch, generally consistent with analyses of in vivo clone data (González-Gaitán et al., 1994; Resino et al., 2002; Baena-López et al., 2005; Worley et al., 2013; Mao et al., 2013; Heemskerk et al., 2014).

To examine the dynamics of tissue shape change, and its underlying cellular contributions, we calculated the average tissue shear and its cellular contributions over the entire tracked region using a coordinate system whose x-axis corresponding to the DV boundary (Fig. 7A, dotted horizontal line) and the y-axis perpendicular to it. In this coordinate system, a positive xx component of shear corresponds to shear parallel to the DV boundary, whereas a negative xx component of shear corresponds to shear perpendicular to the DV boundary. Fig. 7Bii (blue line) shows the xx component of the average tissue shear throughout the tracked region, accumulated over time. These plots reveal that the wing pouch becomes more anisotropic as it grows. Examining the cellular contributions to tissue shear reveals that cell divisions (Fig. 7Bii, orange line; see also Fig. S11) contribute to the change in shape of the wing pouch during growth. However, cell division orientation cannot entirely account for tissue shear: quantitatively similar contributions to tissue shear stem from both cell rearrangements (Fig. 7Bii, red line; see also Fig. S11) and cell shape changes (Fig. 7Bii, green line; see also Fig. S11).

Comparing the outlines of the wing pouch at the beginning and end of the movies (Fig. 7A) suggests that wing pouch tissue is expanding in a direction almost entirely perpendicular to the DV boundary. To quantify expansion along the x- and y-axes separately over time, we determined the xx and yy components of velocity gradient tensor based on the measured shear and area expansion (see supplementary Materials and Methods). Indeed, tissue expansion (Fig. 7Biii,Biv, blue lines) occurs along the y-axis, perpendicular to the DV boundary (Fig. 7Biv) and not along the x-axis (Fig. 7Biii). We decomposed the xx and yy components of the tissue velocity gradient tensor into their respective cellular contributions (see supplementary Materials and Methods). This calculation reveals that expansion of the wing pouch along the y-axis is almost identical to the expansion contributed by cell divisions in all three movies (compare orange and blue lines in Fig. 7Biv). Surprisingly, along the x-axis, the relationship between tissue flow and cell divisions is very different. Expansion due to cell divisions is only slightly smaller along the x-axis than along the y-axis (Fig. 7Biii,Biv, orange lines). Nevertheless, the xx component of tissue velocity gradient is small and negative, i.e. the tissue does not expand in this axis but rather contracts slightly. In this direction, the sum of the other cellular contributions is larger than along the y-axis and balances the contribution from cell divisions. We note that, although the contributions of cell shape changes, rearrangements and extrusions vary in each of the three movies, they always combine to cancel the expansion along the x-axis due to cell division.

To confirm that cell elongation contributes to oriented growth in vivo, we quantified cell elongation in the wing pouch of explanted discs at 96 h, 119 h and 136 h AEL (Fig. 8). Indeed, the xx component of cell elongation becomes more negative over developmental time. This result is consistent with previously published patterns of cell elongation (Mao et al., 2013).

In summary, we use an improved method for long-term cultivation of wing imaginal discs to measure directly the cellular contributions to tissue shape and area changes. This analysis revealed that contributions from cell divisions to tissue shape changes are not nearly as anisotropic as tissue expansion and indicates that additional morphogenetic mechanisms exist to prevent expansion of the wing parallel to the DV boundary.

This work considerably expands our understanding of the cell dynamics in the growing Drosophila wing disc, revealing important contributions from cell rearrangements and cell shape changes to oriented tissue growth. Our analysis provides the foundation for future experiments aimed at understanding how patterned gene expression and tissue mechanics regulate cellular behavior during tissue growth. Furthermore, our transcriptomic data from wing discs cultured in 20E or insulin provide a novel global view of the direct response of wing imaginal tissues to these hormones. These data have in vivo implications for the organismal coordination of growth and development and constitute a valuable resource for the planning of any type of experiment requiring wing disc culture.

