The basic leucine zipper (bZIP) domain protein, Activating transcription factor 3 (Atf3), has been defined as a stress-response gene pivotal to the maintenance of immune and metabolic homeostasis (Chakrabarti et al., 2014; Gilchrist et al., 2006; Jadhav and Zhang, 2017; Rynes et al., 2012; Zmuda et al., 2010), but has also been implicated in the regulation of cytoskeletal dynamics and vesicular trafficking (Boespflug et al., 2014; Donohoe et al., 2018; Yuan et al., 2013). We have reported previously that the fat body-specific overexpression of Atf3 under the C7-GAL4 driver (Koyama and Mirth, 2016) (hereafter abbreviated as C7>Atf3^WT) induced a lean phenotype and decreased lipid droplet size (Rynes et al., 2012). Intriguingly, a closer examination of C7>Atf3^WT larvae in which the fat body was visualized with a help of UAS-GFP and the hemocyte population with Hml:DsRed reporter (hereafter abbreviated as Hml:DsRed, C7>Atf3^WT) revealed a massive attachment of hemocytes to the adipose tissue surface (Figure 1B, D and E), a phenotype not detected in control larvae (Figure 1A, C and E and Figure 1—figure supplement 1A). The fat body-associated hemocytes (hereafter referred to as FBAHs) present in Hml:DsRed, C7>Atf3^WT larvae were densely packed and often formed contiguous monolayers over the entire length of the adipose tissue (Figure 1—figure supplement 1B). In contrast, the characteristic pattern of sessile hematopoietic pockets observed in controls was disrupted in Hml:DsRed, C7>Atf3^WT larvae (Figure 1A and B), while the amount of circulating hemocytes was not significantly affected (Figure 1—figure supplement 1C). Importantly, we found that hemocytes also associated with clones of Atf3 overexpressing adipocytes induced by the heat shock FLPout (hsFLPout) technique (Figure 1F), suggesting that the attachment is an intrinsic property of the fat body cells with excessive Atf3 levels rather than a consequence of the systemic activation of blood cells. In support of this notion, neither the uptake of GFP-positive material originating from the fat body was observed in the hemocytes, nor signs of an encapsulation reaction, such as adhesion of lamellocytes or tissue melanization, were present in the case of C7>Atf3^WT fat bodies (Figure 1—figure supplement 1D–1F).

Together these data indicate that the hemocytes either actively relocate from the sessile hematopoietic pockets or are redirected from the circulation to preferentially adhere to the fat body. Importantly, the interaction of hemocytes with Atf3 expressing adipocytes is not a result of anti-tissue immune response.

The fact that the hemocytes favored the attachment to the C7>Atf3^WT fat body at the expense of body wall sessile pockets indicated that the immune cells associated with Atf3 overexpressing adipocytes may form a de novo hematopoietic compartment reminiscent of the sessile tissue. To determine whether FBAHs would display features of the natural sessile hemocyte population, we focused on their morphological characteristics, mitotic activity, differentiation and ability to participate in the immune responses. Intriguingly, FBAHs, like sessile hemocytes (Lanot et al., 2001; Leitão and Sucena, 2015; Makhijani et al., 2011; Márkus et al., 2009), projected filopodia of varying lengths as well as lamellipodia (Figure 2A–B) and proliferated in situ as demonstrated by the Fly-FUCCI in vivo cell cycle reporter (Zielke et al., 2014) and immunostaining for a mitotic marker phospho-histone-H3 (pH3) (Figure 2C–D).

