Drosophila Epsin mediates a select endocytic pathway that DSL ligands must enter to activate Notch

The Notch signaling pathway plays profound roles in specifying cell fate during animal development. Notch proteins are single-pass transmembrane receptors that are activated by transmembrane ligands of the Delta/Serrate/Lag2 (DSL) family (reviewed by Greenwald, 1998; Mumm and Kopan, 2000; Schweisguth, 2004). During transit to the cell surface, Notch is processed once in the extracellular domain, at the `S1' site, yielding a heterodimer composed of the N- and C-terminal fragments. Binding of DSL ligands to Notch induces a second extracellular domain cleavage at the `S2' site, close to the transmembrane domain. The remaining, C-terminal portion of the receptor, called NEXT, is then cleaved at the `S3' site within the membrane, releasing the intracellular domain from the cell surface. Once liberated, the intracellular domain enters the nucleus, where it acts together with the sequence-specific DNA-binding protein Suppressor of Hairless (SuH) to activate the transcription of target genes.

Although much is known about how Notch transduces signals once the receptor undergoes the ligand-dependent S2 cleavage, the mechanism by which DSL ligands engage Notch and trigger this cleavage are less well understood. Recent evidence suggests that Clathrin-mediated endocytosis of DSL ligands is required for this activating event. First, Dynamin, which functions to pinch off invaginating coated pits to form endocytic vesicles, appears to be required for Notch signaling prior to transmembrane (S3) cleavage of the receptor (Seugnet et al.,1997; Struhl and Adachi,2000). Second, the E3-Ubiquitin ligases Neuralized (Neur) and Mind bomb (Mib) interact physically with Delta (Dl), promote Dl ubiquitination and internalization, and enhance its signaling activity(Deblandre et al., 2001; Lai et al., 2001; Pavlopoulos et al., 2001; Yeh et al., 2001; Itoh et al., 2003; Le Borgne and Schweisguth,2003a). Third, deleted or mutated forms of Dl that cannot be efficiently endocytosed have reduced, or no, signaling activity(Parks et al., 2000; Itoh et al., 2003). Finally,Notch signal transduction appears to be correlated with trans-endocytosis of the Notch extracellular domain into the signal-sending cells(Parks et al., 2000).

It has been proposed that endocytosis of DSL ligands on signal-sending cells that are bound to Notch on adjacent signal-receiving cells induces, via mechanical stress, the S2 or S3 cleavage of the receptor, thus activating signal transduction (Parks et al.,2000). Alternatively, it has been suggested that endocytosis may function to cluster DSL ligands, or to clear previously shed ectodomains of Notch, to ensure robust activation of the receptor, or to package DSL ligands into signaling exosomes (reviewed by Le Borgne and Schweisguth, 2003b).

To resolve the role of endocytosis in signaling by DSL ligands, we sought to identify and manipulate endocytic proteins that are required specifically in signaling cells for the activation of Notch in receiving cells. As we will describe, we have found that Liquid Facets (Lqf), the sole DrosophilaEpsin (Cadavid et al., 2000),plays just such a role, allowing us to examine the relationship between DSL endocytosis and signaling.

Epsins are endocytic proteins that bind phosphatidylinositol(4,5)-bisphosphate [PtdIns(4,5)P₂] in the plasma membrane as well as Clathrin, AP-2 Adapter Complex, and other accessory proteins in coated pits (reviewed by Wendland,2002). Epsins were initially thought to be core components of the endocytic machinery because of the dominant-negative effects of truncated Epsin proteins on endocytosis in mammalian cells(Chen et al., 1998; Ford et al., 2002), their essential role in yeast endocytosis(Wendland et al., 1999), and their inherent capacity to induce membrane curvature(Ford et al., 2002) and bind other core components such as Clathrin and AP-2(Chen et al., 1998; Owen et al., 1999; Rosenthal et al., 1999; Wendland et al., 1999; Drake et al., 2000). More recently, however, the identification of Ubiquitin-interacting motifs (UIMs)in Epsins, as well as in other proteins involved in membrane trafficking(Hofmann and Falquet, 2001),have led to the suggestion that Epsins belong to a family of cargo-selective adapters that link mono-ubiquitinated cell-surface proteins with the endocytic machinery (Wendland,2002).

Here, we report that DSL ligands must normally be endocytosed in signal-sending cells via the action of Lqf to activate Notch on the surface of signal-receiving cells. Surprisingly, however, bulk endocytosis of DSL ligands appears normal in the absence of Lqf. We resolve this apparent paradox by providing evidence that Lqf is unique amongst adapters that target mono-ubiquitinated cargo proteins for internalization, in that it allows them to enter a special endocytic pathway that DSL ligands must enter to acquire signaling activity. We also show that this requirement can be bypassed by introducing the internalization signal that normally mediates internalization and recycling of the Low Density Lipoprotein (LDL) receptor. On the basis of these results, we hypothesize that Epsin-mediated endocytosis might be required to allow DSL proteins to be recycled rather than degraded following internalization, possibly to convert them from inactive pro-ligands into active ligands.

The Dl-coding sequence, and variants thereof, were introduced into flies as UAS>CD2,y+> transgenes(Jiang and Struhl, 1995), and expressed following excision of the >CD2,y+> Flp-out cassette in clones of cells, or in the germline. In many experiments, the encoded forms of Dl were epitope tagged, either with six tandem copies of the Myc epitope(^MYCDl), or HRP (^HRPDl); both tags were inserted in the extracellular domain at the position indicated by the slash (/) in the sequence ESYDSVTFDA/QGATTQAR, 17 amino acids upstream of the transmembrane domain. We could not detect any difference in the subcellular distribution or biological activity of either Myc- or HRP-tagged Dl, relative to untagged Dl. The relevant sequences, or joins, of the various chimeric forms of Dl are as follows (Dl sequences are shown in bold, followed by a slash after the stop transfer signal):

• Dl, AACVVFCMKRKRKR/AQE...

