Regulation of AP-3 function by inositides. Identification of phosphatidylinositol 3,4,5-trisphosphate as a potent ligand.

As part of the growing effort to understand the role inositol phosphates and inositol lipids play in the regulation of vesicle traffic within nerve terminals, we determined whether or not the synapse-specific clathrin assembly protein AP-3 can interact with inositol lipids. We found that soluble dioctanoyl-phosphatidylinositol 3,4,5-trisphosphate (DiC8PtdIns(3,4,5)P3) was only 7.5-fold weaker a ligand than D-myo-inositol hexakisphosphate in assays that measured the displacement of D-myo-[3H]inositol hexakisphosphate. In functional assays we found that both of these ligands inhibited clathrin assembly, but DiC8-PtdIns(3,4,5)P3 was more potent and exhibited a larger maximal effect. We also examined the structural features of DiC8-PtdIns(3,4,5)P3 that establish specificity. Dioctanoyl-phosphatidylinositol 3,4-bisphosphate, which does not have a 5-phosphate, and 4,5-O-bisphosphoryl-D-myo-inosityl 1-O-(1,2-O-diundecyl)-sn-3-glycerylphosphate, which does not have a 3-phosphate, were, respectively, 2-fold and 4-fold less potent than DiC8-PtdIns(3,4,5)P3 as inhibitors of clathrin assembly. Deacylation of DiC8-PtdIns(3,4,5)P3 reduced its affinity for AP-3 almost 20-fold, and also dramatically lowered its ability to inhibit clathrin assembly. The deacylated products of the soluble derivatives of phosphatidylinositol 3,4-bisphosphate and phosphatidylinositol 4,5-bisphosphate were both not significant inhibitors of clathrin assembly. It therefore appears that the interactions of inositides with AP-3 should not be considered simply in terms of electrostatic effects of the highly charged phosphate groups. Ligand specificity appears also to be mediated by hydrophobic interactions with the fatty-acyl chains of the inositol lipids.

There is increasing evidence that certain inositides (members of a large family of inositol phosphates and inositol lipids) act at several levels to regulate both the recruitment and the activation of proteins involved in vesicular transport. PtdIns(3,4,5)P 3 , 1 and the polyphosphoinositide 3-kinase that synthesizes this lipid both feature prominently in this area (1,2). For example, intracellular trafficking of tyrosine kinase receptors depends upon interactions between the internalized receptors and the 3-kinase (3). Additionally, PtdIns(3,4,5)P 3 interacts either directly or indirectly with several GTPases that play important roles in vesicle traffic (1). Finally, in yeast, the Vps34 protein that is essential for Golgi to vacuole transport is a phosphoinositide 3-kinase (4).
Interest in this field is not restricted to the inositol lipids. Other studies into the roles of the inositides in protein traffic have ascertained that certain inositol polyphosphates can bind with high affinity to the clathrin assembly proteins, AP-2 and AP-3 (5)(6)(7)(8)(9). AP-2 is believed to play a general role in endocytosis, since it has been found on vesicles budding from the plasma membranes of all types of cells (10). AP-3 is believed to play a more specialized role in synaptic vesicle biogenesis and recycling (11,12), since its expression is restricted to neuronal cells, and within neuronal cells to presynaptic terminals (13)(14)(15). One important consequence of inositol polyphosphates binding to either AP-2 or AP-3 is inhibition of their ability to promote clathrin assembly (5,8,9). Some attention has been drawn to the particularly potent inhibition brought about by InsP 6 , which is also one of the more abundant of the inositol polyphosphates (16). However, it has also been established that inositol lipids can bind to AP-2 and inhibit AP-2-mediated clathrin assembly (5,17). Whether inositol phosphates, inositol lipids, or both, are natural ligands for AP-2 in vivo remains to be established.
