Common and Distinct Genetic Properties of ESCRT-II Components in Drosophila

Background Genetic studies in yeast have identified class E vps genes that form the ESCRT complexes required for protein sorting at the early endosome. In Drosophila, mutations of the ESCRT-II component vps25 cause endosomal defects leading to accumulation of Notch protein and increased Notch pathway activity. These endosomal and signaling defects are thought to account for several phenotypes. Depending on the developmental context, two different types of overgrowth can be detected. Tissue predominantly mutant for vps25 displays neoplastic tumor characteristics. In contrast, vps25 mutant clones in a wild-type background trigger hyperplastic overgrowth in a non-autonomous manner. In addition, vps25 mutant clones also promote apoptotic resistance in a non-autonomous manner. Principal Findings Here, we genetically characterize the remaining ESCRT-II components vps22 and vps36. Like vps25, mutants of vps22 and vps36 display endosomal defects, accumulate Notch protein and – when the tissue is predominantly mutant – show neoplastic tumor characteristics. However, despite these common phenotypes, they have distinct non-autonomous phenotypes. While vps22 mutations cause strong non-autonomous overgrowth, they do not affect apoptotic resistance. In contrast, vps36 mutations increase apoptotic resistance, but have little effect on non-autonomous proliferation. Further characterization reveals that although all ESCRT-II mutants accumulate Notch protein, only vps22 and vps25 mutations trigger Notch activity. Conclusions/Significance The ESCRT-II components vps22, vps25 and vps36 display common and distinct genetic properties. Our data redefine the role of Notch for hyperplastic and neoplastic overgrowth in these mutants. While Notch is required for hyperplastic growth, it appears to be dispensable for neoplastic transformation.


Introduction
Appropriate cell/cell signaling requires both coordinated activation and inactivation of cell surface signaling receptors. Usually, the receptors are activated by ligand binding upon which they induce an intracellular response including ubiquitination of the receptor which provides the signal for receptor internalization by endocytosis [1][2][3]. Endocytosis also controls the steady-state levels of cell surface receptors independently of ligand occupation. After endocytosis, the cell surface receptors are present at the early endosome. Because the intracellular domain of activated signaling receptors is exposed to the cytosol, the receptors are still able to signal. In fact, signaling from the endosomal location appears to be the preferred mode of several signaling pathways as it brings the receptor in close proximity to intracellular signaling complexes [4][5][6][7][8]. To fully inactivate the signaling receptors, a second form of internalization at the limiting membrane of the early endosome is necessary to form the multi-vesicular body (MVB) [3,[9][10][11][12][13][14]. In the MVB, the receptors are completely detached from the cytosol and stop signaling. Finally, the MVB fuses with lysosomes for proteolytic degradation.
Genetic studies in yeast have identified fifteen class E vps (vacuolar protein sorting) genes required for MVB formation [15]. These genes encode the components of four ESCRT (Endosomal Sorting Complex Required for Transport) protein complexes (reviewed by [3,9]). Hrs (Vps27) and STAM (Hse1) form ESCRT-0, which initiates the recruitment of the signaling receptor (the cargo) to the early endosome and delivers it to ESCRT-I. From there, the cargo is transferred to ESCRT-II and then to ESCRT-III. At ESCRT-III, the receptors are internalized into MVBs [3,9]. Loss of class E vps function in yeast leads to accumulation of ubiquitinated proteins on the limiting membrane of enlarged endosomes [12]. Biochemical studies in mammalian cells have revealed a similar function for endosomal protein sorting [3,9].
