Drosophila Cornichon acts as cargo receptor for ER export of the TGFα-like growth factor Gurken

During Drosophila oogenesis, signals from the oocyte to the overlying follicular epithelium govern the polarization of the maturing egg and the establishment of the future embryonic body axes(Roth, 2003). These signals are mediated by the TGFα-like growth factor Gurken(Neuman-Silberberg and Schüpbach,1993). Eggs laid by females homozygous for null alleles of grk are ventralized and lack anteroposterior polarity(Gonzalez-Reyes et al., 1995; Roth et al., 1995; Schüpbach, 1987). Grk consists of an extracellular growth factor domain containing one EGF repeat, a transmembrane domain and a short C-terminal cytoplasmic tail, thus resembling vertebrate TGFα (Derynck et al.,1984) and the other Drosophila TGFα family members Spitz (Rutledge et al., 1992)and Keren (Reich and Shilo,2002). The mature, secreted forms of all three DrosophilaTGFα homologues are likely to be generated by intramembrane proteases of the Rhomboid family (Lee et al.,2001; Urban et al.,2001; Urban et al.,2002). Specifically, Grk appears to be cleaved by Rhomboid-2/Brho(Ghiglione et al., 2002; Guichard et al., 2000). Processing of Spitz occurs in the Golgi apparatus, where its protease Rhomboid1 resides, and is therefore strictly dependent on the prior export of Spitz from the endoplasmic reticulum (ER), which is in turn mediated by Star(Lee et al., 2001; Tsruya et al., 2002; Urban et al., 2001). By contrast, we will show here that Grk processing occurs before export from the oocyte ER.

Generation of the Grk signals depends on the presence of Cornichon [Cni(Roth et al., 1995)] within the germline. cni encodes a small hydrophobic protein that is the founding member of a family of conserved eukaryotic proteins(Hwang et al., 1999; Powers and Barlowe, 1998; Roth et al., 1995). Erv14p,one of the two S. cerevisiae Cni homologues, was identified as an integral membrane protein of COPII-coated ER-derived vesicles(Belden and Barlowe, 1996). The COPII coat consists of several subunits that are assembled into a multimolecular coat on the surface of the ER(Barlowe et al., 1994; Bednarek et al., 1995) and serves as an external scaffold organizing the assembly of anterograde transport vesicles at the ER exit sites(Bonifacino and Glick, 2004). Direct or indirect interactions with COPII components can provide an efficient mechanism for the recruitment of cargo proteins into vesicles leaving the ER(Barlowe, 2003; Kuehn and Schekman, 1997).

Erv14p is itself recruited into such vesicles through interactions with the COPII coat (Powers and Barlowe,2002). Loss of Erv14p results in a bud site selection defect caused by inefficient membrane transport of the bud site selection protein Axl2p (Powers and Barlowe,1998; Roemer et al.,1996). In erv14Δ yeast cells, Axl2p fails to be sorted into COPII vesicles and accumulates in the ER, while other cargo molecules are secreted at normal rates. Thus, only a subset of secreted proteins depends on Erv14p for ER export(Powers and Barlowe, 1998; Powers and Barlowe, 2002). In wild-type oocytes, freshly synthesized Grk protein is efficiently and rapidly cleared from the large, continuous ER spanning the oocyte. Consistent with Cni acting as a Grk cargo receptor, reduction in Cni activity causes diffuse mislocalization of Grk protein within the ER(Herpers and Rabouille,2004).

Here, we provide biochemical and genetic evidence for an involvement of Cni in Grk ER export and present data explaining why Cni function is essential for the spatial and temporal specificity of Grk signalling.

The Drosophila strains cni^AR55, cni^AA12, Df(2L)H60(Roth et al., 1995), grk^DC (Schüpbach,1987), grk^2B6, grk^HF48(Neuman-Silberberg and Schüpbach,1993), shi^TS (shi¹)(Grigliatti, 1973) and Df(2L)JS7(Sekelsky et al., 1995) have all been described previously.

