Dynamic de novo adipose tissue development during metamorphosis in Drosophila melanogaster

The adipose tissue is a central organ for the control of systemic metabolism across various animal orders, including invertebrates and vertebrates (Gesta et al., 2007). It plays essential roles in storing and mobilizing energy substrates and functions as a pivotal signaling center for inter-organ communications that regulate energetic metabolism, systemic growth and immune responses at the organismal level (Droujinine and Perrimon, 2016; Rosen and Spiegelman, 2014). The adipose tissue predominantly, but not exclusively, comprises adipocytes, which accumulate large lipid droplets as energy storage. Adipocytes and other cell types are enclosed by layers of a basal lamina, thus assembled into a discrete organ. Whereas molecular mechanisms controlling homeostasis in mature adipose tissue have been extensively studied, much less is known about developmental mechanisms underlying adipose tissue organogenesis (Sebo and Rodeheffer, 2019).

The adipose tissue correlate in insects is the fat body (Buys, 1924), which accumulates not only lipid droplets but also glycogen and protein granules, and controls a wide array of physiological activities, including systemic control of metabolism and growth, innate immunity and adult courtship behaviors (Arrese and Soulages, 2010; Boulan et al., 2015; Franz, et al., 2018; Li et al., 2019). Drosophila (Sophophora) melanogaster develops two distinct fat body types – the larval fat body (LFB) and adult fat body (AFB) – both of which are solely composed of fat cells. The LFB develops during embryogenesis and persists until 2-3 days after adult eclosion, serving as a major energy reservoir in feeding larvae and as a nutriment reserve during metamorphosis and in newly eclosed adult flies (Aguila et al., 2007, 2013; Butterworth et al., 1988). The AFB is found in the newly eclosed imago and functions during adult life (Johnson and Butterworth, 1985). The LFB and AFB are distributed throughout the body and reside in the body cavity between the gut and the body wall (Fig. S1A,B; Miller, 1950; Rizki, 1980). Fat cells are ensheathed by connective tissues that form a characteristic one-cell thick sheet-like structure (Dai et al., 2017). Both types of Drosophila fat body have served as model systems for studying the significance of flexible lipid mobilization and the impacts of deregulated energetic metabolism in the face of various types of physiological or pathological perturbations (Alfa and Kim, 2016; Baumbach et al., 2014; DiAngelo and Birnbaum, 2009; Storelli et al., 2019).

Our understanding of developmental processes underlying fat body organogenesis has been provided chiefly by studies of LFB development in Drosophila melanogaster (Hoshizaki, 2005). Although it is widely accepted that both LFB and AFB are mesodermal in origin, they arise from distinct cell lineages (Lawrence and Johnston, 1986). The larval fat cells are specified by positional information in the embryo, leading to the formation of fat cell progenitors in a metameric manner (Azpiazu et al., 1996; Riechmann et al., 1998). Histochemical studies showed that these progenitors proliferate to form clusters and are integrated into a continuous tissue covering a broad region of the larval body (Miller et al., 2002). Recent studies also shed light on many aspects of mechanisms underlying the establishment of tissue architectures and intracellular structural organizations of the LFB cells (Dai et al., 2018; Diaconeasa et al., 2013; Ugrankar et al., 2019; Zang et al., 2015; Zheng et al., 2020). In contrast to the LFB, the early developmental trajectories giving rise to the AFB in Drosophila have been mostly unexplored for decades, except for a few studies (Ferrus and Kankel, 1981; Hoshizaki et al., 1995; Lawrence and Johnston, 1986).

The AFB has several features that are distinct from the LFB. AFB cells can contain multiple nuclei (Doane, 1960) and deposit different storage granules (Butterworth et al., 1965). In addition, the AFB will be optimized for serving adult flies, the dispersion of which requires energy mobilization, in contrast to larvae, which are specialized for eating foods and accumulating biomass. Interestingly, de novo formation of the AFB during metamorphosis has been reported only in limited species, including derived dipteran species (reviewed by Dean et al., 1985; Trager, 1937). Thus, analyzing developmental mechanisms of the AFB may reveal alternative molecular mechanisms for establishing and mobilizing energy stores, and the nature of plastic development of fat tissues both in vertebrates and invertebrates (Gesta et al., 2007; Sakers et al., 2022).

Here, we have developed and combined genetic tools that allow Gal4 expression in the AFB lineage during metamorphosis, enabling us to label precursor cells that give rise to the AFB and to manipulate gene functions in these cells.

The precursor cells emerged from central segments, dispersed to the abdomen and head, and assembled into the mature AFB. At the cellular level, this transformation involved directional migration, continuous proliferation and homotypic cell fusion. At the genetic level, some of the key events in AFB organogenesis appeared to be governed by ecdysone signaling and the Serpent transcription factor.

The AFB is found in newly eclosed adult flies (Johnson and Butterworth, 1985) but not in mature third instar larvae. Thus, major developmental events related to the AFB must take place during metamorphosis. In this study, we mainly focused on the abdomen, in which the AFB predominantly resides (Fig. 1A; Miller, 1950). We first tested whether known fat body Gal4 drivers were able to highlight the AFB in young adult flies (0-1 day after adult eclosion; see Materials and Methods); however, all of the Gal4 lines that we tested drove GFP expression in the LFB, thus obscuring AFB development due to close proximity between the LFB and AFB. Although larval fat cells undergo remodeling, involving autophagy induction and tissue dissociation, with progressive cell death during metamorphosis (Butterworth et al., 1988; Jia et al., 2017; Liu et al., 2013; Rusten et al., 2004), a large population of larval fat cells is persistent even at adult eclosion. To overcome this limitation, we adopted two strategies. First, we took advantage of the Gal80 protein, which inhibits Gal4 (Ma and Ptashne, 1987; Pfeiffer et al., 2010), in conjunction with cis-elements of genes that are expressed in the later development of the LFB (Fig. S1D). Second, we tested Gal4 lines associated with genes controlling the early development of the LFB.

