Relationships between EEA1 binding partners and their role in endosome fusion

Homotypic fusion between operationally defined early endosomes has been shown to require phosphatidylinositol 3 (PI3)-kinase activity (Jones and Clague, 1995; Li et al., 1995; Spiro et al., 1996). The generation of phosphatidylinositol 3-phosphate (PI3P), at the early endosome, is necessary to recruit a vesicle tethering factor, EEA1, which is required prior to SNARE complex assembly on the vesicle fusion pathway (Christoforidis et al., 1999; Clague, 1999; Mills et al., 1998). EEA1 interacts with PI3P via a double zinc finger FYVE domain at its C terminus (Burd and Emr, 1998; Gaullier et al., 1998; Patki et al., 1998). This domain of EEA1 is also implicated in interactions with a number of endosomal proteins. An interaction with the early endosome-specific small GTPase, rab5, has been shown by a variety of techniques (Simonsen et al., 1998). At high levels of rab5-GTP, the fusion pathway is refractory to PI3-kinase inhibition (Jones et al., 1998; Li et al., 1995), presumably because rab5 can recruit sufficient EEA1 to the membrane independently of PI3P. A direct interaction between the EEA1 FYVE domain and the t-SNARE syntaxin 13 has been proposed, on the basis of biosensor measurements (McBride et al., 1999), whilst an interaction with syntaxin 6 has been shown by 2-hybrid screening and biochemical techniques (Simonsen et al., 1999). Furthermore, an IQ motif proposed to bind calmodulin is juxtaposed N-terminally to the FYVE domain (Mu et al., 1995).

Rab5 was the first specific factor shown to participate in endosome fusion (Gorvel et al., 1991), whilst SNARE proteins have long been presumed to feature, because NSF function is required for fusion (Diaz et al., 1989). The identity of the relevant SNAREs and exactly how they are linked to rab5 function is an important question. Recently, McBride et al. presented evidence that syntaxin 13 is required for early endosome fusion and proposed that EEA1, syntaxin 13 and NSF coexist in a large oligomer at some point on the fusion pathway (McBride et al., 1999).

Colombo et al. have shown a requirement for calmodulin in fusion of endosomes derived from J774 macrophage cells (Colombo et al., 1997). This result parallels data from studies of yeast vacuole fusion, which bears some resemblance to the endosome fusion event. In this system calmodulin is believed to cooperate with protein phosphatase 1 to complete a late step in vacuole fusion (Peters et al., 1999; Peters and Mayer, 1998).

The multiplicity of proposed EEA1 binding partners raises questions about potential cooperativity or competition. In this paper, we used an overlay technique to show interaction of calmodulin with the IQ domain of EEA1 and with syntaxin 13, but not with syntaxin 6. We examined the influence of protein binding partners on the interaction of EEA1 with PI3P and also investigated the role of these proteins in endosome fusion.

Syntaxin 13 and 7 polyclonal antibodies were kindly provided by Richard Scheller and Rytis Prekeris, syntaxin 6 monoclonal antibody was purchased from Transduction Laboratories.

Bacterial expression vectors for GST- and His6-tagged proteins were used. pGEX-2T-rab5 (canine) was provided by Kirill Alexandrov (Dortmund, Germany), pGEX-KG-syntaxin 6 (human, 1-233) was provided by Sharon Tooze (London, UK) and pGEX-KG-syntaxin 7 (human, 1-236) and 13 (rat) were provided by Richard Scheller (Stanford, USA). A baculovirus construct for EEA1 was provided by Harald Stenmark (Oslo, Norway). GST-EEA1 (1098-1411) and GST-EEA1-ΔIQ(1098-1411) were subcloned from pGemT into the BamHI/EcoRI sites of pGEX4T2. The IQ deletion mutant, EEA1-ΔIQ (1098-1411), lacking residues 1286-1306, was generated using sense and antisense primers (5′-ATAGAAAAGCTTCAAACCAAAGTA-TTAGAATTGC-3′ and 5′-P-AACCGTTGCTTCTAATACTGCAA-TTTCACCC-3′ according to the Quickchange mutagenisis protocol (Stratagene). It was then subcloned into the BamHI/EcoRI sites of pGEX4T2.