Antibody staining was performed as described(Peri et al., 1999). The antisera used were rabbit anti Grk, mouse anti Grk 1D12 (DSHB), mouse anti myc 9E10 (Dianova), rabbit anti myc A-14 (Santa Cruz), mouse anti Drosophila Golgi 9C1 (Abcam) and mouse anti KDEL-receptor (Abcam). For the shi^TS experiments, freshly dissected ovaries were incubated for 3 hours at either 25°C or 32°C in Schneider's Drosophila tissue culture medium (Sigma) before fixation and staining. Confocal imaging was performed using a Zeiss LSM 510 microscope. Ultrastructural analysis was carried out as described(Wilsch-Braeuninger et al.,1997).

cnir and dEmp24 cDNAs were amplified from an ovarian two-hybrid library (Grosshans et al.,1994). All Grk transgenes are derived from a Grk genomic rescue construct containing bases 49777-54827 of P1 DS02110. The region encoding Grk amino acid 236-294 (transmembrane and cytoplasmic domain) was replaced using PCR generated restriction sites in domain swaps with the transmembrane or transmembrane and cytoplasmic domains of Yl (Grk-Yl TM, Grk amino acids 1-235+ Yl amino acids 1801-1825; Grk-Yl TMC, Grk amino acids 1-235 + Yl amino acids 1801-1984). In three other constructs, the cytoplasmic tail of Grk was replaced with the cytoplasmic tails of Yl (pGrk-YlCyt, Grk amino acids 1-275 +Yl amino acids 1826-1984), dEmp24 (pGrk-EmpCyt, Grk amino acids 1-275 + CG3564 amino acids 194-208) or cni (pGrk-CniCyt, Grk amino acids 1-275 + Cni amino acids 100-145).

The pcni::Cnir plasmid was cloned by replacing the cni-coding region with a corresponding genomic fragment of cnir in the published rescue construct (Roth et al.,1995). The same template was used to introduce a single myc epitope tag at the Cni C terminus. All constructs were cloned into pCasper4(Pirrotta, 1988).

For western analysis, ovaries were ground on ice in lysis buffer (50 mM Tris pH 7.5, 1 mM EDTA, 150 mM NaCl, 2 mM DTT, 1%Triton X-100, 1% SDS 2 mg/ml Aprotinin, 1 mg/ml Leupeptin, 0.5 mg/ml Pepstatin, 10 mM PMSF) and loaded at one to three ovaries per lane. Protein was detected using the A14 and 9E10 anti-myc antisera (Santa Cruz) or the anti-Grk monoclonal 1D12 (DSHB).

Yeast two-hybrid experiments were performed using the system of James et al. (James et al., 1996). The control pair Staufen RNA binding domain 5/CG18501 was a gift from Uwe Irion and Daniel St Johnston. Pulldown experiments were performed by bacterially expressing the first 57 amino acids of Cni and Cnir or lacZ ORF, respectively,fused to the maltose binding protein (MBP) of pMal-c2e (NEB). These fusion proteins were subsequently tested for precipitation from bacterial lysates by incubation with GST (pGEX2T, Pharmacia) or a GST-Grk fusion (GST and Grk amino acids 179-245) prebound to Glutathion-Sepharose beads (Pharmacia). The beads were spun down, washed in PBS with 300 mM NaCl and 1% Triton X-100, and MBP fusions in the pellet detected by western blots using an anti-MBP monoclonal antibody (Sigma).

Owing to a failure in Grk signalling during oogenesis, eggs produced by cni mutant females lack both anteroposterior (AP) and dorsoventral(DV) polarity (Roth et al.,1995). However, unlike mutations in grk, loss of cni function also causes defects in adult somatic tissues. Flies homozygous for the amorphic allele cni^AR55 are subviable(20% of the expected number hatch), possess rough eyes(Fig. 1A-C), their postvertical, interocellar and ocellar bristles are largely missing(Fig. 1A,B), and the wings have a truncated vein 2 (Fig. 1E,F).