We generated fly lines with Gal80 under the control of several cis-elements of fat-body-related genes (Figs S1E,F, S2A-D, Table S1) and tested whether these transgenes could inhibit GFP expression in the LFB induced by Cg-Gal4, which is expressed in the fat body and hemocytes (Asha et al., 2003). Among these Gal80 strains, we focused on the Gal80 gene induced by a cis-element of the ppl gene (hereafter referred to as ppl-4kb-Gal80; Buch et al., 2008; Zinke et al., 1999). ppl-4kb-Gal80 potently decreased GFP signals in the LFB at the late third instar stage (Fig. 1B). The suppression of Gal4 activity in the LFB persisted until the late pharate adult stage, allowing the visualization of the AFB (Fig. 1C,D). When a GFP marker was driven by ppl-4kb-Gal80 and c833-Gal4 (a known fat body Gal4 driver, Hrdlicka et al., 2002), tissues with a sheet-like architecture were seen in the abdomen from 60-65 h after puparium formation (APF) onward (flies were grown at 25°C unless otherwise noted). The tissues were efficiently stained with Nile Red, a neutral lipid marker, indicating that they indeed comprised the AFB (Fig. 1E). The accumulation of numerous lipid droplets in the AFB was detected as early as 66 h APF (Fig. 1F). At 60 h APF, GFP-positive cells contained only a few lipid droplets (Fig. 1F, Fig. S3A), suggesting that the AFB starts to deposit lipids at 60-65 h APF. Lipid droplets enlarged progressively at 82 h APF and 0-8 h after adult eclosion (Fig. 1F, Fig. S3). It is noteworthy that the specificity of this Gal4/Gal80 combination is rather limited. c833-Gal4 expresses is expressed in some neurons, including peripheral sensory class IV neurons (Grueber et al., 2002; Shimono et al., 2009), and in surrounding tissues, including part of the adult epidermis, as revealed by the expression of nuclear-localized fluorescent protein markers (Fig. S2E). In addition, ppl-4kb-Gal80 suppressed Gal4-driven gene expression in hemocytes at the third instar and white prepupa stage (Fig. S4).

We also examined strains in Gal4 libraries driven by short cis-elements encoded in the fly genome (Jenett et al., 2012; Kvon et al., 2014). seven-up (svp) is expressed during the early development of the LFB (Hoshizaki et al., 1994). Two svp Gal4 strains (svp[31F09]-Gal4 and svp[31H02]-Gal4) exhibited transient Gal4 expression early in the AFB lineage, as revealed by the G-TRACE system (Fig. S2F; Evans et al., 2009). To sustain the transient Gal4 expression induced by svp[31F09]-Gal4, we employed the Ay-Gal4 system – a lineage-tracing method (Fig. S2G; Ito et al., 1997). This svp[31F09]-Gal4 plus Ay-Gal4 combination drove UAS-dependent markers in the AFB more specifically than the c833-Gal4 plus ppl-4kb-Gal80 combination, although it induced Gal4-dependent marker expression in a small population of cells that moved with weak directionality, possibly hemocytes, and a few adult nephrocytes in the abdomen (Fig. S5D; Na et al., 2015).

While observing the AFB labeled with these Gal4-based tools, we noticed that various genotypes of flies expressing only fluorescent markers exhibited reduced or absent AFB signals with variable degrees of penetrance (Table S2; Fig. S5A-C). Those adult flies with diminished marker signals displayed reduced or missing Nile Red signals (n=11; Fig. S5D,E), suggesting that AFB development was disrupted in these flies. Unexpectedly, even Canton-S, a widely used wild-type strain, but not Oregon-R, showed disorganized or lost AFB tissues with ∼30% penetrance (Fig. S5F,G, Table S2). To examine whether defects in AFB development are common in wild-type strains, we tested 39 strains in the Drosophila Genetic Reference Panel, which were established as inbred lines from the Raleigh, USA population (Huang et al., 2014). These strains rarely showed AFB defects, with 1.4% penetrance on average (n for each line was approximately 140; Table S2). Thus, the aberrant AFB development observed in wild-type genetic backgrounds might be due to genetic variations that might arise after establishment as laboratory strains. In addition, cellular toxicity due to fluorescent protein expression might contribute to AFB defects, as strains with multiple fluorescent markers appeared to have high AFB defect rates (Table S2: Ansari et al., 2016; Kintaka et al., 2016). Taken together, we established genetic combinations that predominantly express Gal4 in the developing AFB, which starts to accumulate lipid droplets at 60-65 h APF.

The genetic tools described above enabled us to follow cells giving rise to the AFB as early as 15 h APF, revealing their directional migration and continuous proliferation. As these cells were located beneath the integument in a quasi-two-dimensional manner (Fig. S6A,B, Movie 1), their developmental dynamics were readily observed using live imaging without dissection. We hereafter refer to these cells that give rise to the AFB (∼65 h APF) as AFB precursor (AFBp) cells. They extended and retracted many prominent filopodia-like and lamellipodia-like protrusions, which may support their directional and persistent motility (Fig. 2A).

In the abdomen, AFBp cells were first seen at the ventroanterior end at ∼15 h APF, from which site they migrated posteriorly and eventually reached the posterior end, covering almost the entire area of the ventral aspect of the abdomen by 40 h (Fig. 2B,C, Movie 2). These data suggest that AFBp cells originate and emigrate from more-anterior segments: the thorax or head. At ∼30 h APF, a subset of ventral AFBp cells started to migrate dorsally, entering pleurites (Fig. 2D). The onset timing of the dorsal switching differed along the anteroposterior (AP) axis, with AFBp cells in anterior segments starting the dorsal migration earlier. To quantitatively analyze their migration patterns, we tracked AFBp cells at 27-29 and 33-35 h APF, and measured angle and displacement values of nuclear movements at 2 min intervals (Fig. 2E). At 27-29 h APF, their migration steps were biased toward the ventroposterior direction. At 33-35 h APF, their migration direction reversed along the dorsoventral (DV) axis (Fig. 2E). AFBp cells distant from the ventral midline (VML) more frequently moved dorsally than those in other regions at 33-35 h APF, but not at 27-29 h APF (Fig. S6C). Both at 27-29 and 33-35 h APF, AFBp cell migration along the AP axis was biased toward the posterior direction (Fig. 2E). We also measured displacement values of migration events subdivided by four directional categories (dorsal, ventral, anterior and posterior) and found that anterior movements were only slightly smaller than migrations toward other directions (Fig. S6D). These data suggest that the directional bias toward the posterior direction, rather than differences in the displacement length of movements toward each direction, is a significant driver of the posterior migration of AFBp cells.

The observed directional tendency toward the VML at earlier times (27-29 h) might be imposed by rearrangements of surrounding tissues. From 25 to 35 h APF, the length of the migration area along the DV axis progressively decreased (Fig. S6E,F, Movie 3), possibly suggesting that other tissues physically confine the migration area for AFBp cells.

As described above, dorsally migrating AFBp cells proceeded through a narrow path in each hemisegment and entered the pleurites. Leading AFBp cells passed the abdominal spiracle at 40 h (Fig. 3A, Movie 4) and reached the dorsal midline by 50 h APF (Fig. 3B, Movie 5). In the tergite, there were regions lacking AFBp cells (dashed lines in 55 h, Fig. 3B), and these regions will be occupied by dorsal oenocytes (Koch, 1945). The directional bias of AFBp cell migration at the pleurites varied over time. A gradual reversal in their migration direction occurred around abdominal spiracles during 47-53 h APF (Fig. 3C,D). The displacement length of dorsal migration appeared to decrease during the same period (Fig. S6G). In contrast to AFBp cells, LFB cells and hemocytes in the same region displayed only small movements without directional bias (Fig. S7B-D).

The characteristic sheet-like architecture of the AFB was achieved by 65 h APF across all abdominal regions (Fig. 4A,B, Movie 6). After the migration period, AFBp cells became flattened but continued to extend fine filopodia-like protrusions in various directions (Fig. 4C, Movie 7). These cells increased contact areas with neighboring cells as fine protrusions were retracted. Eventually, the characteristic monolayer was established.