The phenotypic consequences of loss of class E vps genes in the context of a multi-cellular organism have just recently been unveiled. In Drosophila, mutants in hrs, erupted (ept, encoding the ESCRT-I component vps23) and vps25 (a component of ESCRT-II) have recently been described. These mutants are characterized by enlarged endosomes which contain increased protein levels of Notch, Delta, EGFR, Patched, Smoothened, and Thickveins (the Drosophila TGFb type 1 receptor) [16][17][18][19][20][21]. Despite these common endosomal defects, hrs, ept and vps25 display different phenotypes at the organismal level. While hrs mosaics do not display any obvious adult phenotypes, ept and vps25 mosaics are characterized by overgrown adult eyes and heads, and overgrown larval imaginal discs due to hyperplastic proliferation. Hyperplastic proliferation refers to increased proliferation and overgrowth; however, hyperplastic cells still maintain epithelial polarity and will eventually stop proliferating [22]. Interestingly, this hyperplastic growth does not occur in ept and vps25 mutant tissue itself. Instead, it occurs in wild-type cells immediately abutting the mutant tissue [18][19][20][21]. This non-autonomous hyperplastic proliferation is caused by increased Notch activity at the ept and vps25 endosomes which stimulates neighboring cells to undergo proliferation by activating the Jak/STAT pathway [23][24][25]. Increased Notch activity has not been observed in hrs mutants despite the accumulation of Notch protein, explaining the lack of hyperplastic overgrowth in hrs mutants.
In addition to non-autonomous hyperplastic growth in genetic mosaics, ept and vps25 mutations can cause neoplastic overgrowth. Neoplastic cells lose epithelial polarity and fail to stop proliferating giving rise to significant overgrowth [22]. ept and vps25 mutants show neoplastic overgrowth if almost the entire imaginal disc is mutant [18,20,26]. Neoplastic overgrowth can also be induced in vps25 mosaic tissue, if apoptosis is blocked in vps25 mutant cells [19,21]. Under both conditions, neoplastic growth occurs in an autonomous manner, i.e. in the mutant tissue [19,21]. These findings were significant for a better understanding of tumor formation caused by inactivation of Tsg101 (tumor susceptibility gene 101), the human vps23 homolog, which has been implicated in cervical, breast, prostate and gastrointestinal cancers [27][28][29][30][31].
In addition, although vps25 mutant cells undergo apoptosis, before they die they can increase the apoptotic resistance of neighboring cells through up-regulation of the apoptosis inhibitor Diap1 (Drosophila Inhibitor of Apoptosis Protein 1) [21].
Except for vps25, a genetic analysis of the ESCRT-II components for endosomal protein sorting in metazoan organisms has not been reported. Here, we characterize and compare the mutant phenotypes of the individual components of the ESCRT-II complex, vps22 (also called larsen [32]), vps25 and vps36 in Drosophila. The ESCRT-II complex is a heterotetramer composed of two Vps25 subunits, and one subunit each of Vps22 and Vps36 [33,34]. We show that mutant cells of the three ESCRT-II components display endosomal defects and accumulate Notch protein. Moreover, imaginal discs predominantly mutant for the three ESCRT-II components show characteristics of neoplastic tissue growth. However, despite these common defects, the phenotypic consequences of loss of vps22, vps25 and vps36 in mosaic animals are distinct. vps22 and vps25, but not vps36 mosaics show non-autonomous hyperplastic growth. In contrast, vps25 and vps36, but not vps22 mosaics strongly increase apoptotic resistance. We demonstrate that these differences are caused by selective Notch activation. vps22 and vps25 clones display high Notch signaling activity, while vps36 clones do not, suggesting that hyperplastic growth depends on Notch signaling. However, neoplastic growth may be independent of Notch signaling. Thus, despite their intimate physical relationship, the individual ESCRT-II components are genetically not equivalent.

Common phenotypes I: ESCRT-II mutants contain enlarged endosomes accumulating ubiquitinated proteins
Because vps25 mutants in Drosophila are characterized by enlarged early endosomes [19][20][21] (see also Fig. 1B0), we tested whether mutants in the other two ESCRT-II components also contain abnormal endosomes. As endosomal marker we used an antibody raised against Hrs [17]. Mutant clones of vps22 and vps36 in eye imaginal discs contain enlarged Hrs-positive particles, representing abnormal early endosomes (Fig. 1A0,C0). The FK1 and FK2 antibodies recognize ubiquitin-conjugated proteins, but not unconjugated ubiquitin [35][36][37]. Using the FK antibodies, we found that the enlarged Hrs-positive particles accumulate ubiquitin-conjugated proteins (shown for FK1 in Fig. 1A90-C90). Similar data were obtained for FK2 antibody (data not shown). In addition, an antibody that recognizes both conjugated and unconjugated ubiquitin also detects increased abundance of ubiquitin in vps22, vps25 and vps36 mutant clones (Fig. 2A0, C0). Thus, vps22, vps25, and vps36 mutant cells contain abnormally large early endosomes that accumulate ubiquitinconjugated proteins.

