Glycosphingolipids control the extracellular gradient of the Drosophila EGFR ligand Gurken

Glycosphingolipids (GSLs) are ubiquitous components of eukaryotic cell membranes. They consist of a variable oligosaccharide chain attached to a ceramide lipid backbone (Cer) that tethers them to the lumenal leaflet of membranes. GSLs are mainly synthesized from Cer in the Golgi apparatus by a stepwise process in which unique glycosyltransferases add monosaccharides to a growing lipid-linked oligosaccharide chain. They are subsequently exported towards the plasma membrane, their principal location, where they are enriched together with cholesterol in membrane microdomains. The expression of a particular GSL is differentially regulated according to the developmental stage, the cell type and its differentiation state. The role of vertebrate GSLs has mostly been addressed in vitro (reviewed by Degroote et al., 2004; Sillence and Platt, 2004). These studies indicate that cell-surface GSLs participate in adhesion through the binding in trans of lectins or other GSLs. GSLs can also bind in cis to directly modulate the activity of receptor tyrosine kinases (RTKs) at the plasma membrane. GSLs are also thought to be involved in vesicular transport along the exocytic and endocytic pathways, sorting proteins into different compartments. Lastly, the presence of GSLs in membrane microdomains, which are considered as signaling platforms, may underlie many of their functions.

There is, however, little in vivo evidence to support any of these presumed functions. In S. cerevisiae, mutants abolishing all GSL synthesis fail to show defects in intracellular trafficking(Lisman et al., 2004). In C. elegans, GSLs appear to be dispensable throughout life(Griffitts et al., 2005). In mammals, the vast majority of GSLs are built on glucosylceramide (GlcCer), and synthesis branches at the level of the third glycosyl residue to yield three classes (lacto-, globo- and ganglioseries, Fig. 1A). Knockout of the mouse GlcCer synthase gene (Ugcg) leads to early embryonic lethality, for unclear reasons; assessing the effects of knocking out downstream glycosyltransferases is complicated by redundancy in these genes and between different GSLs (reviewed by Sabourdy et al., 2008). Nonetheless, disrupting the ganglioseries pathway produces mice that display neurological abnormalities after birth. Interestingly, in humans, mutations affecting GSL synthesis and degradation trigger severe neuropathologies (reviewed by Kolter and Sandhoff, 2006). Therefore, GSLs are at least required for proper function of the adult nervous system, but no firm link has yet been established between this requirement and their proposed cellular roles.

Drosophila melanogaster GSLs are simpler in structure than their vertebrate counterparts, with a single biosynthetic pathway described to date,giving rise to a family of differentially elongated molecules(Seppo et al., 2000). We previously identified Egghead (Egh) and Brainiac (Brn) as glycosyltransferases responsible for GSL biosynthesis in the fly, catalyzing the addition of the second and third glycosyl residues of the GSL oligosaccharide chain(Fig. 1B)(Schwientek et al., 2002; Wandall et al., 2003; Wandall et al., 2005). We also showed that there is no redundancy in these enzyme functions and no alternate biosynthetic pathway. Hence, egh and brn mutants are devoid of elongated GSLs and provide a useful model system for studies of GSL functions in vivo. Importantly, mutations in each gene are lethal and cause identical phenotypes during oogenesis and embryogenesis that are reminiscent of loss-of-function in the Notch receptor and EGF RTK (EGFR) pathways(Goode et al., 1996a; Goode et al., 1996b). Since the expression of a GSL-dedicated human galactosyltransferase in Drosophila egh mutants rescues their viability and fertility in a brn-dependent fashion (Wandall et al., 2005), these data indicate that Drosophila GSLs are essential for development, perhaps by modulating signaling.