Studying the cellular basis of tissue development requires prolonged, high time resolution imaging that is ideally entirely in vivo. However, the Drosophila wing grows during larval stages, when animals are mobile and feeding. Although completely in vivo methods for visualizing wing disc growth continue to improve (Heemskerk et al., 2014), the currently available frame rate is still far too low to study cell dynamics. Analyzing whether cell boundary exchanges that are noisy and dynamic have an orientation, for example, demands the integration of high time resolution data over as much time as possible. Explant culture is the only option, provided that the patterning systems that drive oriented growth can be maintained.

Although previous studies have used insulin to sustain discs long term in culture (Zartman et al., 2013; Handke et al., 2014; Mao et al., 2011, 2013; Legoff et al., 2013; Heller et al., 2016; Tsao et al., 2016; Strassburger et al., 2017), we show that the expression of numerous genes involved in patterning declines when discs are cultured without the steroid hormone 20E. 20E was previously thought to promote imaginal growth by repressing the expression of the translational inhibitor 4E-BP (Herboso et al., 2015); however, we show here that, in addition, 20E broadly affects many different signaling pathways, each one of which is crucial for growth. These data indicate that the inclusion of 20E in culture media is vitally important to the goal of replicating in vivo growth patterns. Although high concentrations of 20E promote eversion, lower concentrations promote growth in the absence of insulin. This condition actually supports proliferation for longer than insulin-containing media: up to ∼24 h in 20E, compared with only ∼7-10 h in insulin (Fig. 1) (Zartman et al., 2013; Handke et al., 2014; Tsao et al., 2016).

How does the growth of 20E-cultured discs compare with growth in vivo? Although discs can proliferate in culture for up to 24 h, they slow down sooner (after ∼13 h) if they are continuously imaged. Nonetheless, for the first 13 h of live imaging, our direct quantification of the tissue growth pattern agrees well with results of indirect methods for following growth in vivo (González-Gaitán et al., 1994; Resino et al., 2002; Baena-López et al., 2005; Mao et al., 2013; Worley et al., 2013; Heemskerk et al., 2014). We also confirm data from fixed tissues (Baena-López et al., 2005) showing that cell division orientation has a slight bias toward a direction perpendicular to the DV boundary, at least in central regions of the pouch (compare orange lines in Fig. 7Biii,Biv). We estimate that the overall cell doubling time in continuously imaged explanted discs is about three times longer than the estimated rate in well-fed animals (Garcia-Bellido and Merriam, 1971; Bryant and Levinson, 1985; González-Gaitán et al., 1994; Heemskerk et al., 2014). However, phototoxicity associated with continuous imaging seems to affect not just the duration but also the rate of proliferation: estimating the rate from the density of PH3-stained nuclei in discs that were not continuously imaged suggests a value that is closer to that of freshly explanted discs (Fig. 1).

Interestingly, in vivo, if animals are starved after the attainment of a critical weight, the release of insulin-like peptides from the brain is reduced (Géminard et al., 2009) and growth arrests in the larval – but not imaginal – tissues (Britton and Edgar, 1998; Cheng et al., 2011). The rate of imaginal proliferation during this so-called ‘sparing’ condition is ∼60% lower than in the well-fed state (Cheng et al., 2011). Thus, proliferation of explants given 20E but not insulin may be closer to this in vivo low-insulin state. Importantly, although the ablation of insulin-producing cells in the brain reduces the size of the emerging adult wings, these wings are nevertheless well-proportioned (Rulifson et al., 2002). Thus, the orientation of tissue growth and its underlying cell dynamics is likely to be preserved in discs cultured with 20E in the absence of insulin. Perhaps the slight but reproducible lag in the division rate at early times of 20E culture (Fig. 1Biii, Fig. 7Bi) reflects a transition phase when discs taken from well-fed animals must switch to an insulin-independent/20E-dependent mode of growth.

The mechanism underlying the sparing of imaginal tissues is unknown. Sparing of the central nervous system involves a locally provided ligand that activates downstream signaling in the insulin pathway (Cheng et al., 2011). Our transcriptomic data provide no evidence that wing discs can produce their own insulin-like ligands or that they can fully activate insulin signaling in the absence of exogenous ligands. Nonetheless, wing discs in culture continue to proliferate in 20E, even when no insulin-like ligand is present. This finding raises the possibility that 20E is also sufficient to support their proliferation in vivo during starvation. Although 20E promotes imaginal growth, it inhibits that of larval tissues (Delanoue et al., 2010), potentially explaining the opposite response to starvation of these two tissue types.

