A cellular process that includes asymmetric cytokinesis remodels the dorsal tracheal branches in Drosophila larvae Free

Tubular structures are ubiquitous across animal phyla, and in many cases they form branched continuous epithelial or endothelial networks. At the cellular level, most tubes have the shape of a wrapped epithelial sheet that encloses a lumen on the apical side. The integrity and impermeability of the tube walls is maintained through adherens junctions (AJs) and septate junctions (SJs), while the lumen is supported by pressure and occasionally by a secreted apical extracellular matrix. Although mitosis in isolated, single cells is a well-understood process (Harris and Tepass, 2010), less is known about cell division within flat or tubular epithelia, where the forces generated by the actomyosin cortex are balanced by E-Cadherin-mediated intercellular adhesion (Bourdages and Maddox, 2013). The cells within epithelial sheets must undergo divisions and rearrangements without disrupting the intercellular junctions that are necessary for barrier function (Blanpain and Fuchs, 2009). Tissue homeostasis and injury repair require morphogenetic movements similar to those in early development (Martin and Parkhurst, 2004) as well as specialized stem cells (Giangreco et al., 2009; Rock and Hogan, 2011). Several recent studies (Founounou et al., 2013; Guillot and Lecuit, 2013; Herszterg et al., 2013) have highlighted the dynamic reorganization of AJs during cytokinesis in the epithelia of Drosophila embryos (Guillot and Lecuit, 2013) and the Drosophila pupal notum epithelium (Founounou et al., 2013; Herszterg et al., 2013). The actomyosin ring that drives cytokinesis deforms the AJ of the neighboring cells, detaching from them once the tension exceeds a certain threshold (Founounou et al., 2013; Guillot and Lecuit, 2013). As shown by laser ablation and genetic experiments, the tensile stress in the surrounding cells also influences the geometry of the emerging junction in a non-autonomous manner (Herszterg et al., 2013).

The Drosophila tracheal system is one of the established models for the study of branched tubular epithelial networks (Ghabrial et al., 2003; Affolter and Caussinus, 2008). Depending on their diameter, tracheal branches may alternatively be multicellular, unicellular or subcellular tubes (Uv et al., 2003); this diversity makes them well suited for the study of tubulogenesis and epithelial morphogenesis. The embryonic tracheal system starts to develop in a segmental fashion as ten invaginating placodes on either side of the embryo. From an invaginated sac-like structure, several primary sprouts grow out, subsequently leading to the formation of secondary and terminal branches in a tightly regulated sequence (Samakovlis et al., 1996; Ghabrial et al., 2003; Uv et al., 2003). Each hemisegment contains a basic unit called tracheomere, consisting of the dorsal trunk (DT) and six other branches (Fig. 1A). Tracheomeres are connected to their neighbors along the anteroposterior axis by specialized fusion cells (Tanaka-Matakatsu et al., 1996), and within each hemisegment, the left-side dorsal branch (DB) connects with its right-side counterpart. Branchless (Bnl)/FGF signaling is crucially involved at all stages of tracheal development, including the initial budding, primary, secondary and terminal branching (Sutherland et al., 1996), air sac primordium (ASP) induction (Sato and Kornberg, 2002) and pupal abdominal trachea (PAT) migration (Chen and Krasnow, 2014). The elongation of the dorsal branch does not occur in the absence of Dpp signaling (Chen et al., 1998), because, by itself, Bnl/FGF only induces filopodial extensions (Ribeiro et al., 2002). Dpp signaling is required to suppress Spalt (Sal) in the DBs, enabling the formation of a unicellular tube through cell intercalation (Ribeiro et al., 2004; Neumann and Affolter, 2006).

Most studies have focused on the embryonic tracheal system, partly because it is easily accessible to live imaging. The tracheal system of Drosophila larvae displays a set of characteristics that sets it apart from other models of tubulogenesis. In common with most larval tissues, the cells of the tracheal system grow in size by endoreplicating their DNA (becoming polyploid) and are largely replaced during the pupal stage (Manning and Krasnow, 1993; Edgar and Orr-Weaver, 2001). However, specific regions of the tracheal system, i.e. the entire Tr2 (where the ASP originates), the dorsal branches (DBs) of Tr3-Tr5 and all the spiracular branches (SBs), were found to retain their proliferative potential (Guha and Kornberg, 2005; Guha et al., 2008; Sato et al., 2008; Weaver and Krasnow, 2008; Pitsouli and Perrimon, 2010). Although some populations of polyploid cells have been shown to re-enter the mitotic cycle (Fox et al., 2010), the cells that the DBs and SBs comprise are likely diploid cells, and functionally similar to those in the imaginal disks (Sato et al., 2008; Weaver and Krasnow, 2008; Djabrayan et al., 2014).