Following SDS-PAGE, proteins were transferred to PVDF membrane using a Genie blotter. The filter was placed in 100% methanol for 10 seconds, left to dry for 15 minutes, soaked in Tris-buffered saline (TBS; 137 mM NaCl, 3 mM KCl and 5 mM Tris, pH 7.5) solution for 10 minutes and then blocked for 1 hour at room temperature in TBS containing 1% (w/v) milk powder. It was then incubated for 45 minutes with biotinylated calmodulin (100 ng/ml, b-CaM, Calbiochem) in overlay buffer, TBS, pH 7.5, 0.05% Tween-20, 0.1% dried milk and divalent metal ions as indicated. The filter was then washed twice with TBS, 0.05% Tween-20. Avidin-horseradish peroxidase (10 ⁻¹⁰ M, Sigma A-3151) in overlay buffer was incubated with the membrane for 30 minutes before washing twice with TBS, 0.05% Tween-20 and twice with TBS. Peroxidase was detected using enhanced chemiluminescence (Pierce Supersignal).

GST-tagged proteins were expressed in E. coli (BL21) and batch-purified with glutathione-sepharose (Pharmacia), whilst His6-tagged proteins were purified on Ni²⁺-NTA columns (Pharmacia), both according to the respective manufacturer’s instructions.

Cell-free fusion assays were carried out essentially as previously described (Jones and Clague, 1995; Jones et al., 1998). Briefly, confluent BHK-21 cells grown on plastic were washed with PBS. Subsequently, either 4 mg/ml avidin or 1.8 mg/ml biotinylated horseradish peroxidase (HRP) in Dulbecco’s PBS supplemented with 1 mM CaCl2 and 1 mM MgCl2 were incubated with the cells for 9 minutes at 37°C. This incubation was followed by extensive washing at 4°C, before removal of the cells from the dish with a cell scraper and homogenization in 3 mM imidazole/HCl, pH 7.4, 250 mM sucrose, 1 μg/ml pepstatin, 2 μg/ml aprotinin, 2 μg/ml leupeptin. Post-nuclear supernatants were obtained by centrifugation at 1,500 g for 15 minutes.

Post-nuclear fractions containing avidin or biotin-HRP loaded endosomes were then combined in a mixture containing 10 mM Hepes-KOH, pH 7.0, 1.2 mM MgCl2, 50 mM potassium acetate, 0.8 mM DTT, biotin-insulin and ATP regenerating system. The mixture was then incubated at 37°C for 30 minutes. Fusion assays were stopped by lysis on ice for 15 minutes with 0.25% Triton X-100. Fusion results in the formation of an avidin-biotin-HRP complex, which was immunoprecipitated with anti-avidin antibodies bound to protein A-sepharose. The relative amounts of immunoprecipitated HRP were quantified by determination of the HRP activity with o-dianisidine as substrate.

The method of Schiavo et al. for measuring lipid binding was used (Schiavo et al., 1997). 0.2 μm unilamellar liposomes were prepared using an extrusion device. Lipid composition of the vesicles was 63% egg lecithin, 20% phosphatidylserine, 15% phosphatidylethanolamine, 2% PI3P, reconstituted in 20 mM Hepes-KOH, pH 7.6, 100 mM KCl, 0.2 mM DTT together with trace amounts (1 μCi/175 μg lipid) of L-3-phosphatidyl(N-methyl-3H)choline (Amersham). Glutathione beads or Ni ²⁺-NTA beads were coupled to GST-EEA1(1098-1411) or His ₆-EEA1, respectively. Washed beads with EEA1 attached were incubated at room temperature for 30 minutes in 20 mM Hepes-KOH, pH 7.6, 100 mM KCl, 0.2 mM DTT, 1 mM MgCl ₂, 1 μM ZnCl ₂ together with 100 μl of ³H-labelled liposomes and other factors as indicated. Beads were washed three times by centrifugation at 1500 g for 2 minutes and the amount of radioactivity remaining with the beads was determined by scintillation counting.

A Bio-Gel P-6 polyacrylamide gel column (BioRad) was used according to the manufacturer’s instructions, to exchange rab5 storage buffer for low-magnesium nucleotide exchange buffer (20 mM Hepes/KOH, pH 7.2, 10 mM EDTA, 1 mM DTT, 5 mM MgCl ₂). Nucleotides (GDP, GTP or GTPγS) were added to rab5-containing samples to a final concentration of 10 mM. These were incubated for 30 minutes at 25°C and the exchange reaction was then stopped by the addition of magnesium chloride to a final concentration of 15 mM. Rab5 was applied in various nucleotide states to the in vitro lipid-binding assay.