Under well-fed conditions, insulin signaling must be able to combine with 20E signaling to further increase imaginal disc growth, but this feature is not well-reproduced in culture (Fig. 1Biv) (Handke et al., 2014; Strassburger et al., 2017). Drosophila has seven insulin-like peptides that interact not only with the single receptor, but also with other proteins that modulate their activities (Honegger et al., 2008; Arquier et al., 2008; Okamoto et al., 2013). Perhaps specific insulin peptides or their binding proteins allow proper interfacing with 20E signaling to promote growth.

Although the mechanisms dictating cell division orientation in this tissue have been studied (Mao et al., 2011, 2013; Legoff et al., 2013), the extent to which the observed slight bias in cell division orientation could account for the orientation of tissue growth has remained unclear. The Triangle Method (Merkel et al., 2017; Etournay et al., 2015) for decomposing tissue shape changes into quantitative contributions from all types of cellular events provides a measurement tool to connect the cell and tissue scales. Our results reveal that the observed bias in cell division orientation in the central region of the disc is insufficient to explain the anisotropy of tissue growth. Whereas expansion of the tissue perpendicular to the DV axis is driven primarily by cell divisions, in the orthogonal direction, cell shape changes and T1 transitions fully cancel the contribution to expansion by cell divisions, such that the wing does not expand along an axis parallel to the DV boundary.

Previous efforts to quantify cell rearrangements have used live imaging of discs cultured in insulin and have concluded that they occur either with very low frequency (Gibson et al., 2006; Legoff et al., 2013), or with a significant frequency but no pattern or orientation at the tissue scale (Mao et al., 2013; Heller et al., 2016). In contrast, we see that the contribution of cell rearrangements to tissue growth is quantitatively similar to that of oriented cell division and helps restrict growth parallel to the DV boundary. This discrepancy could result from either the differing culture conditions (lack or presence of 20E) or the inherent difficulties in defining and counting cell rearrangements. A classic T1 transition (Weaire and Rivier, 1984) occurs when two pairs of cells exchange neighborships, i.e. the connecting bond between two cells shrinks as all cells come together to make a four-way vertex; then, a new bond is formed such that the two cells that were originally separated come into contact, whereas the two that were originally together stay separate. But sometimes the four-way vertex simply resolves back without changing the original neighborship, and sometimes the neighborship fluctuates back and forth several times. How and when do you count the rearrangement? The method that we use overcomes the problem of having to classify a ‘true’ T1 event based on observation time windows. It considers all changes in neighborship, regardless of the length of time they persist, and quantifies their accumulated contribution to tissue shape change. Thus, we are not simply quantifying the number and orientation of T1s but the tissue shape change that they cause. This approach could be key to our ability to detect and quantify the important contribution of cell rearrangements to oriented growth of the wing disc.

What mechanisms might allow cell rearrangements, cell shape changes and cell extrusions to cancel expansion due to cell division specifically in one direction? We propose that a mechanical constraint prevents wing pouch expansion parallel, but not perpendicular, to the DV boundary. In this case, area growth due to cell division would lead to anisotropic expansion with corresponding anisotropic stresses that could be associated with oriented cell shape changes and rearrangements. Consistent with this hypothesis, we observe that the sum of expansion due to rearrangements, shape changes and extrusions is more reproducible than their individual cellular contributions. Long-range anisotropic stresses could coordinate these cellular processes and account for their mutual dependence. A similar dependence between cellular contributions is observed in the pupal wing, where the constant tissue area is maintained by epithelial tension via extracellular matrix connections to the cuticle (Etournay et al., 2015). We do not yet know whether the larval wing disc is mechanically constrained. It will be interesting to explore whether the extracellular matrix has anisotropic properties or whether the folding pattern of the tissue influences growth orientation.

The fact that Fat PCP mutant wings are associated with less anisotropic clone shapes and more weakly oriented cell divisions (Baena-López et al., 2005; Mao et al., 2011) has led to the idea that Fat orients growth through its effect on cell division orientation. We find that shear due to cell divisions cannot fully account for the anisotropy of growth, however. This result suggests either that there are other orienting factors, or that Fat PCP has more profound effects on growth anisotropy than can be accounted for by cell division.