Staining against the plasmatocyte marker NimrodC1 (NimC1) (Kurucz et al., 2007a) and the crystal cell-specific BcF6:GFP reporter (Tokusumi et al., 2009) showed that like in the sessile tissue (Leitão and Sucena, 2015), the majority of the FBAHs were plasmatocytes, interspersed by a smaller number of crystal cells (Figure 2E). Surprisingly, a portion of the BcF6:GFP-positive crystal cells also stained for NimC1 and showed Hml:DsRed levels similar to plasmatocytes, suggesting that these cells are intermediate hemocytes in the process of plasmatocyte-crystal cell trans-differentiation (Figure 2E–E'). Moreover, we observed blackened crystal cells among FBAHs (Figure 2F) when combining the Hml:DsRed, C7>Atf3^WT with the Bc¹ mutation (Rizki et al., 1980). Since the runaway melanization cascade results in cell death (Lanot et al., 2001; Neyen et al., 2015), we speculate that their differentiation likely occurred on the adipocyte surface. Notably, both BcF6:GFP expressing and melanized crystal cells in the Bc¹ background were mostly located within the FBAH clusters and were surrounded by plasmatocytes (Figure 2E–F).

Finally, to evaluate whether the FBAH population can participate in an immune response as previously reported for the sessile hemocytes (Márkus et al., 2009; Vanha-Aho et al., 2015), control and Hml:DsRed, C7>Atf3^WT larvae were infected with Leptopilina boulardi parasitoid wasps. We found that the FBAHs detached from the Atf3 overexpressing adipose tissue 48 hr after parasitic infection (Figure 2G–I), and lamellocyte differentiation and capsule formation (Figure 2J, L, N and P) was not impaired to a noticeable extent when compared to control larvae (Figure 2J, K, M and O). Together, these data support the notion that hemocytes associated with the fat body in C7>Atf3^WT larvae display a hematopoietic program reminiscent of sessile blood cells and engage in the innate immune defense. The C7>Atf3^WT FBAH model thus represents a unique opportunity to dissect cellular and molecular mechanisms of hemocyte-tissue homing.

The Drosophila larval adipocytes are the primary source of the basement membrane components for most internal organs (Pastor-Pareja and Xu, 2011), including their own. A crucial step in the formation of basement membranes is the proper deposition and assembly of the ECM components which rely on an intricate cellular and membrane trafficking machinery. Hindering ECM protein secretion and release can lead to fibrotic accumulation (Shahab et al., 2015; Zang et al., 2015). Given the described role of Atf3 in cellular trafficking in epithelial cells (Donohoe et al., 2018), we asked whether the attachment of hemocytes to the Atf3 overexpressing fat body could be caused by changes in the structure or composition of the basement membrane. Indeed, the visualization of the Vkg::GFP reporter, a fusion protein of Collagen IVα2 and GFP (Morin et al., 2001), as well as immunostaining against Laminin and the Collagen XV/XVIII-type protein Multiplexin (Mp) revealed accumulation of ECM below the basement membrane in Atf3 overexpressing clonal adipocytes (Figure 3A–C). Transmission-electron microscopy (TEM) further showed that in contrast to the mostly smooth cell membranes of control adipocytes (Figure 3E), the C7>Atf3^WT fat bodies formed deep pericellular folds containing electron-dense material (Figure 3F). These data suggest that the ECM proteins are trapped in pericellular spaces, which might attract hemocytes to the fat body surface.