We sought to confirm earlier distinct reports that early endosome fusion requires release of intravesicular calcium and calmodulin activity (Colombo et al., 1997; Holroyd et al., 1999). The configuration of the endosome fusion assay that we have used here is insensitive to the sequestration of free Ca²⁺ with EGTA, but completely inhibited when either the membrane-impermeable, ‘fast’ calcium chelator, BAPTA, or the membrane-permeable EGTA-AM, is included, as previously described (Holroyd et al., 1999) (Fig. 1A). This indicates a requirement for release of lumenal Ca²⁺ for fusion to occur. As calmodulin may be a relevant effector of elevated calcium we tested for its role in early endosome fusion. We included the structurally unrelated calmodulin antagonists, W7 and calmidazolium chloride, in the fusion assay, both of which exerted substantial inhibition (Fig. 1B). Excess calmodulin in the assay (3-50 μM) provided a constant stimulation (approx. 30% increase in fusion signal, not shown), which probably corresponds to maximal fusion efficiency in the experimental configuration we have used.

We have previously shown a requirement for EEA1 in endosome fusion (Mills et al., 1998). The IQ motif of EEA1 is predicted to bind calmodulin (Mu et al., 1995). We have therefore utilised an overlay technique to characterise biotin-calmodulin binding to EEA1 and a range of other proteins. Although this technique will not reflect all potential calmodulin interactions with cellular proteins, we found that biotin-calmodulin bound to recombinant GST-EEA1(1098-1411) and to full-length His6-EEA1 (data not shown) in a Ca²⁺-dependent manner (Fig. 2A). Specificity was indicated by the failure of biotin-calmodulin to bind GST alone or several other GST-tagged proteins. Furthermore, in bacterial lysate and Sf9 cells, induction of EEA1 expression led to the appearance of an intense biotin-calmodulin binding signal whilst other proteins in the mixtures, present at similar amounts on the lanes, did not bind biotin-calmodulin (data not shown). The overlay assay provided novel indications of calmodulin binding to syntaxin 13 and VAMP2 (a v-SNARE) but not syntaxins 6 (not shown) or 7. Calmodulin binding to syntaxin 13 was calcium-dependent, whilst binding to VAMP2 was relatively insensitive (Fig. 2A).

We also found a further enhancement of the bound calmodulin signal if 1 μM Zn²⁺ was included in the overlay buffer together with 100 μM Ca²⁺ (Fig. 2B). We do not believe that the enhancement due to zinc is related to reconstitution of the correctly folded FYVE domain, because we see similar enhancements with other proteins that are supposed not to bind zinc, e.g. syntaxin 13 (Fig. 2B).

We removed the IQ domain from the C-terminal fragment (ΔIQ-1098-1411) of EEA1 and showed that the calmodulin binding signal is now lost in the overlay assay (Fig. 2C).

EEA1 has previously been reported to bind both syntaxin 6 and syntaxin 13, but the relative strength of binding to these syntaxins has never been directly compared (McBride et al., 1999; Simonsen et al., 1999). We incubated equimolar amounts of GST-tagged syntaxins, immobilised on glutathione-sepharose, with His6-EEA1. Following recovery of the sepharose beads, interaction of EEA1 with syntaxin 6 was readily detected, whilst interactions with syntaxin 7 or 13 could only be detected at much longer exposure times (Fig. 3).

We wished to compare the requirement of endosome fusion for the proposed EEA1 binding partners syntaxin 6 and syntaxin 13. We have added antibodies against syntaxins 6, 7 and 13 to the endosome fusion assay. Syntaxin 7 serves as a control, as it also displays an endosomal location, but has not been shown to interact significantly with EEA1 by any methodology. In accord with the results of McBride et al. (McBride et al., 1999), we find that syntaxin 7 antibodies are without effect (not shown), whilst syntaxin 13 antibodies are inhibitory. In contrast to their results we find that anti-syntaxin 6 (in this case a monoclonal antibody) inhibits the assay to a similar extent as anti-syntaxin 13 (Fig. 4A). Pre-absorption of the respective antibodies with their cognate GST-syntaxins nullified the inhibitory effects of each antibody. Titrations of the antibodies revealed that in neither case could we achieve inhibition of the fusion assay beyond the levels shown in Fig. 4A. We also found no further inhibitory effect by combining both anti-syntaxin 13 and anti-syntaxin 6 in the assay.