Common phenotypes II: Accumulation of Notch and Delta proteins ESCRT-II mutant clones
It has been reported that vps25 mutants accumulate Notch protein in endosomes [19][20][21] (see also Fig. 2B,E). Thus, we analyzed vps22 and vps36 mutants for accumulation of Notch protein. Consistently, vps22 and vps36 mutant clones accumulate Notch protein in punctate particles that, due to their colocalization with ubiquitylated proteins, correspond to enlarged endosomes ( Fig. 2A,C,D,F). This was found using antibodies that recognize both the extracellular domain of Notch ( Fig. 2A90-C90) and the intracellular domain of Notch ( Fig. 2D9-F9). In addition, an antibody raised against the Notchligand Delta also detects increased abundance of Delta protein in mutant clones (Fig. 2G9-I9).
Common phenotypes III: ESCRT-II mutant clones are apoptotic vps25 mutant clones in eye imaginal discs are extremely apoptotic [19][20][21] (see also Fig. 3B). To test whether this applies to vps22 and vps36 mutant clones, we performed immunolabeling using an antibody that recognizes the cleaved and thus activated form of Caspase-3 [38]. In 3rd instar eye imaginal discs of wildtype larvae, apoptotic cell death does not occur [39,40]. However, vps22 and vps36 clones contain increased caspase activity (Fig. 3A,C). Thus, as for vps25, loss of vps22 and vps36 causes the apoptotic death of the affected cells.  [20,26]. Eye discs predominantly mutant for a gene were obtained using the eyFlp-cell lethal system [26,41]. In this system, all cells which are not mutant for the gene of interest are eliminated by homozygosity of the cell lethal mutation or by induction of apoptosis using GMR-hid [41]. When we applied this technique to ESCRT-II mutants, the resulting mutant eye discs are overgrown compared to normal discs (Fig. 4). This overgrowth is particularly striking for vps25 mutant discs, consistent with previous reports [26], but also vps22 and vps36 mutant discs are significantly larger than normal discs (Fig. 4).
We labeled these discs with phalloidin, a marker for cortical actin. In wild-type discs, phalloidin labeling reveals regular cellular architecture (Fig. 4A). However, discs predominantly mutant for vps22, vps25 and vps36 show disorganized cellular architecture. Similar observations have been made for other neoplastic tumor suppressor genes including ept (vps23), avalanche, rab5 and lgl [26]. Therefore, vps22, vps25 and vps36 mutant discs display neoplastic tumor characteristics.
Distinct phenotypes I: vps22 and vps25 mosaics, but not vps36, display non-autonomous overgrowth Next, we tested whether vps22 and vps36 mosaics -similar to vps25 -cause non-autonomous overgrowth which is the result of increased Notch signaling activity [19][20][21]. In Figure 2, we showed that vps22 and vps36 mutant clones contain increased Notch protein levels. Therefore, we expected that vps22 and vps36 mosaic animals would display the same non-autonomous overgrowth phenotype as vps25 mosaics (Fig. 5C,K). Surprisingly, that was only observed for vps22 mosaics, but not for vps36 mosaics. Eyes and heads of vps22 mosaics are significantly larger compared to wild-type controls (Fig. 5A,B,I,J). The overgrowth is also detectable in eye-antennal imaginal discs, the larval precursors of the adult eyes. vps22 mosaic eye-antennal discs are significantly larger compared to wild-type controls (Fig. 5Q,R). This overgrowth is non-autonomous because, as shown in Figure 3, vps22 mutant clones are apoptotic and cannot be recovered in mosaic vps22 eyes (mutant vps22 tissue is marked by the absence of red eye  pigment, i.e. are phenotypically white 2 ; note the lack of white 2 tissue in Fig. 5B compared to the wild-type control in Fig. 5A).