• Dl^ΔC, AACVVFCMRRR/

• Dl^LDL+, /GSRLRNINSINFDNPVYQRTT

• Dl^LDLm, /GSRLRNINSINADAAVAQRTT

• Dl^R+, /GSWIPSFYNVVTGKTLALPNLIALQHIPLSPAGVIAKRPAPIALPNSCAA

• Dl^Ubi+, /GSPMQIFVRTLTGRTITLEVEPSDTIENVRARIQDREGIPPDQQRLI...

Dl^Rm is identical to Dl^R+ except that the two Lysines are changed to Arginine. Dl^Ubim is identical to Dl^Ubi+except that the Isoleucine at position 44 is changed to Alanine. The Lysines in the Dl stop transfer sequence were deleted for the Dl^LDL+,Dl^LDLm, Dl^Rm, Dl^Ubi+ and Dl^Ubimchimeric proteins. For Dl^ΔC and the Dl^R+ chimera,both possible forms (with and without the Lysines in the stop transfer sequence) were assayed; no difference in behavior was observed for either protein.

lqf¹²²⁷ (this study), lqf^ARI,lqf^BT (Overstreet et al.,2003); P[w⁺],lqf⁺(Cadavid et al., 2000); hrs^D28(Lloyd et al.,2002); Tubα1-Gal4, UAS-GFPnls(Struhl and Greenwald, 2001); wg^CX4, C765-Gal4, Vg-boundary-lacZ(Zecca et al., 1996); nub-Gal4, UAS-y⁺(Calleja et al., 1996); omb-lacZ, dpp-lacZ (Nellen et al., 1996); UAS-CD8-GFP, Tubα1-G80(Lee and Luo, 1999); UAS-wg, UAS-dpp, UAS-hh (K. Basler, unpublished); UAS-fng,UAS-Ser (Panin et al.,1997); UAS-Neur (Lai et al., 2001); Dl-lacZ, arm-lacZ, Ubi-GFP (from Bloomington stock center).

y hsp70-flp; lqf¹²²⁷ FRT2A/Ubi-GFP (or arm-lacZ) FRT2A [lacZ reporter genes on X, II or III; mwh was used in cis with lqf¹²²⁷ to mark hairs in the adult cuticle; clones of lqf^BT and lqf^ARI cells were generated in the same way and gave indistinguishable results].

These clones were the same as above except for the added presence of the appropriate UAS-Dl, UAS-neur, C765-Gal4 and/or nub-Gal4transgenes. We note that Neur does not enhance the endocytosis or signaling activity of either the Dl^LDL+ or Dl^R+ chimeric proteins,presumably because it does not recognize their cytosolic domains as substrates. Therefore, we could not use ectopic Neur expression to create a sensitized background to test whether Dl^LDL+ or Dl^R+endocytosis is compromised in lqf^- cells, in contrast to the situation with native Dl.

y hsp70-flp Tubα1-Gal4 UAS-GFPnls (or UAS-CD8-GFP)/y hsp70-flp; lqf¹²²⁷FRT2A/Tubα1-Gal80 FRT2A [UAS-X transgenes and lacZ reporters on X, II or III; clones of lqf^BTcells (as well as control clones of lqf⁺ cells) that express markers; and/or UAS-X transgenes were generated in the same way]. Dpp, Hh and Wg signaling were assayed by, respectively, omb-lacZ expression (Grimm and Pflugfelder, 1996; Nellen et al., 1996), dpp-lacZ and Collier expression(Basler and Struhl, 1994; Vervoort et al., 1999), and Dl and Senseless expression (Micchelli et al., 1997; Nolo et al.,2000).

y hsp70-flp Tubα1-Gal4 UAS-GFPnls/y hsp70-flp; UAS-Dl (or UASSer)/+; lqf¹²²⁷ Tubα1-Gal80 FRT2A/arm-lacZ FRT2A.

y hsp70-flp/y hsp70-flp UAS-^HRPDl; hrs^D28(wg^CX4) FRT40/ P[w⁺],lqf⁺Tubα1-Gal80 FRT40; lqf¹²²⁷ C765-Gal4/lqf¹²²⁷(or +); clones of hrs^- lqf^- cells expressing ^MycDl were obtained similarly, using a UAS-^MycDl transgene in cis with hrs^D28FRT40 in place of the UAS-^HRPDl transgene in X, and gave the same result.

Clones were generated by heat shocking first or second instar larvae at 37°C for 60 minutes.

Imaginal discs were fixed and stained by standard procedures (e.g. Jiang and Struhl, 1995; Zecca et al., 1996), using mouse α-Dl (Developmental Studies Hybridoma Bank, DSHB), guinea pigα-Dl (Parks et al.,2000), guinea pig α-Hrs(Lloyd et al., 2002), mouseα-Wg (DSHB), mouse α-Cut (DSHB), guinea pig α-Lqf(Overstreet et al., 2003),rabbit α-Col (Vervoort et al.,1999), rat α-Ci (DSHB), rat α-Ser(Panin et al., 1997), and commercially available mouse α-Myc, rabbit α-HRP, and rabbit anti-βGal. Two protocols were used to stain cell surface expression (both gave similar results): (1) living discs were incubated in Drosophilatissue culture media with the first primary antisera at 4°C for 20 minutes, rinsed several times with ice-cold media, fixed in the absence of detergent, and then subjected to the standard protocol using additional primary antisera in the presence of detergent; and (2) discs were fixed and processed for immunofluorescent staining by the standard protocol, except in the absence of detergent until after the secondary antisera was removed.

