Orchestrating morphogenesis: building the body plan by cell shape changes and movements

During morphogenesis, the body is assembled and remodeled by cell shape changes and movements (Gillard and Röper, 2020). These events are powered by forces generated by the contractile actomyosin cytoskeleton, which are converted to shape changes and movements by linking the cytoskeleton to the plasma membrane at cell-cell and cell-matrix junctions. The linkage of cell junctions to the cytoskeleton is key for morphogenesis in many ways. First, the linkage is essential for stable cell adhesion itself as, in its absence, adhesion proteins like cadherins are not maintained at the plasma membrane. Second, this linkage is crucial for establishing and maintaining apical-basal polarity, a key feature of epithelia. Finally, all the stereotypical cell shape changes and movements of epithelia during morphogenesis – from apical constriction to convergent elongation to tissue spreading via collective cell migration – depend on this linkage.

In the past decade, a paradigm shift has occurred in our view of how cell-cell adhesion is coupled to the cytoskeleton. Shaped by experiments in both cultured cells and intact animals, the simple static textbook model in which an apical belt of cadherins directly links to an underlying actin ring via β- and α-catenin (Fig. 1A) was supplanted by a much more complex and dynamic view (Fig. 1B; reviewed in Charras and Yap, 2018; Pinheiro and Bellaïche, 2018). First, it has become clear that, in many situations, the entire apical surface of a cell is underlain by a dynamically contractile cortical actomyosin network, linked to adherens junctions around the apical circumference (Martin and Goldstein, 2014). Second, our understanding of the role of α-catenin has evolved. While α-catenin can bind actin (Rimm et al., 1995), strong interaction is a feature of α-catenin homodimers, but not of α-catenin in complex with β-catenin and E-cadherin (Drees et al., 2005; Yamada et al., 2005). Recently, this picture was further modified, with experiments confirming that α-catenin helps mediate cadherin-actin linkage as a monomer (Desai et al., 2013) and revealing that α-catenin acts as a force sensor (Yonemura et al., 2010). Indeed, when subjected to pulling forces from the actomyosin cytoskeleton, α-catenin undergoes a conformational change shifting it into a high-affinity actin-binding state, thus exhibiting catch bond behavior (Fig. 1B; Buckley et al., 2014; Yonemura et al., 2010). This work also shifted focus from α-catenin as the sole linker, revealing roles for multiple junction-cytoskeletal linker proteins that provide distinct mechanical properties to junctions in different cell types. These include Vinculin, Canoe (the Drosophila homolog of afadin), ZO-1, Ajuba and many others (Fig. 1B; Mege and Ishiyama, 2017).

Now, our challenge is to define how cells use this diverse toolkit to accomplish the different dynamic events of morphogenesis (for more comprehensive reviews of the underlying mechanics, see Pinheiro and Bellaïche, 2018; Yap et al., 2018; Miao and Blankenship, 2020). Here, we analyze three key cell behaviors central to embryonic development: apical constriction, convergent extension and collective cell migration. We explore our understanding of these events in Drosophila, a powerful model for studying these events (Harris, 2018), briefly mentioning related insights from other model animals. The breadth of research in the area means our coverage is not comprehensive, so we provide references to more-comprehensive reviews in each area of focus.

One of the first morphogenetic events during embryonic development is gastrulation, a process in which the blastula, a ball of cells, undergoes major rearrangements to form the embryonic germ layers: ectoderm, endoderm and mesoderm. Fate-determining genes act early, giving cells different transcriptional programs and, thus, different fates, leading to unique cell behaviors at the mid-blastula transition. Although gastrulation mechanisms are not the same in all animals (Leptin, 2005), some cell shape changes are commonly used across the animal kingdom. One such cell shape change is apical constriction, a process in which constriction of the apical end of a cell leads it to take on a wedge shape and move inward. Apical constriction is used broadly throughout morphogenesis. It can be employed by individual cells, as in Drosophila neural stem cell invagination, or can involve small groups of cells, leading to invagination to form a pocket, as seen in bottle cells leading Xenopus gastrulation, or in C. elegans endoderm or Drosophila salivary gland precursors. Finally, apical constriction can be employed by programmed groups of cells along one body axis, as in the vertebrate neural tube or, as we discuss below, during Drosophila mesoderm invagination (Fig. 2).

The first step in Drosophila gastrulation is ventral furrow formation, which occurs in the first cell cycle after the maternal to zygotic transition (Farrell and O'Farrell, 2014). During this process, ∼1200 cells along the ventral midline (a group roughly 70 cells long and 18 cells wide) apically constrict in a coordinated way (Fig. 2), leading to invagination of a tube of cells that will form the embryonic mesoderm (Kam et al., 1991; Sweeton et al., 1991; Turner and Mahowald, 1977). Ventral furrow cells constrict asymmetrically, with cells exhibiting more constriction along the dorsal-ventral (DV) axis than along the anterior-posterior (AP) axis, where tension is highest. Ventral furrow cells go from a columnar to a trapezoidal shape, shifting their nuclei more basally, thereby forming a groove along the ventral midline (Gelbart et al., 2012; Sweeton et al., 1991). Mesoderm internalization is specified and regulated by the mesoderm-specific transcription factors Twist and Snail; disrupting either triggers invagination failure and loss of presumptive mesodermal cells (Leptin, 1991; Simpson, 1983). Twist and Snail are, in turn, regulated by high nuclear levels of the transcription factor Dorsal (Kosman et al., 1991; Stathopoulos and Levine, 2002) in response to maternal Toll signaling.

Effective mesodermal cell internalization requires intricate regulation of actin and non-muscle myosin II (referred to as myosin hereafter), as they provide the force that drives apical constriction. Full internalization to form the ventral furrow also requires cell volume conservation, shortening the apical-basal axis, basal expansion and basal myosin downregulation (Dawes-Hoang et al., 2005; Krueger et al., 2018). Once presumptive mesodermal cells internalize, the mesoderm is covered by lateral ectodermal cells flanking the furrow. Only then does this tube of cells collapse, undergoing a transient epithelial-to-mesenchymal transition (EMT) by modifying adherens junctions, switching expression from E-cadherin to N-cadherin. Mesoderm collapse and monolayer formation, but not basal spreading, require fibroblast growth factor (McMahon et al., 2010).