Overall, this work opens new avenues towards an understanding of wing morphogenesis that integrates different scales – from the larval hormonal networks to the morphogen-dependent patterns of cell dynamics from which tissue shape emerges.

The culture experiments described in Fig. 1 and the transcriptomic analyses were performed using wild-type OregonR Drosophila. For live-imaging experiments, we used animals in which the endogenous E-Cadherin gene had been fused with GFP (Huang et al., 2009). To enable temperature-inducible GAL4-dependent gene expression, we combined the apterous-GAL4 (Marois et al., 2006), phantom-GAL4 (Ono et al., 2006) and engrailed-GAL4 (DeVido et al., 2008) loci with the temperature-sensitive GAL4 repressor GAL80^ts driven by the tubulin promoter. UAS-EcR.B1 (BDSC 6469) and UAS-EcR.B1-DeltaC655.W650A (BDSC 6872) (Cherbas et al., 2003) were acquired from Bloomington Stock Center (Bloomington, Indiana, USA). The line used to induce RNAi against neverland was obtained from the group of R. Niwa (Yoshiyama et al., 2006). To visualize FoxO, we used the dfoxo-v3-mCherry knock-in line generated by genomic recombineering of the endogenous foxo (Kakanj et al., 2016).

Flies were grown on standard media containing cornmeal, molasses agar and yeast extract under a 12 h light/dark cycle at 25°C (except when gene expression was induced using the temperature-sensitive GAL80, described below).

For all disc culture experiments, we used wing discs from larvae at 96 h AEL. Flies were allowed to lay eggs on apple juice agar plates (supplemented with yeast paste) for 1-2 h at 25°C. Agar pieces containing ∼10-12 eggs each were placed into food vials and grown at 25°C. The middle of the egg collection window was considered to be 0 h AEL.

Early experiments (data not shown) indicated that overall disc morphology was best preserved in Grace's Insect culture medium, compared with Schneider's or Shield's and Sang. Thus, Grace's was used for all experiments. Grace's medium (Sigma, G9771) was prepared without sodium bicarbonate but with the addition of 5 mM BisTris. The pH was adjusted to 6.6-6.7 at room temperature, and the prepared liquid media was stored at 4°C for no longer than one month. On the day of the experiment, we added 5% fetal bovine serum (FBS; ThermoFisher/Invitrogen, 10270098) and Penicillin-Streptomycin (Sigma P4333, 100× stock solution ) to impede microbial growth. Note that although other groups have included female fly extract in the media (Zartman et al., 2013; Handke et al., 2014; Tsao et al., 2016), we found it to be less reproducible than FBS (data not shown). Furthermore, recent evidence indicates that fly extract is associated with non-physiological calcium waves in explanted wing discs (Balaji et al., 2017).

Stock solutions of either 20E or insulin were added just prior to the start of the experiment. 20E (Sigma, H5142) was added as a 1:1000 dilution from an ethanol stock solution (stored at −20°C, prepared fresh weekly) to give a final concentration of 20 nM in the media (except when otherwise noted). Bovine insulin (Sigma, I5500) was prepared as a 10 mg/ml stock solution in acidified water (stored at −20°C) and added at a final concentration of 5 µg/ml. This concentration of insulin is comparable to the levels used in previous studies aimed at prolonging proliferation of discs in culture (Zartman et al., 2013; Handke et al., 2014). Given that our data on the proliferation index in discs cultured in insulin alone (Fig. 1Bii) quantitatively agree with the results of these previous studies, the other differences between our culture media and theirs do not seem to be as important as hormonal content.

To prepare larvae for dissection, we first floated them out of the food by adding 30% sucrose. Larvae were transferred using a brush or wide-tip transfer pipette into glass dishes. Excess food was removed by washing in sucrose, followed by distilled water. They were then surface sterilized by immersion in 70% ethanol for 1-2 min. Finally, they were washed once more with sterile water and then media. Except for the live-imaging experiments, discs were dissected in hormone-free media and then immediately transferred from the dissecting well into media containing hormone. For each experiment, dissections were performed over a period of no more than 45 min, and the start time of the culture was considered to be the midpoint of that dissection window. The number of discs analyzed per sample was limited by how many could be manually dissected as carefully as possible during that time window.