Transgenes expressing Cni tagged with a single C-terminal myc epitope from its endogenous promotor rescue the cni germline and somatic defects(Fig. 1G-I; data not shown). The tagged Cni protein is detectable in the germline from germarium stages onwards and becomes enriched within the oocyte during early and middle stages of oogenesis (Fig. 2A). Cni is also detectable in the somatic follicular epithelium(Fig. 2A), in the embryo, and in male and female somatic tissues (not shown). To address the subcellular localization of Cni, we turned to the cells of the squamous follicular epithelium, the large size and flat geometry of which minimizes colocalization artefacts. Myc-tagged Cni expressed in these cells using the CY2 driver line is found in small, discrete, dot-like structures that are only partially overlapping with the ER (Fig. 2B), labelled here using a GFP genetrap of the ER resident protein Protein Disulfide Isomerase (PDI-GFP)(Morin et al., 2001) (70% of all Cni-dots scored in three fields of view associated with PDI-GFP, n=291). However, whether this represents true colocalization is difficult to judge as the ER extends widely throughout the cytoplasm. Although only 8.7% of all Cni-containing structures were also stained with an antibody against the p120 Golgi protein (Fig. 2C) (n=346 Cni-dots scored in three fields of view), 91%were positive for staining against the KDEL-Receptor, which is involved in retrieving KDEL-tagged proteins from the Golgi to the ER(Fig. 2D) (n=305 Cni-dots scored in three fields of view). As biochemically determined for Erv14p (Powers and Barlowe,1998), the subcellular localization of Drosophila Cni is thus consistent with cycling in and out of the ER. Although almost all Cni-containing dots were KDEL-receptor positive, the converse is not the case. Only 20.4% of the KDLR-positive dots contained Cni (n=641). Drosophila Golgi units have been shown to be biochemically heterogenous. with different enzyme and cargo content even at the same cisternal level. Presence in only a subset of retrograde vesicles may therefore reflect a prior targeting of Cni only towards specific Golgi units(Yano et al., 2005).

The Drosophila genome contains a second cni-like gene (CG17262)which we named cni-related (cnir). Cni and Cnir share 28.5%identical and 43.1% similar amino acids, but are each more closely related to specific vertebrate Cornichons, here exemplified by the human proteins. Cnirel and Cni4 form a branch distinct from other metazoan Cornichons(Fig. 3B). Overall structural properties are conserved between all family members, with a cytoplasmic N terminus and three transmembrane domains, as determined for Erv14p(Powers and Barlowe, 2002) and modelled here for Cni using the TMHMM and TMpred algorithms(Fig. 3C).

cnir is deleted by Df(2L)JS7(Sekelsky et al., 1995). Heterozygosity for the deficiency causes synthetic lethality in an amorphic cni background (Table 1). In combination with the hypomorphic allele cni^AA12 rare and severely malformed Df(2L)JS7 cni^AR55/cni^AA12 escapers could be recovered. The synthetic lethality was completely rescued by introduction of a transgene driving expression of cnir under control of the cni promotor and 5′ and 3′ untranslated regions(pcni::Cnir), demonstrating that the initial enhancement of the cniphenotype to lethality was caused by the reduction of the total level of cni-like genes (P<0.001, Fisher's exact test). The construct also suppressed the loss of postvertical and ocellar bristles and the wing venation defects of amorphic cni^AR55/cni^AR55 flies(Fig. 1D,F), but not the rough eye phenotype and the loss of the interocellar bristles(Fig. 1D).

In contrast to rescue constructs expressing native or myc-epitope tagged cni from either the cni or bicoid promotors(Fig. 1I and data not shown),analogous constructs containing cnir were unable to compensate for the loss of cni during oogenesis(Fig. 1J; data not shown). Cnir is therefore able to substitute for Cni in some contexts, whereas other processes, most notably Grk signalling by the oocyte, specifically require Cni function.