Proteins extracted from mature third instar wing discs carrying multiple clones of wild-type or lqf^- cells co-expressing ^MYCDl and Neur (MARCM technique) were separated by SDS-PAGE electrophoresis, and blotted to Nitrocellulose for western analysis by standard protocol (see Struhl and Adachi,2000). Commercially available mouse α-Myc was used to detect ^MYCDl.

In screens for mutations affecting wing pattern(Jiang and Struhl, 1995; Chen and Struhl, 1999), we obtained six alleles of a single complementation group that cause phenotypes similar to those caused by the loss of Notch signaling, namely severe wing notching, wing vein thickening and bristle tufts(Fig. 1B,C). All six alleles fail to complement existing alleles of liquid facets (lqf),and are associated with nonsense or missense mutations in the lqf-coding sequence (Overstreet et al., 2003) (data not shown). One new allele, lqf¹²²⁷, truncates the coding sequence after amino acid 119 in the middle of the ENTH domain, the most N-terminal conserved domain,and abolishes Lqf protein expression in vivo(Fig. 1A; data not shown). We refer to this allele as lqf^-, and use it for all experiments described below unless stated otherwise. A transgene containing the intact lqf gene (Cadavid et al., 2000) rescues the lethality of lqf^-homozygotes, as well as all of the mutant phenotypes associated with lqf^- clones (see Materials and methods; data not shown). Lqf encodes the sole ortholog of vertebrate Epsin1(Cadavid et al., 2000); a second Drosophila protein, sometimes referred to as Dm Epsin2(Overstreet et al., 2003),lacks several conserved domains found in Lqf and vertebrate Epsin1, and appears instead to be the Drosophila ortholog of vertebrate EpsinR, a functionally distinct Epsin-related protein (see Mills et al., 2003).

In imaginal wing discs, signaling by the DSL ligands Delta (Dl) and Serrate(Ser) specifies the wing margin at the dorsoventral (DV) compartment boundary,and can be assayed by boundary-specific expression of wing margin genes (or their protein products), such as cut, wingless (wg) and vestigial (vg) (reviewed by Irvine and Vogt, 1997). lqf^- clones resemble Dl^-Ser^- clones or N^- clones in that they cause the loss of cut, wg and vg boundary-specific expression when they abut or cross the DV compartment boundary(Fig. 1D; data not shown),corroborating the Notch-related phenotypes of lqf^- clones observed in the adult wing.

The loss of margin gene expression in lqf^- clones is not cell autonomous. Instead, wild-type cells can rescue the expression of margin specific genes in adjacent lqf^- cells (e.g. cut; Fig. 1D′). Similarly, we observed non-autonomous rescue of lqf^-clones in the adult, where the presence of wild-type cells can rescue the ability of neighboring lqf^- cells to form single bristles(Fig. 1C). In both respects, as well as in others (see supplementary material), lqf^-clones resemble Dl^- Ser^- clones, but differ from N^- clones, which show a strictly cell-autonomous loss of Notch target gene expression.

Collectively, these data establish an obligate role for Lqf in Notch signaling, and implicate Lqf in sending, rather than receiving, DSL signals.

To determine whether Lqf is required in signal-sending cells, we used the MARCM technique (Lee and Luo,1999) to generate lqf^- clones that express either Dl or Ser under Gal4 control (Materials and methods).

Notch is normally expressed in both the D and V compartments of the wing primordium, but is modified in D cells by the action of the glycosyltransferase Fringe (Fng) so that it responds preferentially to Dl signaling from V cells (de Celis and Bray,1997; Fleming et al.,1997; Panin et al.,1997; Blair, 2000). Ser is expressed predominantly in D compartment cells, and signals in the opposite direction, activating unmodified Notch in V cells. Clones of cells that express Dl under Gal4 control activate Notch strongly in adjacent wild-type cells only when located in the D compartment, as monitored by the expression of margin-specific genes like cut(Fig. 2A). Conversely,Ser-expressing clones activate Notch strongly only when located in the V compartment (Fig. 2D). In both cases, the levels of exogenous Dl and Ser expression are several fold higher than the peak levels of endogenous Dl and Ser generated along the DV boundary,and this overexpression autonomously inhibits the activation of Notch in cells within the clones (de Celis and Bray,1997; Micchelli et al.,1997) (Fig. 2A,D).

Clones of lqf^- cells that overexpress either Dl or Ser fail to induce margin gene expression, irrespective of where they are located within the wing primordium (Fig. 2B,E; data not shown). Indeed they behave like simple lqf^- clones in blocking normal margin gene expression when they abut, or cross, the DV compartment boundary(Fig. 1D). Thus, Lqf is required in DSL signal-sending cells to activate Notch in adjacent,signal-receiving cells.

To determine whether Lqf is required in signal-receiving cells for Notch activation or signal transduction, we used a modification of the MRCM technique to determine whether clones of lqf^- cells can transduce ectopic Dl or Ser signals sent by adjacent cells that are wild-type for lqf. To do this, we generated `twin spots' comprising adjacent daughter clones in which one clone expresses Dl or Ser under Gal4 control and the other clone is lqf^- (see Materials and methods). We find that Dl-expressing clones located in the D compartment induce cut expression in surrounding cells, even when the responding cells belong to adjacent lqf^- clones(Fig. 2C). The same is true for Ser-expressing clones in the V compartment(Fig. 2F). Thus, wild-type cells that express ectopic Dl or Ser can activate Notch in adjacent lqf^- cells, indicating that Lqf is not essential in signal-receiving cells to transduce DSL ligands. In separate experiments, we confirmed that lqf^- cells can transduce normal levels of endogenous Dl sent by neighboring wild-type cells by assaying Cut expression in clones of lqf^- cells that ectopically express Fng in the V compartment (see supplementary material).