We have previously studied how the coordinated action of cell signaling and cell migration leads to dorsal branch formation during embryogenesis [reviewed by Affolter and Caussinus (2008); Maruyama and Andrew (2012)]. Here, we investigate dorsal branch remodeling during larval development, a dynamic cellular process that includes mitosis, unlike the rearrangement of the embryonic dorsal branches.

The tracheal system of the early third instar larva is very similar to that of the preceding L2 stage, aside from changes in the branching structure of the anterior spiracles and the expansion of the DT lumen. As previously shown (Weaver and Krasnow, 2008), the cells in the dorsal branches of the anterior tracheomeres (Tr2 to Tr5) are able to proliferate, similar to the cells from the spiracular branches (Pitsouli and Perrimon, 2010). During early L3, the autocellular junctions that were established in the embryonic stage through cell intercalation (Ribeiro et al., 2004) are still apparent (Fig. 1B,I). However, within about 24 h, the DBs transform into multicellular tubes (Fig. 1C), the cellular complexity of which continues to increase until the wandering larva (WL) stage (Fig. 1D).

In order to gain an overview of tracheal remodeling at the cellular level, we made use of Raeppli, a novel technique for cell labeling based on the differential expression of four spectrally separable fluorescent proteins (Kanca et al., 2014). We used two copies of Raeppli for a stronger signal and a broader color palette, under the control of the Tub-Gal80^ts;UAS-Flp,Act-Gal4 induction system (Kanca et al., 2014). The animals were first reared at 18°C until L2 stage and were then transferred to the non-permissive temperature of 29°C, unlocking Raeppli and irreversibly forcing a stochastic color selection for each copy of the Raeppli construct. After staging, the L3 larvae were imaged twice, at 20 and 48 h, respectively, using cold anesthesia (supplementary material Fig. S1) to suppress their movements during image acquisition (Fig. 1E,F). The neighboring red- and green-labeled cells (Fig. 1G, yellow arrows) had undergone three or four rounds of division between the imaging sessions, consistent with a cell cycle of ∼8 h (Fig. 1E,F). At first glance, the narrow and elongated shape of the DB cells would seem to favor a ‘sausage-chain’ mode of cell division that would produce largely non-overlapping clonal populations. Instead, we observed a considerable degree of overlap between the two labeled populations (Fig. 1H, yellow arrow).

Dorsal branch cells show a peculiar architecture, acquired through cell intercalation in the embryo. The cells are long, thin, wrapped around the lumen and sealed by a long autocellular AJ (Fig. 1B). Puzzled by the ability of tube cells to divide, we wanted to study their behavior in detail and with high spatial resolution during mitosis. We observed a shortening of the autocellular junctions in the DBs and a corresponding lengthening of the intercellular junctions (Fig. 2A,B) before mitosis was initiated. This cellular behavior, which we will refer to as junctional remodeling, resembles the embryonic intercalation process, albeit in the opposite direction. Along the entire length of a dorsal branch, the cells expanded their intercellular junctions, with those closer to the DT progressing at a faster pace than those closer to the terminal cell (data not shown). Unlike in the embryonic tracheal system, where the migration of the tip cell is a major driving force for the intercalation process (Caussinus et al., 2008), the stalk cells in the larval DBs do not move relative to the DT or the terminal cell, so the junctional remodeling process is likely cell autonomous.

We have investigated the position and structure of the intercellular AJs during the remodeling process, using a functional GFP knock-in version of DE-Cadherin (Huang et al., 2009) in combination with an antibody against Dlg, a component of the septate junctions that is localized to the basolateral side. The position of the nuclei was visualized with btl-mCherryNLS, and for increased sensitivity we also performed anti-GFP antibody staining. Surprisingly, DE-Cadherin was found to be concentrated in two belts: a primary one at the apical side of the cell, likely part of the AJ; and an additional one in a region more basal than the septate junctions, where Dlg is localized (supplementary material Fig. S2A). The second belt of DE-Cadherin was especially prominent in the DTs, whereas in the DBs it usually appeared as a row of punctae (supplementary material Fig. S2C,C′). This distinctive ‘sandwich’ pattern of DE-Cadherin and Dlg was discernable throughout the tracheal system, including the fusion cells (supplementary material Fig. S2A), but was not apparent in other tissues, such as the larval epithelium or the wing disk (data not shown). We examined staged 20-28 h L3 larvae, when the first round of cell division takes place in the DBs. Although in all cases junctional remodeling preceded cell division, we identified three distinct pathways for cytokinesis.