During Drosophila oogenesis, activation of the EGFR pathway primarily depends on Gurken (Grk), an EGFR ligand similar to vertebrate TGFα, that is secreted by the oocyte (reviewed by Nilson and Schupbach, 1999). The EGFR-Grk couple acts twice to polarize the follicular epithelium as well as the future embryo along both anteroposterior (AP) and dorsoventral (DV)axes (Gonzalez-Reyes et al.,1995; Roth et al.,1995). Despite ubiquitous expression of EGFR in follicle cells,its activation is spatially restricted by asymmetric Grk localization. In early oogenesis, grk mRNA and protein are enriched at the posterior pole of the oocyte (Fig. 1C),and Grk activates EGFR in neighboring follicle cells, inducing them to adopt a posterior fate. At mid-oogenesis, these cells signal back to the oocyte,resulting in a reorganization of its cytoskeleton, a redistribution of oocyte maternal determinants along the AP axis, and the movement of the nucleus towards the anterior oocyte cortex. As grk RNA remains associated with the oocyte nucleus, a new restricted source of Grk is created to limit the highest activation of EGFR to the adjacent follicle cells, instructing them to assume a dorsal identity (Fig. 1C) (Schupbach,1987; Neuman-Silberberg and Schupbach, 1993). Respiratory appendages(Fig. 1D) are eggshell structures derived from dorsolateral follicular cells and their examination is an excellent means to monitor EGFR signaling. Indeed, mild Grk or EGFR loss-of-function causes a fusion of the respiratory appendages owing to the absence of the dorsal-most cells (weak ventralization, Fig. 1E). By contrast, a more severe reduction in EGFR signaling abrogates the formation of these structures(complete ventralization, Fig. 1F).

Here, we address the role of GSLs in EGFR signaling during oogenesis using egh and brn mutants. First, we show that GSLs exert a temporal control on the level of activation of EGFR. We find no evidence of a role for GSLs in the direct modulation of EGFR activity, but instead we show that GSLs act at the level of the EGFR ligand Grk. Despite reports of GSL function in trafficking, our results indicate that GSLs are dispensable for Grk export to the plasma membrane and for its secretion. However, by observing, for the first time, the gradient of secreted Grk, we show that GSLs control Grk diffusion in the extracellular space.

Flies were raised on standard medium at 25°C, or at 18°C for crosses involving the cold-sensitive brn^fs107 allele(Goode et al., 1992). Other mutant alleles were: brn^1.6P6(Goode et al., 1996b), egh^62d18, molecularly characterized by Wandall et al.(Wandall et al., 2005), the top¹ allele of Egfr(Clifford and Schupbach, 1994), cni^AA12 and cni^AR55(Roth et al., 1995) and grk^DC (Queenan et al.,1999). GAL4 and UAS lines used: nos-GAL4(Rorth, 1998), e22c-GAL4 (Duffy et al.,1998), UASP-brn (Chen et al., 2007) and UASP-mbgrk^myc(Ghiglione et al., 2002),which we remobilized to the third chromosome. brn^1.6P6 and egh^62d18 were recombined to FRT101, germ line clones generated with the hsFLP ovo^D1 system(Chou and Perrimon, 1992), and follicle cell clones obtained using the e22c-GAL4,UAS-FLP stock(Bloomington Stock Center).

The brn^fs107 allele was characterized by isolating genomic DNA from adult homozygous mutant flies, amplification by PCR of the full brn coding sequence and sequencing. Mutations were confirmed in two independent PCR reactions. Cloning details and primer sequences are available upon request.

Whole-mount in situ hybridization for grk on ovaries was carried out as described (Vanzo and Ephrussi,2002). For rho1 and aos mRNAs, amplification was skipped and Fast Red (Roche) was used as a fluorescent substrate for the alkaline phosphatase. Probes were prepared according to the manufacturer's instructions (DIG RNA Labeling Kit, Roche).

Localization of Grk by immunoelectron microscopy was performed as described(Herpers and Rabouille, 2004). Conventional immunostaining was as described(Wandall et al., 2005). Primary antibodies used: mouse monoclonal anti-MacCer(Wandall et al., 2005)undiluted, mouse monoclonal anti-Grk (1D12, DSHB) 1:200, rat polyclonal anti-EGFR (a kind gift from B. Shilo, Weizmann Institute of Science, Rehovot,Israel) 1:100. Fluorescently conjugated secondary antibodies were from Jackson ImmunoResearch. DAPI (Sigma) and fluorescently conjugated phalloidin(Molecular Probes) were always included in the staining procedure. Note that for MacCer staining, we used, as a detergent, Tween 20 at 0.05% instead of Triton X-100 at 0.1%. Extracellular immunostaining was as reported(Strigini and Cohen,2000).

Confocal data were acquired with a Leica TCS microscope as single images collected from the focal plane in which the oocyte nucleus is visible, or as image stacks for dorsal views.