In the next step of gastrulation, the posterior midgut invaginates in a process very similar to that of the ventral furrow – these invaginating cells become endoderm. Posterior midgut cells apically constrict using a mechanism similar to that employed by ventral furrow cells, but instead of Twist and Snail regulating events, the transcription factors Huckebein and Tailless are key regulators (Harbecke and Lengyel, 1995; Murakami et al., 1999; Strecker et al., 1986; Weigel et al., 1990).

The mesoderm-specific transcription factors Twist and Snail initiate a cascade of events driving apical constriction (Fig. 2A). At the site of invagination, Twist activates expression of the secreted ligand Folded gastrulation (Fog), which in turn triggers coordinated apical constriction of presumptive mesodermal cells (Costa et al., 1994; Oda and Tsukita, 2001). Mesodermal cells also express two G-protein coupled receptors (GPCRs): mesoderm-invagination signal transducer (Mist), the expression of which is activated by Snail, and Smog. In the absence of Fog, these receptors are inactive. Fog binding leads to activation of Concertina, a Gα_12/13 class G-protein α subunit. Activated Concertina then binds Rho guanine nucleotide exchange factor RhoGEF2, which in turn activates the small GTPase Rho1 (Barrett et al., 1997; Dawes-Hoang et al., 2005; Häcker and Perrimon, 1998; Jha et al., 2018; Kerridge et al., 2016; Manning et al., 2013; Parks and Wieschaus, 1991). Ric8, Gβ13F and Gγ1 also play roles (Hampoelz et al., 2005; Kanesaki et al., 2013; Peters and Rogers, 2013). Fog signaling can be downregulated via phosphorylation of the GPCR by the GPCR kinase Gprk2, allowing the β-arrestin Kurtz to bind the GPCR and direct its removal via endocytosis (Chai et al., 2019; Fuse et al., 2013; Jha et al., 2018). RhoGEF2 apical accumulation is also promoted by T48, a transmembrane protein regulated by Twist. Rho1 activation triggers downstream cytoskeletal modulation: activation of Rho kinase (Rok) leads to myosin activation and apical constriction (Kerridge et al., 2016; Kölsch et al., 2007). Intriguingly, a strikingly similar pathway modulates apical constriction of both the posterior midgut and the salivary gland later in embryonic development; in both cases, Fog activation via different transcriptional circuits leads to medioapical accumulation of Rok and myosin (Bailles et al., 2019; Chung et al., 2017). In the posterior midgut, a novel mechanical positive-feedback pathway then leads to a wave of contractility (Bailles et al., 2019).

Apical constriction was initially thought to be driven by a purse string-like contraction, in which an F-actin/myosin ring underlying the adherens junctions at the apical end of the lateral cell interface constricted to reduce the apical area of the cell. However, the purse string model suggested that cells constrict via myosin localized at adherens junctions, but this proved to be incorrect. In twist and snail mutants, myosin becomes concentrated at adherens junctions, but apical constriction fails (Martin et al., 2009). Moreover, when researchers examined myosin in wild-type embryos, they found that actin and myosin are not restricted to adherens junctions but instead are found in a network covering the apical cortex (Fig. 2A,E; Martin et al., 2009). This was the first clue that apically constricting mesodermal cells exhibit ‘radial polarity’, with both medioapical and junctional pools of actin and myosin.

Live imaging also revealed that constriction is not continuous; instead, pulsatile constriction of ventral furrow cells is observed, with repeated phases of constriction and relaxation. Stabilizing some of the apical area lost at each cycle drives apical constriction (Martin et al., 2009). A similar pulsatile actomyosin network helps drive Drosophila germband extension and dorsal closure (discussed below). Subsequent computational analysis revealed that presumptive mesodermal cells gradually transition from non-constricting to ‘un-ratcheted’ pulsed constriction, in which area loss is not captured, to ratcheted pulses. This final persistent ratcheted pulse state results in collective constriction (Xie and Martin, 2015). Twist is required for the un-ratcheted to ratcheted pulse transition and for establishing persistent ratcheted pulses; it is hypothesized to act through the Rho pathway by enhancing apical Rok localization (Xie and Martin, 2015).

In ventral furrow cells, myosin, actin, Rok and adherens junction proteins all display radial cell polarity, with both medioapical and junctional pools (Mason et al., 2013); this is distinct from the planar cell polarity these cells exhibit later in gastrulation. Apical constriction is driven by assembly and disassembly of actomyosin coalescences over the medioapical cell surface. These are regulated by the opposing actions of Rok and myosin phosphatase, which control actomyosin coalescence and pulses via the phosphorylation/activity state of myosin (Dawes-Hoang et al., 2005; Mason et al., 2013; Vasquez et al., 2016, 2014). Myosin apical recruitment requires a balance between active, GTP-bound Rho and inactive, GDP-bound Rho. This cycling is regulated by guanine nucleotide exchange factors (GEFs), which activate Rho, and GTPase activating proteins (GAPs), which inactivate Rho. RhoGEF2 and the Rho GAP Cumberland work in opposition to regulate Rho activity, and the GEF/GAP ratio is important for the ratchet behavior of apical constriction and its spatiotemporal restriction to the apical cortex (Mason et al., 2016). Collective invagination of mesoderm cells also requires attaching the medioapical cytoskeletal network to adherens junctions at discrete sites (discussed in detail below), to pull the plasma membrane inward with every pulse (Martin et al., 2009). Radially polarized filamentous actin (F-actin), with F-actin minus ends positioned medioapically and plus ends enriched where filaments connect to adherens junctions, is essential, allowing the force transmission that drives cell shape changes (Coravos and Martin, 2016). Both apical and basal actin organization in the presumptive mesoderm are regulated in part by RhoGEF2, and by the nonreceptor tyrosine kinase Abelson, which negatively regulates the actin polymerase Enabled (Fox and Peifer, 2007; Jodoin and Martin, 2016). Finally, recent evidence suggests that as mesodermal cells initiate EMT, they assemble apical-to-basal myosin cables, with the force they exert also playing a role in tissue invagination (Gracia et al., 2019).