The most common form, henceforth ‘type 1’, takes place after the disappearance of the autocellular junctions (Fig. 2C,D). By contrast, ‘type 2’ cytokinesis occurs in cells that have retained part of their self-junction (Fig. 2E-G). Finally, we refer to those rare events where the dividing cell is separated into unequal compartments as ‘type 3’ cytokinesis (Fig. 2H,I,O). We found that during the transition to a multicellular tube, the intercellular junctions flanking the autocellular junction elongated until they made contact, unzipping the self-junction until it completely disappeared. A new intercellular junction developed at the site, with an orthogonal orientation to the former autocellular junction. Therefore, along most of their length, DBs transitioned from a unicellular to a multicellular tube through extensive junctional rearrangement. The de novo intercellular junctions that separated the two daughter cells developed on the opposite side of the lumen, with respect to the position of the former autocellular junctions (Fig. 2P). Therefore, the nuclei of the newly remodeled branch ended up in pairs, with a zigzag arrangement along the length of the tube (Fig. 2D,P). Owing to this geometry, the uppermost daughter cell on the right and the lowermost daughter cell on the left arrived at a similar height and on opposite sides of the lumen (Fig. 2D). This process can explain the clonal overlap observed in the Raeppli experiments (Fig. 1H).

Type 2 cytokinesis refers to the formation of a de novo junction with the shape of a circumlumenal ring, within a cell that has retained part of its autocellular junction (Fig. 2F,G). During mitosis the nuclei adopted a rounded shape (Fig. 2Q′,S1) and closely associated with one side of the autocellular junction (Fig. 2Q′). Unlike in epithelial cytokinesis, the presence of a rigid apical side imposed restrictions on the range of possible cellular movements, as the forces generated by cell division were not sufficient to deform the lumen. The adherens junctions of epithelial cells are known to furrow towards the centrally positioned actomyosin ring (Bourdages and Maddox, 2013). However, the self-junction of a unicellular tube may only furrow in one direction without compromising epithelial integrity. The rounded prophase nucleus has a diameter close to that of the lumen (Fig. 2R′,S1), so it was able to approach only one side of the junction. This naturally emerging asymmetry ensured that the furrowing of the membrane and the growth of the new junction separated the daughter nuclei (Fig. 2E,L).

Type 3 cytokinesis (Fig. 2H,I) is a subtype of type 2 cytokinesis that leads to faulty abscission. The position of the nuclei relative to the intercellular junctions varies dynamically along the DB, depending on the length of the cells. Because of the smaller distances that were involved, the shorter cells completed the transition from autocellular junctions to intercellular ones faster than the longer cells. Incorrect abscission was observed when the nucleus from a longer cell was close to the intercellular junction with a shorter cell (Fig. 2H). During type 2 cytokinesis, the daughter nuclei remained aligned on one side of the autocellular junction (Fig. 2Q,Q′). By contrast, the daughter nuclei of cells undergoing type 3 cytokinesis were pushed out of alignment (Fig. 2R′,S4,S5) by the ongoing junctional remodeling (Fig. 2I). We could not pursue the ultimate fate of the bi-nucleated cells. However, as most larval tracheal cells are naturally polyploid, we assume they can remain functional. The anucleated compartments were very small compared with the other DB cells, so they probably do not compromise the integrity of the multicellular tube either. For an easier comparison with epithelial cytokinesis (Bourdages and Maddox, 2013), we also include a representation of the tracheal tube as a cylindrical projection (supplementary material Fig. S3A-C), with a duplicated autocellular junction for type 2 and type 3 cytokinesis (supplementary material Fig. S3B,C).