We re-examined the Grk protein distribution in cni ovaries by light and immunoelectron microscopy, comparing the results with wild-type and a grk mutation (grk^DC) that produces a non-secreted Grk protein (Queenan et al.,1999). In wild-type stage 10 ovaries, Grk is concentrated between the oocyte nucleus and the adjacent oolemma(Fig. 4A)(Neuman-Silberberg and Schüpbach,1996). Immunoelectron microscopy shows that during vitellogenesis the oolemma and the apical sides of the follicle cells have enlarged surfaces covered by microvillous processes to facilitate rapid yolk uptake by the oocyte (King, 1970)(Fig. 4D). Grk protein could at that stage be detected at the oocyte surface, especially on the microvilli. Grk was also found at the microvilli covering the apical surface of the follicular epithelium and within follicle cells(Fig. 4D), confirming Grk release from the oocyte (Peri et al.,1999). grk^DC is representative of a distinct class of amorphic grk alleles with amino acid substitutions affecting a conserved alanine residue within the transmembrane domain(Queenan et al., 1999). These mutations apparently block proteolytic release of the extracellular growth factor domain (see below, Fig. 6C), which has been shown to be necessary for Grk activity(Ghiglione et al., 2002). Grk distribution in grk^DC oocytes is indistinguishable from wild type before the onset of vitellogenesis (stage 10A), when the oocyte increases its endocytosis activity. Grk protein can then also be found in posterior and ventral parts in homo- or heterozygous grk^DCoocytes, from where it is completely absent in wild-type ovaries(Fig. 4B). At the ultrastructural level internalized Grk protein associates with the membranous cortex of yolk granules in heterozygous grk^DC ovaries(Fig. 4E). Yolk granules are endosomal derivatives that grow by fusion with endocytic vesicles internalizing yolk proteins (DiMario and Mahowald, 1987). Both the timing of the Grk^DCmislocalization and the association of the mutant protein with the cortex of the yolk granules therefore suggest that the Grk^DC protein is transiently inserted into the plasma membrane, and then reinternalized during yolk uptake.

By comparison, Grk is more smoothly distributed within cni mutant oocytes and seems to diffuse away from a source near the nucleus(Fig. 4C). At the ultrastructural level, the comparison with grk^DC and wild-type ovaries shows that Grk protein in cni oocytes is neither clustered at the cortex of yolk granules nor enriched at the plasma membrane adjacent to the oocyte nucleus (Fig. 4F). In contrast to earlier suggestions(Queenan et al., 1999), the mutant Grk^DC protein in wild-type ovaries therefore appears to reside in a different endomembrane compartment than wild-type Grk protein in cni mutant egg chambers.

To confirm a requirement for Cni during Grk transport to the plasma membrane, we attempted to trap the apparently membrane-tethered Grk^DC protein at the oocyte surface in the presence or absence of Cni by inactivating a temperature-sensitive allele of the Drosophiladynamin homologue shibire (shi^TS)(Chen et al., 1991; Grigliatti,1973; van der Bliek,1991).

Incubation of dissected shi^TS ovaries heterozygous for grk^DC at a restrictive temperature (32°C) resulted in the expansion of Grk protein staining around the anterior margin of these oocytes (Fig. 5D). Such an expansion was not observable in control ovaries that contained only wild-type grk (Fig. 5E) or were heterozygous for shi^TS (not shown). Incubation at the permissive temperature never resulted in Grk accumulation at the plasma membrane, irrespective of the presence of the grk^DC allele(Fig. 5A,B). These observations support our model that grk^DC encodes a cleavage-resistant protein that localizes to the plasma membrane prior to its reinternalization.

By contrast, Grk^DC protein could not be trapped at the plasma membrane of oocytes from shi^TS/shi^TS; grk^DC Df(2L)H60/cni^AR55 females. Both at permissive and restrictive incubation temperatures for shi^TS, all Grk protein remained diffusely distributed throughout the oocyte (Fig. 5C,F), in a pattern indistinguishable from the distribution of wild-type Grk in homozygous cni ovaries. Thus, Cni is required for the transport of mutant Grk^DC protein and, by extension, wild-type Grk to the oocyte plasma membrane.

We generated a fully functional (not shown) C-terminally tagged version of Grk by replacing the intracellular domain, which is dispensable for function(Queenan et al., 1999), with five myc epitopes (Grk5myc). When Grk5myc was expressed from the endogenous promotor, only a short fragment corresponding in size to the expected C-terminal cleavage remnant (23 kDa) could be detected in western blots by antisera directed against the C-terminal epitope tags(Fig. 6A). Thus, at near endogenous expression levels Grk protein is quantitatively present in the processed form. As the same band was observed in extracts from flies mutant for cni (Fig. 6A), Grk processing does not require Cni function.