Our time-lapse imaging suggests that AFBp cells are derived from the thorax or head. To better characterize the site of origin, we attempted to visualize migrating AFBp cells in the thorax and head. Whereas the thick ventral integument of thoracic segments prevented imaging AFBp cells within the thorax, AFBp cells in the head could be imaged in dorsal and frontal views. To observe AFBp cells in the head, we used the c833-Gal4 and ppl-4kb-Gal80 combination, as svp[31F09]-Gal4-based labeling methods failed to visualize AFBp cells in the head for unknown reasons (n=7 flies with the developed AFB in the abdomen, Fig. S8A). In mature adult flies, the AFB was located beneath the vertex, the frons and the stem region of the proboscis (Fig. 5A, Fig. S8B-D; Miller, 1950). In the frontal view, AFBp cells seemed to emerge from the border between the head and thorax at ∼25 h APF, when reared at 29°C (Fig. 5B). They migrated in the anterior direction at the region between the compound eye and proboscis, and a subset of cells entered the proboscis (Fig. 5C, Movie 8). Moreover, the displacement length of anterior movements might be slightly larger compared with migration events toward other directions (Table S3). When imaged from the dorsal view, AFBp cells appeared to emerge at the posterior edge of the vertex and exhibited anteriorly biased motility (Movie 9). Taken together, precursors in the head migrated anteriorly, as opposed to abdominal AFBp cells, which moved posteriorly.

These lines of evidence suggest that AFBp cells originate from the thorax; however, there might be a distinct population of abdominal cells that give rise to AFBp cells, and these resident precursors would be concealed by a large number of cells coming from the thorax. To address this possibility, we disrupted the motility of AFBp cells during the early development of the AFB and tested whether hypothetical resident AFBp cells could be detected in the abdomen. We took advantage of mutant variants of Rac1, which acts as a major regulator of cell motility by controlling cytoskeletal dynamics (Luo et al., 1994; Petrie and Yamada, 2015; Pocha and Montell, 2014). A constitutively active form of Rac1 was expressed in the AFBp cells only after puparium formation to avoid defects before metamorphosis using the TARGET system (McGuire et al., 2004). When the c833-Gal4 and ppl-4kb-Gal80 combination was activated by shifting to a restrictive temperature (29°C) after puparium formation, GFP signals within AFBp cells were first detected at 20-25 h APF. When constitutively active Rac1[V12] was ectopically induced, abnormally large cell clumps were seen at the anterior region of the ventral side, suggesting aberrant homotypic cell-cell adhesions between AFBp cells migrating out of the thorax (Fig. 5D; Fig. S9). These clumps disappeared by 40 h APF. In the ventral aspect of the abdomen of flies expressing Rac1[V12], a small population of migrating AFBp cells was seen (data not shown). We speculate that these cells started expressing the marker after the entry into the abdomen and were unlikely to be AFB primordial cells within the abdomen, because their emerging sites were inconsistent among the individual flies that we observed (n=6 flies).

In newly eclosed adult flies, the abdominal AFB contains ∼18,000 and 12,000 nuclei in the female and male abdomen, respectively (Johnson and Butterworth, 1985). Achieving the numerous nuclei would require multiple rounds of cell division for each precursor cell, but how and when these cells divide has not been explored. Our time-lapse imaging captured AFBp cells undergoing cell division during and after migration.

The cell division of migrating AFBp cells involved several distinct steps. Actively migrating cells retracted their prominent protrusions and rounded (Fig. 6A). Concurrently, nuclear-localized fluorescent marker signals increased in area and decreased in intensity, indicating the onset of nuclear membrane breakdown. In these rounded cells, mitotic spindles were observed, as visualized by Clip::GFP, a microtubule marker based on the microtubule-binding domain of human CLIP170 (Fig. 6B; Stramer et al., 2010). After cytokinesis, the resultant daughter cells recovered their polarized shapes and started to migrate individually (Fig. 6B). Quantification of the nuclear doubling time of ventral AFBp cells at 25-60 h APF (Fig. 6C) revealed their continuous proliferation, perhaps with more rapid nuclear division rates at earlier developmental periods (doubling time for 25-40 h≈7.9 h; for 45-60 h≈14.6 h). Assuming that the nuclear doubling rate at 15-25 h APF, when we did not quantify cell proliferation rate due to the paucity of AFBp cells, is the same as that at 25-40 h APF, precursors may undergo approximately four rounds of nuclear division at 15-60 h APF.

Cell division events continued even after AFBp cells were arranged as the monolayer tissue (Movie 10). During this post-migration proliferation period, the size of the AFB was relatively unchanged, but the cell and nuclear sizes of its cells were progressively decreased (Fig. 6D, Fig. S10A-F). Adult fat cells showed a rapid decrease in their division rate after 70 h APF, and, by ∼75 h APF, almost all had ceased their cell cycle progression (Fig. 6E). There were 3.3 times as many AFBp cells at 75 h APF as at 61 h APF, implying ∼1.7 rounds of nuclear division events per cell after 61 h APF, on average.

AFB cells in dipteran species, including Drosophila melanogaster, are multinucleated (Day, 1943; Doane, 1960). Multinucleation was also seen in the larval fat body in non-dipteran species (Nakahara, 1918); however, how multinucleated fat body cells are formed has been unclear. Multinucleated cells were seen both at the tergite and sternite. Most of them contained two or four nuclei at 85 h APF and 0-8 h after eclosion (Fig. 7A, Fig. S10G; Doane, 1960). Our time-lapse imaging captured the occurrence of cell-cell fusions between neighboring AFB cells and the resultant multiple nuclei in a single cell from 70 to 85 h APF (Fig. 7B,C, Movie 11). These data argue that homotypic cell-cell fusion between AFB cells contributes to the formation of multiple nuclei found during imago. We analyzed the relationships between cell size and multinucleation states, and found that cellular area size, but not nuclear area, was correlated with nuclear numbers per cell not only at 85 h APF but also 0-8 h after adult eclosion (R²=0.383 and 0.581, respectively; Fig. 7D, Fig. S10H). The correlation at 0-8 h after eclosion is not trivial, as AFB cells grew after 85 h APF (Fig. S10I), suggesting that multiple nuclei confer onto AFB cells the capacity to grow larger.

Anatomical studies on the development of the AFB in dipteran species suggested that certain types of hemocytes might give rise to adult fat cells (Jones, 1962; Sakurai, 1977; Whitten, 1964). Given that recent studies have shown that Drosophila hemocytes show trans-differentiation capacity within hemocyte lineages (Csordás et al., 2020), we examined whether hemocytes directly contribute to the AFB.

