To identify new examples of adult tissues in Drosophila that share cytoplasm, we ubiquitously expressed Cre and UAS-dBrainbow (Hampel et al., 2011; Figure 1A), a Cre-Lox-based system that randomly labels cells with only one of three fluorescent proteins. We used animals heterozygous for UAS-dBrainbow to ensure single-labeling of cells. We ubiquitously expressed Cre, which does not require heat-shock induction, from early embryonic stages (before cells endocycle to become polyploid). Cre-mediated excision occurs independently of Gal4 expression and Gal80^ts repression of dBrainbow. Therefore, we can ensure that multi-labeled cells only arise by cytoplasm sharing between cells not related by cell division or incomplete cytokinesis (Figure 1B). We examined a wide range of tissues (Figure 1—figure supplement 1A). From our screen, we discovered that the rectal papilla is a new example of a tissue with cytoplasm sharing. Adult Drosophila contain four papillae, each with 100 nuclei of genome content between 8 and 16C (Fox et al., 2010), that reside in the posterior hindgut (Figure 1C). Each papilla is a polarized epithelial cone with the apical region facing the gut lumen and the basal region surrounding a central canal that connects to the fly’s hemolymph (Figure 1D). The papillar structure supports its function to reabsorb water, ions, and small molecules from the gut lumen and recycle them back to the hemolymph (Cohen et al., 2020). Knowing that adult papillar cells share cytoplasm, we next used our dBrainbow system to identify when papillar cells begin to share relative to other developmental events that we previously identified (Figure 1E). Using both fixed and live imaging of whole organs, we found that at 62 hours post-puparium formation (HPPF), each papillar cell contains only one dBrainbow label (Figure 1F). By contrast, at 69HPPF, multi-labeled cells are apparent (Figure 1F’,H–H’). We quantitatively measured papillar sharing across the tissue (Figure 1—figure supplement 1B, Materials and methods) and found that cytoplasm sharing initiates over a narrow 6 hr period (68-74HPPF, Figure 1G). Our results suggested that at least RNA and possibly protein passes between papillar cells to facilitate cytoplasm sharing. To directly test if protein is shared, we photo-activated GFP (GFP^PA) in single adult papillar cells and observed in real time whether GFP^PA spreads to adjacent cells. We find the principal papillar cells, but not the secondary cells at the papillar base (Garayoa et al., 1999; Figure 1—figure supplement 1C), share protein across an area of at least several nuclei (Figure 1I–I’). We next tested whether a larger protein can be shared between papillar cells. We used rectal papillae RNA-sequencing data (Leader et al., 2018) to identify proteins that are endogenously expressed, cytoplasmic, and relatively large. We therefore generated flies expressing a UAS-inducible, photoactivatable GFP fused to Glyceraldehyde 3 phosphate dehydrogenase 2 (UAS-Gapdh2-GFP^PA). This construct should produce a tagged protein of 62.3 kDa. We found that Gapdh2-GFP^PA protein is shared between cells, as it never stops at a papillar cell–cell boundary, though it may move at a slower rate than GFP^PA (Figure 1—figure supplement 1D). Therefore, proteins as large as ~62 kDa (the size of GFP-tagged Gapdh2) can move across an area covered by multiple papillar nuclei. Additionally, the movement of our Gapdh2 transgenic protein indicates that papillar cells likely share endogenously expressed proteins. These results indicate that papillae undergo a developmentally programmed conversion from 100 individual cells to a single giant multinuclear cytoplasm that shares the products of ~1200 genomes.

We next examined whether cytoplasm sharing requires either programmed endocycles or mitoses. We have previously shown that larval papillar cells first undergo endocycles, which increase cellular ploidy. Then, during metamorphosis, pupal papillar cells disassemble polytene chromosomes and undergo polyploid mitotic cycles, which increase cell number (Fox et al., 2010; Stormo and Fox, 2016; Stormo and Fox, 2019). Both endocycles and mitoses occur well prior to the start of papillar cytoplasm sharing (Figure 1E). Papillar endocycles require the Anaphase-Promoting Complex/Cyclosome regulator fizzy-related (fzr) while the papillar mitoses require Notch signaling (Schoenfelder et al., 2014). Knockdown of fzr significantly disrupts cytoplasm sharing (Figure 1—figure supplement 1E,F,H). We hypothesize that endocycles are required for differentiation of the papillae, which later enables these cells to trigger cytoplasm sharing. In contrast, blocking Notch signaling, which initiates papillar mitotic divisions (Fox et al., 2010), does not prevent sharing (Figure 1—figure supplement 1E,G,H). Thus, papillar cytoplasm sharing requires developmentally programmed endocycles but not mitotic cycles.