Several previous studies have described the use of soluble SNARE protein domains to inhibit cell-free assays of membrane trafficking, presumably due to competition with the membrane-anchored, endogenous form, for binding partners (Lowe et al., 1997; McBride et al., 1999). We prepared the soluble domains of syntaxin 6, 7 and 13 as GST-fusion proteins. Addition of the proteins in this form failed to influence the endosome fusion assay (data not shown). McBride et al. have shown that removal of the GST-tag is required for syntaxin 13 inhibition of endosome fusion (McBride et al., 1999). We confirmed this result, but additionally show an inhibitory effect due to syntaxin 6 soluble domain (once it is cleaved from GST, Fig. 4B), in conflict with the observations of McBride et al. For both syntaxin 6 and 13 soluble domains, we observe maximal inhibitory effects at concentrations approximately 2 orders of magnitude below those published by McBride et al. for syntaxin 13 inhibition. The concentration range that we have used is similar to that used in other studies that have observed effects of the relevant SNARE domain (Lowe et al., 1997).

We investigated the relationship between binding of PI3P by EEA1 and the protein factors described above, which interact with the same region of EEA1 as the lipid. We used a lipid binding assay that has previously been used to establish the specificity of the PI3P interaction with FYVE domains (Gaullier et al., 1998; Schiavo et al., 1997). GST-EEA1 (1098-1411) attached to glutathione-sepharose, or full-length His6-EEA1 attached to Ni²⁺-NTA, was incubated with radiolabelled liposomes containing 2% PI3P and counts were measured following recovery of the sepharose beads. We checked the concentration dependence of the EEA1 fragment on the lipid binding assay, and chose a concentration for subsequent experiments from the linear part of the curve. Incubation of the EEA1-loaded beads with calmodulin inhibited binding between EEA1 and liposomes. The ΔIQ mutant bound lipid as effectively as the wild-type sequence, but in this case, calmodulin no longer inhibited the EEA1 interaction with lipid (Fig. 5A). This result is consistent with the lack of binding of calmodulin to this fragment observed in the overlay experiments. Rab5-GTPγS but not rab5-GDP also inhibited lipid binding by about 50% (Fig. 5C). Incubation with GST-syntaxin 6 or 13 had no effect on EEA1 lipid binding (Fig. 5B). Incubation of BHK post-nuclear supernatants under the conditions used for endosome fusion, did not show any shift in the membrane cytosol distribution of EEA1 when incubated with up to 50 μM calmodulin (not shown).

We have shown that calmodulin binds to the IQ domain of EEA1. We have primarily used a calmodulin-overlay technique for this purpose, although the influence of calmodulin on EEA1 binding to lipids also supports our conclusions. Using the overlay technique, we also identified the EEA1 partner syntaxin 13 and VAMP 2 as calmodulin binding partners. Calmodulin binding to VAMP 2 has recently been confirmed using alternative techniques (Quetglas et al., 2000).

A role for Ca²⁺ binding proteins in endosome fusion is indicated by the inhibition due to both the ‘fast’ chelator BAPTA and the membrane permeable EGTA-AM (Holroyd et al., 1999; Fig. 1A). We speculate that calmodulin may be a relevant sensor of the released intravesicular Ca²⁺ based on the sensitivity of the assay to calmodulin antagonists found by ourselves and others (Fig. 1B; Colombo et al., 1997). Intravesicular release of Ca²⁺ combined with calmodulin is also required for late endosome-lysosome fusion (Pryor et al., 2000) and yeast vacuole fusion (Peters and Mayer, 1998). The picture emerging suggests that this may be a widely used, if not universal means of promoting intracellular fusion on the endosomal pathway. Although the relevance of the interaction of calmodulin with EEA1 and syntaxin 13 to early endosome fusion remains to be directly tested, one can envisage that either protein could serve to localise calmodulin to the site of membrane docking and presumably of Ca²⁺ release.