First, we imaged hemocytes in the abdomen during mid-metamorphosis (from 24 to 50 h APF) by using srp.Hemo-Gal4, a pan-hemocyte Gal4 driver line (Brückner et al., 2004; Evans et al., 2014). There were two types of hemocytes: plasmatocytes and possibly crystal cells (Banerjee et al., 2019). Plasmatocytes were the predominant class of hemocytes and exhibited massive intracellular inclusions, some of which contained muscle debris (Fig. S11A,B; Ghosh et al., 2020; Regan et al., 2013). Crystal cells showed no large intracellular inclusions, stronger GFP signals induced by srp.Hemo-Gal4 and rounded bulges from the cell membrane at 48 h APF (Fig. S11C). These hemocytes showed distinct cell morphologies and distributions compared with those of AFBp cells (Fig. S11D-F). In addition, plasmatocytes in the abdomen showed active migrations with relatively small directional biases at 25-45 h APF, but then their migration activities decreased (Fig. S7A,D). We also examined whether or not the AFB after its formation (72-75 h APF) shows hemocyte-Gal4-driven GFP signals. We observed flies expressing GFP under the control of srp.Hemo-Gal4 or pxn-Gal4 (Stramer et al., 2005; n=3 for each condition) and found no GFP signals in most AFB cells (Fig. S12). These observations indicate that hemocytes are not the cellular origin of the AFB.

Our imaging revealed that AFBp cells occur early in metamorphosis and the AFB grows after 65-70 h APF and persists in imagoes. Another fat tissue, the LFB, which is composed of 2500 cells in female larvae, starts to undergo cell death early in metamorphosis (Butterworth et al., 1988). The LFB histolysis continues until ∼3 days after adult eclosion, resulting in the almost complete loss of the LFB cells. By taking advantage of the stochastic AFB defects in wild-type genetic backgrounds, we analyzed the effect of the presence or absence of AFB on the progression of LFB histolysis in young adults. At 24 h after eclosion, y*, w* flies with or without the AFB had comparable numbers of LFB cells [635.2±48.3 and 530.2±110 cells, respectively (average±95%CI); Fig. S13A]. By 96 h after eclosion, almost all LFB cells were lost in both conditions. Collectively, the AFB will not be a major regulator of LFB histolysis. We also attempted to evaluate the effect of the LFB cell ablation on AFB development. rpr, an apoptosis-inducing gene (White et al., 1996), was driven by Lsp2-Gal4, an LFB-specific driver that starts to be expressed at the mid-to-late 3rd instar stage (Fig. S13B; Okamoto et al., 2009), resulting in massive LFB cell death in early stages of metamorphosis (Fig. S13C). The LFB cell ablation appeared to retard overall metamorphic development and induced large ectopic lipid droplets in various tissues (Fig. S13D-G). The massive precautious LFB cell death may cause the uncontrolled efflux of lipids into the hemolymph and aberrant interorgan lipid circulations. Although the AFB development was severely perturbed (n=10/10 flies; Fig. S13E,F), these systemic defects precluded a specific assessment of how the LFB affects the AFB development and may highlight the physiological importance of the gradual LFB histolysis during a long period (i.e. from early in metamorphosis to 3 days after adult eclosion; Butterworth et al., 1988).

Finally, we examined the roles of genes that are expected to be crucial for the metamorphic development of the AFB. Ecdysteroids are the central humoral regulator of developmental transitions in insects, including molt and inter-molt events (Thummel, 2001; Truman and Riddiford, 2019). Ecdysteroids are incorporated by target cells and trigger ecdysteroid responses mediated by a heterodimer complex composed of the nuclear receptor EcR and its co-receptor Ultraspiracle (USP; Yao et al., 1993). During metamorphosis, ecdysteroid signaling controls the progression of metamorphosis by promoting breakdown of various larval tissues, triggering remodeling of persistent cells or promoting differentiation of imaginal cells set aside during embryogenesis (Brown et al., 2006; Ninov et al., 2007). Ecdysteroids are essential in the progression of the metamorphic remodeling of the LFB (Jia et al., 2017; Liu et al., 2013; Rusten et al., 2004).

We examined the roles of ecdysteroid signaling in AFB development by expressing dominant-negative forms of EcR isoforms (EcR[DN]), which suppress EcR-dependent gene activation by competing for USP with endogenous EcR receptors (Cherbas et al., 2003; Hu et al., 2003). EcR[DN] expression was temporally induced only after puparium formation, similar to the Rac1 mutant experiment (Fig. 5D, Fig. S9). Expression of various EcR isoforms with dominant-negative mutations led to striking reductions in AFB area in the abdomen when flies were observed at 65-70 h APF (aged at 29°C) and 0-1 day after adult eclosion (Fig. S14, Table S4). Time-lapse imaging revealed that the defects occurred late in the AFB development (Fig. 8A). When imaged at 45-50 h APF (29°C), AFBp cells expressing EcR.A[W650A] or EcR.B1[W650A] in the tergites showed minor defects. At 50-70 h APF, these cells exhibited atrophic changes, and most of them were progressively lost (Movie 12). The developmental defects caused by EcR.A[W650A] were partially rescued by expressing EcR.A[WT] and EcR.B1[WT], but not p35, a pan-caspase inhibiting protein (Zhou et al., 1997, 1998; Fig. 8B, Fig. S15). These results suggest that the effect of EcR[DN] is specific to ecdysteroid signaling and that caspase activation does not play a role in the atrophic changes. Immunostaining signals of a previously reported monoclonal antibody against EcR (Ag10.2; anti-EcR common; Talbot et al., 1993) were detected in the nucleus of the AFB cells (Fig. S16A-E), and the signals were reduced with EcR knockdown by the co-expression of EcR[dsRNA] and Dcr-2 (Fig. S16F,G). Taken together, these data suggest that ecdysteroid signaling promotes AFB organogenesis or adult fat cell survival through transcriptional activation late in metamorphosis, which may contrast with roles for ecdysone signaling in the induction of histolytic processes in larval fat cells during metamorphosis.

We also examined the role of the srp gene, a GATA-type transcription factor expressed in the earliest stage of the LFB development (Sam et al., 1996; Tremblay et al., 2018). srp loss of function results in the absence of larval fat cell precursors by late stage 15, possibly owing to precocious cell death (Sam et al., 1996). Moreover, it has been reported that forced srp expression in the mesodermal lineage results in the formation of ectopic larval fat cells, which merge with endogenous fat cells to form an expanded LFB (Hayes et al., 2001). First, we attempted to systemically knockdown srp function after puparium formation using tub-Gal4 and tub-Gal80[ts]. We tested five RNAi lines, of which two RNAi lines, srp[HMS01083] and srp[HMS01298], caused almost complete loss of the Nile Red-positive AFB at 96 h APF (29°C) with complete penetrance (Fig. S17A; Table S5). These two RNAi constructs do not share target sequences. When driven by c833-Gal4 and ppl-4kb-Gal80, srp[HMS01298] resulted in a dramatic loss of AFBp cells at 50 h, when aged at 29°C (Fig. S17B), reminiscent of lost larval fat cells in srp mutant embryos. When srp[HMS01298] expression was induced later in development (from 24 h APF aged at 19°C onward), migrating AFBp cells frequently showed thread-like cellular assemblies (Fig. 8C, Movie 13), suggesting aberrant homotypic cell-cell adhesions among AFBp cells. An independent knockdown induced by srp[HMS01083] resulted in web-like assemblies of AFBp cells (Fig. 8D, Fig. S17B, Movie 14). These similar defects in AFB assembly, likely due to aberrant adhesions among AFBp cells, suggest that the aberrant AFB developmental phenotypes are in fact due to Srp depletion and not to off-target effects. The co-expression of p35 did not rescue the AFB loss, indicating that active caspases do not contribute to the loss of AFBp cells caused by srp knockdown (Fig. 8E, Fig. S18).