As our dBrainbow approach only identifies cytoplasm sharing events that do not involve incomplete division/cytokinesis, we examined whether sharing results from fusion pore formation, as in skeletal muscle. A well-studied model of such cell–cell fusion in Drosophila is myoblast fusion, which requires an actin-based podosome (Richardson et al., 2007; Sens et al., 2010). We conducted a candidate dBrainbow-based RNAi screen (77 genes, Figure 2A, Table 1) of myoblast fusion regulators and other plasma membrane components. Remarkably, 0/15 myoblast fusion genes from our initial screen regulate papillar cytoplasm sharing (Figure 2A, Figure 2—figure supplement 1A, Table 1). Furthermore, dominant-negative forms of Rho family GTPases have no impact on dBrainbow labeling (Figure 2—figure supplement 1B), providing additional evidence against actin-based cytoplasm sharing. Instead, we found 8/77 genes, including subunits of the vacuolar H+ ATPase (Vha16-1), ESCRT-III complex (Vps2), and exocyst (Exo84) (Figure 2A) are required for papillar cytoplasm sharing. Through additional screening, the only myoblast fusion regulator required for papillar cytoplasm sharing is singles bar (sing), a presumed vesicle trafficking gene (Estrada et al., 2007; Figure 2—figure supplement 1A). Given the enrichment of our candidate screen hits in membrane trafficking and not myoblast fusion, we further explored the role of membrane trafficking in cytoplasm sharing.

We conducted two secondary dBrainbow screens to find specific membrane trafficking pathway components that regulate papillar sharing. First, a focused candidate membrane trafficking screen revealed additional components (12/36 genes screened, Figure 2B, Table 2) including three more vacuolar H+ ATPase subunits, five more exocyst components, and the Dynamin GTPase shibire (shi) (Figure 2B,D,E,H). Second, we screened constitutively-active and dominant-negative versions of all 31 Drosophila Rabs. Sharing requires only a small number of Rabs, specifically the ER/Golgi-associated Rab1, the early endosome-associated Rab5, and the recycling endosome-associated Rab11 (Figure 2C,D,F–H). Given our identification of the membrane vesicle recycling circuit involving shi, Rab5, and Rab11, we focused on these genes. Two unique RNAi lines for each gene show consistent sharing defects, and most of these knockdowns completely recapitulate the pre-sharing state (Figure 2H). Despite exhibiting strong cytoplasm sharing defects, shi, Rab5, and Rab11 RNAi papillae appear morphologically normal, with only minor cell number decreases (Figure 2—figure supplement 1C). These results suggest that membrane recycling GTPases regulate a specific developmental event associated with cytoplasm sharing, and not papillar morphogenesis. In agreement with these GTPases acting during development, rather than as part of an ongoing transport process, GTPase knockdown after sharing onset does not block cytoplasm sharing (Figure 2—figure supplement 1D–F). Together, our screens reveal that membrane trafficking, particularly Dynamin-mediated endocytosis and early/recycling endosome trafficking, regulates papillar cytoplasmic sharing.

To better understand how membrane trafficking GTPases initiate cytoplasm sharing during development, we examined endosome and Shi localization during sharing onset. We imaged a GFP-tagged pan-endosome marker (myc-2x-FYVE), overexpression of which should not alter endosome shape or localization (Gillooly et al., 2000; Wucherpfennig et al., 2003), and a Venus-tagged shi before and after sharing. Endosomes are evenly distributed shortly before sharing, but become highly polarized at the basal membrane around the time of sharing onset (Figure 3A–A’,C, Figure 3—figure supplement 1A). This basal endosome repositioning requires Shi (Figure 3B–C, Figure 3—figure supplement 1A) and the change in endosome localization is attributed to Rab5-positive early endosomes (Figure 3—figure supplement 1B–B’’). Additionally, Shi localization changes from apical polarization to a uniform distribution during sharing onset (Figure 3D–E). These localization changes indicate that membrane trafficking factors which regulate cytoplasm sharing are highly dynamic during cytoplasm sharing onset.