To establish if there is a causal link between the accumulation of the basement membrane components and the attachment of hemocytes, we decided to interfere with the proper deposition of ECM proteins by silencing the matricellular chaperone SPARC (Shahab et al., 2015), and to reduce the amount of specific core ECM components in the C7>Atf3^WT adipose tissue. As expected, inhibition of SPARC enhanced accumulation of ECM aggregates in C7>Atf3^WT fat body cells which was associated with membrane blebbing (Figure 3—figure supplement 1A–1B). In contrast, the knockdown of Col4a1 (Collagen IVα1) and Trol (Perlecan) strongly suppressed fat body association of hemocytes, while silencing Vkg partially inhibited blood cell attachment (Figure 3—figure supplement 1C–1E and 1I) and silencing of Laminin-A, -B1 and -B2 (LanA, LanB1, LanB2) had no effect on the presence of FBAHs (Figure 3—figure supplement 1F-1I). It is important to note that similar to SPARC silencing, inhibition of Col4a1 and Trol caused marked alterations to fat body structure and impacted animal viability, as reported previously (Pastor-Pareja and Xu, 2011; Shahab et al., 2015). Surprisingly, downregulation of Mp manifested by a noticeable reduction of Mp levels (Figure 3—figure supplement 2A–2B), completely abolished hemocyte association to the Atf3 overexpressing fat body (Figure 3G-K and Figure 3—figure supplement 2C) without any adverse effect on tissue integrity, although the pericellular membrane folds with ECM material were still present (Figure 3—figure supplement 2D–2E). These results indicate that not the membrane folds, but their specific content is responsible for FBAH adhesion. To confirm this notion, we combined Mp knockdown with silencing of SPARC, which we found to exacerbate pericellular ECM accumulation (Figure 3—figure supplement 2F–2H). While the structure of C7>Atf3^WTSPARC^RNAiMp^RNAi fat body remained disrupted and blebbing was still present (Figure 3—figure supplement 2G) the amount of FBAHs was significantly decreased compared to C7>Atf3^WTSPARC^RNAi adipose tissues (Figure 3—figure supplement 2H). Importantly, Mp knockdown in Atf3 overexpressing adipocytes resulted in restoration of the stereotypical pattern of the sessile hematopoietic pockets at the body wall (Figure 3L–O), supporting a notion that upon fat body-specific Atf3 overexpression hemocytes residing in the sessile pockets relocate and expand on the adipose tissue surface. These results demonstrate a functional link between ECM accumulation and hemocyte attachment to the fat body, and uncover Mp as a key component of this interaction.

Since FBAHs phenocopy sessile hemocytes in their morphology and behavior, we asked whether they share a common mechanism of tissue attachment. While little is known about the interaction that anchors the hemocytes to the epidermis, a loss of the plasmatocyte-specific phagocytosis receptor Eater (Kocks et al., 2005) was shown to completely abolish sessile pockets, increasing the number of circulating hemocytes (Bretscher et al., 2015; Melcarne et al., 2019). To address the requirement of Eater in FBAH attachment, we generated C7>Atf3^WT larvae homozygous for the eater¹ loss of function allele. Strikingly, these larvae lacked both sessile hematopoietic pockets and FBAHs (Figure 4A–E). To validate a specific requirement for Eater in the hemocytes, we first silenced its expression in C7>Atf3^WT fat bodies, which did not interfere with hemocyte-fat body interaction (Figure 4F–G). Next, we generated larvae which aside from C7-GAL4 also carried the HmlΔ-GAL4 hemocyte-specific driver (Sinenko and Mathey-Prevot, 2004). While the expression of Atf3 in both the fat body and the hemocytes did not markedly impact hemocyte adhesion (Figure 4H), simultaneous Atf3 overexpression and Eater knockdown impaired FBAH formation (Figure 4I–J). These data define Eater as a hemocyte-specific adhesion molecule that acts both in the context of the sessile pockets and the hemocytes on the adipose tissue surface.

The notion of a common hemocyte adhesion mechanism prompted us to test if Mp is necessary for anchoring hemocytes to the sessile pocket as it is for FBAH formation. To this end, we silenced Mp with the a58-GAL4 (larval epidermis), mef2-GAL4 (body wall muscles) and elav-GAL4 (neurons) (Galko and Krasnow, 2004; Lin and Goodman, 1994; Ranganayakulu et al., 1996) drivers to account for the fact that the pockets reside between the larval epidermis and the body wall muscle layer, and their maintenance depends on the activity of peripheral neurons (Makhijani et al., 2011). While loss of Mp in the epidermal cells drastically disrupted the pattern of sessile tissue, resulting in a phenotype strikingly resembling that of eater¹ (Figure 4K–O), Mp knockdown in the muscles and the neurons did not result in a similar dispersion of the sessile tissue (Figure 4—figure supplement 1A–1D). Interestingly, although effective in suppressing FBAHs on Atf3 overexpressing adipose tissues, knockdown of Col4a1 and Trol in the epidermis did not mimic the loss of sessile tissue structure inflicted by Mp knockdown (Figure 4—figure supplement 1E–1F). Furthermore, silencing Mp specifically in hemocytes had no visible impact on the integrity of the sessile hematopoietic pockets (Figure 4—figure supplement 1G–1H).