The number of stalk cells in the DBs usually varies between three and six (Caussinus et al., 2008), and we have found examples of up to seven per branch. Because junctional remodeling and mitosis are parallel processes that take place on a similar time scale, cell division occurs at different time points along the dorsal branches. Both the de novo junctions and the new intercellular junctions (formed at the sites of previously autocellular junctions) eventually evolve into similar, alternating half-ring structures. The localization of Myosin II (MyoII) was examined using the Sqh-GFP reporter, and as expected we found accumulations of MyoII at the sites of mitosis (Fig. 3E). We also observed dense accumulations of MyoII (the midbodies) closely associated with the de novo junctions that resulted from both type 1 (Fig. 3G) and type 2 (Fig. 3I) cytokinesis. Because the circumlumenal rings produced by type 2 cytokinesis eventually merge with the neighboring intercellular junctions (Fig. 3J), thereby losing their characteristic shape, we had to compromise between recording as many divided cells per branch as possible and being able to distinguish unambiguously between type 1 and type 2 cytokinesis events. Our statistical analysis of cell division included 13 complete DBs that fulfilled these criteria (Fig. 3L,M). Given that the elongation of intercellular junctions is an ongoing and irreversible process, the cells whose nuclei are located above an intercellular junction, or have already entirely abolished their autocellular junction (Fig. 3F) must eventually undergo type 1 cytokinesis. We refer to these cells as possessing ‘prospective type 1 junctions’. The converse is true of cells possessing ‘prospective type 2 junctions’: although at the time of dissection those cells had the potential of undergoing type 2 cytokinesis, given a sufficiently delayed mitosis they may have yet remodeled their junctions sufficiently to undergo type 1 cytokinesis instead. Therefore, the number of cells containing prospective type 2 junctions should be taken as an upper limit. One of the 13 analyzed branches is shown in Fig. 3C-E, together with a color-coded graphical representation (Fig. 3C′-E′) and a summary of the various types of cytokinesis and cell lengths (Fig. 3K). According to our data, on average more than two-thirds of DB stalk cells undergo type 1 cytokinesis, while the remaining ones undergo type 2 cytokinesis. We did not find any examples of type 3 cytokinesis in our sample of 13 DBs, consistent with them being a rare event. The length of the cells is strongly correlated (P<0.001) with the cytokinesis type (Fig. 3M), although the two populations largely overlap, as would be expected if junctional remodeling is not a checkpoint for mitotic entry.

In dividing cells, the actomyosin cortex is reorganized as part of the well-known process of mitotic rounding (Kunda et al., 2008; Kondo and Hayashi, 2013). Another actin cytoskeleton-based process, the interkinetic nuclear migration (INM), has been observed to re-position nuclei at the apical side in pseudostratified epithelia (Meyer et al., 2011). We observed that during mitosis the nuclei are closely associated with the autocellular junction (Fig. 2R,S1; 2R,S4), whereas the nuclei of pre-mitotic cells and those belonging to non-remodeling branches are located on the opposite side of the tube, relative to the autocellular junction (supplementary material Fig. S2D′). Actin localization during mitosis was explored using the F-actin probe Utrophin-GFP (Herszterg et al., 2013). Compared with the neighboring cells, mitotic cells showed a strong accumulation of actin (Fig. 4A), both at the level of the intercellular junctions (Fig. 4A′,S2) and of the cell cortex (Fig. 4A′,S3,S4). We also observed the accumulation of actin behind the deformed autocellular junction, which is the origin of the de novo junction during type 2 cytokinesis (Fig. 4B). In epithelial tissues, the contraction of the actomyosin ring deforms the cell junctions and is the main driving force of cytokinesis (Bourdages and Maddox, 2013). The ingression of the actomyosin ring forms a persistent midbody, close to the site of the emerging junction that eventually separates the two daughter cells (Herszterg et al., 2013). We observed a similar accumulation of MyoII at the site of de novo junction formation during type 2 (Fig. 4C) and type 1 (Fig. 4D) cytokinesis.

To better understand the formation of asymmetrical junctions during cytokinesis, we also wanted to acquire movies of the DBs from live specimens. For this purpose, we developed a so-called humid box (see Materials and Methods; supplementary material Fig. S4) and used fast-spinning disk confocal microscopy to reduce the image blurring that is induced by larval movement. Compared with our cold anesthesia method and the recently reported chemical anesthesia setup (Heemskerk et al., 2014), the humid box has several advantages. Most importantly, the larvae can continue to develop during imaging, for up to 12 h. As the lower halves of the larvae are embedded in agarose, only moderate squeezing is sufficient to restrain them, allowing for very high resolution imaging through the transparent cuticle (provided the scanning frequency is above 10-15 Hz). We found that for imaging sessions of less than 30 min, such as for screening purposes, the larvae can also be mounted on glass slides. However, for longer acquisitions, the humid chamber (loaded with a few grams of 7% agarose) was necessary as a source of moisture and as a heat sink. Time points blurred through larval movement were discarded (some 20-50%, depending on the movie) and we used the Volocity software for 3D rendering.