In ovary lysates from flies overexpressing Grk5myc in the germline under control of the maternal α-Tubulin Gal4 or nanos driver lines both the cleavage remnant and a longer protein species (ca. 70 kDa) could readily be detected (Fig. 6B and data not shown). The 70 kDa band corresponds to an uncleaved precursor form, as it can be detected both by a monoclonal antibody directed against the extracellular growth factor domain (Fig. 6B,right panel) as well as by antibodies against the intracellular, C-terminal epitope tag (Fig. 6B, left panel). Formation of the C-terminal Grk cleavage remnant is independent of cni also under overexpression conditions(Fig. 6B, left panel). However,in the absence of Cni, another species accumulates (48 kDa), which is recognized by the extracellular but not the C-terminal antisera, and therefore appears to be the N-terminal cleavage product corresponding to the mature growth factor. Supporting this interpretation, a band of this size has previously been identified as the active secreted Grk ligand in cell culture experiments (Ghiglione et al.,2002).

Together, these experiments suggest that cni is not required for proteolytic processing of Grk, and that in wild type, the generation,secretion and eventual degradation of the mature growth factor is a rapid process with low steady state levels of the protein. Although the cleavage process is saturable when Grk is overexpressed in wild type, causing accumulation of the uncleaved precursor, the N-terminal cleavage product is detected only after overexpression in a cni mutant background.

We next asked where processing of Grk occurs along the secretory pathway. We generated a C-terminally tagged version of Grk that resembles Grk5myc, but in addition carries the point mutation that induces the A to V amino acid exchange found in grk^DC (GrkDC5myc). In contrast to the functional cleavable Grk5myc protein, GrkDC5myc expressed from the endogenous promotor accumulates in the ovary as a high molecular weight form (70 kDa, Fig. 6C) corresponding in size to the precursor band found in the overexpression situation(Fig. 6B). This supports the notion that the grk^DC point mutation prevents proteolytic cleavage of the Grk precursor. In lysates from cni heterozygous control ovaries, the GrkDC5myc signal is smeared out into several high molecular weight bands, which is indicative of glycosylation of the protein in the Golgi (Fig. 6C). Strikingly, this smearing of the GrkDC5myc band does not occur in the absence of Cni. We therefore checked the glycosylation state of the N-terminal Grk5myc cleavage product accumulating in cni mutant ovaries(Fig. 6B). The molecular weight of this fragment could be decreased (Fig. 6D) both by treatment with Endoglycosidase H (EndoH) and Protein N-glycosidase F (PNGaseF). Although PNGaseF is capable of removing all N-linked sugar modifications, EndoH sensitivity is characteristic for the high mannose type modifications added already in the ER, which will subsequently be trimmed and replaced by the final sugar modifications upon entry into the Golgi (Kornfeld and Kornfeld,1985). Thus, in the absence of Cni, proteolytic cleavage of the Grk precursor occurs normally, but the extracellular/lumenal fragment that in wild type forms the mature, secreted growth factor accumulates in the ER and retains its pre-Golgi glycosylation signature.

Proteolytic processing of Grk therefore differs from that of Spitz which has to be exported from the ER to be processed by Rhomboid1 within the Golgi(Lee et al., 2001; Urban et al., 2001; Urban et al., 2002). By contrast, Grk is processed at the ER level, and requires Cni function for efficient export of the mature ligand. These in vivo observations are consistent with tissue culture data showing that the presumptive Grk protease Rhomboid2 (Guichard et al.,2000) is able to process Grk retained within the ER(Ghiglione et al., 2002).

Secretion of Grk depends both on the presence of Cni protein and the Grk transmembrane domain, and it has therefore been suggested that an interaction within the membrane might be responsible for the role of Cni in Grk secretion(Queenan et al., 1999). Unlike Grk, the Drosophila vitellogenin receptor Yolkless (Yl) is efficiently transported to the plasma membrane of cni mutant oocytes as seen from the formation of yolk vesicles in these cells(Fig. 4D)(DiMario and Mahowald, 1987). We therefore generated transgenic flies expressing domain swap constructs from the endogenous grk promotor, in which either the Grk intracellular domain, the transmembrane domain, or both, were replaced with the corresponding fragments of Yl. Although Yl itself is not proteolytically released, it possesses, like Grk, an alanine residue at the position mutated in the cleavage resistant allele grk^DC (Ala245Val). All three constructs were able to rescue the oogenesis phenotype of grkmutant flies (Fig. 7D; data not shown). As Grk activity depends on prior proteolytic release of the extracellular growth factor (Ghiglione et al., 2002), these domain swaps do not interfere with the processing of Grk despite the reported substrate selectivity of Rhomboid type proteases (Urban and Freeman,2003). By contrast, signalling by all three fusion proteins still depends on Cni function, as neither construct was able to suppress the cni oogenesis phenotype (Fig. 7E; data not shown). The region crucial for the role of Cni in Grk secretion shown above must therefore lie within the extracellular domain of Grk.