To compare specimens of different genotypes, egg chambers were stained in parallel and imaged with identical acquisition settings. The extent of extracellular Grk diffusion towards the posterior pole was expressed as a ratio of the length of Grk staining in the extracellular space (delimited by F-actin staining) over the total length of the extracellular space along the AP axis (anterior- and posterior-most points were at the boundary with nurse cells and posterior polar follicle cells, respectively). These distances were measured using the Freehand tool from ImageJ, on single images of the longest longitudinal section taken at the level of the dorsal midline.

For dorsal views of the extracellular Grk gradient, quantification of pixel intensity at a defined level was performed with ImageJ on projections of 1μm-spaced confocal sections encompassing the whole domain of detectable Grk expression (typically ∼35 sections, roughly corresponding to the dorsal half of the egg chamber circumference). Average plots from the number of samples indicated in the text were generated using Microsoft Excel.

Ovarian extracts were prepared as described(Ghiglione et al., 2002). The rabbit anti-Myc (A-14, Santa Cruz) antibody was used at 1:1000, the mouse monoclonal anti-α-Tubulin (DM 1A, Sigma) at 1:10,000 and HRP-conjugated secondary antibodies (Jackson ImmunoResearch) at 1:5000. Western blots were developed with ECL reagents (Amersham).

Most mutations in the glycosyltransferases Egh and Brn lead to lethality,or to female sterility when germ line clones are generated, owing to a block in early oogenesis related to the formation of compound egg chambers (egg chambers containing multiple oocyte-nurse cell complexes, see Fig. 1C and asterisk in Fig. 3B)(Goode et al., 1996a; Goode et al., 1996b). As a result, very few eggs are laid, and those that are exhibit weakly ventralized eggshells, suggesting a defect in EGFR activity during DV patterning of the follicular epithelium.

To examine the requirement for GSLs in EGFR signaling, we used the viable and cold-sensitive brn^fs107 allele, in addition to the lethal egh and brn alleles. We molecularly characterized brn^fs107 and found a single nucleotide substitution changing the conserved tyrosine residue at position 147 to an asparagine. Even when raised at 18°C, brn^fs107 mutant females are fertile and rarely display early oogenesis defects, but produce eggs with fused respiratory appendages (RAs) (Fig. 1G, arrow). At this temperature, the brn^fs107eggshell phenotype is highly penetrant and this allele behaves genetically as a null for this phenotype (Goode et al.,1992). Because of its viability and robust eggshell defect, the brn^fs107 allele is useful to study the role of GSLs in late oogenesis events such as DV patterning.

To assess the level of EGFR activity in the follicular epithelium of the brn^fs107 mutant during DV patterning, we looked at the expression of argos (aos), which requires amplification of EGFR signaling, and of rhomboid1 (rho1; rho -FlyBase), which is induced prior to the amplification phase(Wasserman and Freeman, 1998). In wild-type stage 11 egg chambers, aos mRNA is expressed in follicle cells in a triangle above the oocyte nucleus, being more abundant at the anterior margin of the epithelium (Fig. 2A,A′, arrow). rho1 transcripts are present as two L-shaped stripes, in a pattern complementary to that of aos(Fig. 2C,C′). In stage 11 brn mutants, aos and rho1 were still expressed, but at lower levels. aos mRNA was only detected in the anterior-most follicle cells (Fig. 2B,B′, arrow), whereas rho1 transcripts were distributed in a single broad domain covering the anterior third of the dorsal follicular epithelium (Fig. 2D,D′, bracket). Strikingly, these expression domains resemble the wild-type aos and rho1 patterns from stage 10a egg chambers (see Peri et al.,1999; Queenan et al.,1997; Ruohola-Baker et al.,1993). This suggests that the initial activation of EGFR by Grk proceeded normally, but that the amplification process itself did not take place. These observations fit nicely with the fusion of the RA in the brn mutant because the amplification step and aos expression are necessary for the splitting of the RA(Wasserman and Freeman, 1998). These results provide evidence that GSLs are required during DV patterning of the follicle cells to achieve high levels of EGFR activity.