To determine what membrane remodeling events underlie GTPase-dependent cytoplasm sharing, we turned to ultrastructural analysis. Adult ultrastructure and physiology of papillar cells has been examined previously in Drosophila (Wessing and Eichelberg, 1973) and related insects (Gupta and Berridge, 1966). These cells contain elaborate membrane networks that facilitate selective ion resorption from the gut lumen, facing the apical side of papillar cells, to the hemolymph, facing the basal side. Still, little is known about developmental processes or mechanisms governing the unique papillar cell architecture. We looked for changes in cell–cell junctions and lateral membranes that coincide with cytoplasm sharing, especially to determine if there is a physical membrane breach between cells. We identified several dramatic changes in membrane architecture. First, apical microvilli-like structures form during sharing onset (Figure 3F–F’’). Just basal to the microvilli, apical cell–cell junctions are straight in early pupal development and compress into a more curving, tortuous morphology around the time of cytoplasm sharing onset (Figure 3—figure supplement 1C–C’’). One of the most striking changes, coincident with Shi re-localization, is formation of pan-cellular endomembrane stacks surrounding mitochondria. These stacks are likely sites for active ion transport, such as that mediated by the P-type Na⁺/K⁺-ATPase, coupled to mitochondria for ATP (Figure 3G–G’’; Berridge and Gupta, 1967; Patrick et al., 2006). Thus, massive apical and intracellular plasma membrane reorganization coincides with both cytoplasm sharing and Shi/endosome re-localization. We next assessed whether the extensive membrane remodeling requires Shi, Rab5, and Rab11. In shi and Rab5 RNAi animals, microvilli protrude downward, instead of upward (Figure 3H–J). Additionally, apical junctions do not compress as in controls (Figure 3—figure supplement 1D–F). Notably, membrane stacks are greatly reduced (Figure 3K–M). shi RNAi animals exhibit numerous trapped vesicles, consistent with a known role for Dynamin in membrane vesicle severing (Damke et al., 1994; Hinshaw and Schmid, 1995; Figure 3L, inset). Together, we find that Shi and endosomes extensively remodel membranes during papillar cytoplasm sharing.