For mesoderm invagination to occur, the cytoskeleton must be coupled to adherens junctions. As noted above, in the classic view, the cytoplasmic tail of the transmembrane protein E-cadherin binds β-catenin, which in turn binds α-catenin, which directly binds F-actin. Over time, however, this simplified model has evolved into a complex network of mechanosensitive proteins.

Drosophila mesoderm invagination occurs immediately after cellularization, during which cells form and polarize. The localization of E-cadherin and Armadillo (Drosophila β-catenin) is dynamic during this period, with adherens junctions moving from a subapical position during cellularization to a tight apical localization (Harris and Peifer, 2004; Dawes-Hoang et al., 2005; Kölsch et al., 2007; Weng and Wieschaus, 2016). This apical transition is regulated by Traf4, a Twist target, and by myosin contractility pulses (Mathew et al., 2011; Weng and Wieschaus, 2016). Adherens junction proteins themselves are, not surprisingly, essential for effective apical constriction. Inactivating them, by Armadillo knockdown, disrupts the tethering of the medioapical actomyosin network to the cell membrane, interrupting cell shape changes; similar effects are seen when F-actin assembly is reduced by reducing expression of the formin Diaphanous (Dawes-Hoang et al., 2005; Homem and Peifer, 2008; Martin et al., 2010; Mason et al., 2013).

The linkage between adherens junctions and actin is complex. Canoe works in parallel or together with core adherens junction proteins to link them to the cytoskeleton. In the absence of Canoe, ventral furrow cells initiate constriction but cell shape change arrests prematurely, as the medioapical actomyosin condensation detaches from adherens junctions (Fig. 2A; Sawyer et al., 2009). α-Catenin remains at circumferential adherens junctions, but embryos lacking Canoe also have membrane extensions connected to the actomyosin balls, the tips of which are enriched in α-catenin (Sawyer et al., 2011, 2009). Similar membrane extensions are seen following Armadillo knockdown (Martin et al., 2010). This suggests that parallel mechanisms attach junctions to the cytoskeleton, with some connections retained when individual mechanisms are knocked out or knocked down. Alternatively, reducing Canoe or Armadillo may disrupt lateral junction ‘cross linking’, leading to membrane extension where point contacts of adherens junctions with the cytoskeleton are maintained.

Although most analyses have focused on actin and myosin, it should be noted that microtubules also play a role in mesoderm apical constriction. Disrupting microtubules does not affect adherens junction assembly or myosin contractility, but does disrupt stable adherens junction attachment to the medioapical actomyosin network, perhaps by blocking the ability to repair disrupted attachments (Ko et al., 2019). A parallel role for microtubules in apical constriction is seen during salivary gland apical constriction, although there the microtubules stabilize the medioapical myosin network (Booth et al., 2014).

Harmony among a complex group of cells is required to produce elegant and effective collective cell movement. At the tissue level, gradients of genes and proteins form across the embryo to allow effective and coordinated ventral furrow formation (Heer et al., 2017; Lim et al., 2017). Both Twist and Snail are expressed in all presumptive mesodermal cells, but their transcriptional targets exhibit fine-scale expression gradients. For example, spatiotemporal visualization of T48 and Fog transcripts revealed a gradient along the DV axis that depends on the timing of transcriptional activation (Lim et al., 2017). These gradients reflect and may dictate gradients of myosin activity. Twist also activates expression of the E3-ubiquitin ligase Neuralized in the presumptive mesoderm, while Snail represses expression of Bearded, a Neuralized inhibitor, there. The resultant restriction of Bearded to the ectoderm is crucial for preventing medioapical myosin activation there, ensuring that only mesoderm cells undergo apical constriction (Perez-Mockus et al., 2017). The target of Neuralized in the mesoderm remains to be identified.

Force propagation between cells also plays a role in coordinated invagination. An early example came from the observation that mesodermal cells do not apically constrict ‘isotropically’. Instead, they constrict more rapidly along the DV versus the AP axis (Fig. 2C). This anisotropic apical constriction is triggered by higher tension along the AP axis, resulting at least in part from the shape of the embryo and thus that of the mesoderm primordium, which contains more cells along the AP than the DV axis (Martin et al., 2010). These data fit well with classic transplant experiments in which patches of cells with mesodermal fate were generated: round patches of cells exhibited isotropic constriction, while elongated patches constricted anisotropically (Leptin and Roth, 1994). A more comprehensive modeling approach verified that an asymmetric constricting domain in an elastic tissue exhibits anisotropic constriction (Doubrovinski et al., 2018; Spahn and Reuter, 2013).

Mechanical forces within cells propagate across the tissue via adherens junctions, which also stabilize shape change. Connections between cells are robust, with redundant paths of cytoskeletal connectedness allowing adjustment to failed connections at individual junctions (Yevick et al., 2019). Cells outside the furrow also contribute. Lateral and dorsal cells each have distinctive fates, cytoskeletal networks and, thus, mechanical properties. Lateral cells are stiff, and dorsal cells are soft and stretchy, properties that are important to allow mesodermal cells to constrict and invaginate in a coordinated manner (Rauzi et al., 2015). Within the mesoderm, things are even more complex. The mesoderm is stiffer along the AP axis and more flexible along the DV axis, and this helps to determine tissue shape and orientation (Yevick et al., 2019). Finally, cell mitotic entry also regulates cell and tissue shape (Grosshans and Wieschaus, 2000; Mata et al., 2000; Seher and Leptin, 2000). After Drosophila embryos complete cellularization, groups of cells undergo mitotic entry in a spatially and temporally patterned manner, forming mitotic domains (Foe, 1989). Normally, mesodermal cells do not initiate mitosis until after invagination, but prematurely inducing mitosis in these cells interferes with medioapical Rho signaling, which in turn affects myosin and reverses apical constriction (Ko et al., 2020).