A decrease in EGFR signaling could originate from a defect at any level of the pathway. Cell culture studies have shown that vertebrate GSLs physically interact with EGFR, modulating its activity in cis(Miljan et al., 2002; Yoon et al., 2006). Given that the GSL biosynthetic pathway is active in the follicular epithelium(Wandall et al., 2005), we might expect GSLs to be required in follicle cells. Nonetheless, clonal analysis has suggested that egh and brn are required in the germ line (Goode et al.,1996a). It is thus possible that GSLs act in both tissues.

To investigate this, we extended our previous analysis of the activity of the GSL biosynthetic pathway during oogenesis. Since antibodies against elongated GSLs or Egh and Brn are not available, we used an antibody that specifically recognizes the non-elongated biosynthetic intermediate produced by Egh, mactosylceramide (MacCer, Fig. 3A) (Wandall et al.,2005). Theoretically, MacCer detection indicates that Egh is active. However, MacCer staining was never seen in wild-type ovaries(Wandall et al., 2005) (data not shown), suggesting (1) that Egh is not expressed in this tissue or (2)that Egh is expressed but that Brn immediately elongates MacCer. In the second case, abrogating brn function should lead to MacCer accumulation. When we examined MacCer distribution in brn^fs107 ovaries(Fig. 3B and data not shown),we found MacCer staining in the germarium (the structure producing the egg chambers) and early egg chambers. MacCer was again detected at stage 8, when DV patterning of the follicular epithelium is initiated, and persisted at least until stage 11. MacCer was present in follicle cells, confirming our previous observations, but was also strongly detected in both nurse cells and oocyte. This pattern is not allele-specific, as brn^1.6P6follicle cell and germ line clones generated at different time points gave the same result (data not shown). Therefore, the GSL pathway is active in both germ line and somatic follicle cells at the time when DV patterning is being established.

Next, we tested whether GSLs from follicle cells or the germ line participate in EGFR signaling. We produced egg chambers with large clones of follicle cells mutant for the lethal alleles brn^1.6P6 or egh^62d18, without affecting the germ line. In these experiments, the percentage of mutant follicular epithelia was high (65.8% for brn^1.6P6 as determined by MacCer staining; n=38 stage 10 and 11 egg chambers; data not shown), yet none of the eggs derived from such chambers exhibited defects in the DV polarity of their eggshells(Table 1A). In agreement with Goode and colleagues (Goode et al.,1996a), germ line clones of these alleles led to the formation of eggs with fused RAs (Table 1B). Furthermore, restricted expression of a UAS-brn transgene to the oocyte of brn^fs107 mutants completely rescued the ventralized eggshell phenotype (Table 1C). Altogether, these results indicate that GSLs are required in the germ line, but not in follicle cells, to support EGFR activity in follicle cells. This suggests a role in the ligand-producing cell, arguing against a direct modulation of EGFR activity in cis in this context.

Given the requirement for GSLs in the germ line, we examined whether the distribution of grk mRNA or Grk protein would be altered in brn mutant oocytes (see Introduction). In stage 10a egg chambers, the localization of grk transcripts in brn^fs107mutant oocytes was indistinguishable from that in the wild type: grkmRNA remained in the dorso-anterior (DA) corner, tightly associated with the oocyte nucleus (Fig. 4A,B).

By contrast, we detected stage-specific changes in Grk protein distribution. In wild-type egg chambers at stage 8, Grk was concentrated within the oocyte at its DA corner (Fig. 4C,C′). It was also present at the interface between the oocyte and the follicular epithelium, just above the oocyte nucleus, as a result of directional transport (Herpers and Rabouille, 2004). A small fraction of Grk is seen in adjacent follicle cells, reflecting uptake of secreted Grk(Ghiglione et al., 2002; Peri et al., 1999; Queenan et al., 1999). At stages 10a and 10b, the same pattern as previously was observed(Fig. 4E-G′), but in addition Grk was prominently localized in the extracellular space between the DA corner of the oocyte and the follicular epithelium, in dot-like structures. It was detected there at a considerable distance from the oocyte nucleus,indicative of Grk diffusion from the site of its secretion.

In the brn mutant, there was no difference to the wild type at stage 8 (Fig. 4D,D′), but from stage 10a, all brn mutant egg chambers displayed abnormal Grk localization (Fig. 4H-I′). Although Grk was still present near the oocyte nucleus, it was noticeably absent from the extracellular space. Moreover, we saw very little evidence of internalized Grk(Fig. 4I′, arrow and arrowheads; data not shown). These findings indicate that GSLs are dispensable for initial Grk secretion at the DA corner at stage 8, but that they are implicated either in sustained Grk secretion or in Grk stabilization once released at stage 10a.