We imaged the transition between a shrinking autocellular junction and a new intercellular junction for several hours in a E-Cad-GFP;btl-mCherryNLS larva, at 10 min intervals (Fig. 4E, white arrow). The new multicellular junction formed orthogonal to the old self-junction (Fig. 4E, t=1:20:00) and eventually evolved into a complete half-ring (Fig. 4E, t=2:50:00). Within the same time frame, the upper cell was observed to undergo type 2 cytokinesis (Fig. 4E, red arrow). The mitotic nucleus was closely associated with the autocellular junction and during nuclear envelope breakdown the mCherryNLS was dispersed into the cytoplasm (Fig. 4E, t=0:00:00), revealing the shape of the cell. Autocellular junction deformation and the budding of the de novo type 2 junction were also apparent (t=0:20:00 and 0:30:00). Although junctional remodeling continued throughout the duration of the movie (Fig. 4E, orange and white arrows) the type 2 de-novo junction failed to extend all the way around the lumen. These results confirm that the extension of the circumlumenal junction can be a protracted affair. Therefore, a large amount of junctional remodeling may take place in the meantime.

We next explored the distribution of MyoII in the DBs of ubi-H2A-EYFP;Sqh-GFP,E-Cad-tomato larvae (Fig. 4F). During cytokinesis, the actomyosin ring (t=18:00 to t=21:00, white arrows) was observed to shrink and eventually form a midbody that remained closely associated with the autocellular junction (Fig. 4F, t=1:04:00). The midbody came to reside at the apical side of the cell (close to the lumen of the tube), similar to what was observed in other epithelial tissues (Founounou et al., 2013; Guillot and Lecuit, 2013; Herszterg et al., 2013). Furthermore, the midbody was located close to the tip of the deformed autocellular junction, as we observed in a separate short movie (Fig. 4G). The de novo junction emerged as an extension from the vertex of the furrowed autocellular junction (Fig. 4G), in agreement with our data from immunostaining (Fig. 4C,D).

Tracheal remodeling is a coordinated process that takes place in only a subset of branches: the contrast between the remodeling seen in the DB of Tr5, and the lack thereof in the DB of Tr6, is particularly striking. We investigated the activity of the Dpp pathway, a known regulator of DB fate in the embryo (Ribeiro et al., 2004), in staged larvae using dad-GFP and brk-Gal4,UAS-mCherry as reporters. High levels of Dad and low levels of Brk result from increased Dpp signaling, and dad-GFP was shown to be an effective Dpp signaling readout (Ninov et al., 2010; Weiss et al., 2010). By 3 h into L3, the Dpp pathway was found to be activated within the spiracular branches of all tracheomeres (supplementary material Fig. S5A,D), as well as in the dorsal branches of Tr2-Tr5 (supplementary material Fig. S5A-C). The first cell divisions in the DB and DT of Tr2 take place about 18 h into L3 and Dpp signaling was detectable several hours before the event (supplementary material Fig. S5E). We observed a wave of Dpp activity from anterior to posterior that preceded and was strongly correlated with the remodeling of tracheal system throughout the third instar (Fig. 5L-N). Low levels of Dpp signaling were seen in the non-remodeling DBs and DTs of abdominal tracheomeres (Tr6-Tr10) towards the end of the third instar (supplementary material Fig. S5P).