These results highlight a requirement for Eater and Mp to facilitate hemocyte-basement membrane interactions both in the case of the naturally occurring sessile hematopoietic pockets and the de novo hematopoietic compartment on Atf3 overexpressing fat bodies.

The Drosophila Mp consists of an N-terminal Thrombospondin-like domain, followed by a Collagen triple helix domain, an NC1 trimerization domain and the C-terminal Endostatin-domain, all of which are present in its mammalian counterparts, Collagens XV and XVIII (Heljasvaara et al., 2017). Mp expression initiates during late embryonic development, mostly in the heart tube and the central nervous system, and is needed for motoaxonal pathfinding (Harpaz et al., 2013; Meyer and Moussian, 2009). These traits are distinct from the ubiquitous Col4a1 and Vkg, suggesting that Mp-containing basement membranes may have specific functions in cell adhesion and migration. Given the fact that hemocyte attachment to the epidermis as well as to C7>Atf3^WT fat bodies was dependent on Mp, we asked whether its presence in the basement membrane may be sufficient to promote immune cell attachment. To this end, we generated a C-terminally GFP-tagged Mp transgene (UAS-Mp::GFP) and overexpressed it with the C7-GAL4 driver in the adipose tissue. Compared to control, Mp levels markedly increased in the basement membrane of C7>Mp::GFP fat body (Figure 5A–B) as determined by immunostaining with a Mp-specific antibody (Harpaz et al., 2013). However, GFP signal was restricted to the adipocyte cytoplasm (Figure 5—figure supplement 1A), suggesting that the C-terminal end of Mp undergoes proteolytic cleavage. Surprisingly, Mp::GFP overexpression in the fat body did not induce hemocyte adhesion (Figure 5A–B). Instead, it led to the complete dispersal of the sessile hematopoietic pockets, phenocopying homozygous eater¹ mutants (Figure 5C–E, compare to Figure 4A). In addition, circulating hemocyte numbers in C7>Mp::GFP larvae significantly increased compared to controls, reaching similar levels as in eater¹ mutants (Figure 5F). Immunoblots from the cell free hemolymph revealed a notable increase of the Mp protein in the circulation of C7>Mp::GFP larvae relative to controls (Figure 5G). These results indicate that while fat-body-produced Mp incorporates into the adipocyte ECM, it is also released into the hemolymph where it may interact with hemocytes and interfere with their binding to the basement membrane due to a saturation effect.

To circumvent the hemolymph overload, we induced Mp::GFP expression only in clones, which resulted in local Mp accumulation over the targeted adipocytes and in their intercellular spaces (Figure 6A–E). Strikingly, clonal adipocytes overexpressing Mp::GFP were surrounded by hemocytes, which displayed previously established characteristics of FBAHs, namely tight clustering and projection of filopodia and lamellipodia, the ability to undergo mitosis on the tissue surface and the presence of crystal cells (Figure 6C–E). Moreover, Mp::GFP expression in the pouch region of the wing disc using the nubbin-Gal4, UAS-mCherry driver (nub>mCherry) promoted hemocyte adhesion to this specific domain (Figure 6F–G). Of note, this hemocyte epithelial tissue association was not provoked by apoptosis of imaginal cells (Figure 6—figure supplement 1A–1B). Together these results demonstrate that Mp is not only necessary but also sufficient to facilitate hemocyte attachment to tissue surfaces.

The necessity of Eater and Mp to secure the immune cells to the basement membrane and the results from Mp overexpression experiments prompted us to test if the two proteins might physically interact. To this end, we performed co-immunoprecipitation experiments with tagged Eater::GFP and Mp::FLAG proteins in S2 cells, and found that the two proteins indeed co-precipitated (Figure 6H). While we cannot exclude that the binding might be indirect, these results present an argument that Mp acts as an interacting partner of Eater on the basement membranes of C7>Atf3^WT fat bodies, and on the epidermal surface covering the sessile hematopoietic pockets (Figure 6I).