There is growing evidence that a large number of proteins that have been implicated in vesicular traffic within nerve terminals participate in inositide metabolism or bind to inositide metabolites. These include synaptotagmin, dynamin, and synaptojanin. Dynamin is the GTPase believed to be involved in the pinching off of a budding clathrin-coated vesicle during synaptic vesicle recycling. The pleckstrin homology domain of dynamin has been found to bind to Ins(1,4,5)P 3 and PtdIns(4,5)P 2 (18). Synaptotagmin, a synaptic vesicle membrane protein, has been found to bind to inositol polyphosphates and inositol lipids (19,20), and antibodies that block this binding have been found to inhibit transmission at the squid giant synapse (21). Synaptojanin (p145/dephosphin) is a nerve terminal enriched inositide 5-phosphatase, which inter-* This work was supported by National Institutes of Health NINDS Grant NS29051 (to E. M. L.), NIGMS Grant GM31278 (to J. R. F.), and NINDS Grant NS29632 (to G. D. P.), and by NEN Life Sciences. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
* acts with the Src homology 3 domain of Grb2 and amphiphysin and is rapidly dephosphorylated following synaptic activity (22)(23)(24). An essential role of inositides in neuronal secretion is also believed to account for the long-standing observation that ATP is essential for the priming of vesicles to be fusion competent (25). A series of studies have established that the ATP requirement is for the synthesis of inositol lipids (26,27).
As part of the growing effort to understand the role inositol phosphates and inositol lipids play in the regulation of vesicle traffic within nerve terminals, we set out to determine whether or not the synapse-specific clathrin assembly protein AP-3 can also interact with inositol lipids. This is of particular relevance since cellular levels of inositol lipids respond quickly to a range of extracellular agonists. For example, levels of PtdIns(4,5)P 2 have been known for many years to fall by up to 50% during the initial few seconds of receptor-dependent activation of phospholipase C (28). Such an effect is very transient, and lipid levels typically return to near-stimulated levels within 1-2 min. On the other hand, receptor-mediated activation of the PtdIns(4,5)P 2 3-kinase brings about much more profound changes in cellular levels of PtdIns(3,4,5)P 3 . Its concentration rapidly increases several fold when cells are activated by a wide range of agonists that control many inflammatory, trophic, and growth responses (for reviews, see Refs. 2 and 29). These credentials comprise a specific focus for attempts to understand the molecular actions of this putative second messenger. Thus, several converging lines of inquiry have led us to test the possibility that PtdIns(3,4,5)P 3 may be a physiologically important ligand for AP-3. Once it was clear that this was indeed the case, we went on to uncover some important information concerning the determinants of ligand specificity.

Methods
Inositol Polyphosphate Binding-The experiments to determine the K d and B max values for the binding of [ 3 H]InsP 6 to GST-AP-3 were performed exactly as described previously (9), except that nonspecific binding was determined at a concentration of 100 M; single-site binding parameters were determined by nonlinear regression using Slide-writePlus, version 2 (Advanced Graphics Software, Carlsbad, CA). The experiments in which 5 nM [ 3 H]InsP 6 was displaced from GST-AP-3 by various ligands were also performed as described previously (9).