The results of both sets of experiments lead us to conclude that Lqf is only required in signal-sending cells, and not in signal-receiving cells, to activate the Notch transduction pathway.

To assess whether Lqf might be required more generally for generating,modulating or transducing extracellular signals, we have tested whether lqf^- cells are capable of sending and receiving three other extracellular signals, each representing a different family of secreted ligands: Decapentaplegic (Dpp), Wingless (Wg) and Hedgehog (Hh). As in the case of DSL ligands, we generated clones of lqf^- cells that ectopically expressed each of these ligands, or which were located in regions where these signals are normally transduced, and assayed for target gene expression. For each ligand, we were unable to detect any change in signaling activity, in either sending or receiving cells, as a result of abolishing lqf function (Fig. 2G,H, data not shown; see Materials and methods).

To determine whether lqf^- cells behave normally in terms of their ability to grow, proliferate and interdigitate with surrounding cells, we used twin-spot analysis to compare the behavior of lqf^- and Dl^- Ser^- clones with their wild-type, sibling clones. We find that lqf^-and Dl^- Ser^- clones behave similarly in terms of clone size, cell density and the wiggliness of their borders, indicating that the mutant cells of both genotypes grow, divide and interdigitate normally (Fig. 2H). Thus, aside from the failure to send DSL-signals, lqf^- cells appear indistinguishable from wild-type cells, suggesting a dedicated requirement for Epsin in sending DSL signals.

To ascertain whether Lqf might be required for DSL ligands to reach the cell surface, we examined the abundance of endogenous Dl on the surface of lqf^- mutant cells using living disc, or non-detergent,staining protocols (see Materials and methods). In wild-type wing discs, Dl expression peaks along the DV compartment boundary and the presumptive wing veins. We were unable to detect a change in the surface abundance of Dl in clones of lqf^- cells relative to neighbhoring wild-type cells, except that large clones that abut or cross the DV boundary are associated with reduced surface expression (data not shown). This reduction,however, can be attributed to the loss of DSL signaling associated with the clone [which is normally required for peak Dl expression along the DV boundary(Micchelli et al., 1997)]; we observe a similar reduction when the staining was performed in the presence of detergent to detect both cytosolic and surface expression of Dl (e.g. Fig. 4C).

We repeated this experiment using the Gal4 technique to drive uniformly high levels of Dl throughout the wing discs, again using both living disc and non-detergent protocols, and could not detect any reduction in the amount of Dl on the surface of lqf^- cells relative to their wild-type neighbors (Fig. 5A;data not shown). Thus, it appears that DSL ligands reach the surface of lqf^- cells normally, but cannot activate Notch on the surface of neighboring cells.

Intact Dl and Ser normally accumulate in intracellular puncta, some of which co-localize with the endosomal marker Hepatocyte growth factor-regulated tyrosine kinase substrate (Hrs) (Kooh et al., 1993; Lloyd et al.,2002) (data not shown), as well as at the apical cell surface. By contrast, C-terminally truncated forms of Dl that lack the intracellular domain (Dl^ΔC) accumulate predominantly at the cell surface and, like C-terminally truncated forms of Ser (Ser^ΔC)(Sun and Artavanis-Tsakonas,1996), cannot activate Notch on the surface of neighboring cells(Itoh et al., 2003) (data not shown). If such truncated DSL ligands fail to signal because they cannot be endocytosed, replacement of the missing Dl cytosolic domain with heterologous domains that contain other internalization signals should rescue both endocytosis and signaling activity. Moreover, mutations in the internalization signals of these domains should eliminate their rescuing activity. We have tested and confirmed these predictions with two heterologous domains, each containing a different internalization signal.

First, we replaced the missing intracellular domain of Dl^ΔC with a 21 amino acid peptide from the Low Density Lipoprotein (LDL) receptor that contains either the wild-type internalization signal FDNPVY, or a mutant signal, ADAAVA(Chen et al., 1990). The LDL peptide contains two Lysines; these were replaced by Arginine to avoid their serving as possible acceptors for ubiquitination. Both the wild-type(Dl^LDL+) and mutant (Dl^LDLm) chimeric proteins were labeled by the insertion of six copies of the Myc epitope tag in the juxtamembrane portion of the extracellular domain (see Materials and methods). When expressed in the wing disc, Dl^LDL+ shows a similar subcellular distribution to wild-type Dl, accumulating on both the apical cell surface and in intracellular puncta (Fig. 3A; data not shown). Dl^LDL+-expressing clones, like wild-type Dl-expressing clones, can induce cut activity in surrounding cells (Fig. 3A),indicating that the chimeric protein has signaling activity. However, they differ from wild-type Dl-expressing clones in that they only induce cut when located close to the DV boundary (compare Fig. 3A and Fig. 2A). Hence, we infer that Dl^LDL+-expressing clones have reduced signaling activity relative to wild-type Dl-expressing clones, and require the additional boost provided by endogenous signaling from neighboring wild-type cells to activate cut.

By contrast, Dl^LDLm accumulates predominantly on the apical cell surface, but not in intracellular puncta, and lacks signaling activity(Fig. 3C). Indeed, clones of cells overexpressing Dl^LDLm that abut the DV boundary block normal Notch signaling across the boundary, as would be expected if Dl^LDLmcan inhibit Notch transduction within the same cell (like wild-type Dl; Fig. 2A), but is devoid of the capacity to activate Notch in adjacent cells.