We tested for potential corresponding protein interactions using a yeast two-hybrid system selecting for adenine and histidine autotrophy(James et al., 1996) with bait constructs containing different portions of the Grk extracellular domain extending to the beginning of the transmembrane domain (amino acids 74-245,179-245, 215-245). Introduction of a prey construct containing the Cni ORF into this background conferred the ability to grow under stringent selection conditions, indicating an interaction between the two proteins(Table 2). Thus, the membrane proximal 31 residues of the Grk extracellular domain are sufficient to mediate binding to Cni in a yeast two-hybrid assay. Interaction was also observed with a prey construct encoding only the first 57 amino acids of Cni. By contrast,neither of the Grk bait constructs nor the CG18501 control were able to bind to a prey construct containing Cnir. These two hybrid data were confirmed by pull-down experiments, where a GST fusion protein containing amino-acids 197 to 245 of Grk could specifically co-purify a MBP fusion construct containing the N-terminal 57 amino acids of Cni, but not one with the corresponding Cnir domain or a lacZ control (Fig. 6E). This difference between Cni and Cnir in their ability to bind Grk may underlie the strict requirement for Cni during Grk secretion, even though the two Drosophila Cni-like proteins exhibit redundancy in other contexts.

Yeast Emp24p is a cargo receptor cycling between the ER and the intermediate compartment (Schimmoller et al., 1995). Exit of Emp24p and associated cargo from the ER is in part mediated by binding of Emp24p to components of the COPII coat through diaromatic amino acid pairs in the C-terminal cytoplasmic tail. We replaced the intracellular domain of Grk with the short cytoplasmic tail of one of the D. melanogaster Emp24 homologues (CG3564 amino acids 194-208). A transgene expressing this fusion protein from the endogenous promotor(pGrk-EmpCyt) fully rescued the loss of grk, indicating that the transgene produced normal amounts of active Grk ligand (not shown). Interestingly, it also restored some Grk signalling activity in the absence of cni (Fig. 7F). Eggs laid by homozygous cni females containing one copy of pGrk-EmpCyt had normal anteroposterior polarity. Some eggs also possessed recognizable dorsal appendage material, indicating low to intermediate levels of Grk signalling in these egg chambers (Fig. 7F). Thus, fusing Grk to a domain known to mediate selective recruitment into COPII vesicles partially alleviates its dependence on Cni.

Similar results were achieved using an analogous transgene replacing the Grk intracellular domains with a Cni fragment consisting of the C terminus after the second predicted transmembrane domain (Cni amino acids 100-145,pGrk-CniCyt). The transgene fully rescued the loss of grk (data not shown) and restored some signalling activity in the absence of cni. Eggs laid by cni mutant females carrying one copy of this transgene had normal anteroposterior polarity and showed slight and variable rescue of the dorsoventral axis (Fig. 7G). The C-terminal domains of Cni-like and Emp24-like proteins may therefore be functionally equivalent.

If Cni were only functioning as a cargo receptor for ER export of Grk,massive overexpression of Grk should result in bulk flow ER to Golgi transport and might thus overcome the requirement for cni. To test this assumption, we analyzed the egg phenotypes produced by those grkoverexpression lines that were used for the western blots described above. Expression of grk with the help of the maternal α-Tubulin Gal4 driver leads to a strong increase in the amount of Grk protein in stage 9 egg chambers when compared with endogenous Grk levels(Fig. 8A,C,E). When overexpressed in a wild-type background, the bulk of grk mRNA is still transported to the vicinity of the nucleus (data not shown). Grk protein remains asymmetrically distributed, although the region with high Grk protein levels within the oocyte is more expanded when compared with wild type(Fig. 8A,C). This might be due to the saturation of the mechanisms normally responsible for retention of the protein near its site of translation and the subsequent secretion through a few local ER exit sites (Herpers et al., 2004). The resulting egg chambers maintain DV polarity although they are severely dorsalized. The operculum, the dorsal-most chorion structure that is specified in follicle cells receiving maximal Grk levels, is expanded while the dorsal appendages normally specified at more lateral positions experiencing slightly lower Grk signalling are shifted to the ventral side of the egg(Fig. 8B,C).