Flies were kept at 25°C on a diet consisting of 0.8% (wt/vol) agar, 8% cornmeal, 1% soy meal, 1.8% dry yeast, 8% malt extract, and 2.2% sugar beet syrup supplemented with 0.625% propionic acid and 0.15% Methylparaben (Sigma-Aldrich). In standard crosses, 10 virgins were crossed with five males. The crosses were flipped daily, and the larvae were analyzed 6 days after egg laying (AEL). Driver lines crossed to w¹¹¹⁸ served as controls.

Clones overexpressing specific transgene were generated with the FLPout technique utilizing heat shock-induced FLP expression which removes an FRT-flanked stop cassette separating a constitutive Act5C promoter from the GAL4 coding sequence (hsFLP, Act5C>y⁺>GAL4). Clonal cells were marked by the expression of UAS-GFP or UAS-myr-mRFP. To generate adipocyte clones, first instar larvae (24 hr AEL) were heat-shocked in a 37°C water bath for 20 min, and were afterwards kept at 25°C until processing on day six AEL.

The following Drosophila melanogaster strains were used: w¹¹¹⁸ (BDSC; RRID:BDSC_3605), w; Hml:DsRed (Makhijani et al., 2011), w; C7-GAL4 (Rynes et al., 2012), w; C7-GAL4, UAS-Atf3^WT (Rynes et al., 2012), w; C7-GAL4, UAS-GFP, Hml:DsRed (recombined in this study), w; C7-GAL4, UAS-GFP, Hml:DsRed, UAS-Atf3^WT (recombined in this study), w; nub-GAL4, UAS-mCherry (BL 63148, Bloomington Drosophila Stock Center), w; hsFLP, Act5C>y⁺>GAL4, UAS-GFP (Sekyrova et al., 2010), w; hsFLP, Act5C>y⁺>GAL4, UAS-GFP, UAS-Atf3^WT (Donohoe et al., 2018), w; hsFLP, Act5C>y⁺>GAL4, UAS-myr-mRFP, UAS-Atf3^WT (recombined in this study), w;; Ubi-GFP.E2f1^1-230, Ubi-mRFP1.NLS.CycB^1-266 (Zielke et al., 2014), w;; BcF6:GFP (Tokusumi et al., 2009), w; Bc¹ (Rizki et al., 1980), w; P[PTT-un1]vkgG454 (vkg::GFP, Morin et al., 2001), w; UAS-Mp^RNAi (v38189, VDRC), w; UAS-Mp^RNAi II (v35431, VDRC), w; UAS-Col4a1^RNAi (v28369, VDRC), w, UAS-trol^RNAi (v22642, VDRC), w; UAS-vkg^RNAi (v16986, VDRC), w; UAS-LanA^RNAi (v18873, VDRC), w; UAS-LanB1^RNAi (v23119, VDRC), w; UAS-LanB2^RNAi (v42559, VDRC), w; UAS-SPARC^RNAi (v16677, VDRC), w; UAS-Mp^RNAi, UAS-SPARC^RNAi (recombined in this study), w; Hml:DsRed; a58-GAL4, UAS-GFP (Galko and Krasnow, 2004, recombined in this study), w; Hml:DsRed; mef2-GAL4, UAS-GFP (Ranganayakulu et al., 1996, recombined in this study), w, elav-GAL4; Hml:DsRed, UAS-GFP (Lin and Goodman, 1994, recombined in this study), eater¹ (Bretscher et al., 2015), w; C7-GAL4, UAS-GFP, Hml:DsRed, UAS-atf3^WT; eater¹ (recombined in this study), w; HmlΔ-GAL4, UAS-GFP (Sinenko and Mathey-Prevot, 2004), w;; UAS-eater^RNAi (v4301, VDRC), w;; UAS-Mp::GFP (this study).