To analyze the possible role of Dpp signaling in the remodeling of dorsal branches, we overexpressed either the pathway repressor Dad (UAS-dad) or the constitutively active receptor (UAS-Tkv^QD) under the control of btl-Gal4. As DB formation in the embryo requires Dpp signaling (Ribeiro et al., 2004), we used tub::Gal80^ts for temporal control of the UAS transgenes. The animals were grown at 18°C for 2 days and then allowed to continue development at 29°C until the WL stage. During L3, the activity of the btl promoter is known to be relatively low outside of the remodeling DBs and SBs (Weaver and Krasnow, 2008), so we used a tub>CD2>Gal4 cassette under btl-Gal4,UAS-Flp control to maintain high levels of Gal4 in all the tracheal cells (Fig. 5A-F,S-U). Depending on the larval genotype (tub>CD2>Gal4,UAS-lacZ or CyO,wg-lacZ), β-Gal was present in the nuclei of the tracheal cells (Fig. 5A-F,S-U) or in the wg domain of the wing disk (data not shown). Dad overexpression, driven by the combined btl and tub promoters, resulted in a two- to threefold reduction in the number of cells in the Tr2-Tr4 DB (Fig. 5A-D), as well as a mosaic pattern of large and small nuclei that is not seen in the controls (Fig. 5D, yellow arrow). We observed a similar effect, albeit of lower magnitude, when Dad was overexpressed under the control of the btl promoter alone (Fig. 5I-K). The remodeling of the DB from Tr2 was found to be less sensitive to Dad overexpression, compared with the DBs of Tr3-Tr5 (Fig. 5A). The DBs of dad-GFP (Fig. 5L-N) and btl-mCherry (Fig. 5O-R) larvae are shown as controls. We also investigated the link between Dpp downregulation and junctional remodeling during the mid-L3 stage (Fig. 5E-H). The characteristic elongation of the intercellular junctions was severely delayed by the presence of high levels of ectopic Dad (compare Fig. 5E,F with Fig. 1C). Low levels of Dad overexpression had little impact on the timing of junctional remodeling (Fig. 5G,H), consistent with the results from WL stage (Fig. 5I-K).

Conversely, ectopic expression of Tkv^QD did not induce cell proliferation in the anterior DBs beyond the limits of that particular niche, as opposed to the dramatic proliferation that can be induced by PI3K overexpression (data not shown). Both the high (Fig. 5S-U) and the low (Fig. 5V-X) levels of induction resulted in similar outcomes. Importantly, the overexpression of Tkv^QD did not induce proliferation in the abdominal Tr6-Tr10 DBs either (data not shown). Therefore, we conclude that Dpp signaling is necessary but not sufficient for remodeling in the DBs.

The adult tracheal system of Drosophila comprises regions that have been carried over from the larval stage, as well as new structures that originated from pools of imaginal cells. The precursor cells that will go on to form the air sacs (Sato and Kornberg, 2002), the spiracles (Pitsouli and Perrimon, 2010), the abdominal trachea (Chen and Krasnow, 2014) and the coiled tracheolar cells (Weaver and Krasnow, 2008) retain their proliferative potential until the third instar, when they begin to undergo multiple rounds of division. During the pupal stage, they divide, migrate and differentiate, thereby replacing histolyzed tissues such as the abdominal tracheomeres (Chen and Krasnow, 2014) and forming new structures, such as the air sacs. Alone among tracheal metameres, remodeling in Tr2 also includes the DT and VB. The special identity of Tr2, where the ASP originates (Sato and Kornberg, 2002), is specified by a Hox code (Sato et al., 2008; Djabrayan et al., 2014). It is interesting that the cells in the DB of Tr2 divide a few hours earlier than those from the other anterior tracheomeres and generally have thicker branches by the WL stage (Fig. 5O-R).

As previous studies have shown, the cells in the DBs of Tr2-Tr5 will continuously divide throughout the third instar (Sato et al., 2008; Weaver and Krasnow, 2008). The embryonic transition from a multicellular tube to a unicellular tube with autocellular junctions takes place in the absence of cell division (Ribeiro et al., 2004; Neumann and Affolter, 2006) and is an extreme example of tubular epithelial morphogenesis. We sought a new model system that would incorporate cell division as part of the developmental process. To this end, we investigated the larval dorsal branches at the cellular level, with the aid of high-resolution images from dissected larvae and live imaging methods that we have developed specifically for this task.