Discs were cultured in glass wells (Electron Microscopy Sciences; 70543-30) in 500 µl of growth media, with 10-20 discs per well, at 25°C in humidified chambers. Media was exchanged approximately every 2 h, except during the long incubations of 16-24 h, when there was a maximum incubation without media change of 8 h. At the end of culture, discs were fixed by exchanging the media with 4% paraformaldehyde (PFA; 8% PFA stock solution in PBS diluted into Grace's media). Discs were incubated for 20 min at room temperature and stained as described below in the immunofluorescence section.

The data presented in Fig. 1 represent a compilation of results from multiple days (labeled in different colors). Two replicates were carried out for each hormonal condition and time point. All discs were included in the analysis, except in rare cases when a disc was distorted during mounting.

Three biological replicates were performed for each hormone condition and time point. For each, ∼20-30 discs were dissected and cultured as described above and then transferred to a microfuge tube in a volume of 15-20 µl (growth media). Discs were frozen immediately in liquid nitrogen and stored at −80°C until all samples were ready for RNA isolation. A Qiagen RNeasy Mini kit was used to isolate RNA for sequencing. Samples were lysed by adding Buffer RLT+2-mercaptoethanol and vortexing for 20-30 s on ice. The manufacturer's instructions were followed for the rest of the protocol. Further purification was achieved with an ethanol precipitation.

mRNA was isolated from 1 µg total RNA by poly-dT enrichment using the NEBNext Poly(A) mRNA Magnetic Isolation Module according to the manufacturer's instructions (New England Biolabs). Final elution was performed in 15 µl of 2× first strand cDNA synthesis buffer. After chemical fragmentation by incubating for 15 min at 94°C, the sample was directly subjected to the workflow for strand-specific RNA-Seq library preparation (Ultra Directional RNA Library Prep, New England Biolabs). For ligation, custom adaptors were used (Adaptor-Oligo 1: 5′-ACACTCTTTCCCTACACGACGCTCTTCCGATCT-3′, Adaptor-Oligo 2: 5′-P-GATCGGAAGAGCACACGTCTGAACTCCAGTCAC-3′). After ligation, adapters were depleted with an XP bead purification (Beckman Coulter), adding beads in a ratio of 1:1. Indexing was carried out during the following PCR enrichment (15 cycles) using custom amplification primers carrying the index sequence indicated with ‘NNNNNN’ (primer 1: Oligo_Seq 5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCT-3′; primer 2: 5′-GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT-3′; primer 3: 5′-CAAGCAGAAGACGGCATACGAGATNNNNNNGTGACTGGAGTT-3′). After two more XP beads purifications (1:1), libraries were quantified using Qubit dsDNA HS Assay Kit (Invitrogen). Equimolar amounts of each sample were pooled and distributed amongst four lanes of an Illumina HiSeq 2500 sequencer for 75 bp single-end sequencing. On average we achieved 37 million reads per sample.

Short-read data was trimmed using Cutadapt (Martin, 2011) and aligned using TopHat2 (Kim et al., 2013). Differential gene expression analysis was performed for all pairwise comparisons between conditions and time points using Cufflinks (Trapnell et al., 2012). Our threshold for significance was q<0.01. We considered a gene to be expressed if it had a normalized count of fpkm (fragments per kilobase of transcript per million mapped reads)>5.

Raw sequencing data and fpkm normalized expression data files have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE92933.

The identification of hormone-responsive gene sets was achieved by selectively filtering the differentially expressed genes table from Cufflinks using custom R scripts. First, we identified the genes in this table for which levels were significantly different (q<0.01) between no culture (freshly dissected discs from the beginning of the experiment, 96 h AEL) and culture without hormone (4 h). We split this culture-responsive subset into two groups based on whether their levels in uncultured discs were abnormally low or high after culture without hormone. To identify hormone-activated genes, we first identified the genes for which expression was abnormally low in culture with no hormone (relative to uncultured control). Amongst these, we identified genes whose levels were significantly elevated (either at 4 h or 9 h) when discs were cultured with hormone (compared with culture with no hormone). Hormone-repressed genes had the opposite pattern: abnormally high when cultured without hormone but significantly lower in culture with hormone. Additional information is provided in the supplementary Materials and Methods.