Early anatomical studies in dipteran species revealed that the AFB newly develops by adult eclosion (Evans, 1935; Pérez, 1910; Robertson, 1936; Wigglesworth, 1949); however, how the AFB develops during metamorphosis has remained unclear. Here, we have identified highly migratory precursor cells and characterized the cellular dynamics underlying AFB development at the single-cell resolution. Precursor cells underwent a long-distance migration, continuous proliferation and structural transformation to form the monolayer. These cellular and tissue-level events appeared to be temporally and regionally controlled (Fig. 9). AFBp cells were assembled into the AFB and started lipid deposition by approximately 65 h APF, consistent with reports of the accumulation of lipid droplets in the AFB during metamorphosis in dipteran species other than the fruit fly (Evans, 1967; Wiesmann, 1962). Thus, AFB development in flies will provide a unique experimental platform in which most of the developmental events of an adipose tissue can be observed in vivo.

We established genetic tools to highlight the developing AFB during metamorphosis. First, we employed a 4 kb cis-element in ppl-Gal4 to induce Gal80 in the LFB (Buch et al., 2008; Zinke et al., 1999). The ppl-4kb-Gal80 system might be helpful for labeling other cell types when used with Gal4 drivers expressed both in cells of interest and the LFB. Second, we found that two svp-Gal4 strains in Gal4 libraries (Jenett et al., 2012; Kvon et al., 2014) showed a transient expression in an AFB lineage at an early stage. Although our AFB Gal4 tools provided important insights into cellular mechanisms, their utility was limited due to undesired expression in neighboring tissues and the difficulty in precisely timed control of Gal4 function (see Materials and Methods). Nonetheless, the basic strategy will be useful for further refinements to genetically dissect mechanisms underlying AFB development.

Our data argue that AFBp cells emigrate from the thorax and undergo a long-distance journey to disperse across the body. The broad distribution is likely crucial for circulating energy metabolites and signaling molecules to respond quickly to environmental challenges and to avoid adverse effects due to locally accumulated metabolites. The migration-based distribution strategy appears to contrast with those of other mesodermal organs. Larval fat cell precursors differentiate in a metameric manner in the embryo, followed by cell proliferation and assembly into the LFB (Azpiazu et al., 1996; Hoshizaki et al., 1994; Miller et al., 2002). Larval and adult muscles, another class of mesodermal tissue undergoing metamorphic tissue remodeling, are also locally generated during embryogenesis and metamorphosis, respectively (Bate et al., 1991; Baylies, et al., 1998; Currie and Bate, 1991; Gunage et al., 2017). Thus, on-site differentiation is likely to be the canonical mechanism for the broad distribution of mesodermal organs in the fly.

If AFBp cells were generated in central segments, their adequate distribution across the whole body would be problematic. As the AFB is abundant in the abdomen of the imago, especially in the tergite, most precursors must migrate posteriorly to reside in the abdomen, and a remaining minor population can migrate into the head. Moreover, the abdominal population must preferentially migrate dorsally to reach the tergites, with a subset of the population remaining in the pleurite and sternite. The directional switching around the abdominal spiracles may help to prevent their excessive dorsal colonization. Although we did not address the molecular mechanisms underlying the allocation, these mechanisms could be intrinsic, extrinsic, or a combination of both. Differential gene expression profiles between the head and abdominal AFBs in the adult fly have been reported (Fujii and Amrein, 2002). If the distinction in gene expression profiles is established during early differentiation, intrinsic factors could direct different populations toward different destinations. Such an early genetic separation scenario might be consistent with the fact that svp[31F09]-Gal4 plus Ay-Gal4, a potent Gal4 driver for the abdominal AFB, failed to be expressed in the head AFB.

Extrinsic factors provided by other tissues and organs may control AFBp cell dynamics. The migration area for AFBp cells around the ventral midline progressively narrowed at 25-35 h APF, and a significant subset of AFBp cells started to migrate dorsally at 30 h APF. During the same developmental period, the ventral nest histoblasts, which form imaginal abdominal epidermis, continued to expand ventrally, and replaced the larval epidermal cells by 32 h APF (Ninov et al., 2007; Roseland and Schneiderman, 1979). In addition, adult lateral muscles in the pleurite start to elongate along the dorsoventral axis at 31 h APF (Currie and Bate, 1991). These remodeling events might shape AFBp migration patterns by physically constraining their migration space or by providing scaffolds and attractants for the directed migration. Another candidate tissue controlling AFBp cell migration may be oenocytes, which predominantly reside in the tergites and are functionally connected with the AFB in lipid circulation (Bousquet et al., 2012; Chatterjee et al., 2014; Koch, 1945). Additionally, oenocytes intermingle with fat cells to assemble into the fat body in various insect species (Makki et al., 2014). These facts raise the possibility that oenocytes secrete humoral factors that attract AFBp cells. Although the developmental mechanisms of adult oenocytes are still unclear, our time-lapse imaging revealed that developing adult oenocytes appear to emerge by 35 h APF (Movie 15), almost concomitant with the onset of the dorsal migration of ventral AFBp cells.

Continuous proliferation during the migration period appears to be a prominent feature of AFBp cell dynamics and is inherently necessary to build up the AFB in the limited time frame of metamorphic transformation. Cell proliferation during cell migration has been reported in various cell types, including neural crest cells, neuroblasts and germ cells in mammals (Cantú et al., 2016; Ridenour et al., 2014; Zhang et al., 2007). The proliferation during migration may be advantageous because migrating cells could sense environments along their paths and at destinations. For example, starved fly larvae can progress into metamorphosis when they have passed the critical timing, resulting in small adult flies (Beadle et al., 1938). AFBp cells in these small flies could sense changes in body size and match their tissue mass by modulating their proliferation or cell-size growth.