Surprisingly, a similar strong overgrowth phenotype was not found for vps36 mosaic eyes and heads (Fig. 5D,L). We also found this confirmed at the level of eye imaginal discs. vps36 mosaic eye discs are noticeably smaller than vps22 and vps25 mosaic discs and are comparable in size to wild-type (Fig. 5Q-T). We do note, however, a rough eye phenotype in vps36 mosaics (Fig. 5D). Thus, although vps22, vps25 and vps36 mutants display similar endosomal defects and contain increased Notch protein levels, they affect mosaic animals differently.
Because we observed increased Notch protein levels in vps22, vps25 and vps36 mutant clones (Fig. 2), we analyzed whether Notch accounts for the overgrowth phenotypes in vps22 and vps25 mosaics. This can be tested by determining whether the overgrowth phenotype can be suppressed by reducing the gene dose of Notch. Thus, we analyzed vps22, vps25 and vps36 mosaics in a heterozygous Notch background. Indeed, heterozygosity of Notch suppresses the overgrowth of vps22 and vps25 eyes and heads suggesting that Notch activity is required for the overgrowth phenotype of vps22 and vps25 mosaics (Fig. 5E-G, M-O). The rough eye phenotype observed for vps36 mosaics is not suppressed by heterozygous Notch (Fig. 5D,H).
To further characterize these differences in the overgrowth phenotypes, we analyzed 3 rd instar eye-antennal imaginal discs of the ESCRT-II mutants by BrdU labeling as a marker for cells in Sphase. Anterior to the morphogenetic furrow and in the antennal disc, BrdU-labeling is homogeneous in wild-type discs (Fig. 6A). In contrast, vps22 and vps25 mutant clones do not proliferate very well (Fig. 6B,C). However, the wild-type tissue immediately adjacent to  Fig. 6B9,C9 to wild-type in Fig. 6A9). Thus, similar to vps25 mosaics, vps22 controls cell proliferation non-autonomously. However, vps36 mosaics behave differently. We observe BrdU-positive cells both within and outside of vps36 clones with homogeneous density (Fig. 6D9). There is no apparent increased density of BrdUpositive cells outside of vps36 mutant clones. Thus, vps36 mutant clones appear to be unable to induce non-autonomous proliferation.
Distinct phenotypes II: vps22 and vps25 mutant clones, but not vps36, contain strong Notch activity The analyses presented above demonstrated that mutants of the three ESCRT-II components display endosomal defects with accumulated Notch protein (Figures 1 and 2); yet, only vps22 and vps25 mutant clones trigger non-autonomous proliferation in a Notch-dependent manner, while vps36 clones do not (Figures 5 and  6). One possibility to explain this discrepancy is that only vps22 and vps25, but not vps36 mutations trigger Notch signaling although Notch protein accumulates in all three mutants ( Figure 2). Therefore, we assayed for Notch activity in ESCRT-II mutants using the E(spl)m8 2.61-lacZ reporter transgene that responds well to Notch activity [21].
The E(spl)m8 2.61-lacZ reporter is turned on by normal (wildtype) Notch activity posterior to the morphogenetic furrow in eye imaginal discs (Fig. 7A,C). There is no reporter activity detectable anterior to the morphogenetic furrow. However, in vps22 and vps25 mutant clones located anterior to the morphogenetic furrow, E(spl)m8 2.61-lacZ reporter expression can be detected (arrows in Fig. 7A0 and B0). In contrast, clones of vps36 located anterior to the morphogenetic furrow do not or only very mildy increase reporter activity ( Figure 7C0). This behavior was consistently observed in fifteen eye imaginal discs of each genotype. Thus, the observed differences in non-autonomous overgrowth between vps22, vps25 and vps36 do correlate with de-regulation of Notch activity.