Second, we found serendipitously that replacement of the missing cytosolic domain of Dl^ΔC with a random peptide, R+, of 50 amino acids(Dl^R+; see Materials and methods) also restored normal behavior. Dl^R+ accumulates in intracellular puncta as well as on the apical cell surface; in addition it activates Notch in neighboring cells(Fig. 3D). The R+ peptide contains two Lysines that might potentially serve as acceptors for ubiquitination. Replacement of both Lysines with Arginine blocked the rescuing activity of the R+ peptide. The mutant protein, Dl^Rm, accumulated predominantly only on the cell surface and lacked signaling activity(Fig. 3F); moreover, clones of Dl^Rm that abutted the DV boundary interrupted signaling across the boundary.

To assess the possibility that mono-ubiquitination of native Dl, as well as the Dl^R+ chimera, might suffice to provide an internalization signal, we replaced the missing cytosolic domain of Dl^ΔC with Ubiquitin itself, and with a corresponding mutant form of Ubiquitin in which the Isoleucine at position 44 was mutated to Alanine, which functionally inactivates the internalization signal(Shih et al., 2002). All seven Lysine residues in each Ubiquitin domain were also replaced by Arginine to avoid additional ubiquitination (Terrell et al., 1998). Both the resulting proteins, Dl^Ubi+ and Dl^Ubim, accumulated on the cell surface, as well as in intracellular puncta (Fig. 3G,H; data not shown). However, far fewer puncta were found in Dl^Ubim-expressing cells than in Dl^Ubi+-expressing cells,and only Dl^Ubi+ was able to signal to neighboring cells(Fig. 3G,H). These data indicate that mono-ubiquitination is sufficient for Dl endocytosis and signalling, and suggest that at least one of the Lysines in the R+ peptide serves as Ubiquitin acceptor, allowing the protein to be internalized and to signal. We note that the Dl^Ubi+ protein appears to have only weak signaling activity relative to Dl or Dl^R+, as we could only detect induction of vg boundary-specific expression, but not cut or wg expression (Fig. 3G; data not shown).

We conclude: (1) that the cytosolic domain of Dl is essential for its endocytosis; (2) that mono-ubiquitination is sufficient for Dl internalization; and (3) that Dl endocytosis is essential for signaling activity.

Given that Epsin has been implicated in endocytosis, lqf^- cells may fail to send DSL signals because they are generally impaired for endocytosis. However, Dpp, Hh and Wg signaling, Wg internalization, and cell growth and proliferation are not adversely affected by the absence of Lqf (Fig. 2G,H, Fig. 4D,E;data not shown) suggesting that endocytosis is not significantly impaired overall.

Alternatively, Lqf might be required specifically for the endocytosis of DSL ligands. To assess this possibility, we first examined the effects of lqf^- clones in the developing retina, where they have been reported to cause abnormally high levels of Dl on the cell surface, consistent with impaired Dl endocytosis (Overstreet et al., 2003). We find that such clones do indeed cause elevated surface expression of Dl, but we also observed that endogenous Dl transcription (as assayed using a Dl-lacZ reporter gene), is strongly upregulated in the mutant cells (Fig. 4A), apparently as a consequence of the lack of Notch signaling(Baonza and Freeman, 2001). Furthermore, we can detect Dl staining in intracellular puncta in such lqf^- eye disc clones(Fig. 4B). Thus, the elevated surface accumulation of Dl observed in lqf^- eye clones can be ascribed to elevated Dl expression in the mutant cells, and may not reflect impaired Dl endocytosis.

Second, we generated lqf^- clones in wing discs expressing uniformly high levels of exogenous Dl under Gal4 control, and compared Dl staining in lqf^- cells and their wild-type neighbors. Under this condition, the level of Dl expression does not vary between wild-type and lqf^- cells, simplifying analysis. We were unable to detect any difference in the subcellular distribution of Dl between lqf^- and adjacent wild-type cells. In both cases,Dl was localized predominantly at the cell surface, as well as in similar numbers of intracellular puncta, many of which co-localize with the endosomal protein Hrs (Fig. 5A,A′;data not shown). We obtained the same result in separate experiments in which we assayed the subcellular distribution of endogenous Dl only (i.e. in the absence of overexpressed Dl; Fig. 4C; data not shown).

Third, we reasoned that if the Dl-positive puncta in lqf^- clones are indeed endocytic, the appearance of such puncta should change in the absence of hrs activity, which interferes with the maturation of early into late endosomes, and causes the formation of abnormal endosomal structures (Lloyd et al., 2002). To test this, we generated both hrs^- and hrs^- lqf^- clones. We find that endogenous Dl accumulates in abnormally large puncta in both types of clones, and similar results were obtained when these clones express exogenous Dl under Gal4 control (data not shown). We note that the block in endosomal maturation caused by the removal of Hrs does not interfere with signaling by Dl; nor does it alter the requirement for Lqf. Clones of hrs^- cells that express exogenous Dl induce Cut expression in surrounding cells, whereas corresponding hrs^-lqf^- clones do not (data not shown).

To determine unequivocally whether the abnormal puncta that accumulate Dl in hrs^- and hrs^- lqf^- cells are indeed endosomal, we made use of the finding that Wg secreted from prospective wing margin cells accumulates in similar, abnormally large puncta in hrs^- cells positioned at a distance from the secreting cells (E. S. Seto and H. J. Bellen, personal communication). We obtained the same result in double mutant hrs^- wg^- cells(data not shown), establishing that the accumulation of Wg in these puncta serves as an in vivo marker for endocytosis. We then examined Wg and Dl staining in triple mutant hrs^- wg^-lqf^- clones that express an HRP-tagged form of Dl under Gal4 control (see Materials and methods). In this case, as in corresponding hrs^- wg^- double mutant clones, we observe co-localization of Wg and Dl in large intracellular puncta(Fig. 4D,E).