Some features of the mesoderm apical constriction program are conserved in diverse apical constriction events in other animals. In C. elegans, gastrulation begins quite simply with apical constriction and invagination of two E cells, the endoderm progenitors. This also involves constriction of an apical network of actin and myosin, with some evidence of a transition during which actomyosin contractility becomes coupled to cell shape change (Roh-Johnson et al., 2012). Both Rok and the myosin light-chain kinase MRCK-1 play roles, activating apical myosin and, via the tension generated, reinforcing adherens junctions (Marston et al., 2016). An intriguing genetic screen suggested that many proteins involved in vertebrate neural tube closure also play roles in C. elegans gastrulation, including the actin-regulatory WAVE complex (Sullivan-Brown et al., 2016). The cortex of the one-cell C. elegans zygote also exhibits pulsatile constriction, and has proven a fruitful place to explore feedback mechanisms powering pulsation (Michaux et al., 2018; Nishikawa et al., 2017). Recent work suggests pulsatile contraction of an apical actomyosin network is also seen in vertebrate development, both during compaction of early mouse embryos (Maître et al., 2015) and during Xenopus neural tube closure (Christodoulou and Skourides, 2015). Zebrafish neural tube closure also involves myosin-driven apical constriction of cells connected via N-cadherin (Araya et al., 2019) and, as in Drosophila mesoderm invagination, Abl kinase paralogs play roles in neural tube closure (Koleske et al., 1998). As each system is examined in more detail, it will be interesting to explore mechanistic similarities and differences between apical constriction events in different contexts.

A second morphogenetic event found in diverse animals is convergent extension: a process in which tissues narrow along one body axis and elongate along another. In many animals, this plays a crucial role in turning a ball of cells into an elongated, bilaterally symmetric embryo (Keller and Sutherland, 2020; Shindo, 2018; Sutherland et al., 2019). It can involve epithelial cells, as is the case in Drosophila germband extension, or cells with more mesenchymal character, as in frog or zebrafish embryos, and can occur simultaneously or sequentially in different tissues. Here, we focus on one of the best understood examples: Drosophila germband extension.

Germband extension initiates within minutes of the onset of zygotic transcription, closely following the onset of mesoderm invagination [see Pare and Zallen (2020), for a more comprehensive review]. During this process, the thoracic and abdominal body segments elongate almost twofold along the future AP axis, while narrowing along the DV axis (Fig. 3A). The constraints of the eggshell mean that the posterior end of the embryo moves dorsally, giving the embryo a ‘sideways U’ shape. Pioneering work revealed that this event is driven by acquisition of differential cell fates along the AP axis (Irvine and Wieschaus, 1994; Zallen and Wieschaus, 2004). Further analyses revealed a key role for planar polarity in driving cell intercalation along the DV axis. As germband extension initiates, cells polarize their cell junctions and actomyosin cytoskeletons, setting up reciprocal membrane domains. Myosin and actin become more enriched along cell borders oriented vertically with respect to the AP body axis (referred to here as AP cell borders) while adherens junction proteins, especially Bazooka/Par3, are more enriched along cell borders oriented horizontally with respect to the AP body axis (referred to here as DV borders; Fig. 3B-D). Reducing either Bazooka or myosin slows germband extension (Bertet et al., 2004; Zallen and Wieschaus, 2004). Contractile myosin cables along the AP borders (Fig. 3C), at times spanning multiple cells, shrink AP cell junctions, creating four-cell or multi-cell rosettes (Fig. 3B; Bertet et al., 2004; Blankenship et al., 2006), and rosette resolution completes the process of neighbor exchange. Other events also contribute to axis elongation, including basolateral protrusions (Sun et al., 2017), oriented cell division (da Silva and Vincent, 2007; Wang et al., 2017), DV cell relaxation or AP cell elongation (Butler et al., 2009; Sawyer et al., 2011) and pulling forces from midgut invagination (Collinet et al., 2015; Lye et al., 2015).

Many labs have probed the molecular mechanisms by which planar polarity is established, force is generated and linked to adherens junctions, and adherens junctions are reinforced and remodeled to power these dramatic events. For example, laser ablation studies allowed measurement of cortical tension, confirming that AP borders are subject to more contractile force (Rauzi et al., 2008). The myosin regulator Rok is enriched at AP borders and is crucial for activating myosin there to generate tension (Simões et al., 2010). Rok also phosphorylates Bazooka, reducing its accumulation along AP borders and thus helping establish reciprocal cortical domains – a contractile AP border and a DV border where Bazooka reinforces cell adhesion (Simões et al., 2010). It is presumed that Rho regulates Rok, although the earlier roles of Rho make loss-of-function studies challenging. A novel Rho GEF, Dp114RhoGEF/Cysts, acts at cell junctions to activate Rho and stabilize junctional myosin (Garcia De Las Bayonas et al., 2019; Silver et al., 2019); this contrasts with apical constriction where RhoGEF2 is the relevant player. The Rok regulator Shroom, enriched at AP borders, reinforces Rok enrichment there, but Shroom is not essential for viability, suggesting other parallel regulators (Simões et al., 2014). Intriguingly, computational analysis revealed that multicellular myosin cable formation is more common than expected by chance, and is reinforced by positive feedback between myosin accumulation and tension (Fernandez-Gonzalez et al., 2009). Notably, cables often form at seemingly stereotyped positions within each body segment (Tetley et al., 2016).

As AP borders shrink, junctions must be remodeled and membrane removed. This involves targeted endocytosis of membrane and adherens junction proteins on AP borders, a process requiring dynamin, clathrin (Levayer et al., 2011) and the GTPase Rab35 (Jewett et al., 2017). Endocytosis is regulated by planar-polarized RhoGEF2 accumulation, and by activity of the Rho effector and the formin Diaphanous (Levayer et al., 2011).