The origin of AFBp is a long-standing unresolved issue. Several studies in dipteran species suggest the possibility that a subset of hemocytes transdifferentiates into fat cells and forms the AFB (Jones, 1962; Sakurai, 1977; Whitten, 1964). Our data imply that hemocytes do not provide a significant direct contribution to the AFB. It might be challenging to distinguish polarized and motile AFBp cells from hemocytes, which are often highly migratory and load various types of debris, including lipid-containing structures (Ghosh et al., 2020; Jones, 1962). Quiescent larval cells associated with the epidermis in abdominal segments have also been proposed as progenitors in histological and histochemical studies (Koch, 1945; Wiesmann, 1962; Hoshizaki et al., 1995). Our time-lapse imaging argues that AFBp cells originate from central segments but not from abdominal segments. Adepithelial cells associated with wing and eye-antennal imaginal discs would be the sole previously proposed candidate cellular origins that are consistent with our observations (Hoshizaki et al., 1995); however, a previous wing imaginal disc transplantation study did not report the identification of adult fat cells (Lawrence and Brower, 1982). Thus, the specific site and origin of the AFBp cells is still unclear.

As discussed in the introduction, de novo AFB formation is likely to be specific to derivative dipteran species. In other holometabolous insect species, it is widely accepted that the embryonically developed fat body operates throughout the life cycle, with a large-scale reorganization of intracellular structures during metamorphosis (Dortland and Esch, 1979; Haunerland and Shirk, 1995; Larsen, 1976; Oertel, 1930). Therefore, the de novo formation of the adipose tissue during metamorphosis seems to have evolved in derived species. Our results suggest that AFB development may involve common transcription programs used in LFB development, including Srp and Svp. Larval and adult fat cells may branch from a common ancestral cell lineage, which could make up the embryonic fat body in basal species. The diversification in differentiation programs might involve the modulation of responses to humoral factors (Butterworth, 1972), including ecdysteroids. In a few Lepidopteran species, regionally and biochemically distinguished embryonically generated fat body tissues take contrasting fates during metamorphosis (Haunerland et al., 1990; Shirk and Malone, 1989; Wang and Haunerland, 1992). Although a population of larval fat cells in these species is destined to be degraded during pupal stages, another subset persists into the adult. The elucidation of molecular mechanisms underlying the regional difference might provide a clue to understand how post-embryonic AFB formation has evolved.

Flies were raised and maintained at 25°C on standard cornmeal food in plastic vials unless otherwise specified (Watanabe et al., 2017). Fly strains were obtained from the Bloomington Drosophila Stock Center, the Kyoto stock center, the Vienna Drosophila Resource Center and published resources. Fly strains and plasmids used in this study and exact genotypes are summarized in Table S6.

Precisely staged flies for time-lapse imaging were obtained by collecting white prepupae on the wall of plastic vials using wet paintbrushes at 1 h intervals (defined as 1 h after puparium formation: APF). The flies were stored for further development in a humid incubator. At the time of live-imaging and histochemical staining, flies were placed on double-sided adhesive tape on glass slides, and their puparium was carefully removed using forceps. Flies are eclosed at ∼100 h APF and 80 h APF when aged at 25°C and 29°C, respectively.

For searching for Gal4 stocks that can visualize AFB development, commonly used fat body Gal4 lines were first examined for expression of GFP markers in the LFB and AFB of young adult flies. Gal4 driver lines we tested were Cg-Gal4, r4-Gal4, c833-Gal4, ppl-Gal4 and Lsp2-Gal4 (Asha, et al., 2003; Buch, et al., 2008; Lee and Park, 2004; Hrdlicka, et al., 2002). These Gal4 lines showed strong expression in the persistent LFB. This persistent expression prevented delineating AFB development, even though c833-Gal4 was expressed in the AFB lineage, as revealed by Kaede, a photoconvertible protein emitting green or red fluorescence before and after photoconversion, respectively (Fig. S1C; Ando et al., 2002; Tsuyama et al., 2017). Photoconversion was performed using a macro zoom microscope (MVX10, Olympus) with a mercury lamp and a band-pass filter transmitting ultraviolet wavelength light (U-MWU2, Olympus).

Gal80 lines under the control of cis-elements of fat-body-related genes were established (Fig. S1D). The regulatory cis-element sequences were selected based on genomic positions of genes of interest and neighboring genes using FlyBase (Thurmond, et al., 2018) and amplified by PCR reactions (KOD FX, KFX1-1, Toyobo) from genomic DNA or a plasmid containing a cis-element of the ppl gene (pCK1, Buch et al., 2008; Table S1). Primers used to amplify fragments are summarized in Table S1. Amplified DNA fragments were subcloned into entry vectors (pENTR/D-TOPO, K240020, ThermoFisher; or pCR8/GW/TOPO, K 250020, ThermoFisher) and then were inserted into pBP-Gal80Uw-6 (Addgene 26236) via L-R reaction with GATEWAY LR clonase II plus enzyme mix (12538120, Thermo Fisher). pBP-Gal80Uw-6 was designed to establish constructs carrying Gal80 under the control of regulatory cis-elements of interest (Pfeiffer et al., 2010). Fly strains carrying fat-body Gal80 strains were established by a commercially available injection service company (Best Gene). The landing site used was the attP2 site. Among the Gal80 stocks generated, ppl-4kb-Gal80 was used with c833-Gal4 to image AFBp cells. The DNA sequence used as the cis-element of our ppl-4kb-Gal80 strain contains the ppl-coding sequence and ∼3 kb of flanking sequences (Buch et al., 2008). ppl-4kb-Gal80 also suppressed Gal4-dependent marker expression in hemocytes driven by pxn-Gal4 at the late third instar and white prepupa stage (Fig. S4).

In our search for strains expressing Gal4 early in the AFB lineage, the G-TRACE system was employed to avoid overlooking strains with transient Gal4 expression (Evans et al., 2009). In G-TRACE, a nuclear-targeted GFP marker (Stinger) under the control of a ubiquitin promoter represents lineage signals. In the absence of Gal4 activity, the ubiquitin promoter and the open reading frame of Stinger are separated by a transcriptional termination signal within an FRT cassette. After Gal4 expression, activation of UAS-FLP causes the irreversible removal of the FRT cassette, resulting in persistent Stinger expression as a memorized lineage signal. In addition, UAS-RedStinger was employed to monitor real-time Gal4 function. Gal4 lines related to serpent (srp) and seven-up (svp) were tested for G-TRACE signals. svp[31F09]-Gal4 and svp[31H02]-Gal4 showed GFP signals (lineage) but not RFP signals in the AFB in young adult flies (Fig. S2F), suggesting that these lines expressed Gal4 in precursor cells in the early development of the AFB but did not maintain Gal4 expression through to adult eclosion. svp[31F09]-Gal4 was used to image AFBp cells combined with Ay-Gal4 to sustain Gal4 activity in the AFB lineage (Fig. S2G; Ito et al., 1997).

To visualize early developmental events of adult oenocytes, we first tested PromE-Gal4, a widely used oenocyte-specific Gal4 line (Bousquet et al., 2012) and svp[32C04]-Gal4. We found that svp[32C04]-Gal4 was expressed in adult oenocytes in newly eclosed adult flies in our fat-body Gal4 search. These Gal4 drivers began to be expressed in oenocytes later in metamorphosis. Next, we examined Gal4 lines associated with genes required for early differentiation and delamination of larval oenocytes (Brodu et al., 2004). vvl[17C04] and slm[LP39] strains enabled relatively specific labeling of adult oenocytes from 25 and 30 h APF onward, respectively.