Third instar larvae were collected and washed in PBS and placed in glass dissection dishes filled with PBS on ice for 15 min for immobilization. Imaging was performed with an Olympus SZX-16 microscope fitted with a DP72 camera. GFP and RFP images were captured with the cellSens standard v1.11 software (Olympus, RRID:SCR_014551).

Stereotypical sessile tissue patterns were generated from larvae imaged under identical conditions. One larva was selected as a reference and every other image was aligned to match the reference in Adobe Photoshop CS5.5 (Adobe Systems, Inc, RRID:SCR_014199) using the Puppet warp tool. Aligned images were projected in Fiji v1.52i (Schindelin et al., 2012, RRID:SCR_003070) with ‘Average intensity’ projection, using the ‘Fire’ lookup table to enhance visualization.

For confocal imaging of whole fixed larvae, third instar wandering larvae were washed in 70% ethanol and were then injected with 4% paraformaldehyde with a sharpened glass injection capillary. The fixed larvae were immediately placed on a glass slide with a double-sided tape and imaged with an Olympus FV-1000 confocal microscope, with an UPlanSApo 10x (NA0.40) objective. Images were taken using the multi-area module of the FluoView FV-10ASW (Olympus, RRID:SCR_014215) software. Overlapping images were stitched in Fiji v1.52i and all single Z-planes were exported for both GFP (fat body) and DsRed (hemocytes) channels. Each individual confocal section was examined for GFP expression (fat body) and Hml:DsRed expression (hemocyte). The DsRed channel was locally recolored on each section when it was determined to overlap with the GFP signal to amber using Adobe Photoshop CS5.5, then all Z-planes were maximum projected. Sessile- and lymph gland hemocytes are shown in magenta.

Fat body dissection from third instar larvae was performed in PBS by opening the posterior of the larvae and inverting the carcass. After removal of the intestine, larvae were fixed in 4% paraformaldehyde for 1.5 hr. Larval epidermal fillet samples were prepared according to Brent et al., 2009, fixed in 4% paraformaldehyde, followed by the gentle removal of most body wall muscles. Tissues were blocked in 0.5% BSA (A3059, Sigma-Aldrich) supplemented with 0.1% TritonX-100 (T8787, Sigma-Aldrich) in PBS. Wing disc dissection was performed as described in Donohoe et al., 2018. Primary antibody staining was performed overnight at 4°C on a nutating mixer with anti-α-Tubulin (mouse, 1:200, DSHB, AA4.3, RRID:AB_579593), anti-Hemese (mouse, 1:100, Kurucz et al., 2003), anti-L1 (mouse, 1:100, Kurucz et al., 2007b), anti-NimC1 (mouse, 1:100, Kurucz et al., 2007a), anti-Laminin (rabbit, 1:500, Abcam, ab47651, RRID:AB_880659), anti-phospho-histone H3 (rabbit, 1:500, Cell Signaling, Cat# 9701, RRID:AB_331535), anti-Endostatin (rat, 1:200, Harpaz et al., 2013) and anti-cleaved-Dcp-1 (rabbit, 1:500, Cell Signaling, Cat# 9578, RRID:AB_2721060) antibodies diluted in the blocking solution. After washing, the samples were incubated with the corresponding Cy3- or Cy5-conjugated secondary antibodies (Jackson ImmunoResearch) for 1.5 hr at room temperature and counterstained with DAPI (1 μg/ml, 6335.1, Carl Roth GmbH) to visualize nuclei. The fat bodies were dissected from the carcass and mounted in Dabco-Mowiol (Sigma-Aldrich). Tissues were imaged on an Olympus FV-1000 confocal microscope with UPlanSApo 10x (NA0.40), UPlanSApo 20x (NA0.75), UPlanSApo-O 20x (NA0.85), UPlanFLN-O 40x (NA1.30) and UPlanSApo-O 60x (NA1.35) objectives using the FluoView FV-10ASW software. All fat body images represent Z-projections (unless otherwise indicated), which were generated with Fiji v1.52i.