We found that, prior to mitosis, the characteristic autocellular junctions of the DBs undergo remodeling, likely driven by changes in actomyosin force distribution, as has previously been shown to be the case in epithelial tissues (Lecuit, 2005). The intercalation process is not a prerequisite condition for the formation of the de novo junctions, as cytokinesis can take place in cells that retain part of their self-junctions (Fig. 2G). Mitotic cell rounding within an epithelium has been shown to be a key process in the invagination of the Drosophila tracheal placode (Kondo and Hayashi, 2013), and the mitotic rounding of the DB cells (Fig. 4A) is likely the driving force behind the changes in cell shape, as well as the associated elongation of the intercellular junctions at the expense of the autocellular ones. As we have shown, cell division in the DBs is highly directional along the lumen (Fig. 3C-E). During cytokinesis, the membrane furrows on the side of the cell where the nucleus is located (Fig. 4E) and the new junction extends around the lumen until it connects with another membrane (Fig. 4C,D). The actomyosin rings that drive cytokinesis in the Drosophila epithelia are able to symmetrically deform the AJs of the two cells that flank the emerging junction (Founounou et al., 2013; Guillot and Lecuit, 2013; Herszterg et al., 2013). We find that in the remodeling dorsal tracheal branches, asymmetric de novo junction formation is the norm: presumably, the peculiar geometry of the unicellular tubes favors this outcome. The unicellular tubes that undergo type 2 cytokinesis do not have neighboring cells close to the plane of division that might assist the membrane furrowing process, as in epithelial sheets (Herszterg et al., 2013). Those differences notwithstanding, the actin accumulation behind the deformed autocellular junction (Fig. 4B) and the position of the midbody near the tip of the de novo junction (Fig. 4C,D) bears an uncanny resemblance to the mitotic machinery described in epithelial cells (Bourdages and Maddox, 2013). As all the DB cells are potentially capable of dividing in this fashion, we documented the various ways of completing cytokinesis, as summarized in Fig. 6. According to our results, the intercellular junctions continue to remodel in parallel with cytokinesis (Fig. 6C,D). It follows that there are multiple ways in which cytokinesis may be completed, depending on the position of the intercellular junctions relative to the autocellular junctions (Fig. 6C-F). Our analysis has identified three possible outcomes. In type 1 cytokinesis, the emerging junction intersects with an intercellular junction (supplementary material Fig. S3,A5 and Fig. 6C). In the case of type 2 cytokinesis, the junction connects to the other side of the autocellular junction (supplementary material Fig. S3,B5 and Fig. 6D) and forms a circumlumenal ring (Fig. 2G). Finally, in type 3 cytokinesis, the nuclei are pushed away from the growing junction by the ongoing remodeling process (supplementary material Fig. S3,C3) and the asymmetric junction that normally runs between the daughter nuclei, intersecting with the membrane on the opposite side of the cell, fuses to the proximal side of the cell instead (supplementary material Fig. S3,C5 and Fig. 6F). Through this unequal division the cell is partitioned into a bi-nucleated and an anucleated daughter cell (Fig. 6F).

This apparent lack of specificity does not result in an uncoordinated development. We find that the daughter cells that are the outcome of type 1 (Fig. 6G,H) and type 2 (Fig. 6J-L) cytokinesis continue to undergo junctional evolution (Fig. 2P), ultimately producing a distinct, alternating ladder architecture (Fig. 6I). In conclusion, the remodeling of the dorsal branches has at its core a dramatic junctional rearrangement and a robust mechanism of asymmetric cytokinesis.

We have found that Dpp signaling is activated in all the tracheal regions that undergo remodeling in the third instar, namely the SBs, the anterior DBs and the DT of Tr2 (supplementary material Fig. S5). Low levels of activation were seen in the non-remodeling tracheal branches (supplementary material Fig. S5O,P). Dad overexpression (Fig. 5A-D) reduced the number of cells in the DBs by approximately a factor of three, compared with the controls (Fig. 5L-R). The downregulation of Dpp signaling delays both junctional remodeling and mitosis (Fig. 5E,F), so Dpp is likely required as a permissive factor for cell cycle re-entry in the DB cells. Within the abdominal DBs of late L3 wild-type larvae, medium levels of Dpp signaling do not lead to remodeling (supplementary material Fig. S5P). This is consistent with our inability to induce remodeling ectopically by overexpressing the constitutively active receptor Tkv^QD in those segments. Therefore, in addition to its role in the formation of the embryonic DBs, Dpp signaling is also involved in the later development of the larval trachea, presumably in conjunction with other, so far unidentified, factors.