Overexpression of Grk in a cni mutant background results in uniform high levels of Grk protein within the oocyte(Fig. 8E). Interestingly, the resulting eggs possess variable amounts of dorsal appendage material(Fig. 8F,G), indicating restoration Grk signalling, albeit to lower levels than in the presence of Cni. However, the eggs lack DV and frequently even AP polarity, as can be seen by the patchy induction of dorsal appendage material around the entire egg circumference (Fig. 8F,G) and the presence of a posterior micropyle (Fig. 8G), respectively.

These observations show that the requirement for Cni can be overcome by Grk overexpression. Thus, cni function is not essential for the formation of an active ligand per se, but the Cni-mediated increase in the efficiency of Grk secretion is a prerequisite for the precise temporal and spatial control of Grk signalling.

Recruitment within the ER is essential for the efficient ER exit of many different proteins. Cargo proteins may be concentrated into outgoing COPII-coated vesicles at the ER exit sites either by binding directly to vesicle or coat components, or through indirect recruitment using additional adaptors (Barlowe, 2003; Bonifacino and Glick, 2004; Kuehn and Schekman, 1997). Cni partially colocalizes with the ER resident protein marker PDI-GFP, associates with the KDEL-receptor (a protein involved in retrieval of escaped ER proteins) but is largely excluded from the Golgi. This suggests that Cni, like its yeast homologue Erv14p (Powers and Barlowe, 1998; Powers and Barlowe, 2002), has a pre-Golgi localization.

Using the cleavage-resistant Grk^DC protein, we have demonstrated that loss of cni blocks Grk transport to the oocyte plasma membrane,and in the absence of Cni overexpressed Grk protein accumulates within the oocyte and retains an ER-type glycosylation. Together with the accumulation of Grk within the ER of hypomorphic cni mutant oocytes(Herpers and Rabouille, 2004),these data show that the requirement for Cni in Grk export lies at the level of ER export. Our two hybrid data and Grk domain swap experiments suggest that Cni binds Grk through an interaction between its lumenal hydrophilic loop and the membrane-proximal extracellular spacer between the transmembrane and EGF domains of Grk. This interaction is consistent with the membrane topology proposed by Powers and Barlowe (Powers and Barlowe, 2002) for Erv14p, and suggests that Cni is the cargo receptor for Grk ER export. The inability of Cnir to bind to Grk in the same assay also correlates with its failure to suppress the cni oogenesis phenotype.

Processing of Spitz by Rhomboid1 depends on prior export of the transmembrane precursor from the ER (Lee et al., 2001; Tsruya et al.,2002; Urban et al.,2001). However, Spitz, as well as Grk, can be processed by Rhomboid2 and Rhomboid3 while still within the ER(Ghiglione et al., 2002; Urban et al., 2002). In the absence of Cni, Grk is effectively processed in the oocyte although it cannot leave the ER. Thus, during oogenesis Grk processing must occur within the ER,suggesting that the presumptive Grk protease Rhomboid2(Ghiglione et al., 2002; Guichard et al., 2000) either resides in, or cycles through, the oocyte ER.

Processed Spitz generated before export of the precursor to the Golgi is specifically retained in the ER(Schlesinger et al., 2004). By contrast, our data suggest that Cni serves to specifically ensure efficient export of Grk after cleavage in the oocyte ER. We propose that Grk interacts with Cni prior to its proteolytic processing, and that the proteins remain associated at least until the mature growth factor is recruited into an outgoing vesicle (Fig. 9A). By contrast, soluble Grk protein lacking a membrane anchor would diffuse away into the lumen after synthesis, precluding recruitment by Cni at the ER membrane, explaining why truncated Grk protein lacking a transmembrane domain is not secreted from the oocyte and corresponding grk alleles are nonfunctional (Queenan et al.,1999). The same fate would await Grk released from its membrane anchor through proteolytic processing in a cni mutant oocyte(Fig. 9B).