For the pseudocoloring of hemocyte and adipocyte nuclei, confocal Z-stacks of the DAPI channel were divided into surface sections (~0–8 μm from the surface) and deeper sections (>8 μm from the surface), which were projected and exported separately using Fiji v1.52i. The surface projection containing the hemocyte nuclei was then colored amber, while the deeper projection with the adipocyte nuclei was colored magenta. The two projections were merged in Adobe Photoshop CS5.5.

Flies were reared on the following diet: 8% cornmeal, 1% agar, 4% yeast, 5% saccharose and 0.16% methylparaben. Fifty early third instar larvae were placed in an infection chamber with 100 female L. boulardi wasps for 15 min (Bajgar et al., 2015). After the removal of the wasps, the larvae were kept at 25°C for 24 or 48 hr before bleeding or dissection.

Third instar larvae were bled into 1x PBS containing 0.01% n-phenylthiourea (31056, Sigma-Aldrich) on 12 spot glass slides (HM-101, Hendley-Essex). Allowing hemocytes to adhere for 45 min, samples were then fixed in acetone. Blocking was performed with 0.5% BSA in 1x PBS for 30 min. Hemocytes were stained with anti-Hemese (mouse, 1:100, Kurucz et al., 2003) or anti-L1 (mouse, 1:100, Kurucz et al., 2007b) antibodies and anti-mouse-Cy3 conjugated secondary antibodies (1:500, Jackson Immunoresearch). Nuclei were counterstained with DAPI. Samples were mounted on glass slides in Dabco-Mowiol 4–88 and imaged on an Olympus FV-1000 confocal microscope with UPlanSApo 10x (NA0.40) and UPlanSApo-O 20x (NA0.85) objectives.

Fat bodies were fixed in 2.5% glutaraldehyde diluted in 100 mM phosphate buffer (PB), washed in 100 mM PB and postfixed in 2% osmium tetroxide in PB for 1 hr on ice. Contrasting was performed with 2% uranyl acetate, after which the samples were dehydrated in ethanol and embedded in acetone-resolved araldite. Electron microscopy was performed with an EM 109 (Zeiss) microscope.

To express C-terminally GFP- or FLAG-tagged Eater and Multiplexin proteins under UAS control, eater and Mp cDNA was cloned into pENTR4-dual vector between BamHI and NotI sites, and subsequently recombined using LR Clonase II (11791–020, Life Technologies) into pTWG (Mp and Eater) and pTWF (Mp) vectors, respectively (T. Murphy, Drosophila Genomic Resource Center). The UAS-Mp::GFP transgenic flies were obtained by standard P-element-mediated germline transformation of pTWG-Mp plasmid into w¹¹¹⁸ Drosophila embryos (Genetics Fly Facility, The University of Cambridge, UK). Multiple transformants were recovered and tested, all showing comparable Mp::GFP expression.

Schneider 2 (S2) cells (Drosophila Genomic Resource Center, RRID:CVCL_Z992) were cultured at 25°C in Shields and Sang M3 insect medium (S8398-1L, Sigma-Aldrich) containing 8% fetal bovine serum (Gibco, Life Technologies) without antibiotics. There are no verified reports of Mycoplasma infection in S2 cells (Cherbas and Gong, 2014). S2 cells were only used to express transgenic proteins for biochemical experiments. All functional data were obtained from in vivo studies in a Drosophila melanogaster model. Cells were transfected using TransIT Insect transfection reagent (MIR 6100, Mirus). Expression of UAS-driven cDNAs was induced by co-transfection with a pWA-GAL4 plasmid expressing GAL4 under an actin5C promoter. Cells were lysed 36 hr after transfection in lysis buffer containing 50 mM Tris-HCl (pH 7.8), 150 mM NaCl, 1 mM EDTA (pH 8.0), 1% Triton X-100, 0.01% Igepal, and protease inhibitors (Roche Applied Science). Protein concentration was quantified using Bradford reagent (K015.1, Roth GmBH) according to manufacturer’s instructions.