As in mesoderm invagination, the situation is complicated by the presence of two distinct but connected ‘pools’ of actin and myosin. In addition to the planar-polarized myosin cables at AP cell borders (Fig. 3C), there also is a medioapical pool of myosin, which, like that in the mesoderm, undergoes pulsatile contraction (Fig. 3E; Fernandez-Gonzalez and Zallen, 2011; Rauzi et al., 2010; Sawyer et al., 2011). This pool appears to contribute to cell elongation along the AP axis (Sawyer et al., 2011). Apical myosin also moves by flow to junctional cables at AP cell borders (Levayer and Lecuit, 2013; Rauzi et al., 2010). Intriguingly, pulsatile constriction of the medioapical myosin does not appear to involve a regulated cycle of Rho activity. Instead, a self-regulating feedback loop of Rok activation, myosin recruitment, contraction and cortical disassembly drives the cycle (Munjal et al., 2015). Cells also have mechanisms restricting contractility to the apical membrane, preventing it from leading to epithelial folding (Urbano et al., 2018). Recent work points to a parallel role for a population of myosin and E-cadherin at cell vertices that also helps drive intercalation by vertex sliding (Vanderleest et al., 2018).

Myosin also is important for the final phase of cell rearrangement: the establishment of new cell borders to resolve rosettes (Kasza et al., 2014; Yu and Fernandez-Gonzalez, 2016). Border extension requires myosin activity and occurs in a pulsatile way, with cycles of border elongation and relaxation. Cells anterior and posterior to the rosette vertex appear to play the most important roles (Yu and Fernandez-Gonzalez, 2016), with medioapical myosin contractility one likely source of the force required (Collinet et al., 2015). Mathematical modeling provides a way to assess the plausibility of interpretations of experimental data and develop testable predictions (e.g. Lan et al., 2015; Siang et al., 2018).

One exciting frontier is determining how cells respond to force and defining roles for mechanical signaling. Within cells, both the cytoskeleton and adherens junctions are force responsive. Applied force recruits and stabilizes myosin at adherens junctions, providing a positive-feedback loop to generate cables (Fernandez-Gonzalez et al., 2009). Cells then must strengthen junctions in response. The LIM domain protein Ajuba is recruited to adherens junctions in response to applied tension, with particular enrichment at tricellular junctions (Rauskolb et al., 2019; Razzell et al., 2018). Ajuba loss slows rosette resolution during germband extension; cell adhesion is also compromised at rosette centers, which are likely to be the regions of highest tension (Razzell et al., 2018). However, most embryos complete this process and survive to hatch, suggesting multiple proteins act in parallel to reinforce connections between junctions and the cytoskeleton, allowing cells to resist the force exerted without disrupting tissue integrity. One candidate protein is the junctional-cytoskeletal linker Canoe. It is enriched at AP borders, where it is thought to reinforce connections. Canoe loss destabilizes myosin cable attachment to AP cell junctions and leads to gaps at the center of rosettes, thereby impairing germband elongation. Consistent with an antagonistic relationship between myosin and Bazooka/Par3, as myosin retracts from junctions on AP cell borders, Canoe loss elevates the planar polarization of Bazooka at the center of DV cell borders (Sawyer et al., 2011). Moving forward, we now need to define how the cytoskeleton ‘stores’ and ‘dissipates’ force, allowing transient forces to lead to irreversible shape changes. Recent studies revealed that this may involve actin turnover (Clément et al., 2017) or regulated endocytosis (Cavanaugh et al., 2020). New toolsets and approaches, like combining optical tweezers with light-sheet microscopy, will empower further analyses (Bambardekar et al., 2015).

Cell rearrangement during axis elongation is only one of several processes challenging tissue integrity during and just after germband elongation. Cells must also remodel their junctions as they round up for mitosis (Foe, 1989), and then reverse this to resume columnar architecture. It is known that ∼30% of ventral ectodermal cells apically constrict to invaginate as neural stem cells (neuroblasts), and their neighbors must respond, sealing the resulting gaps (Hartenstein and Wodarz, 2013). These dramatic events require cells to reinforce cell junctions in order to allow them to withstand force without tearing the tissue apart; as such, the ventral ectoderm is sensitive to perturbations to adherens junction proteins or their regulators (Harris and Tepass, 2008; Tepass et al., 1996). Canoe plays a role in this tissue remodeling. In its absence, cells are slow to resume columnar architecture after cell division, but most cells eventually complete the process, such that epidermal epithelial architecture remains intact except along the ventral midline (Manning et al., 2019). The ZO-1 homolog Polychaetoid works in parallel with Canoe. Intriguingly, it is enriched at DV cell borders, while Canoe is enriched at AP cell borders. Strikingly, canoe polychaetoid double mutants rapidly lose junctional integrity, as cells separate and adherens junctions fragment. Bazooka is lost from the cortex before core adherens junction proteins, suggesting it may be their primary target. Intriguingly, the weakest points are tri- and multicellular junctions (Manning et al., 2019), consistent with observations in cultured mammalian cells suggesting this is where force on adherens junctions is the highest (Choi et al., 2016). The junctional LIM-domain protein Smallish (the Drosophila homolog of LMO7) can bind both Bazooka and Canoe, and also acts in this process (Beati et al., 2018). Other proteins acting in parallel to reinforce connections at tri- and multicellular junctions include Ajuba (mentioned above) and the tricellular junction enriched protein Sidekick (Finegan et al., 2019; Letizia et al., 2019).

A final question concerns how cells sense directionality in the tissue to establish planar-polarized distributions of cytoskeletal and junctional proteins. Early studies implicated the transcriptional network establishing initial AP cell fates (Blankenship et al., 2006; Irvine and Wieschaus, 1994; Zallen and Wieschaus, 2004), but the actual effectors remained elusive for almost two decades. A breakthrough came in a search for transmembrane proteins with striped distributions along the AP axis. Strikingly, multiple members of the Toll receptor family, best known for their early roles in DV patterning and later roles in innate immunity, exhibit distinct striped distributions. They act in a combinatorial way to trigger cytoskeletal planar polarization at borders with different ‘Toll codes’ (Paré et al., 2014). However, disrupting the Toll code did not fully eliminate planar polarity, leading to a search for other players. This led to the discovery that the leucine-rich-repeat receptor Tartan and its interacting partner, the teneurin Ten-m, act at parasegmental borders (Paré et al., 2019), opening up new and exciting questions about how these transmembrane receptors direct cytoskeletal polarity. A recent preprint takes this work a step futher, suggesting Toll-8 polarizes Myosin through a physical interaction with the adhesion GPCR Cirl/Latrophilin (Lavalou et al., 2020 preprint). Intriguingly, Latrophilins bind teneurins, and they share roles in axon guidance (Del Toro et al., 2020).