To discriminate between sustained secretion or stabilization of Grk, we first sought to determine whether Grk secretion was blocked in stage 10a brn mutant egg chambers. Grk is synthesized as a transmembrane molecule that is cleaved in the endoplasmic reticulum (ER) by a member of the Rho family, possibly Rho2 (Stet - FlyBase)(Bokel et al., 2006; Ghiglione et al., 2002; Guichard et al., 2000; Urban et al., 2002). Although this processing is not necessary for Grk trafficking to the plasma membrane,it is a prerequisite for its secretion and activity(Bokel et al., 2006; Ghiglione et al., 2002; Peri et al., 1999; Queenan et al., 1999). The cleaved lumenal portion of Grk is then exported to the Golgi apparatus with the help of its cargo receptor Cornichon (Cni)(Bokel et al., 2006; Roth et al., 1995). Importantly, when either step is compromised, Grk mislocalizes in the oocyte cytoplasm. For instance, an uncleaved Grk accumulates in, and is dispersed throughout, the oocyte cytoplasm (Fig. 5D,D′). In a cni mutant, Grk abnormally diffuses into the ER lumen (Fig. 5C,C′).

In brn mutant oocytes, however, Grk did not appear to accumulate or mislocalize in the cytoplasm (Fig. 5B,B′), suggesting that its cleavage and export towards the plasma membrane were normal. Nevertheless, Grk could remain unprocessed in brn mutants and be rapidly targeted for degradation, preventing its accumulation and mislocalization. To investigate this, we performed a western blot on wild-type and brn ovarian extracts. Since endogenous Grk is notoriously difficult to detect by this technique, we overexpressed a construct encoding a Myc-tagged Grk protein in the germ line(Ghiglione et al., 2002). As in the wild type, Grk was still cleaved in the brn mutant(Fig. 5E).

To further explore whether Grk follows its normal exocytic route in the brn mutant, we compared Grk distribution in stage 10a wild-type and mutant oocytes by immunoelectron microscopy. As previously reported(Herpers and Rabouille, 2004),the bulk of intracellular Grk in wild-type samples was found in tER-Golgi units close to the oocyte nucleus (Fig. 5F). In brn mutants, Grk was also detected in these organelles (the morphology of which was normal), where its labeling density was not very different from that in wild-type ovaries(Fig. 5G and data not shown). Moreover, we observed neither an accumulation nor a loss of Grk pre- or post-Golgi (including at the plasma membrane) (data not shown). Altogether,these data argue against an essential role for GSLs in Grk transport to the plasma membrane.

These last results suggest that Grk is secreted in the brn mutant. To confirm this, we compared the phenotypes resulting from Grk overexpression in wild-type versus brn mutant backgrounds(Table 2). Grk overexpression in wild-type oocytes induces a range of gain-of-function phenotypes seen as increasingly dorsalized eggshells(Neuman-Silberberg and Schupbach,1994). In the brn mutant, not only did the eggshells lose their ventralization defect upon Grk overexpression, but they were also dorsalized. This implies that GSLs are not crucial for the secretion of a ligand that is, in addition, functional.

This then raises the question as to why the EGFR pathway is not properly activated in the brn mutant. It has been reported that DA follicle cells from stage 10 wild-type egg chambers have a lower amount of EGFR than the rest of the epithelium (Sapir et al.,1998). The authors suggested that in response to EGFR signaling, Egfr transcription is downregulated. We therefore used the decrease in EGFR levels as a measure of its activation by Grk.

First, we re-examined EGFR distribution in a wild-type background, as well as in a cni mutant as a negative control. As expected, at stage 10a,wild-type egg chambers displayed faint EGFR staining in most follicle cells of the dorsal midline (Fig. 6A,A′, arrowheads in A). In cni mutants, in which Grk is not secreted, EGFR staining was uniform(Fig. 6B-B″),establishing that its asymmetrical expression is a consequence of its own activation.