Distinct phenotypes III: vps25 and vps36 mutations, but not vps22, promote strong apoptotic resistance Paradoxically, despite the fact that vps25 clones are highly apoptotic ( Figure 3B), we originally isolated vps25 mutants based on their ability to suppress apoptosis [21]. Specifically, the eyeablation phenotype caused by expression of the pro-apoptotic gene hid under control of the eye-specific GMR promoter (GMR-hid) is  suppressed in vps25 mosaics (see Fig. 8A,C). The suppression of GMR-hid by vps25 occurs in a non-autonomous manner through up-regulation of the apoptosis inhibitor Diap1 in neighboring cells [21]. Therefore, we tested whether vps22 and vps36 have a similar activity. However, to our surprise, although vps22 mosaics cause a strong non-autonomous proliferation phenotype, they do not suppress GMR-hid (Fig. 8B). In contrast, although vps36 mosaics display no or only a mild non-autonomous proliferation phenotype, they are very strong suppressors of GMR-hid, comparable to vps25 mosaics (Fig. 8D). This observation provides another example of genetic differences between the individual components of ESCRT-II.

Discussion
In this paper, we have characterized and compared the mutant phenotypes of the ESCRT-II components vps22, vps25 and vps36 in Drosophila. So far, a systematic genetic analysis of class E vps genes has only been performed in yeast [12,15]. Endosomal defects in these mutants in yeast are genetically similar. Consistently, endosomal defects caused by mutations in the ESCRT-II components vps22, vps25 and vps36 in Drosophila are similar, too. These mutant endosomes accumulate ubiquitinated proteins and signaling receptors including Notch and its ligand Delta. They also show neoplastic characteristics. However, despite these common endosomal defects, at the organismal level, vps22, vps25 and vps36 mosaic animals display distinct phenotypes. vps22 mosaics are characterized by strong non-autonomous proliferation, but not an increase in apoptotic resistance. vps36 mosaics exhibit the reverse phenotype, i.e. increased apoptotic resistance and no or only weak non-autonomous proliferation. As shown before [21], vps25 mosaics combine both phenotypes. Thus, this analysis shows that although these components are part of the same structural complex, they are not genetically equivalent and display distinct genetic properties.
While the vps22 allele used in this study is a clear null allele [32], one might argue that the vps36 allele is not a null and that the observed differences are due to the hypomorphic nature of vps36. However, such an assumption does not explain why vps36 is a strong suppressor of GMR-hid, while a null allele of vps22 that causes a strong overgrowth phenotype, completely fails to suppress GMR-hid (Fig. 8B). In addition, the common phenotypes (endosomal defects giving rise to enlarged endosomes, accumulation of Notch protein, apoptosis and the neoplastic phenotype) are very similar between vps22 and vps36. Thus, it does not appear that the phenotypic differences observed between vps22 and vps36 are due to allelic strength of the mutants. Rather, they appear to be caused by intrinsic differences of the endogenous genes.

Role of Notch signaling for non-autonomous hyperplastic proliferation
It has previously been shown that inappropriate Notch signaling is required for non-autonomous proliferation in vps25 mosaics [18][19][20][21]. Our data confirm this notion here for vps22 mosaics. vps22 and vps25 mutants contain increased Notch activity and heterozygosity of Notch suppresses the non-autonomous overgrowth phenotype. In contrast, vps36 mosaics do not activate Notch signaling and hence do not cause non-autonomous overgrowth. Thus, Notch activity is required for non-autonomous hyperplastic overgrowth.