For each sample, 1 mg of S2 cell protein lysate was incubated with GFP-Trap beads (gtma-20, Chromotek, RRID:AB_2631406) overnight. Following five washes with Lysis buffer, bound proteins were eluted with Glycine-HCl on 37°C, followed by neutralization. Proteins resolved on 10% SDS-PAGE were detected by immunoblotting with anti-Flag M2 (mouse, 1:1000, F1804, Sigma Aldrich, RRID:AB_262044), anti-GFP (rabbit, 1:5000, TP401, Acris, RRID:AB_2313770) and anti-α-Tubulin (mouse, 1:1000, DSHB, AA4.3, RRID:AB_579593) antibodies, followed by incubation with corresponding HRP-conjugated secondary antibodies (Jackson Immuno Research). Chemiluminescent signal was captured using ImageQuant LAS4000 reader (GE Healthcare, RRID:SCR_014246).

Hemolymph was collected from twenty third instar larvae per replicate in 100 μL PBS. After centrifugation, 5000 rpm for 10 min, the supernatant was precipitated with 100 μL of ice cold acetone for 1 hr at −20°C. Afterwards, the proteins were pelleted and resuspended in 50 μL in Lysis buffer (see above), and equal volumes were resolved on 10% SDS-PAGE. Immunoblotting was carried out with anti-Endostatin (rat, 1:1000, Harpaz et al., 2013) and anti-PPO1 (rabbit, 1:750, Jiang et al., 1997) antibodies.

Fat body associated hemocytes were quantified based on Z-projection images taken with Olympus FV-1000 confocal microscope fitted with UPlanSApo 20x (NA0.75) objective. The area imaged for comparisons is depicted in Figure 1—figure supplement 1B. The area selected is devoid of hemocytes in wild-type larvae, and lacks non-adipose tissue, such as the gonad precursors or the tracheal branches found at the posterior end. Quantification was performed manually using the CellCounter plugin in Fiji v1.52. For quantification of circulating hemocytes images were acquired using the UPlanSApo 10x (NA0.40) objective. One representative image per larva was processed for cell counting. DAPI positive nuclei were counted with CellProfiler (Kamentsky et al., 2011, RRID:SCR_007358), using a customized pipeline. Lamellocyte percentage was determined by comparing the number of L1-positive cells to all counted DAPI-positive nuclei.

For the quantification of sessile hemocytes, whole larval images aligned to a reference (see ‘Whole larval imaging’ section) were analyzed. These images were loaded in Fiji v1.52i as hyperstacks, containing both control larvae and the genotypes of interest. Using the Rectangle tool, an area was selected where no hemocytes were present, where the average pixel intensity was measured for each control larva and averaged across larvae to establish background levels. Using the Rectangle tool, selections corresponding to single sessile stripes in the A4, A5, A6 and A7 segments were made on control larvae, and the average intensity of the Hml:DsRed signal was measured for the relevant genotypes within this selection. From every measurement the previously established background values were subtracted, then for each larva the four corrected intensity values were added together.

Statistical analysis and plotting were carried out with Prism 6 (GraphPad, RRID:SCR_002798), using one-way ANOVA assuming unequal variances with multiple comparisons or unpaired Student's t-test. For comparison of cell counts, one-way Kruskal-Wallis test with Dunn's multiple comparison or nonparametric two-tailed Mann-Whitney test were used. For fat-body-associated hemocytes, every data point represents the number of attached hemocytes within the quantified area, while for circulating hemocytes and sessile hemocyte quantifications, values were normalized to control mean (represented as 1) and shown on the graphs as fold change (F.C.). All experiments represent at least two independent temporal replicates.

Sample-size criteria was estimated post hoc with G*Power 3.1 (Faul et al., 2009). All significantly different datasets exceeded 0.99 Power (1-beta) with respective sample sizes, means and standard deviations.