Clathrin Assembly-GST-AP-3 and clathrin were dialyzed into 10 mM Tris-HCl, pH 8.0, and stored in this buffer after clarification by ultracentrifugation at 100,000 ϫ g for 1 h. Ligands were also dissolved in 10 mM Tris-HCl, pH 8.0. All clathrin assembly assays were done in 100-l aliquots. First, 70 l of reaction mixture containing GST-AP-3 (at a final concentration of 0.14 -0.21 mg/ml) and the appropriate ligand was made in the buffer containing 10 mM Tris-HCl, pH 8.0. To this 10 l of 10 ϫ clathrin assembly buffer (1 ϫ buffer is 0.1 M MES-NaOH, 1 mM EDTA, 1 mM EGTA, 1.5 mM MgCl 2 , pH 6.7) was added to reduce the pH of the solution to 6.7. The contents were incubated on ice for 30 min. Finally, 20 l of clathrin, present at a concentration of 1.5 mg/ml in 10 mM Tris-HCl, pH 8.0, was added to the reaction mixture to induce clathrin assembly. The final volume of the reaction mixture is 100 l. The final concentrations of clathrin and GST-AP-3 in the assembly mixture are 0.3 and 0.1-0.15 mg/ml, respectively, containing an appropriate final concentration of the ligand. The final buffer conditions are 0.009 M Tris-HCl, 0.1 M MES-NaOH, 1 mM EGTA, 1 mM EDTA, 1.5 mM MgCl 2 , pH 6.7. The contents were further incubated on ice for 1 h and centrifuged at 100,000 ϫ g for 20 min at 4°C in a Sorvall RC M120 ultracentrifuge to pellet the assembled cages. The top 80% of the supernatant was removed and analyzed for its clathrin content on 4 -20% acrylamide gradient SDS gels. The amount of clathrin assembled was quantified by comparing the intensity of Coomassie Blue-stained clathrin bands before and after centrifugation. The Coomassie Blue-stained gels were scanned by a Molecular Dynamics laser densitometer. The amount of clathrin assembled in the absence of extraneous ligands was typically in the range of 60 -70%. Each data point in the plots shown represent the average of a minimum of three independent experiments, as explicitly indicated in the figure legends, with error bars drawn corresponding to 1 standard error.

RESULTS
We first determined the characteristics with which our current preparations of GST-AP-3 bound InsP 6 . The binding parameters were estimated by fitting a saturation curve to the data by nonlinear regression analysis. The calculated B max value was 0.62 Ϯ 0.05 mol of InsP 6 /mol of GST-AP-3, and the K d value was 0.67 Ϯ 0.05 M (means and standard errors from five determinations). These values are within the range of those previously reported for GST-AP-3 (8,9).
We next determined the affinities of certain inositide ligands for GST-AP-3, relative to InsP 6 . This was achieved by measuring binding of [ 3 H]InsP 6 to GST-AP-3 in the presence of increasing quantities of nonlabeled, competing ligands. From the resulting displacement curves (e.g. Fig. 1), we calculated IC 50 values (Table I). For example, we found that the IC 50 values for Ins(1,3,4,5)P 4 and InsS 6 were, respectively, 22-fold and 128-fold higher than the IC 50 value for InsP 6 . Thus, in agreement with our earlier study (9), we determined that Ins(1,3,4,5)P 4 and InsS 6 are considerably weaker ligands when compared to InsP 6 . In earlier studies it was found that the physiologically relevant consequence of InsP 6 binding to AP-3 was inhibition of its ability to promote clathrin assembly (8,9). We confirmed this observation in the current study (Fig. 2). Also in agreement with our earlier report (9), we found that the dose-response inhibition curve for InsP 6 reached an asymptote of about 30% inhibition. (Note that we are unable to directly compare this particular result with the data provided by Norris et al. (8), who presented their results in a manner that does not reveal the absolute level of inhibition of clathrin cage assembly by InsP 6 .) In our earlier studies the relative affinities of inositide ligands for AP-3 corresponded with their relative potencies as inhibitors of clathrin assembly. This was also the case in the current work, with regard to InsS 6 being both a relatively weak ligand ( Fig. 1; Table I) and a poor inhibitor of assembly ( Fig. 2; Table I). However, this pattern was not maintained in the case of Ins(1,3,4,5)P 4 . This polyphosphate was as efficacious as InsP 6 at inhibiting clathrin assembly ( Fig. 2; Table I), despite the binding assays indicating that Ins(1,3,4,5)P 4 was a 22-fold weaker ligand ( Fig. 1; Table I).
Our next goal was to study the interaction of PtdIns(3,4,5)P 3 with GST-AP-3. Due to the insolubility of the naturally occurring lipid, for in vitro studies it must be inserted into lipid-or detergent-based micelles. Uncertainties concerning the effective concentration of inositol lipids under these conditions, and the contributory effects of the micelle itself, both hinder the interpretation of any effects that may be observed. To combat this problem, there has been an increasing use of water-soluble analogues of the inositol lipids, which has led to some important, quantitative information on their actions (17,30,35,36). We also adopted this approach. We found that soluble dioctanoyl-PtdIns(3,4,5)P 3 (DiC 8 PtdIns(3,4,5)P 3 ) was only about 7.5-fold weaker a ligand than InsP 6 (Fig. 1).