Similar processes drive other morphogenetic events in Drosophila and other animals. For example, pulsatile junctional and medioapical myosin and cytoskeletal planar polarization drive Drosophila neuroblast invagination (Simões et al., 2017; An et al., 2017), while salivary gland invagination integrates radial planar polarity, cell intercalation and apical constriction (Chung et al., 2017; Röper, 2012; Sanchez-Corrales et al., 2018; Sidor et al., 2020). C. elegans also elongates its body axis, although this occurs later in embryonic development than in Drosophila. Because of this later timing, C. elegans body elongation initiates via events in the epidermis but then involves interplay between the epidermis and underlying muscles, with muscle contractions generating force, initiating cross-tissue mechanical signaling. Force transmission within and between tissues occurs both at adherens junctions and at integrin-based ‘molecular tendons’ or hemidesmosomes linking muscles and epidermis. At the latter, VAB-10/plectin may be the tension sensor (Suman et al., 2019). However, force transmission then acts, as in Drosophila, to planar polarize Par3, which in turn is required to orient actin filaments along the DV axis (Gillard et al., 2019). Bundled actin filaments then stabilize cell shapes (Lardennois et al., 2019). Epidermal cells also secrete an apical extracellular matrix, distinct from that on the basal surface, that transmits mechanical signals (Vuong-Brender et al., 2017). In another parallel with Drosophila, C. elegans ventral nerve cord elongation also involves cell intercalation, via formation and resolution of multi-cellular rosettes. In this context, the classical planar polarity pathway proteins VANG-1/Van Gogh and PRKL-1/Prickle act along with Slit/Robo signaling (Shah et al., 2017). Planar polarized junctional and cytoskeletal proteins may also play roles in vertebrate axis elongation (Butler and Wallingford, 2018; Shindo et al., 2019; Huebner et al., 2020 preprint) and mammalian limb bud elongation (Lau et al., 2015), while α-catenin acts as a mechanosensor in zebrafish convergent extension (Han et al., 2016).

A third common and crucial morphogenetic event re-shaping epithelial tissues is collective cell migration (Begnaud et al., 2016). Collective cell migration is seen in many contexts, with migratory epithelial cells varying widely in their organization. Many have complex morphologies and migrate through or past other tissues, such as the leading edge cells of sprouting vertebrate blood vessels or Drosophila tracheae, or zebrafish lateral line precursors, while others undergo partial EMT, such as Drosophila border cells. Below we focus on cells moving as a sheet to power tissue spreading. These events occur in a wide range of animals and tissues, from epiboly in early zebrafish development and ventral enclosure in C. elegans to palate closure in mammals. We focus on Drosophila dorsal closure, which differs from many examples because the cells move as a sheet that is pulled along, in part, by the neighboring tissue.

Dorsal closure in Drosophila is one of the final morphogenetic movements of embryogenesis (for more comprehensive reviews, see Hayes and Solon, 2017; Kiehart et al., 2017). It is not a tissue-autonomous event; instead, it requires close coordination of two tissues, with each affecting behavior of the other. At the onset of closure, the epidermis encloses the embryo's ventral and lateral surfaces, while the dorsal surface is occupied by a ‘temporary tissue’, the amnioserosa, which eventually undergoes apoptosis (Fig. 4A,B). During closure, the two lateral epidermal sheets stretch dorsally and meet at the dorsal midline, enclosing the embryo in skin. Both tissues play a role: amnioserosa cells undergo pulsed apical constriction of a medioapical actomyosin network (Fig. 4C), while lateral sheets of epidermis move dorsally, first meeting at the anterior and posterior canthi (i.e. the corners) and then zipping together to seal at the dorsal midline. At the epidermal leading edge, an intercellular contractile actomyosin cable links epidermal cells via their adherens junctions (Fig. 4A,B,D). Specialized tricellular junctions form where epidermal cells meet the amnioserosa (Fig. 4A,E). Laser ablation experiments suggest that the pulsatile apical constriction of amnioserosa cells and contractility of the leading-edge cable combine to drive proper closure, elongating lateral epidermal cells in the process (Kiehart et al., 2000). As the epidermal sheets meet, they build new bicellular adherens junctions and the cable is disassembled. This complex interplay between two tissues, with cells in each changing shape in different ways, provides a fascinating model to explore how cell junctions and the cytoskeleton work together.

Amnioserosa cells provide many valuable insights into how apical constriction can re-shape tissues. As noted above, the textbook picture of apical constriction has been substantially revised. A set of key papers in 2009-2010 began to re-shape this view, revealing that amnioserosa cell apical constriction involves a contractile actomyosin network across the entire apical surface of these squamous cells, and further revealing that this network exhibits pulsatile rather than continuous contraction, with periodic myosin assembly and disassembly (Blanchard et al., 2010; David et al., 2010; Solon et al., 2009). This leads to a ratcheted decrease in the apical cell area of individual cells, reducing the area of the amnioserosa and powering closure. Later in closure, pulsatile constriction transitions to sustained constriction (Blanchard et al., 2010; Solon et al., 2009), with cells near the leading edge experiencing accelerated constriction leading to their ingression (Sokolow et al., 2012).

Subsequent work revealed insights into how this pulsatile contractility is regulated. Not surprisingly, myosin is essential for cell shape oscillations and contractility (Franke et al., 2005). Intriguingly, increasing myosin activity elevates the amplitude of cell shape oscillations and increases the time cells spend in the constricted phase of the oscillatory cycle (Duque and Gorfinkiel, 2016). RhoGEF2 may play a role in this (Azevedo et al., 2011). Apical polarity proteins of the ‘Par complex’ – Bazooka/Par-3, aPKC and Par6 – also play a role, forming an apical patch in amnioserosal cells. However, Bazooka and aPKC/Par6 act differentially, with Bazooka promoting the duration of each pulse and aPKC and Par6 promoting the lag time before the next pulse (David et al., 2010). Combining experimentation and modeling, it was suggested that Par proteins shift from mainly junctional to mainly apical over a long time scale, somehow coupled with the short time scale pulsed actomyosin contractions. Feedback loops between myosin and the Par proteins contribute to ratcheted contraction (Durney et al., 2018). In this model, Bazooka antagonizes aPKC to limit aPKC downregulation of myosin (David et al., 2013). Meanwhile, another apical polarity protein, Crumbs, negatively regulates actomyosin contractility via a pathway thought to involve Moesin, Rho, Pak kinases and the Arp2/3 complex (Flores-Benitez and Knust, 2015). Other proteins, more typically viewed as adherens junction proteins, also regulate medioapical myosin contractility: α-catenin and vinculin both accumulate at the apical membrane in a tension-dependent fashion, with α-catenin stabilizing apical actomyosin foci (Jurado et al., 2016). Finally, the formin Diaphanous may play a role (Homem and Peifer, 2008).