We used the fly stocks E-Cad-GFP, E-Cad-tomato, E-Cad-GFP;btl-mCherryNLS, ubi-H2A-EYFP;Sqh-GFP,E-Cad-tomato, ubi-H2A-EYFP;Utr-GFP,Sqh-mCherry/CyO, Brk-Gal4;UAS-mCherryNLS;dad4-EGFPnls, UAS-mCherryNLS, UAS-Tkv^QD, UAS-dad, tub-Gal80^ts, hs-Flp, UAS-Flp,btl-mCherryNLS, UAS-dad,UAS-Flp;btl-Gal4,tub-Gal80^ts, tub-Gal80^ts;UAS-Flp,act-Gal4, UAS-Flp;btl-Gal4,tub-Gal80^ts, tub-FRT>CD2>FRT-Gal4, UAS-lacZ/CyO,wg-LacZ;btl-mCherryNLS, tub-FRT>CD2>FRT-Gal4,UAS-lacZ/CyO,wg-LacZ;UAS-Tkv^QD,btl-mCherryNLS and Raeppli.NLS53D,89A.

Flies were flipped on fresh food every day and maintained at 25°C for 5 days; for the Gal80^ts experiments they were incubated at 18°C for 8 days. The late L2 stage larvae were collected (based on the morphology of the anterior spiracles) and allowed to hatch on agar plates. After 3 h, the freshly hatched L3 larvae were transferred to fresh food vials and incubated at 25 or 29°C until the desired age.

For dissection, the anesthetized larvae (up to 1 h incubation on ice) were transferred in a drop of chilled PBS on a silicone plate, immobilized with needles and opened on the ventral side. After removing everything apart from the tracheal system and the associated imaginal disks, the carcasses were fixed in 1% PFA for 15 min, washed with PBT (1× PBS buffer with 0.3% Triton X-100) and transferred to 24-well plates. We used the following antibodies: mouse anti-Dlg (1:1000, Hybridoma Bank), rabbit anti-GFP (1:2000; Abcam, ab6556), rabbit anti-β-Gal (1:5000; Cappel, 55976) and rabbit anti-mCherry (1:10,000, a gift from the Erich Nigg lab, University of Basel, Switzerland). Anti-mouse and anti-rabbit secondary antibodies, conjugated to Alexa Fluor 488, 568 and 633 were used at a dilution of 1:1000.

Fixed samples were imaged with a Leica SP5 point scanning confocal. Data processing was carried out with Imaris, including the distance measurements using the Point function. Statistical significance was assessed by the Kolmogorov–Smirnov test. Following our observation that Drosophila larvae can survive being cooled to sub-zero temperatures, we constructed a special device in collaboration with the Mechanical and Electronic Workshops at the Biozentrum (Basel, Switzerland) (supplementary material Fig. S1). The device uses a Peltier element to cool an aluminium plate to a temperature of approximately −3°C. The heat from the warm side is dissipated by means of a radiator and a side fan. The plate is shaped to fit on the microscope table and can accommodate a glass slide. The larva was mounted on a slide, using adhesive tape and capillarity to achieve the right amount of pressure (supplementary material Fig. S1D). Two paper stripes were used to protect the larva from glycerol immersion, while a bit of 2% methylcellulose was applied to the coverslip in order to prevent droplet condensation above the sample. Owing to the high thermal conductivity of immersion objectives, a 20× air objective was used for all the experiments. After data acquisition, the larvae were transferred to grape juice plates supplemented with fly food and allowed to continue development in a humidified box, at the appropriate temperature.

Time-lapse imaging was performed with a humid chamber, also constructed by the Mechanical Workshop (supplementary material Fig. S4). A special slide was prepared for each experiment, by casting 7% agarose in a plexiglas mold (supplementary material Fig. S4D). The agarose was trimmed level with the surface using a blade, the tape spacers were added and the larva was mounted on the central patch. After adding the coverslip, the larva was allowed to settle for half an hour into the agarose and then the whole slide was lowered inside the humid chamber. Glycerol was used to seal the interface between the slide and the bottom of the chamber. As the humid chamber does not cool the larvae, we used a fast scanning mode to minimize blurring from the gut peristalsis. Movies were acquired on a PerkinElmer Ultraview spinning disk confocal with a 63× immersion objective, at a frame rate of 15 Hz. Image processing was carried out using the Volocity software and Fiji.

We thank the Mechanical and the Electronic Workshops at the Biozentrum for building our custom devices; the Imaging Core Facility for help with confocal microscopy; and Y. Bellaiche, E. Caussinus, F. Hamaratoglu, M. Müller and A. Ochoa-Espinosa for flies and ideas.