Constructs expressing Grk fused to the cytoplasmic parts of either Cni or dEmp24p are partly able to restore Grk signalling in the absence of Cni, but not to wild-type levels. We suggest that the heterologous cytoplasmic tails,which contain the respective domains shown in yeast to mediate the COPII interactions (Schimmoller et al.,1995; Powers and Barlowe,2002), are rapidly recruiting unprocessed Grk fusion proteins towards prospective vesicle budding sites. Because the ER exit motives are separated from the growth factor part during the processing step, most of the processed protein will in the absence of Cni still escape into the ER lumen,explaining the low rescue efficiency of the fusion proteins. However,proteolytic cleavage would preferentially occur in the vicinity of the outgoing vesicles, locally increasing the concentration of the soluble mature growth factor (Fig. 9C). This appears to be sufficient to ensure inclusion of some processed Grk into the outgoing vesicles in the absence of Cni, but cannot reconstitute wild-type rates of Grk signalling.

Conversely, the hypomorphic mutation cni^AA12 truncates Cni after the first two putative membrane-spanning domains(Roth et al., 1995). It therefore deletes the second, cytoplasmic loop shown to mediate COPII interaction in Erv14p (Powers and Barlowe,2002), but still possesses the first, lumenal loop binding to Grk. The truncated Cni^AA12 protein may therefore remain able to keep processed Grk at the ER membrane. This would limit diffusion of processed Grk to the two dimensions of the ER membrane, rather than the three dimensions of the lumen, thereby enriching it to some degree in vesicles leaving the oocyte ER (Fig. 9D). However, to achieve the full rate of Grk secretion further cargo concentration into outgoing vesicles through interaction of the Grk-Cni complexes with the COPII coat would be required. Consistently, cni^AA12 is clearly a hypomorphic allele with readily detectable remaining Grk signalling activity(Fig. 7B), but in mutant oocytes, Grk protein is diffusely mislocalized within the large, continuous ER and can no longer be found concentrated at ER exit sites(Herpers and Rabouille,2004).

Grk is translated from a localized mRNA and becomes translocated into a giant ER spanning the entire oocyte and containing around 1000 active exit sites. Nevertheless, in the presence of its cargo receptor Cni, Grk is exclusively secreted through a few of these sites and their associated Golgi stacks at the dorsal anterior corner where the grk mRNA is found(Herpers and Rabouille, 2004),giving rise to a spatially tightly confined signal to the neighbouring follicle cells. Concentration-driven bulk flow export from the ER can support secretion if cargo proteins are synthesized at sufficient rates to allow their accumulation within the ER(Martinez-Menarguez et al.,1999). Correspondingly, we have shown that massive overexpression of Grk in the oocyte can in principle restore signalling in the absence of Cni function, most likely via bulk flow ER export. However, in comparison to the wild-type situation the spatial and temporal precision of the Grk signals is lost, with severe consequences for the subsequent steps of pattern formation.

Cni is also required independently from Grk in somatic tissues, where it appears to act redundantly with Cnir. The reduced viability and life span of flies lacking cni function and the synthetic lethality when the gene dose of cnir is reduced in a cni mutant background indicate a more general cellular function of the Cni proteins. Erv14p, the cnihomologue from S. cerevisiae, is involved in recruiting the golgin Rud3p to the cis-Golgi stacks (Gillingham et al., 2004). Besides its function in Axl2p recruitment, Erv14p therefore may play a more general role in establishing cis-Golgi identity. It will be interesting to find out whether Cni proteins in general might have a more fundamental cell biological function, e.g. in establishing Golgi polarity, that may so far have been masked by redundancy and more easily detected phenotypes caused by their roles as cargo receptors.

We thank Jürgen Berger for the Drosophila SEM images,Birgitta Lattemann, Inge Zimmermann and Heinz Schwarz for their help with TEM;Oliver Karst for technical assistance; Andrea Klaes for making transgenic lines; and Jörg Grosshans and Uwe Irion for materials. We are grateful to Anne-Marie Queenan and Trudi Schüpbach for sharing results prior to publication, and we thank Francesca Peri, Katia Litiere, Antoine Guichet and Slawek Bartoszewski for suggestions. The project was supported by SFBs 446,243 and 572 of the Deutsche Forschungsgemeinschaft and the Graduierten Kolleg Genetik zellulärer Systeme (DFG) at the University of Cologne.