We then looked at EGFR distribution in brn mutant egg chambers. We found that throughout oogenesis, the EGFR pattern was identical to that observed in wild-type samples (Fig. 6C-C″, arrowheads; data not shown). Our data thus show that in the brn mutant, Grk is as active in the maintenance of the spatial regulation of EGFR in DA follicle cells as in wild-type egg chambers. It is however possible that the threshold level of Grk needed to maintain this regulation is lower than that required for proper rho1 and aos expression, and that the role of GSLs is in the regulation of the spatial distribution of secreted Grk.

To examine whether GSLs regulate the spatial distribution of secreted Grk,we visualized, for the first time, the extracellular Grk gradient in stage 10a wild-type and brn mutant egg chambers, using conditions that are more sensitive for the detection of extracellular proteins than conventional immunostaining (Strigini and Cohen,2000). We found that this method is suitable to monitor Grk distribution over the dorsal half of wild-type egg chambers, but the predicted ventral pool (Chang et al.,2008; Goentoro et al.,2006) was below detection.

In wild-type samples (Fig. 7A), the highest levels of extracellular Grk were concentrated within a rectangle-like domain along the dorsal midline. Strikingly, in brn samples, extracellular Grk was readily detected and occupied a domain that adopted a triangular shape, with a broad base toward the anterior oocyte cortex (Fig. 7B). We analyzed the Grk gradient in the two backgrounds by measuring the intensity of Grk staining along the DV axis and the dorsal midline.

Along the DV axis at the level of the source (above the oocyte nucleus, Fig. 7C), there was a greater amount of Grk in the brn mutant: the maximal intensity was almost double that of the wild type, and intensity values of 80 and above (dashed line) covered a distance approximately twice as long as in the wild type. It should be noted that follicle cells in this area are not fated to become part of the RA (Dorman et al.,2004). This Grk accumulation was restricted to the region overlying the nucleus in the brn mutant. Just posterior to this region (Fig. 7D), the maximal intensity was equal to that in the wild type, although it was still maintained over a greater area.

Along the dorsal midline, Grk levels in wild-type egg chambers were steeply elevated posterior to the source (Fig. 7E, asterisk) and reached a plateau, before declining. In brn mutants, a plateau of equivalent intensity was also formed, but its length was strongly reduced. In terms of spreading, however, there was still a low amount of Grk (intensity values above 50) that extended as far posteriorly as in the wild type.

The analysis of the extracellular Grk gradient therefore highlighted two major differences in the mutant versus wild-type background: the bulk of Grk diffusion was more efficient along the DV axis at the source and less efficient along the dorsal midline. Since the process that is compromised in the brn mutant is the splitting of the RA that requires high-level EGFR signaling to induce aos expression at the dorsal midline, these data indicate that GSLs play a crucial role in Grk signaling by preventing its diffusion away from the dorsal midline, thus maintaining high levels of extracellular Grk in this region.

Last, we asked whether EGFR, although seemingly normally distributed(Fig. 6), is involved in the altered diffusion of extracellular Grk in brn mutant chambers. To determine how Grk binding to its receptor controls its extracellular movement,we used the top¹ allele of Egfr. This is a hypomorphic viable allele that bears a point mutation in the ligand-binding domain (Clifford and Schupbach,1994), causing homozygous females to lay eggs with fused RAs(Fig. 1E). In our hands, this mutant was subviable and we obtained very few top¹ egg chambers of the right stage, all of them oriented in lateral views. This precluded analysis of the Grk gradient in dorsal views, but allowed us to measure extracellular Grk spreading at the dorsal midline in longitudinal confocal sections (Fig. 7F-K′). High and low levels combined, Grk spanned, on average, 69.4 ± 4.2% of the AP length of the extracellular space in top¹ samples (n=7), versus 54.7 ± 3.4% for wild-type samples (n=17). This result demonstrates that EGFR limits the range of extracellular Grk diffusion. By contrast, this value, although slightly reduced, was not significantly different in the brn mutant(48 ± 4.6%, n=14), indicating that Grk still efficiently binds EGFR.

In conclusion, our experiments show that GSLs shape the extracellular Grk gradient and strongly suggest that this is achieved through the direct modulation of Grk diffusion.