Interaction of EEA1 with both syntaxin 6 and syntaxin 13 has been reported in separate papers (McBride et al., 1999; Simonsen et al., 1999). We made a direct comparison between the two by measuring the binding of EEA1 to GST-tagged syntaxins absorbed on glutathione-sepharose. In our hands and by this method, only syntaxin 6 can be shown to specifically bind EEA1. EEA1 binding to syntaxin 13 can be detected upon longer exposures, when we can also pick up syntaxin 7 binding, so the specificity of this interaction must be questioned. However, we should point out that the interaction between EEA1 and syntaxin 13 was originally detected using a biosensor assay, which measured a high koff for this interaction, implying that the interaction is transient (McBride et al., 1999). The assay that we have used requires repeated washing of beads and would therefore not pick up unstable interactions.

We have now implicated both syntaxins 6 and 13 in early endosome fusion, by the use of antibodies and soluble domains of each protein. In no case could we obtain full inhibition; maximum inhibition was about 50% in each case. Our fusion assay measures fusion between compartments that have been operationally defined by a 9-minute pulse of marker. It is possible that this may include different populations of endosomes that rely on different combinations of SNARE proteins. However, the fact that no enhancement of inhibition by syntaxin 6 antibodies can be seen by further supplementation with syntaxin 13 antibodies, and vice versa, indicates that both proteins are required on the same fusion pathway. The standard expectation for SNARE involvement in endosome fusion would invoke the model that has been derived from studies of synaptic vesicle fusion. In this scenario, a four-helix bundle comprises the core of the fusion apparatus, in which VAMP 2 (a v-SNARE) and syntaxin 1 contribute a helix each, with the other two contributed by SNAP-25 (Poirer et al., 1998; Sutton et al., 1998). Studies of endosome fusion from McBride et al. have inferred the requirement for a large oligomeric structure containing syntaxin 13 and EEA1 (McBride et al., 1999). We suggest that syntaxin 6 may be part of a network of interactions relevant to endosome fusion that includes EEA1, syntaxin 13 and calmodulin.

Syntaxin 6 is an atypical member of the syntaxin family, in that it is also highly related to SNAP-25 (Bock et al., 1996). It largely localises to the trans-Golgi network (TGN) and has been implicated in the TGN to endosome transport (Bock et al., 1997). Examining data from both yeast and mammalian systems, some striking parallels between consumption of TGN-derived vesicles at endosomes (best characterised in yeast) and homotypic early endosome fusion (characterised in mammalian cells) emerge. (1) Both events require syntaxin 6 (or Pep12 in yeast), (2) early endosome fusion requires rab5, EEA1 and rabenosyn 5 whilst vesicle consumption in yeast requires Vps21, the yeast rab5 homologue and Vac1, an EEA1/rabenosyn 5 orthologue (Clague, 1999; Nielsen et al., 2000; Peterson et al., 1999), (3) syntaxin 6 interacts directly or indirectly with mVPS45, the mammalian homologue of Vps45, which is an n-Sec1 family member that interacts with Vac1 (Bock et al., 1997; Peterson et al., 1999). This high degree of overlap leads us to speculate that elements of the homotypic endosome fusion machinery are both delivered by and required for TGN-derived vesicle consumption (Fig. 6). Following vesicle consumption, relevant SNAREs may simply be reprimed to allow fusion with the same class of target membranes, i.e. homotypic endosome fusion.

All the EEA1 protein binding partners we have examined interact with the C-terminal region of EEA1. This is also the site of PI3P binding, via its FYVE domain. As a first step to unravelling the interplay between these interacting components, we studied the influence of the protein partners on EEA1 binding to lipid. We found no instance of positive cooperativity. Calmodulin and rab5-GTP reduce binding of EEA1 to PI3P containing liposomes. Calmodulin can completely inhibit lipid binding whilst rab5-GTP inhibition plateaus at approximately 50%. As EEA1 is dimeric (Callaghan et al., 1999); we suggest that only one lipid binding site per dimer can be occluded by rab5. We propose that EEA1 function may involve competition for binding at its FYVE domain by both lipid and protein. Of the factors that interact with EEA1, rab5-GTP, calmodulin, syntaxins 6 and 13 as well as PI3P have now all been shown to play a role in fusion of endosomes from BHK cells.

We thank Rytis Prekeris and Richard Scheller for reagents against syntaxin 7 and 13, Sharon Tooze for a syntaxin 6 expression vector and Harald Stenmark for EEA1-baculovirus. Ian Mills is the recipient of a Wellcome Trust Prize Studentship. Sylvie Urbé is supported by the North West Cancer Research Fund.