This observation led us to further investigate the effect of this lipid on the function of GST-AP-3. It was particularly notable that DiC 8 PtdIns(3,4,5)P 3 was a more potent inhibitor of clathrin assembly than was InsP 6 ; at a concentration of 10 M, the lipid was about 2-fold more effective (Fig. 2). DiC 8 PtdIns(3,4,5)P 3 also had a far greater maximal effect; up to 90% of total clathrin assembly could be consistently inhibited by this lipid (Fig. 2). This comparison between InsP 6 and DiC 8 PtdIns(3,4,5)P 3 reveals a result that also arose in the comparison of InsP 6 with Ins(1,3,4,5)P 4 , i.e. the rank order of efficacy with which inositides displace [ 3 H]InsP 6 from AP-3 did not always match the rank order of potency with which they inhibit clathrin assembly. We considered that one possible explanation for this difference is that the two assays were routinely carried out at different pH. It was necessary to perform the clathrin assembly assays at pH 6.7 to optimize sensitivity to the limited amounts of several of the available inositides. In contrast, the optimal pH for the binding assays was 7.5, whereupon nonspecific binding was minimized. Nevertheless, we did carry out a complete set of displacement studies in the same buffer as was used in our clathrin assembly assay. The rank-order with which all the ligands displaced InsP 6 from clathrin at pH 6.7 was identical to that obtained at pH 7.5 (data not shown). Therefore, the discrepancy between relative ligand affinity and relative potency at inhibiting clathrin assembly more likely reflects differences in the nature of the measurements. The binding data are carried out under equilibrium conditions, whereas the clathrin assembly measurements are end point determinations.

from GST-AP-3, and for the inhibition of GST-AP-3-mediated clathrin assembly
For displacement of InsP 6 from GST-AP-3, the IC 50 values were calculated from experiments similar to those described in Figs. 1, 4, and 5. Data are means and standard errors from three independent experiments, except in the case of InsS 6 in which the mean and standard error from two independent experiments are given. For inhibition of GST-AP-3-mediated clathrin assembly, the IC 50 values were calculated from the curves shown in Figs. 2, 3, and 6. Each curve was the average of from three to seven independent experiments, as indicated in the respective figure legends. IC 50 is defined as the concentration of ligand required for half maximal inhibition of clathrin assembly. I max is defined as the maximal inhibition of clathrin assembly. further interest that one of the appropriate 5-phosphatase activities can be catalyzed by synaptojanin, which is enriched in nerve terminals where AP-3 is also concentrated (23). We also found that a soluble diether analogue of PtdIns(4,5)P 2 , DiC 11 -O-PtdIns(4,5)P 2 , interacted with GST-AP-3 less effectively than did DiC 8 PtdIns(3,4,5)P 3 (Figs. 3 and  4). The latter exhibited a 23-fold higher affinity for GST-AP-3 ( Fig. 4; Table I) and was a more potent inhibitor of clathrin assembly (e.g. 50 M DiC 8 PtdIns(3,4,5)P 3 inhibited clathrin assembly 4-fold greater than did DiC 11 -O-PtdIns(4,5)P 2 ). This difference could indicate that the 3-phosphate is important for specificity, although we cannot exclude that DiC 11 -O-PtdIns(4,5)P 2 is less effective because of its alternative fatty acids and/or its diether structure. Therefore we directly investigated whether the fatty-acid chains contributed to the interactions of inositol lipids with GST-AP-3. Deacylation of DiC 8 PtdIns(3,4,5)P 3 to yield GroPIns(3,4,5)P 3 reduced its affinity for AP-3 almost 20-fold (Fig. 5A). This important result was confirmed with an independent batch of GroPIns(3,4,5)P 3 kindly supplied by Dr. A. Theibert; its affinity for AP-3 was indistinguishable from our own