Here, we have assessed the function of GSLs in EGFR signaling during DV patterning of the Drosophila egg chamber. In this context, we find no evidence for the reported roles of GSLs in the direct modulation of EGFR activity and in trafficking. Instead, our results show that once Grk is secreted, GSLs are necessary for its efficient diffusion in the extracellular space. Although we cannot dismiss the possibility that GSLs might have a slight influence on Grk secretion, a decrease in Grk secretion cannot explain the observed changes in the extracellular Grk gradient shape in the brn mutant or the higher levels of secreted Grk that accumulate above the source, as compared with the wild type. GSLs are thus crucial for the formation of a Grk gradient that is able to achieve maximal activation of EGFR.

Surprisingly, we find that GSLs are only involved in the final step of Grk signaling during the establishment of the DV axis. Prior to DV patterning, Grk activates EGFR to set the AP axis of the egg (see Introduction). We have observed, however, that AP polarity is not compromised in egh and brn alleles (S.P., unpublished). Our analysis of MacCer distribution during oogenesis also supports the idea that the GSL biosynthetic pathway is not active at the time AP patterning is established(Fig. 3). Grk signaling therefore seems to be more sensitive to GSL function for the determination of DV fates.

Even during this process, there appear to be differential requirements for GSL activity. DV patterning proceeds in two distinct temporal phases of EGFR signaling, but only the second is under the control of GSLs. In a first phase(between stages 8 and 10a), the EGFR pathway establishes embryonic DV polarity and dorsal follicle cell fates. This phase culminates in the induction of rho1 transcription in DA follicle cells and depends on paracrine signaling mediated by Grk. This initial phase is not overtly affected in brn mutants as their embryos have a normal DV axis(Goode et al., 1992) and we found that rho1 expression was still induced(Fig. 2). By contrast, the second phase of EGFR signaling is triggered by rho1 expression and corresponds to an amplification of EGFR activity needed to split the RA(Wasserman and Freeman, 1998). This phase is disrupted in the brn mutant because the expression of rho1 and aos is not upregulated(Fig. 2), as exemplified by the fusion of the RA.

According to Wasserman and Freeman, the amplification phase is independent of Grk and relies on autocrine EGFR signaling(Wasserman and Freeman, 1998). However, we show here that GSLs, unlike the other molecules implicated in this process, act in the germ line to regulate the distribution of extracellular Grk (Table 1 and Fig. 7A-E). This indicates, as previously suggested by others (Peri et al., 1999), that this phase is not solely autocrine and that there is still a need for Grk-mediated paracrine signaling.

At stage 10a, the vitelline membrane is already being deposited as vitelline bodies in the extracellular space between the oocyte and the follicular epithelium. Morphological data show that these bodies have not yet fused, leaving space for a number of interdigitating microvilli emanating from the oocyte membrane and the apical side of the follicle cells(King, 1970). Since the brn mutation specifically affects Grk signaling at this stage, we propose that GSLs play a role in Grk accessibility to follicle cells and that this is likely to be mediated through the microvilli. In support of this, at stages at which elongated GSLs are not essential (AP patterning and the onset of DV patterning), the oocyte plasma membrane is closely apposed to the apical side of the follicle cells. This, therefore, supports our hypothesis that GSLs are only required when Grk is not easily accessible to its receptor.

Our study provides the first experimental description of the wild-type extracellular Grk gradient at stage 10a and uncovers an unexpected feature(Fig. 7A,C-E): at the dorsal midline, past the source, high and steady levels of Grk are maintained over about half of the AP axis length. This result is at odds with a mathematical modeling of the Grk gradient that predicted a shallow decrease from anterior to posterior (Goentoro et al.,2006). However, from a strong and constant level of Grk over half the dorsal midline, we expect the expression domains of the Grk primary target genes to have an identical width along most of the AP axis. This is precisely what is observed for kekkon-1 and pipe, which are clearly EGFR primary target genes (Chang et al.,2008; Peri et al.,2002; Sapir et al.,1998; Sen et al.,1998). We thus suggest that a stripe-shaped source of extracellular Grk along the dorsal midline, rather than a point-like source of Grk above the oocyte nucleus, is more efficient in accommodating patterning across the entire epithelium.