Our extensive ultrastructural analysis did not reveal any clear breaches in the plasma membrane, despite numerous membrane alterations. Adult papillae exhibit large extracellular spaces between nuclei that eliminate the possibility of cytoplasm sharing throughout much of the lateral membrane (Figure 3—figure supplement 2A; Wessing and Eichelberg, 1973; Gupta and Berridge, 1966). Instead, through our GTPase knockdown studies, we identified a striking alteration in the apical cell–cell interface that strongly correlates with cytoplasm sharing. Specifically, shi animals frequently lack apical gap junctions (Figure 3N–O) (p<0.0001) (Figure 3P, Figure 3—figure supplement 1H–H’’). Upon closer examination of control animal development, we find that apical gap junction-like structures arise at cytoplasm sharing onset. There is almost no gap junction-like structure before cytoplasm sharing (Figure 4A–B, Figure 2—figure supplement 1A–A’’). Given our electron micrograph results, we determined which innexins, the protein family associated with gap junctions in invertebrates (Bauer et al., 2005; Phelan et al., 1998), are expressed in rectal papillae. From RNA-seq data (Methods), we determined that ogre (Inx1), Inx2, and Inx3 are most highly expressed (Figure 4C). This combination of innexins is not unique to rectal papillae; the non-sharing brain and optic lobe (Figure 4—figure supplement 1A) also express high levels of all three (Leader et al., 2018). We examined localization of Inx3 (a gap junction component) (Curtin et al., 1999; Richard et al., 2017), and compared it to a septate junction component, NeurexinIV (NrxIV) (Laprise et al., 2009). NrxIV localizes similarly both pre and post-sharing onset (Figure 4D–D’), indicative of persistent septate junctions remaining between papillar cells. In contrast, Inx3 organizes apically only after cytoplasm sharing (Figure 4E–E’, Figure 4—figure supplement 1B–B’). Inx3 also does not localize to cell–cell boundaries in shi RNAi animals (Figure 4C–C'). We tested whether innexins are required for cytoplasm sharing. Knocking down these three genes individually causes mild yet significant cytoplasm sharing defects (Figure 4F). However, we see larger defects in animals expressing dominant-negative ogre^DN (Figure 4F–G; Spéder and Brand, 2014), which contains a N-terminal GFP tag that interferes with channel passage. Also, heterozygous animals containing a ten gene-deficiency spanning ogre, Inx2, and Inx7 have more severe defects (Figure 4F, Df(1)BSC867). Finally, we tested whether cytoplasm sharing is essential for normal rectal papillar function. Rectal papillae selectively absorb water and ions from the gut lumen for transport back into the hemolymph, and excrete unwanted lumen contents (Cohen et al., 2020). One test of papillar function is viability following the challenge of a high-salt diet (Bretscher and Fox, 2016; Schoenfelder et al., 2014). However, with our pan-hindgut driver byn-Gal4 used for all previous experiments, we noted animal lethality with shi, Rab5, and Rab11 knockdown within a few days on control food. We observed melanization and necrosis throughout the hindgut (data not shown) which prevented us from attributing any phenotypes directly to papillar cytoplasm sharing. We therefore identified an alternative driver (60H12-Gal4) with rectum-specific expression during pupation and adulthood (Figure 4—figure supplement 1D–D’). We used this driver to express shi^DN. These animals display similar sharing defects as we find with byn-Gal4 (Figure 4—figure supplement 1E–E’’). Reassuringly, 60H12-Gal4 > shi^DN animals do not show lethality on a control food diet (Figure 4H) allowing us to test rectal papillar physiological function on a high-salt diet. Using either pan-hindgut or papillae-specific knockdown of cytoplasm sharing regulators, we find both shi^DN and ogre^DN animals are extremely sensitive to the high-salt diet (mean survival <1 day, Figure 4H). These results underscore an important function for gap junction proteins, as well as membrane remodeling by Dynamin/Shibire, in cytoplasm sharing.

Our findings identify Drosophila rectal papillae as a new and distinctive example of cytoplasm sharing between multiple nuclei in a simple, genetically tractable system. One defining property of papillar cytoplasm sharing is the lack of an easily observable conduit in the lateral membrane through which cytoplasm can be exchanged. Cytoplasm sharing in a multinucleate tissue/organism frequently involves the creation of a large membrane breach associated with major actin cytoskeleton rearrangement (Kim et al., 2015; Deng et al., 2017; Martin, 2016). However, papillar cytoplasm sharing does not require canonical myoblast fusion regulators nor major actin remodeling factors such as Rho family GTPases. Aside from membrane breaches, other cell types are known to share cytoplasm through the formation of cytoplasmic bridges such as ring canals or plasmodesmata. Such bridge structures assemble as the result of incomplete cytokinesis (Mahowald, 1971; Lůcas and Wolf, 1993). In contrast, papillar cytoplasm sharing does not require mitosis or cytokinesis, and does not contain intercellular bridge structures visible by electron microscopy.

In addition to lacking a large, observable membrane breach, papillar cytoplasm sharing occurs within an intact, polarized epithelium, and apical cell–cell junctions and lateral membranes are retained after the onset of sharing. In contrast, other epithelia known to fuse cytoplasm, such as C. elegans epithelia fused by Epithelial Fusion Failure 1 (EFF-1), dismantle cell–cell junctions (Smurova and Podbilewicz, 2016). Further, cells with ring canals retain cell–cell junctions and lateral membranes (Peifer et al., 1993).