Thus, bulk endocytosis of both endogenous and overexpressed Dl appear normal in lqf^- cells.

Although bulk Dl endocytosis appears unaffected by the absence of Lqf,blockage of a relatively small, but specific, subset of Dl endocytic events might escape detection, and this subset might be crucial for signaling activity. To examine this possibility, we co-expressed Dl together with the E3 Ubiquitin Ligase Neuralized (Neur), under Gal4 control, to drive efficient ubiquitination and internalization of the exogenous Dl(Lai et al., 2001; Pavlopoulos et al., 2001). We reasoned that under these conditions, even modest reductions in the rate of Dl endocytosis might cause an abnormal persistence of Dl at the apical cell surface.

Wing discs that express uniformly high levels of Dl under Gal4 control accumulate high levels of Dl on the apical cell surface. However, in discs that co-express high levels of both Dl and Neur, this surface accumulation is strongly reduced and Dl accumulates instead in an abnormally large number of intracellular puncta. Clones of lqf^- cells generated in such co-expressing discs do not appear to alter the number or general appearance of these Dl-positive puncta, many of which co-localize with Hrs(Fig. 5B′,B″). However, they do affect the level of Dl staining associated with the apical cell surface (as visualized in discs processed either with, or without,detergent). Such lqf^- clones show residual surface staining of Dl, in contrast to neighboring wild-type cells where surface-associated staining is depleted(Fig. 5B). We infer that lqf^- cells cannot endocytose Dl as efficiently as their wild-type neighbors, accounting for why we detect a difference under sensitized conditions in which the rate of surface clearance appears to be limiting.

Significantly, the residual staining of Dl on the surface of lqf^- cells that overexpress Neur and Dl correlates with the failure of these cells to signal. We find that clones of lqf^- cells that overexpress Neur and Dl fail to activate cut in neighboring cells (Fig. 5C), even though clones of otherwise wild-type cells that overexpress Neur and Dl show enhanced Dl signaling(Pavlopoulos et al., 2001). Hence, it appears that the impairment in Dl endoctyosis we detect in lqf^- clones in this sensitized background correlates with an absolute block in signaling activity.

The cytosolic domain of DSL ligands contains multiple Lysines at least some of which serve as acceptors for Ubiquitin(Deblandre et al., 2001; Itoh et al., 2003). Lqf contains two Ubiquitin Interacting Motifs (UIMs)(Hofmann and Falquet, 2001). Hence, mono-ubiquitination of DSL ligands might allow Lqf to target them for a special subset of endocytic events that are required for signaling activity. By contrast, bulk endocytosis of DSL ligands mediated by interactions with other Ubiquitin-binding adaptor proteins might not suffice to confer signaling activity. To test this hypothesis, we investigated whether the signaling activity of the Dl^R+ protein depends on Lqf activity.

Endocytosis and signaling activity of Dl^R+ depends on the presence of at least one of the two Lysines in the R+ peptide comprising the cytosolic domain (Fig. 3D,F). We find that clones of lqf^- cells that express Dl^R+ fail to induce cut expression in adjacent wing disc cells (Fig. 3E). However,Dl^R+ protein in these lqf^- clones accumulates both on the apical surface and in intracellular puncta(Fig. 3E). Moreover, we could not detect any difference in the punctate, cytosolic accumulation of Dl^R+ between lqf^- and wild-type cells in wing discs that generally overexpress Dl^R+ (data not shown). Both results indicate that bulk endocytosis of Dl^R+ is not significantly altered in the absence of Lqf. Because substitution of both Lysines by Arginine blocks internalization and signaling activity of Dl^Rm(Fig. 3F), we infer that Dl^R+ is targeted for internalization solely by ubiquitination at one or both of these Lysines. Hence, we suggest that other Ubiquitin-interacting proteins aside from Lqf can target mono-ubiquitinated cargo proteins, such as Dl^R+ or endogenous Dl, for internalization. However, only Lqf appears able to direct endocytosis of these proteins in a way that allows DSL ligands to signal.

Both endocytosis and signaling activity of Dl^LDL+ depends on the FDNPVY internalization signal (Fig. 3A,C). However, unlike either native Dl or Dl^R+, we find that clones of lqf^- cells expressing Dl^LDL+ can induce cut expression in adjacent wild-type cells (Fig. 3B), indicating that the presence of the LDL internalization signal in the chimeric Dl^LDL+ protein bypasses the requirement for Lqf. As observed for clones of wild-type cells overexpressing Dl^LDL+, the `rescued'lqf^- clones only induced cut when located close to the DV boundary. Nevertheless, their ability to signal, albeit weakly,contrasts with that of lqf^- clones that overexpress native Dl, native Dl plus Neur, or Dl^R+, all of which are devoid of signaling activity. Hence, we conclude that the FDNPVY signal directs internalization of Dl^LDL+ in a manner that permits the protein to acquire signaling activity even in the absence of Lqf activity.

Lqf-dependent endocytosis of DSL ligands might be accompanied by modifications of these ligands, either as a pre-requisite for, or a consequence of, signaling activity. To examine this possibility, we asked whether the size of Dl protein changes as a consequence of Lqf-dependent endocytosis.

Initially, we generated clones of wild-type and lqf^-cells that express Dl tagged by the insertion of six copies of the Myc epitope in the extracellular juxtamembrane domain, and analyzed the profile of Dl peptides that retain the Myc epitope by western blotting (see Materials and methods). Under these conditions, we observed similar, complex profiles of Myc-tagged Dl peptides from both wild-type and lqf^- cells corresponding to full-length Myc-Dl protein, as well as several lower molecular weight peptides (data not shown).