Flies were imaged using Nikon-C1 confocal laser scanning microscopic systems equipped with an inverted microscope (Eclipse Ti, Nikon) or an upright microscope (Eclipse E800, Nikon). Objectives used were 10× (NA 0.3 or 0.4), 20× (NA 0.75), 40× (NA 1.3, oil-immersion) and 60× (NA 1.4, oil-immersion).

For imaging at a single time point without fixation, whole flies or the abdomens dissected away from the thorax were imaged. Whole live flies were placed on glass-bottomed dishes or glass slides and then imaged with an inverted microscope or upright microscope, respectively. The separated abdomens were mounted in 50% glycerol in phosphate-buffered saline, with a piece of plastic tape as a spacer to avoid crushing the abdomen.

For time-lapse imaging, flies were placed on glass-bottom dishes with their puparium removed and imaged using an inverted confocal microscope (Eclipse Ti, Nikon). For observations with oil-immersion objectives, flies were placed on a thin layer of halocarbon oil 27 (H8773, Sigma-Aldrich), which was used as the imaging medium. For ventral view imaging, legs and wings were displaced by using forceps as these structures conceal the abdominal surface. A humid environment was maintained by placing a sheet of paper soaked with water in dishes and using a humidifier in the imaging room. For imaging with long time courses (12-24 h), flies sometimes underwent drying, especially in their extremities (i.e. wings and legs); however, the onset timings of AFB developmental events in those flies were not strikingly affected when compared with flies aged in a humid incubator, with their puparium intact. We confirmed that imaged flies were able to proceed to the mature pharate adult stage after observation. For most flies observed for time-lapse imaging, their sex was determined by external genital structures at the pharate adult stage or the male-specific genitalia rotation (Inatomi et al., 2019), and noted in Table S6. In both sexes, the dynamics of AFBp cells were similar. Although some aspects of AFB development seemed to differ depending on sex (Fig. 6E and Fig. S6F), more detailed studies will be needed to clarify quantitative differences between sexes.

For ectopic expression or knockdown experiments, the TARGET system was employed, where tub-Gal80[ts] is used for timed control of Gal4 activation (McGuire et al., 2004). The Gal80[ts] protein was a temperature-sensitive mutant of Gal80, which inhibits Gal4 under permissive temperatures. After crossing, vials were maintained at a permissive temperature (19°C) to inhibit Gal4 in embryonic and larval stages, and prepupae were collected over 6 h periods (0-6 h APF at 19°C). These flies were aged at 19°C until stages specified in figure legends, and then transferred to a restrictive temperature (29°C). c833-Gal4 plus ppl-4kb-Gal80 was employed as a Gal4 driver. svp[31F09]-Gal4 plus Ay-Gal4-based labeling methods were not applicable with Gal80[ts] because UAS-CD4::tdGFP expression in AFBp cells was undetectable even when flies were aged at 29°C throughout their development. The partially remaining inhibitory function of Gal80[ts] at 29°C may decrease FLP levels driven by svp[31F09]-Gal4 and suppress the FRT cassette removal required for the activation of Ay-Gal4 (Fig. S2G). Female flies were selected for knockdown or ectopic expression experiments because strong auto-fluorescence in the abdomen of male flies aged at 29°C often interfered with visualizing AFBp cells with fluorescent markers.

For imaging AFBp cells in the head, the c833-Gal4 plus ppl-4kb-Gal80 combination was used as svp[31F09]-Gal4-based labeling methods failed to visualize AFBp cells in the head for unknown reasons (n=7 flies with the fully developed AFB in the abdomen, Fig. S8A). tub-Gal80[ts] was employed to suppress c833-Gal4 expression in the larval salivary gland during embryonic and larval stages. Without timed induction with tub-Gal80[ts], AFBp cells in the head were often concealed by large floating debris with strong fluorescent signals derived from the histolyzed salivary gland (Farkas and Mechler, 2000).

For assessing the rate of AFB development defects in large numbers of wild-type strains that were relatively recently established as inbred lines, Drosophila Genome Reference Panel (DGRP) strains were used (Huang et al., 2014). Thirty-nine strains that belong to the core 40 strains were first tested by using stereomicroscopic observations without dissection (MS5, Leica), and flies that appeared to display AFB defects and were hard to judge were further examined by subsequent dissection (Table S2). The accuracy of the observation procedure above was assessed as follows. Canton-S female flies (7-8 days after adult eclosion; n=134) were first examined using a stereomicroscope without dissection by taking advantage of the fact that adult flies with defects in AFB development tend to display the deflated transplanted abdomen. Subsequently, these flies were dissected as described in the Immunostaining section and examined to determine whether the AFB was developed by observing them in bright-field view. Without dissection, most Canton-S flies were correctly judged in terms of whether they developed the AFB or not (n=87 and 19 flies with the fully developed or underdeveloped AFB, respectively). Some Canton-S flies were hard to judge without dissection, and subsequent observations of dissected flies showed that most of these flies exhibited defects in AFB development (n=21/28).

Digital images were processed with a control application for Nikon C-1 confocal systems (EZ-C1, Nikon) and the Fiji software (Rueden et al., 2017; Schindelin et al., 2012). For cell migration tracking, the TrackMate plugin in Fiji was used (Tinevez et al., 2017). Z-projected image sequences were obtained at 1-3 min intervals and analyzed. Positions in the x-y coordinates of randomly selected cells or all of the cells within a ROI were tracked. When possible, the positions of each cell were automatically recorded using the semi-automatic tracking function in TrackMate. When tracked cells were too close to neighboring cells, locations were manually recorded to avoid generating erroneous links. AFBp cells that underwent cell division were excluded from our analysis. Tracked positions in the x-y coordinates were imported into R software (R Core Team, 2021) to calculate the displacement length and angle value (from −180° to 180°; anterior direction: ±180°, posterior direction: 0°, dorsal direction: 90°, ventral direction: −90°) for each step. Steps were classified into four directional categories with equal angles (i.e. 90°) to address whether migration parameters differed depending on particular directions. In presenting migration angle distributions using rose diagrams, migration steps smaller than 0.5 μm/min were excluded, as movements with small displacement values might be due to tissue movements or be caused by imprecise manual recording of motionless cells. Our quantification was not largely affected when all recorded steps were included.

For quantifying the nuclear doubling time of AFBp cells, flies expressing Stinger or RedStinger induced by svp[31F09]-Gal4 plus Ay-Gal4 were imaged. Nuclear division events in z-projected image sequences were manually counted by taking advantage of reduced signal intensities of nuclear markers, possibly due to nuclear envelope breakdown. Total nuclear numbers were automatically quantified by thresholding images and counting extracted masks using Fiji. For calculating nuclear doubling time in each time window, total nuclear numbers were divided by time frame numbers, multiplied by imaging hours, then divided by the number of observed division events.