Amnioserosa cells also provide a superb model of how adherens junctions respond to force. As these cells undergo pulsatile constriction, cell junctions must be reinforced while simultaneously reducing in circumference as the cells apically constrict. α-Catenin stabilizes junctional E-cadherin, linking medioapical actomyosin contractility and junction stability (Jurado et al., 2016). The force exerted on adherens junctions provides a mechanical cue – actomyosin relocalizes to adherens junctions from the medioapical cortex when junctions are stretched (Sumi et al., 2018). This stimulates constriction, which ruffles junctions and increases E-cadherin density, and in response, the endocytosis of adherens junctions material increases, re-straightening junctions. Recent work in cultured mammalian cells has provided intriguing mechanistic links by which RhoA regulates mechanosensitive endocytosis (Cavanaugh et al., 2020). However, not all the action is at the apical end of the cell: integrin-containing structures resembling focal contacts are found at the amnioserosa cell basal surface and, strikingly, these are important for dorsal closure, providing mechanical resistance to the force exerted apically (Goodwin et al., 2016, 2017). Thus, mechanical crosstalk between these two force-anchoring systems is crucial for proper closure.

The other active players in dorsal closure are epidermal leading edge cells. Although it was initially thought that epidermal cells ‘crawl’ over an amnioserosa cell substrate, it was later revelaed that, in fact, they maintain connections to peripheral amnioserosa cells throughout, assembling specialized cell junctions between the two tissues in which both integrins and E-cadherin play a part (Narasimha and Brown, 2004; Wada et al., 2007). As closure begins, leading edge cells – which initially have a square apical area – gradually elongate along the DV axis and become planar polarized. The most prominent feature of this planar polarity is assembly of a supracellular contractile actomyosin cable along the border with the amnioserosa (Young et al., 1993), which laser-ablation studies revealed is under tension generated by myosin-driven contractility (Kiehart et al., 2000). The cable is linked cell-to-cell by specialized tricellular adherens junctions (Gorfinkiel and Arias, 2007; Manning et al., 2019). Cell-cell signaling via the Jnk, Dpp and Wnt pathways specifies and regulates leading edge cell planar polarization (Rios-Barrera and Riesgo-Escovar, 2013), as does AP patterning (Rousset et al., 2017). Another key cue is provided by the nectin relative Echinoid. As in other tissues, an actin cable assembles at the border between cells expressing Echinoid (in this case the epidermis) and those in which Echinoid is absent (in this case the amnioserosa) (Laplante and Nilson, 2011; Lin et al., 2017, 2007).

Balanced contractility along the leading edge cable requires feedback regulation of myosin activity, which involves both Rho and Rok; Rac also plays a role (Harden et al., 1999; Magie et al., 1999). Recent work revealed intriguing roles for both the ELMO-MBC complex, a Rho/Rac GEF, and the RhoGAP Rhogap19d in actin cable assembly and re-establishment of adherens junctions between epidermal cells as they zip together at the canthi (Toret et al., 2018). Even more surprising was the role identified for the cytohesin Steppke, a GEF for Arf family small GTPases (West et al., 2017; Zheng et al., 2019). Steppke acts in the lateral epidermis behind the leading edge, restraining myosin contractility at lateral cell adherens junctions as part of a negative-feedback loop, thus allowing tissue relaxation as contractile force from the amnioserosa and leading edge cable stretch the lateral epidermis. The mechanisms by which the Arf-GEF activity of Steppke regulates myosin remain unclear. Myosin phosphatase, which removes activating phosphates on myosin, also plays a negative regulatory role. In mutants lacking its regulatory myosin-binding subunit, active myosin accumulates along the lateral edges of leading edge cells rather than being restricted to their leading edge, disrupting dorsal closure (Mizuno et al., 2002). Together, these analyses are beginning to reveal a complex network of feedback regulatory mechanisms at play.

The leading edge also provides a model of how cell junctions adapt to contractile force. Wild-type leading edge cells each pull on their neighbors, resulting in a homeostatic system with relatively uniform cell width. Reducing cadherin-catenin complex function disrupts dorsal closure (Gorfinkiel and Arias, 2007), but given their ubiquitous roles in cell adhesion, it is difficult to sort out primary and secondary effects of cadherin and catenin disruptions. However, other junction-actin linker proteins, although not essential for cell adhesion, strengthen linkage in cells under tension, and thus their loss affects dorsal closure. These include three proteins that bind both actin and junctional proteins: Canoe, Polychaetoid and Girdin. Girdin is a junctional protein that interacts with Par3 and strengthens adherens junction:actin interactions. girdin mutants have defects in cell shape changes, and, most strikingly, in assembly of the leading edge actin cable (Houssin et al., 2015). In embryos with Canoe reduction or Polychaetoid loss, leading edge cells still assemble an actin cable, but the balance of contractility among leading edge cells is disrupted, with some cells hyper-constricted and others hyper-elongated (Choi et al., 2011; Manning et al., 2019). This may reflect loss of cadherin-actin linkage at leading edge tricellular junctions, and/or more complex effects on sensing and thus regulating actomyosin-based tension. In dorsal closure, as in other events, Canoe is an effector of the small GTPase Rap1 (Boettner et al., 2003). Another protein whose loss leads to similar defects in balanced contractility is the actin polymerase Enabled, which binds to the plus-ends of actin filaments and stimulates elongation (reviewed by Edwards et al., 2014). Enabled is enriched at leading edge tricellular junctions (Gates et al., 2007), positioned at the interface between adherens junction proteins and actin (Manning et al., 2019), suggesting actin plus ends are enriched at tricellular junctions. Enabled loss disrupts even contractility along the cable, leading to hyper-constricted and hyper-elongated cells (Gates et al., 2007). Canoe and Polychaetoid are required for proper Enabled localization during dorsal closure, and strong genetic interactions suggest that Canoe, Polychaetoid and Enabled act together in regulating or anchoring the actin cytoskeleton during dorsal closure (Choi et al., 2011; Manning et al., 2019). These data are consistent with work in cultured mammalian cells, suggesting that actomyosin-mediated force on adherens junctions is highest at tricellular junctions (Choi et al., 2016). Recently, the LIM-domain protein Ajuba entered the mix: like Enabled, it is enriched at leading edge tricellular junctions, and in embryos lacking Ajuba some leading edge cells are hyper-constricted and others hyper-elongated. In more ventral epidermal cells, Ajuba colocalizes with Steppke and may work with it to restrain myosin contractility in these cells (Rauskolb et al., 2019).