The Venn diagrams presented in Fig. 2 were generated using the VennDiagram package of R (Chen and Boutros, 2011) after identifying the genes that were unique or common in each pairwise comparison. We provide the gene lists that were used to create these graphs in Tables S1-S5. These lists also include information about whether the expression of each gene in culture without hormone is abnormally low or high, relative to its levels in uncultured discs.

We identified enriched Gene Ontology (GO) terms (Biological Process and Cellular Component) within each hormone-responsive dataset using the R-package ClusterProfiler (v 1.9) (Yu et al., 2012). To reduce redundancy, we used the simplify function of this package; similarity cutoff=0.95 and BH-corrected P-value cutoff=0.01. We also required enriched GO terms to have more than five genes included in the data. In Figs 3, 4 and Fig. S2, we only present the subcategories that have more than one enriched GO term.

For selected genes, we plotted the change in expression levels across all conditions (Fig. 5; Figs S2, S3, S6) using the pheatmap package in R. We took the log₂ of the normalized counts (fpkm) in each condition divided by that in the uncultured wing discs (at 96 h AEL, time 0 h).

For the 589 genes regulated by 20E (either exclusively or also by insulin), we manually classified functional groupings after reviewing related literature, GO term assignments and FlyBase annotations. These groupings are described in the supplementary Materials and Methods.

In Figs S3 and S6, TOR targets were selected based on previous studies (Teleman et al., 2008; Guertin et al., 2006). The FoxO targets were selected based on the experimental verifications in published studies (Puig et al., 2003; Teleman et al., 2008; Olson et al., 2013; Chen and Boutros, 2011; Lee et al., 2010). Most direct FoxO targets identified using adult females (Alic et al., 2011) were either not differentially expressed in our culture conditions or not insulin sensitive [i.e. Su(Hw), Akt1, Pi3K68D, Indy; data not shown]. Note, however, that FoxO target genes have been shown to be variable, depending on tissue and developmental stage (Teleman et al., 2008; Alic et al., 2011).

In Fig. 5, 20E targets were taken from previous publications (Beckstead et al., 2005; Gauhar et al., 2009).

Apterous-GAL4, GAL80^ts/CyO-GFP females were crossed with males of UAS-EcR or UAS-EcR-W650A, and their progeny were kept at 18°C (permissive for GAL80 repressor function) on Bromophenol Blue-containing food for 7-8 days. Vials were then moved to 29°C (restrictive) to induce GAL4-dependent transcript of EcR constructs for 24 h. Upcrawling larvae were selected for dissection based on the absence of GFP and the presence of blue food in the gut (indicating that they are still many hours from pupariation) (Andres and Thummel, 1994).

To remove circulating 20E, UAS-CD8-GFP;;phantom-GAL4, GAL80^ts/TM3-GFP females were crossed with males of the neverland targeting line, and their progeny were kept at 18°C. Vials were shifted to 29°C, and discs from upcrawling larvae lacking the TM3-GFP balancer were dissected 2 or 4 days later. As controls, we used progeny of UAS-CD8-GFP;;phantom-GAL4, GAL80^ts/TM3-GFP females crossed to w¹¹¹⁸ flies in parallel. The progeny of the control cross all pupariated by 2 days after the shift to 29°C.

Primary antibodies used were: anti-Ci-full length [1:100, rat, Developmental Studies Hybridoma Bank (DSHB) 2A1 conc], anti-Dally-like (1:100, mouse, DSHB 13G8 conc), anti-Dachsous (1:1000, mouse; Merkel et al., 2014), anti-Delta (1:800, mouse, DSHB C594.9B, conc), anti-Engrailed (1:100, mouse, DSHB 4D9 conc), anti-Discs-large (1:100, mouse, DSHB 4F3), anti-E-Cadherin (1:100, rat, DSHB DCAD2 sup), anti-Flamingo (1:200, rabbit; Sagner et al., 2012), anti-Hh (1:500, rabbit; Eugster et al., 2007), anti-mCherry (1:500, ThermoFisher, PA5-34974), anti-Patched (1:100, mouse, DSHB Apa1 conc), anti-pHistoneH3-Ser10 (1:1000, rabbit, Cell Signaling, #9701), anti-pMAD (1:1000, rabbit, Epitomics, 1880-1), anti-Wg (1:100, mouse, DSHB 4D4).