For quantification of nuclear division rates in the ventral abdomen from 25 h through 65 h APF, AFBp cells in A3-A6 segments were observed at 5 min intervals for 115 min with a 20× dry objective (NA 0.75; n=43 flies in total). Oil-immersion objectives were not used for ventral view imaging as imaged flies often died after imaging. The damage might be due to halocarbon oil penetrating flies through narrow slits on their surface caused by forceps used for displacing their legs and wings. Owing to the limited spatial resolution of the 20× objective we used, precise recordings of cell numbers were often challenging in areas with a dense population of AFBp cells. We manually analyzed data for 25 h and 40 h APF when AFBp cells were sparse and dense, respectively, and found that unadjusted values were slightly overestimated by 3.6±3.9% and 6.4±1.3% (mean±s.d.; n=6 and 4, respectively) for 25 h and 40 h, respectively.

For measuring the nuclear division rate in the tergite, flies were collected and imaged with a 60× objective at 12 min intervals at 53-60 h or 61-81 h APF (n=6 flies for each period). For quantifying cell fusion events in the tergite, the AFB of flies expressing RedStinger and CD4::tdGFP was imaged using a 40× objective at 61-85 h APF at 6 min intervals (n=3 flies). After identifying multi-nucleated adult fat cells within the field of view in the last frame, fusion events were manually and retrospectively detected in the image sequences. Identified fusion events in our quantification were partial fusion events even in the field of views due to too strong GFP signals obscuring cell-cell boundaries and muscle contractions that caused imprecise z-projection matching and large positional changes of cells.

The abdomen of adult flies was teased open in phosphate-buffered saline (PBS) on sylgard-based dissection stages (SYLGARD 184 Silicone Elastomer Kit, Dow Corning). As most LFB cells float in hemolymph, they were able to be collected in solution by pipetting. Recovered LFB cells were suspended in 200 μl of PBS, and 50 μl of cell suspension was dropped on a slide glass. Cells were counted, and the measured cell number was multiplied by four to obtain estimated numbers of LFB cells in the abdomen.

The area size of AFB tissues, lipid droplets, nuclei and cells in fluorescence images was measured using ImageJ. For AFB area quantification, images (A2-4 segments) were taken with a 20× objective. For cell and organellar area measurements, a 60× objective was used. If possible, regions of interest were selected from binarized images using the Wand Tool in ImageJ. Otherwise, regions were manually selected. For AFB tissue and lipid droplet area measurements, Weka Trainable Segmentation (Arganda-Carreras et al., 2017) was used for segmentation. Classifiers were trained with representative images of the AFB in the tergite and used to obtain the probability map of the class of AFB. The probability map was converted into an 8-bit image and then binarized. For nuclear area measurements, images were binarized using auto-threshold (Otsu's method; Otsu, 1979) or Weka Trainable Segmentation, as used in AFB area measurements. For cell size measurements, most regions of interest were manually selected.

To analyze relationships between nuclear numbers in each cell and cell or nuclear area size, AFB cells labeled with CD4::tdGFP (cell membrane) and RedStinger (nucleus) were imaged with a 60× objective to achieve high signal-to-noise ratios. Nuclear numbers per cell may be slightly underestimated as the short working distance of the objective might occasionally be not enough to visualize the entire depth of thick AFB tissues.

Flies were dissected by using insect pins (Minutein Pins, 26002-10, Fine Science Tools) and micro scissors on sylgard-based dissection stages. The persistent larval fat cells were blown out by pipetting. Specimens were then fixed with 3.7% formaldehyde in PBS for 15-30 min at 25°C. Fixed tissues were permeabilized with 3% Triton X-100 in PBS for 15 min. Primary antibodies used were Ag10.2 (α-EcR common; 1:40; DSHB), DDA2.7 (α-EcR common; 1:10; DSHB), 15G1a (α-EcR.A; 1:10; DSHB), AD4.4 (α-EcR.B1; 1:10; DSHB) and an antibody against GFP (1:300; A11122, Molecular Probes). The EcR antibodies were detected using a secondary antibody against mouse IgG conjugated with Alexa 546 (1:500; A-11030, ThermoFisher). The GFP antibody was used to augment GFP marker signals with a secondary antibody against rabbit IgG conjugated with Alexa 488 (1:1000; A-11034, ThermoFisher). Specimens were mounted in 50% glycerol in PBS with a piece of plastic tape as a spacer to avoid crushing the specimen. We failed to obtain reliable immunostaining signals in AFBp cells before the sheet formation (∼ 60 h APF at 25°C) as most of the AFBp cells were lost by pipetting, and the body wall was too fragile to be spread with insect pins. Concerning 15G1a, fixation with methanol or immunoreaction enhancers (Signal Enhancer Hikari for Immunostain, 02363-071, Nacalai; Can Get Signal Immunostain, NKB-401, Toyobo) were tested; however, staining signals in the nucleus were not enhanced.

Nile Red was dissolved in acetone at 1 mg/ml (stock solution, stored at 4°C) and used to stain neutral lipids at 1 μg/ml (144-0811, WAKO). Hoechst 33342 was used to counterstain nuclei at 100 ng/ml (H 3570, Molecular probes). Fixed flies with 3.7% formaldehyde were stained with Nile Red and Hoechst 33342 for 3 min at room temperature. Flies were dissected, fixed and mounted as described in the Immunostaining section.

Statistical analyses were performed using R. The R codes used in this study are available upon request. Figure panels were constructed using the Inkscape software (Inkscape Project, https://inkscape.org/).

To prepare temporal-color coded images, the Temporal-Color Code function in Fiji was used (https://imagej.net/Temporal-Color_Code). Migration angle distributions were illustrated using rose diagrams, which were plotted as bar charts with polar coordinates using the ggplot2 package in R (Wickham, 2016). For time-lapse imaging using pharate adult flies after the onset of muscle contraction, shifts in the x-y coordinates caused by their motions were automatically matched using the Linear Stack Alignment with SIFT function in Fiji (Lowe, 2004).

Distributions of lower-level parameters (θ_i) and hyperparameters were estimated using a Jags script for Bayesian binomial inference described by Kruschke (2015). Hyperpriors with weak information were used [dbeta(1, 1) and dgamma(1.105125, 0.1051249)]. Maximum a posteriori estimation values and equal-tailed intervals (ETI) of posterior distributions were represented as summary statistics.

Fly stocks and genomic datasets were provided by the Kyoto Stock Center, the Bloomington Stock Center, the Vienna Drosophila Resource Center (VDRC, www.vdrc.at), the TRiP at Harvard Medical School (Grant: National Institutes of Health, Office of the Director R24 OD030002), the Drosophila Genomics Resource Center, FlyBase, P. Martin, A. Gould, C. Han, S. Hayashi, M. Miura, K. Emoto, M. Pankratz, T. Igaki, Y. Shimada, R. Niwa and N. Okamoto. We also thank J. Hejna for polishing the manuscript, and K. Oki, M. Futamata and H. Imai for technical assistance.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.200815.reviewer-comments.pdf