Given the retention of cell junctions and absence of clear intercellular bridges, channels, or breaches in lateral membrane, our data lead us to propose that a specialized function of gap junction proteins facilitates cytoplasm sharing between neighboring cells in an otherwise intact epithelium (Figure 4I). Although gap junctions typically transfer molecules of <1 kDa, elongated proteins up to 18 kDa are observed to pass through certain vertebrate gap junctions (Cieniewicz and Woodruff, 2010). Alternatively, gap junction-mediated cell to cell communication has been previously implicated in fusion of placental trophoblasts and osteoclasts (Firth et al., 1980; Dunk et al., 2012; Schilling et al., 2008), so we cannot rule out an indirect role for gap junctions in papillar cells, such as through regulation/recruitment of a fusogenic protein (Petrany and Millay, 2019). Future work beyond the scope of this study can determine if, for example, papillar gap junctions exhibit a specialized structure to directly facilitate exchange of large cytoplasmic contents. As for the connection between membrane remodeling and gap junction formation, Rab11 has been previously reported to recycle gap junction components in Drosophila brain and mammalian cell culture (Augustin et al., 2017). Dynamin2 was also implicated in gap junction plaque internalization in mammalian cells (Gilleron et al., 2011). However, neither of these factors has been previously implicated in gap junction establishment. We show that Dynamin is required for gap junction formation in papillar cells. Future studies will determine the exact role of Dynamin in gap junction establishment. Another clue for future study is that papillar cytoplasm sharing is developmentally regulated, occurring over a brief 6 hr window, and requires membrane remodeling by trafficking GTPases and gap junction establishment (Figure 4I, Figure 4—figure supplement 1H). Our results argue that papillar sharing is triggered by a permanent structural rearrangement rather than an active transport mechanism, as the membrane remodelers we identified are required specifically during developmental membrane remodeling.

The mechanisms we report here may be relevant to other emerging roles for membrane remodeling and cytoplasm sharing in the literature. Here, we identify a close relationship between the formation of membrane stacks and cytoplasm sharing. Basolateral membrane infoldings to expand cellular surface area are a common feature of absorptive cells (Pease, 1956). The mammalian kidney tubule cells exhibit similar basolateral membrane extensions to which ion transporters such as the Na+/K+-ATPase are localized (Maunsbach, 1966; Molitoris et al., 1992; Avner et al., 1992; Pease, 1955; Sjöstrand and Rhodin, 1953). Our results suggest that the same membrane remodeling factors that regulate cytoplasm sharing are required for the formation of membrane stacks. To our knowledge, this is the first study to reveal factors involved in basolateral membrane infolding biogenesis. Additionally, our results may also explain other examples of cytoplasm sharing where the underlying mechanism remains to be determined, such as transient cytoplasm sharing in the zebrafish myocardium (Sawamiphak et al., 2017). Together, our studies indicate that the Drosophila papillar epithelium represents a distinctive example of cytoplasmic sharing to generate giant multinucleate cells.

Our results have several implications for functions and regulation of multinucleation. Here we show that the membrane and junctional changes associated with cytoplasm sharing are required for normal Drosophila rectal papillar function. Papillae in other insects are known to undergo visible movement upon muscle contraction, which may facilitate cytoplasm movement (Lowne, 1869). Arthropod papillar structures are subject to peristaltic muscle contractions from an extensive musculature (Rocco et al., 2017), which aid in both excretion and movement of papillar contents into the hemolymph (Habas mantel and Mantel, 1968). Further, relative to other hindgut regions, the rectum appears to have specialized innervation (Cohen et al., 2020) and regulation by the kinin family of neuropeptides, which are hypothesized to provide additional input in to muscle activity in this critical site of reabsorption (Audsley and Weaver, 2009; Lajevardi and Paluzzi, 2020). We speculate that these muscle contractions aid in vigorous movement of papillar cytoplasm, which includes ions and water taken up from the intestinal lumen. The movement of these papillar contents may facilitate both cytoplasm exchange between papillar cells and the interaction of ions and ion transport machinery with intracellular membrane stacks. This idea is supported by our finding that animals lacking a large common papillar cytoplasm die when fed a high-salt diet.

Given the importance of insect papillae in pathogen biology, the knowledge that this common anatomical structure is a shared cytoplasm can impact both human disease intervention and agricultural pest control. Papillae occur in both primitive insect orders such as Zygentoma and Odonata and also in Lepidopterans, Hymenopterans, and Dipterans, the latter of which exhibit the most prominent and elaborate structures (Palm, 1949). Furthermore, electron micrographs of the hindgut of the mosquito, Aedes aegypti, and the ant, Formica nigricans, show striking ultrastructural similarity to Drosophila, and these studies leave open the possibility that multinucleation may be conserved in insect papillae (Hopkins, 1967; Wessing and Eichelberg, 1973; Garayoa et al., 1999). Cytoplasm sharing is a known mechanism that facilitates pathogen spread (Eugenin et al., 2009), and papillae are an avenue of entry for numerous pathogens including kinetoplastids and mosquito viruses (Gu et al., 2010; Filosa et al., 2019). Thus, our findings may impact strategies to prevent diseases such as African sleeping sickness, or to target agricultural pests that threaten agricultural production.