Secondary antibodies used were: Alexa Fluor-488 goat anti-rabbit (1:1000, ThermoFisher, A11034), Alexa Fluor-555 goat anti-mouse (1:1000, ThermoFisher, A21424), Alexa Fluor-647 goat anti-rat (1:500, ThermoFisher, A21247).

Discs were dissected in PBS and then fixed and stained according to standard methods (detailed in the supplementary Materials and Methods).

Imaging of fixed samples was performed using an Olympus FV1000 laser-scanning confocal microscope fitted with an Olympus BX61 inverted stand and motorized xyz stage, driven by FV10-ASW 1.7 software. Wing discs were imaged using either an Olympus UApochromat 40×1.35NA oil immersion or an Olympus UPlanSApochromat 60×1.35NA oil immersion objective.

All discs were imaged and analyzed, except in very rare cases when the discs were damaged or misshapen during mounting.

z-stacks of wing discs stained with PH3 were analyzed in 2D after a maximum intensity projection. PH3+ nuclei were segmented using Fiji Weka segmentation plugin (Arganda-Carreras et al., 2017). A training set containing images from each condition of the experiment was used to generate a common classifier, which was then applied to all images. Probability masks were thresholded by likelihood and size. The wing pouch was identified morphologically by the innermost folds in the epithelium. Compartment and wing pouch size in Fig. S7 was measured on a 2D projection, using Fiji to identify the region of interest manually (guided by E-cadherin or EcR staining). Plotting was performed in R. For all box plots, the hinges correspond to the first and third quartiles; lines indicate the highest and lowest values within 1.5*IQR (inter-quartile range); outlier points are black, except in Fig. S7Aiii, where they are blue.

In Fig. 1C, apical cell areas were determined and plotted using Tissue Analyzer (Aigouy et al., 2016) to segment cells based on E-Cadherin staining, and Tissue Miner (Etournay et al., 2016) to determine the area and position of each cell.

The changes to patterns of protein production/localization upon in vivo perturbation of 20E signaling were analyzed in a 2D maximum intensity projection. Using Fiji (Schindelin et al., 2012), a line 50 pixels wide (∼20 µm) was drawn either from dorsal to ventral in the anterior and posterior compartments or from anterior to posterior in the dorsal and ventral compartments. Absolute intensity values along these lines were averaged across all samples (imaged on the same day, under the same acquisition settings). Plotted is the mean and standard deviation of intensity along these lines. At least five discs were imaged per biological condition; exact numbers are shown in the figures.

Discs were dissected directly in growth media containing hormone. For imaging, discs were gently immobilized under a porous filter (Whatman cyclopore polycarbonate membranes; Sigma, WHA70602513) in a 35 mm glass-bottomed Petri dish (Mattek, P35G-1.0-14-C) using double-sided adhesive tape (Tesa 5338, doppelband fotostrip, ∼100 µm thick) as a spacer between the glass and the filter. To construct this chamber, we first punched a hole (6 mm) in the tape, and then adhered the tape to the glass. The discs were transferred in ∼10 µl media to the center of the hole in the tape spacer. Care was taken to keep the tape dry. Discs were carefully arranged to be apical-side down (towards the coverslip) using forceps. Most of the media was then removed, and the filter (cut approximately to the size of the tape) was quickly placed over the sample and firmly adhered to the tape with forceps. The dish was then filled with 2-3 ml media. The height of these chambers is greater than the height of the discs, but the presence of the filter isolates the discs from flows and somewhat constrains disc movements so that they usually remain in the field of view. As an alternative, we also tried to immobilize discs using methylcellulose (Aldaz et al., 2010) dissolved in our culture medium, but discs are very hard to position in this media and often do not proliferate.

During imaging, we used a syringe pump (PHD ULTRA, Harvard Apparatus) to exchange the media automatically and continuously during the course of imaging at a rate of 0.03 ml/min. The start of the movie was considered to be time 0 h. This time corresponds to 45-60 min from the start of dissection (time required for sample preparation and microscope setup).