We then repeated this experiment using wild-type and lqf^- cells that overexpress Neur and Myc-tagged Dl, the sensitized condition under which we can detect residual surface expression of Myc-tagged Dl in lqf^-, but not in wild-type, cells. In this case, the profile of Myc-tagged Dl is remarkably simple. Wild-type cells show two bands, one corresponding by size to full-length Myc-tagged Dl(∼105 kDa) and the other to a Myc-tagged cleavage product of ∼50 kDa. By contrast, lqf^- cells show only a single band,corresponding to full-length Myc-tagged Dl(Fig. 6). Thus, the failure to clear Dl from the cell surface of lqf^- cells is associated with an apparent failure in Dl processing. These results provide evidence for a Lqf-dependent cleavage of Dl that correlates with Lqf-dependent endocytosis and signaling activity.

We note that the expected size of the Myc-tagged extracellular domain of Dl is ∼75 kDa, whereas that of the complementary, Myc-tagged portion of the ligand containing the transmembrane and cytosolic domains is ∼40 kDa. Hence, the 50 kDa Myc-tagged cleavage product we detect must be composed of a C-terminal portion of the extracellular domain, and possibly some or all of the transmembrane and cytosolic domains as well. The relationship of this truncated peptide to the active ligand is presently unknown. It could comprise part, or all, of the active ligand, or alternatively, a non-signaling C-terminal fragment cleaved off in the process of generating an N-terminal signaling fragment. Alternatively, it might be a degradation product generated as a consequence of the activation of Notch by Dl.

Epsins, including the sole Drosophila Epsin Lqf, contain a series of discrete functional domains that implicate them in Clathrin-mediated endocytosis. These include the N-terminal ENTH domain, which can induce membrane curvature in response to PtdIns(4,5)P₂ binding,as well as binding sites for Clathrin, AP-2 and accessory proteins (e.g. Eps15), and multiple Ubiquitin Interactions Motifs (UIMs). Recent evidence has suggested that Epsins are members of a class of structurally related proteins that function as cargo selective adapters that target substrate proteins for Clathrin-mediated endocytosis (Mishra et al., 2002; Wendland,2002).

We have demonstrated an absolute, cell-autonomous requirement for Lqf in generating functional DSL ligands. However, we have not been able to detect any other role for Lqf in cell-cell signaling, such as in receiving DSL ligands, or in sending, receiving, or controlling the distribution of other extracellular signals, notably Hg, Wg and Dpp. Furthermore, cells devoid of Lqf activity appear to grow, proliferate and interdigitate in a manner that is indistinguishable from cells devoid of Dl and Ser, the two DSL ligands in Drosophila. These results suggest that Epsin function in Drosophila may be essential solely for the production of active DSL ligands.

Surprisingly, we also failed to detect an effect of removing Lqf on the steady state accumulation of Dl in endocytic compartments. However, we were able to detect a modest effect on Dl internalization in a sensitized background in which we greatly enhance Dl endocytosis by overexpressing Neur,an E3-Ubiquitin ligase that ubiquitinates Dl. Strikingly, high levels of Dl accumulate in endocytic vesicles of such Neur overexpressing cells whether or not they have Lqf, but only cells that have Lqf can signal. We therefore infer that Epsin is required for a discrete and apparently small subset of the endocytic events that normally internalize DSL ligands; however, it is this subset that is crucial for generating active DSL signals.

The selective requirement for Epsin in sending DSL ligands is reminiscent of that for the Presenilin/γ-secretase complex in transmembrane cleavage and signal transduction by Notch (Struhl and Greenwald, 1999; Struhl and Adachi, 2000). Selectivity in the case of the Presenilin/γ-secretase complex does not reflect a dedicated role in Notch proteolysis, but rather an unusual property of the Notch transduction mechanism, namely that ectodomain shedding activates the pathway by inducing transmembrane proteolysis (Mumm and Kopan,2000; Struhl and Adachi,2000). Similarly, selectivity in the case of Epsin may reflect an unusual requirement for DSL ligands to signal, and not a dedicated role of Epsin in confering their signaling activity.

It is generally thought that Epsins target cargo proteins for endocytosis via mono-ubiquitin internalization signals(Wendland, 2002). We replaced the cytosolic domain of Dl with a random peptide (R+) that contains two Lysines, and showed that the presence of at least one Lysine is essential for both the endocytosis and signaling activity of the chimeric Dl^R+ligand. To test the possibility that the presence of Lysine targets Dl^R+, as well as wild-type Dl, for endocytosis by serving as an Ubiquitin acceptor, we replaced the cytosolic domain of Dl with a non-Lysine containing form of Ubiquitin. The resulting Dl^Ubi+ ligand could be endocytosed and had at least partial signaling activity, but not if the Ubiquitin domain contained an additional mutation that blocks its ability to be targeted for endocytosis. Finally, and critically, we demonstrated that the signaling activity of Dl^R+, like that of wild-type Dl, depends on Lqf. Collectively, these findings implicate mono-ubiquitination as the internalization signal required to target DSL ligands for endocytosis by Epsin.

Significantly, bulk endocytosis of the chimeric Dl^R+ ligand,like that of wild-type Dl, appears to be unaffected in cells devoid of Lqf,even though signaling activity is abolished. Hence, we infer that Lqf is not the only adapter protein that can target mono-ubiquitinated substrate proteins for endocytosis. Nevertheless, Lqf appears to be unique amongst all such adapter proteins in its ability to direct internalization of mono-ubiquitinated DSL ligands in a manner that confers signaling activity. We therefore suggest that Epsin has a dedicated role in directing mono-ubiquitinated cargo proteins into a particular endocytic pathway, one that DSL ligands must enter in order to acquire signaling activity. As we detail in the next section, we suggest that Epsin might direct DSL ligands specificially into a recycling pathway.