It is puzzling that despite their intimate physical association in the ESCRT-II complex [33,34], loss of vps22, vps25 and vps36 affects Notch signaling differently. One possibility to explain these differences is that these mutants form distinct endosomal microenvironments which may affect signaling from the endosome differently. The resolution of our labeling technologies may not be sufficient to pick up these differences in the endosomal microenvironment, but the fact that we do observe genetic differences suggests that microenvironmental differences may exist. There is precedence for such a conclusion. Although hrs mutants contain abnormal endosomes leading to accumulation of Notch protein, they do not trigger Notch activity and hence no significant growth defects [16]. Further support of the idea that Notch needs to be in a particular microenvironment at the early endosome in order to be activated comes from a study that analyzes that act upstream of the ESCRT machinery in the endosomal pathway, namely shibire, avalanche and Rab5. Mutations in these genes also result in accumulation of Notch protein, but do not activate the pathway [8].
Class E vps genes have also been reported to function outside of endosomal protein sorting. As such they are involved in virus budding, transcriptional control, cell cycle progression, mRNA localization and apoptosis [31,32,[42][43][44][45][46][47][48]. Therefore, it is possible that the observed genetic differences of the ESCRT-II components may be caused by distinct requirements in addition to and independently of endosomal function and possibly independently of the ESCRT-II complex and the remaining ESCRT machinery. Future work will be necessary to dissect the roles of the ESCRT-II components in processes unrelated to endosomal processing. Role of Notch signaling for autonomous neoplastic proliferation While inappropriate Notch signaling correlates well with nonautonomous hyperplastic growth, it does not correlate with autonomous neoplastic growth. Imaginal discs entirely mutant for vps22, vps25 and vps36 all display overgrowth and loss of cellular architecture, hallmarks of neoplastic behavior [22]. The neoplastic phenotype has been attributed to either increased Notch signaling or to mis-localization of the apical transmembrane protein Crumbs [49]. However, vps36 mutant discs display a very robust neoplastic phenotype, but do not activate the Notch signaling pathway significantly, suggesting that activation of Notch is not required for neoplastic growth in vps36 mutant discs. This observation is consistent with previous findings that mutations in the neoplastic tumor suppressor genes avalanche and Rab5 do not activate Notch signaling [8,49]. We have not analyzed a genetic requirement of crumbs for the neoplastic phenotype in vps22, vps25 and vps36 mutants, but that would be an interesting experiment in the future.
It is clear that the endosomal defects in ESCRT-II mutants not only affect Notch signaling. Other membrane proteins are also affected which may contribute to the neoplastic phenotype. For example, in the case of hrs and vps25, other signaling receptors such as EGFR, Tkv, Ptc and Smo accumulate at endosomes [16,19]. However, it was also shown that these accumulated proteins are largely derived from the pool of unliganded receptors, suggesting that the endosomal defect affects receptor turnover [16] which does not necessarily cause receptor activation. The only receptor known to be activated at the endosome in a ligandindependent manner is Notch [8]. Future work will be necessary to dissect the role of Crumbs and other signaling pathways for developing the neoplastic phenotypes.

Drosophila genetics and generation of mutant clones
For this comparative analysis, we used the following mutant alleles of the ESCRT-II components. vps22 5F8-3 (also known as lsn 5F8-3 ) was previously described by Irion and St. Johnston (2007) [32]. It carries a premature termination codon at residue 2, likely encoding a null allele. vps25 N55 has a premature termination codon at residue 93. We have previously characterized this allele as a null allele [21]. vps36 L5212 was also characterized by Irion and St Johnston (2007) [32]. A P-element transposon is inserted in the first exon 29 base pairs upstream of the initiator ATG.
Fly crosses were conducted using standard procedures at 25uC. The following stocks were used: vps22 5F8-3 and vps36 L5212 [32]; vps25 N55 [21]; N 264-39 [50]; GMR-hid eyFlp [51][52][53]. For generation of mutant clones, the vps mutant alleles were crossed to eyFlp; FRT P[ubi-GFP]. To generate imaginal discs predominantly mutant for vps22, vps25 or vps36, we used the eyFlp cell lethal technique [41]. The vps mutant alleles were crossed to eyFlp; FRT cl GMR-hid flies. cl indicates an anonymous cell lethal mutation that kills when homozygous [41]. The use of the FRT depended on the location of the vps gene in the genome. Mutant clones are marked by absence of GFP. The complete genotypes are indicated in the legend to the figures.