Imaging was performed using a Zeiss spinning-disc microscope consisting of an AxioObserver inverted stand, motorized xyz stage with temperature control set to 25°C, Yokogawa CSU-X1 scanhead, and a Zeiss AxioCam MRm camera (2×2 binning), all controlled by Axiovision software. Discs were imaged through a Zeiss C-Apochromat 63×1.2NA water immersion objective heated to 25°C with an objective heater. To capture the whole pouch, we acquired a 2×2 tiled region (10% overlap). Each region consisted of a z-stack of 65-85 frames, spaced 0.5 µm apart. Tiled z-stacks were captured every 5 min. We kept light exposure as low as possible to achieve a segmentable image. We used a power meter (PT9610, Gigahertz-optik, Munich, Germany) to measure the power of the laser through a 10×/0.45 NA objective within a week of the experiment. For imaging, we used a laser power of 0.04-0.05 mW and an exposure time of 350 ms per image.

Raw data were first processed with a low-pass frequency filter to remove high-frequency noise, followed by a rolling-ball background-subtraction algorithm using Fiji (Schindelin et al., 2012). The apical plane was projected onto 2D using a custom-made algorithm. We identified the two manifolds formed by the disc proper layer and the peripodial layer, respectively, in the 3D z-stacks, based on maximal brightness of the E-Cadherin-GFP signal and subject to hard constraints on the distance between the two layers and the slope of each manifold. We employed the algorithm of Wu and Chen (2002) to determine simultaneously the two manifolds that are optimal, i.e. as bright as possible under the given constraints. This algorithm and its application to the wing disc will be described in detail in a separate upcoming manuscript (details available upon request).

Projected tiles were stitched using the Grid/Collection Stitching Fiji plugin (Preibisch et al., 2009). Segmentation and cell tracking was performed on the projected images of the time lapse using Tissue Analyzer (Aigouy et al., 2016). We oriented the tissue to have the DV boundary as the horizontal x-axis using TissueMiner (Etournay et al., 2016).

The analysis of tissue shear and its cellular contributions was performed on the region of the central wing pouch that was trackable throughout the entire course of our movies. The cells belonging to this region in the first frame of the time-lapse movie are visualized in Fig. S10. We defined this region using Tissue Miner (Etournay et al., 2016) by first manually marking the region between the two innermost folds of the pouch in the last frame of the time-lapse movie, and then backtracking these cells and their lineages to the first time point, discarding cells that moved in/out of the field of view.

Analysis of cellular contributions to changes in tissue size and shape were calculated using the Triangle Method (Merkel et al., 2017) and Tissue Miner (Etournay et al., 2016). Further details are provided in the supplementary Materials and Methods.

For the plot shown in Fig. 8, we dissected wing discs from E-cadherin-GFP-expressing larvae at 96 h, 119 h and 135 h AEL. Note that the E-cadherin-GFP flies are slightly developmentally delayed, taking ∼135 h to pupariate, compared with ∼120 h for OregonR. Discs were mounted as for live imaging, in 20E-containing culture media, and imaged within 1 h of dissection. z-stacks were acquired, processed and projected, as described for live imaging. The cells belonging to the central pouch region (defined as the region between the two innermost folds) were manually identified and marked using TissueMiner (Etournay et al., 2016). This region should correspond roughly to the same regions analyzed in the movies presented in Fig. 7. Tissues were oriented so that the DV boundary would be the horizontal x-axis, as for live imaging. In Fig. 8, we plot for each disc the area-weighted average of triangle elongation in the identified region of analysis. The box plots in Fig. 8 summarize the data for each age group: hinges correspond to the first and third quartiles; lines indicate the highest and lowest values within 1.5*IQR (inter-quartile range).

RNA sequencing was performed by Annekathrin Kränkel and Andreas Dahl in the Deep Sequencing Group SFB655 at the Biotechnology Center of the Technische Universitaet Dresden. The processing of the RNAseq data was performed with considerable help from Holger Brandl of the Bioinformatics facility of the MPI-CBG. We thank Marko Brankatschk, Christian Dahmann, and Savraj Grewal for critical review of the manuscript prior to submission.