It is notable that substitution of the cytosolic domain of Dl with a peptide carrying the FDNPVY internalization signal from the LDL receptor yields a chimeric Dl^LDL+ ligand that is endocytosed and has signaling activity. However, in this case, Lqf is not essential for signaling. One interpretation of this result is that mono-ubiquitinated DSL ligands are normally targeted for endocytic pathways that preclude their signaling activity, unless they are diverted from entering these pathways by association with Lqf, or by the presence of a heterologous internalization signal such as FDNPVY. In both cases, endocytosis would take place via an alternate pathway compatible with signaling activity.

Why must DSL ligands on the surface of signal-sending cells be endocytosed in order to activate Notch on the surface of signal-receiving cells? We can distinguish two general classes of explanation. In the first(Fig. 7A), activation of Notch is triggered by early events in the process of DSL endocytosis that occur while the ligands are still on the cell surface, prior to the pinching off of coated vesicles. In the second (Fig. 7B), internalization of DSL proteins is a necessary prerequisite for endocytic recycling, which is required for subsequent signaling activity.

Most previous models fall into the first class. One such internalization model proposed that DSL/Notch binding creates a physical bridge between the sending and receiving cell that is mechanically stressed by endocytosis of the ligand, causing conformational changes in Notch that elicit either S2 or S3 cleavage (Parks et al., 2000). Another proposed that recruitment of DSL ligands into coated pits increases their local abundance on the cell surface(Le Borgne and Schweisguth,2003b). Both of these models are difficult to reconcile with our finding that Lqf is essential for Dl signaling but not for bulk endocytosis of Dl. This result indicates that DSL endocytosis in signal-sending cells is not sufficient, per se, to activate Notch in signal-receiving cells. Instead, as we suggest above, it appears that DSL ligands have to enter, or traffic through, a special Lqf-dependent endocytic pathway to activate Notch.

For such internalization models to accommodate our results, it seems necessary to posit that productive interactions between DSL ligands and Notch require a special micro-environment that is associated only with a particular subclass of coated pits or other specializations(Fig. 7A). Mono-ubiquitinated cargo proteins might be excluded from such structures, unless chaperoned there by Lqf. Thus, only DSL ligands that gain entry, whether via Lqf, or by the targeting mediated by the LDL receptor signal, would be able to activate Notch on the abutting surface of the receiving cell. Furthermore, one would have to posit the existence of accessory molecules that are provided by the sending cell, sequestered in these specializations, and essential for DSL-dependent activation of Notch on the receiving cell, whether by mechanical stress, DSL clustering, or some other means.

We are directed to the second general class of explanation, in which recycling is the key element, by the ability of the internalization signal from the LDL receptor to bypasses the requirement for Lqf. In general,mono-ubiquitination acts as a sorting signal in the endosomal system that leads to delivery of membrane proteins to late endosomes and eventually lysosomes (reviewed by Hicke and Dunn,2003). By contrast, the FDNPVY signal is associated with rapid recycling back to the cell surface after entry into endosomes(Chen et al., 1990; Matter et al., 1993). Hence,Epsin-binding to mono-ubiquitinated DSL proteins during endocytosis might allow those DSL proteins to escape degradation by altering their sorting, thus allowing them to enter a recycling pathway. Passage through this pathway would be essential to confer signaling activity.

Why might recycling be necessary for DSL ligands to acquire signaling activity? One possibility is that recycling allows DSL ligands to be stripped of the bound ectodomain of Notch so that they can be re-used. Multiple rounds of recycling might then enhance the level of active DSL ligands on the surface of signal-sending cells above a critical threshold necessary to activate Notch transduction in the signal-receiving cell. According to this view, one might expect that massive overexpression of DSL ligands would be able to bypass the requirement for Lqf. However, our results suggest that this is not the case:we estimate that, in our experiments, overexpressed Dl accumulates on the cell surface at levels up to tenfold higher than peak accumulation of endogenous Dl, yet is unable to rescue DSL signaling activity in cells devoid of Lqf.

Alternatively, recycling of nascent DSL proteins may be important to convert inactive `pro-ligands' into active ligands(Fig. 7B). Conversion might entail recruitment of DSL proteins into signaling exosomes(Le Borgne and Schweisguth,2003b). However, Dl signaling appears to be unaffected in cells devoid of Hrs, despite impairment in the maturation of early to late endosomes, and in the formation of multi-vesicular bodies from which exosomes might derive (Lloyd et al.,2002). Another possibility is that DSL proteins need to be processed in order to be converted to active ligands, a hypothesis that is consistent with our evidence that Lqf-dependent endocytosis of Dl correlates with a specific proteolytic cleavage of the ligand. Lqf would be required in this scenario to allow DSL ligands to enter a recycling pathway in which the required processing event can occur. The only specificity one needs to invoke in this model is that of Epsin to allow mono-ubiquitinated cargo proteins to gain access to a recycling pathway. The conditions necessary to convert DSL pro-ligands into active signals (e.g. low pH) might exist generally in early endosomes or recycling endosomes.

We thank A. Adachi, C. Bonin and X.-J. Qiu for technical assistance; J. Jiang, C.-M. Chen and J. A. Fischer for providing lqf mutations; A. Wong for alerting us to the possibility that our mutations were lqfalleles; K. Basler, H. Bellen, E. Lai and L. Luo, for other fly stocks; L. Hicke for Ubiquitin DNAs; M. Muskavitch, J. A. Fischer and K. Irvine for antisera; and Q. Al-awqati, R. Axel, B. Grant, I. Greenwald, M. Robinson, D. Shaye and T. McGraw for advice and comments on the manuscript. G.S. is an HHMI Investigator; W.W. is supported by funding from the HHMI.