Besides contributing to the shape of the Grk gradient, high Grk levels along the dorsal midline might serve to upregulate rho1 expression,leading to higher EGFR activity, aos expression and splitting of the RA primordium. Indeed, we have shown that in the absence of elongated GSLs,weak rho1 expression is retained but it is not upregulated or refined. As mentioned above, Grk signaling is necessary for this step. Since,in the brn mutant, there is a reduction in the high levels of extracellular Grk along the dorsal midline, we propose that a low Grk threshold is sufficient to initiate and maintain rho1 transcription(as well as the spatial regulation of EGFR, Fig. 6C-C″), whereas a higher Grk threshold increases rho1 expression levels.

What could be the basis for the discrepancy between our results and the mathematical modeling of the Grk gradient? In the latter(Goentoro et al., 2006), EGFR expression was assumed to be uniform throughout the follicular epithelium. However, we showed that at stage 10a, EGFR levels are lower along part of the dorsal midline (Fig. 7) in a region coincident with that of high Grk levels. Furthermore, we found that decreasing Grk binding to EGFR increased Grk spreading(Fig. 7F-K′). Therefore,at the dorsal midline, the reduction in EGFR levels might saturate receptor occupancy. This could allow a large quantity of Grk to remain unbound,facilitating its movement toward the posterior pole.

Our most striking result is that GSLs shape the extracellular Grk gradient and play a role in Grk diffusion without apparently interfering with the regulation of Grk diffusion by EGFR (Fig. 6, Fig. 7F-K). But what could that role be? Grk movement in the extracellular space between the oocyte and the follicular epithelium is complicated by the formation of the vitelline membrane (see above). Grk could either be released into the extracellular space or it could remain associated with the oocyte plasma membrane and localize to its microvilli. We could not distinguish between these alternatives, as immunofluorescent staining is of insufficient resolution and the extracellular space was not well preserved in our immunoelectron microscopy experiments. Others have nevertheless reported the presence of Grk on microvilli (Bokel et al., 2006). GSLs could therefore be important for Grk targeting to microvilli versus flat portions of the membrane. This, however, is unlikely because Grk still activates EGFR in the brn mutant(Table 2 and Fig. 6), indicating that it can encounter its receptor. By contrast, what Grk fails to do in the mutant context is to concentrate along the dorsal midline at a distance from its point of secretion. This suggests that GSLs function in the planar transport of Grk along the AP axis, from one oocyte microvillus to the next, a hypothesis supported by the fact that we found the oocyte microvilli to be oriented parallel to the AP axis (S.P., unpublished).

An intriguing property of secreted Grk in the brn mutant context is that it is detected by extracellular staining and not conventional immunostaining. In an effort to understand the basis for this, we found that secreted Grk is sensitive to the presence of detergent and to temperature,suggesting that its conformation relies on the presence of GSLs once it reaches the cell surface (see Fig. S1 in the supplementary material). Interestingly, GSLs induce a conformational change in the amyloidβ-protein upon its release from the plasma membrane (reviewed by Ariga et al., 2008). It is thus possible that under our experimental conditions, Grk conformation is not fully restored, modifying its ability to diffuse.

In this case, how could the two processes be linked? There is increasing evidence that the spreading of secreted molecules depends on elaborate events involving their multimerization and/or incorporation into higher-order structures such as lipoprotein particles(Gallet et al., 2006; Panakova et al., 2005; Zeng et al., 2001). It is therefore tempting to speculate that a change in secreted Grk conformation that depends on plasma membrane GSLs reflects its packaging into special structures that are required for its efficient transport along microvilli. Since mammalian GSLs can be shed from the plasma membrane and are found circulating with secreted lipoprotein particles (see Clarke, 1981; Lauc and Heffer-Lauc, 2006),GSLs could enhance Grk spreading by delivering it to these particles.

Another, non-mutually exclusive means by which GSLs could affect Grk diffusion is linked to their enrichment in plasma membrane microdomains. Since GSLs can interact with proteins through their oligosaccharide chain, GSLs could bind Grk, or a Grk co-factor, sorting Grk into such domains. It has been reported that Grk is potentially palmitoylated(Miura et al., 2006), and palmitoylation is one of the signals that target proteins to membrane microdomains (reviewed by Pike,2004). Because Grk recruitment to these domains has not been addressed and because the detection of extracellular Grk by biochemical means in ovaries has so far eluded our attempts (S.P., unpublished), understanding how GSLs regulate Grk diffusion will have to await the generation of better tools.