The sharing of cytoplasm also has the potential to neutralize detrimental genomic imbalances between nuclei caused by aneuploidy. Our prior work (Schoenfelder et al., 2014) revealed that papillae are highly tolerant of chromosome mis-segregation, and our work here suggests this tolerance may be due in part to neutralization of aneuploidies through cytoplasm sharing. This finding may also be relevant to the study of multinucleate tumors, such as those found in pancreas, bone, and fibrous tissues (Doane et al., 2015; Hasegawa et al., 2017; Mancini et al., 2017), or to conditions of aberrant organelle inheritance (Asare et al., 2017). Finally, we note that our study reveals that, even in a well-studied model organism such as Drosophila, we still have yet to appreciate the full diversity of tissue organization strategies. Our Brainbow-based approach could be applied to other contexts to identify other tissues with cytoplasm sharing, including those with gap junction-dependent but membrane breach-independent cytoplasm sharing. Collectively, our findings highlight the expanding diversity of multicellular tissue organization strategies.

Flies were raised at 25°C on standard media (Archon Scientific, Durham, NC) unless specified otherwise. See Table 4 for a list of fly stocks used. See Table 3 for a full list of fly lines screened in primary and secondary screens. See Table 5 for panel-specific genotypes.

The UAS-Gapdh2-GFP^PA construct was generated by gene synthesis (Twist Biosciences). The GFP was placed at the C-terminus with a 12-amino acid fusion linker (GSAGSAAGSGEF) (Waldo et al., 1999) codon-optimized for Drosophila. This insert was then cloned into the pBID-UASC-FG vector modified to remove the FLAG tag and extraneous cloning sites. Transgenic flies were generated at Duke University. brachyenteron (byn)-Gal4 was the driver for all UAS transgenes with the exception of the screen in Figure 1—figure supplement 1A, which used tub-Gal4, and the shi knockdown in Figure 4H, which used 60H12-Gal4. 60H12-Gal4 expresses only in the papillar cells and not the rest of the hindgut, and use of this driver blocks cytoplasm sharing using UAS-shi^DN (Figure 4—figure supplement 1D–E’’). For all Gal4 experiments, UAS expression was at 29°C, except in Figure 1F–H, where it was at 25°C. If byn-Gal4 expression of a given UAS-transgene was lethal, the experiment was repeated with a temperature-sensitive Gal80^ts repressor transgene and animals were kept at 18°C until shifting to 29°C at an experimentally-determined time point that would both result in viable animals and permit time to express the transgene prior to sharing onset.

For salt feeding assays, age- and sex-matched siblings were transferred into vials containing 2% NaCl food made with Nutri-Fly MF food base (Genesee Scientific) or control food (Schoenfelder et al., 2014). Flies were monitored for survival each day for 10 days.

For fixed imaging, tissues were dissected in PBS and immediately fixed in 3.7% formaldehyde + 0.3% Triton-X for 15 min. Immunostaining was performed in 0.3% Triton-X with 1% normal goat serum (Fox et al., 2010). The following antibodies were used: Rabbit anti-GFP (Thermo Fisher Scientific, Cat#A11122, 1:1000), Rat anti-HA (Roche, Cat#11867423001, 1:100), Rabbit anti-Inx3 (generous gift from Reinhard Bauer, 1:75), [Lehmann et al., 2006], 488, 568, 633 secondary antibodies (Thermo Fisher Scientific, Alexa Fluor, 1:2000). Tissue was stained with DAPI at 5 μg/ml and mounted in VECTASHIELD Mounting Media on slides.

All image analysis was performed using ImageJ and FIJI (Rueden et al., 2017; Schindelin et al., 2012).

Statistical analysis was performed in GraphPad Prism 8. Detailed statistical tests and methods are reported in Table 6.

Some additional methodological details, including animal genotype, applied to only a specific figure panel. Please see Table 6 for this information.