The final step in dorsal closure is zippering together the two lateral epidermal sheets, which initiates at the canthi at the anterior and posterior ends. Both leading edge and amnioserosa cells produce filopodial and lamellipodial protrusions that aid zippering (e.g. Millard and Martin, 2008; Nowotarski et al., 2014). In fact, the leading edge can be viewed as having undergone a partial EMT, with protrusive activity unleashed despite the cells remaining adhered to their amnioserosa neighbors. Bazooka and the lateral polarity proteins Scribble and Dlg are lost from the leading edge and, in the case of Bazooka, are lost from leading edge tricellular junctions (Bahri et al., 2010; Laplante and Nilson, 2011; Pickering et al., 2013). Intriguingly, Bazooka instead localizes to leading edge cell AP borders and is required to recruit the lipid phosphatase Pten to these sites. The lipid PIP3 then becomes enriched at the leading edge; accordingly, downregulating PIP3 delays zippering and reduces filopodial activity (Pickering et al., 2013). As cells meet at the canthi, their partial EMT must be reversed, with filopodial activity dampened and cell junctions re-established. The ELMO-MBC complex and Rhogap19d play roles in this transition, via effects on Rac and Rho (Toret et al., 2018). Tension at the leading edge must continue to be modulated even as the two epidermal sheets meet, and Bazooka plays a complex role in this. Its recruitment to new junctions is stimulated by tension and, once localized there, Baz stabilizes adherens junctions and negatively regulates myosin contractility (Das Gupta and Narasimha, 2019). The kinase Pak is required to relocalize Scribble and Dlg to re-forming cell junctions, and to re-establish septate junctions as the sheets meet (Bahri et al., 2010), perhaps explaining the role of many septate junction proteins in the last steps in dorsal closure (Hall and Ward, 2016). Pak also regulates adherens junction re-establishment.

Recent work challenged our view of dorsal closure at the tissue coordination level. Initial studies of dorsal closure suggested amnioserosa cell apical constriction acts together with leading edge actin cable contractility to power closure (Hutson et al., 2003; Kiehart et al., 2000). However, two recent studies have challenged this view. Using different experimental approaches to genetically disrupt leading edge cable contractility, these studies revealed that amnioserosa cell apical constriction is both necessary and sufficient to drive closure (Ducuing and Vincent, 2016; Pasakarnis et al., 2016). However, in the absence of the leading edge cable, closure proceeds abnormally, with a highly irregular leading edge and very abnormal leading edge cell shapes, and cell matching at the midline is drastically impaired. There also has been a challenge to the idea that amnioserosa cell apical constriction is the sole force acting in that tissue. While apoptosis was initially viewed as a mechanism to remove amnioserosa cells after closure was complete, a subset of amnioserosa cells initiate apoptosis earlier, driving them to apically constrict before their neighbors (Toyama et al., 2008). An additional role for apoptosis recently emerged, revealing caspase activation triggers a substantial reduction in cell volume, thereby hastening closure (Saias et al., 2015). Further analyses will help reveal the full role of apoptosis in dorsal closure.

Analysis of similar events in other animals echoes many of these themes. For example, C. elegans ventral enclosure is topologically very similar, with two sheets of epidermal cells elongating to meet at the ventral midline. As in dorsal closure, α-catenin plays a complex role during this event, with its regulated actin binding contributing to both ventral enclosure and body axis elongation (Shao et al., 2017; 2019; Vuong-Brender et al., 2018). Rho regulation is also important for C. elegans ventral enclosure, with the non-canonical RhoGAP HUM-7/Myo9 modulating Rho activity (Wallace et al., 2018). One apparent difference with dorsal closure is the prominent role of basolateral protrusions in this epithelial cell type, mediated by a VAB-1/Eph to cdc42 pathway (Walck-Shannon et al., 2016, 2015). Extending comparisons with other events involving tissue stretching will provide further insights into similarities and differences.

It is an exciting time for this field of research! Long-standing questions about how cells change shape and move are yielding new and often surprising answers. Each tissue and cell type is different, but some common themes are emerging. Textbook pictures of cell junctions and their connections to the actomyosin cytoskeleton have been re-written. Specifically, it is now known that arrays of actin and myosin can be junctional or apical, isotropic or planar polarized, and connections between cell junctions and the cytoskeleton involve a complex network of interacting proteins that allow mechanosensing and provide robustness. Feedback regulation provides homeostasis, regulating contractility and strengthening cell-junctions under load. Moving forward, it will be exciting to see how new genetic, cell biological, quantitative imaging and computational modeling tools are combined to provide answers to the emerging new questions. It will also be exciting to see what similarities and differences emerge as we explore cell shape changes and movements in new tissues and across the spectrum of animal diversity.

We apologize to the many colleagues whose work we could not cover owing to length constraints. We thank Rodrigo Fernandez-Gonzalez, Tony Harris, and the editor and reviewers for thoughtful comments on the manuscript.