material. The GroPIns(3,4,5)P 3 was also considerably less effective than DiC 8 PtdIns(3,4,5)P 3 at inhibiting clathrin assembly (Fig. 6A), both in terms of relative IC 50 values (Table I, approximately 10 M for DiC 8 PtdIns(3,4,5)P 3 , 45 M for GroPIns(3,4,5)P 3 ) and in terms of the maximal degree of inhibition achieved (approximately 90% for PtdIns(3,4,5)P 3 , 33% for GroPIns(3,4,5)P 3 ). Further evidence for the importance of the fatty acids was obtained by comparing the effects of DiC 8 PtdIns(3,4)P 2 with Gro-PIns(3,4)P 2 , and DiC 11 -O-PtdIns(4,5)P 2 with GroPIns(4,5)P 2 . In both cases the deacylated compounds were not significant inhibitors of assembly (Fig. 6, B and C), and they were at least 10-fold weaker ligands in the binding assays than the corresponding lipids (Fig. 5, B and C). It therefore appears that the interactions of inositides with AP-3 should not be considered simply in terms of electrostatic effects of the highly charged phosphate groups. Ligand specificity appears also to be mediated by hydrophobic interactions with the fatty-acyl chains of the inositol lipids. DISCUSSION This study provides important new data in support of the emerging consensus that inositol lipids are important regulators of vesicle traffic in nerve terminals (1). For example, we have demonstrated that among those inositides that inhibit AP-3-mediated clathrin assembly, DiC 8 PtdIns(3,4,5)P 3 is the most potent. Since Gaidarov et al. (17) recently concluded that PtdIns(3,4,5)P 3 is the most potent inositide inhibitor of AP-2mediated clathrin assembly as well, our results indicate that PtdIns(3,4,5)P 3 might be a general inhibitor of endocytic events. There is additional, indirect evidence that supports this idea. Such information comes from work with wortmannin, a fungal metabolite used in many experiments to inhibit PtdIns(3,4,5)P 3 synthesis by the phosphoinositide 3-kinase (37). Spiro et al., (38) recently reported that wortmannin increased the rate of internalization of occupied transferrin receptors in K562 cells. This result is consistent with PtdIns(3,4,5)P 3 acting to constrain endocytosis, which could be anticipated if the lipid inhibited clathrin assembly in vivo. While this effect has not been seen in several other studies (see Ref. 39), this could reflect the ability of wortmannin to target multiple stages in the complex process of receptor trafficking. For example, the wortmannin-sensitive 3-kinase also participates in the post-endocytic trafficking of internalized receptors (3). However, a mutant platelet-derived growth factor receptor has been created that is specifically deficient in the 3-kinase binding sites that mediate these post-endocytic events (3). This mutant retains its ability to be endocytosed, and offers itself as a useful tool in the specific study of the initial stages of receptor internalization. Indeed, wortmannin was found to increase the rate of internalization of this mutant receptor (3). This result is again consistent with the idea that PtdIns(3,4,5)P 3 inhibits some general aspect of the initial stages of endocytosis of receptors.
As well as improving our insight into the roles of inositol lipids in vesicle traffic in general, the demonstration that AP-3 has high affinity for PtdIns(3,4,5)P 3 provides some new information specific to this particular protein. For example, it prompted us to consider that AP-3 may share some sequence homologies with other PtdIns(3,4,5)P 3 -binding proteins. In this respect, the recent discovery of a novel PtdIns(3,4,5)P 3 -binding protein, centaurin-␣, has led to the identification of a putative consensus motif (PX 13-14 (R/K)X 4 (R/K)XKX 5 FX 6 E) for dictating the high affinity for the 3,4,5-trisphosphate cluster and which is shared by synaptotagmin and a protein (GAPInsP 4 -BP), which is a candidate for mediating Ca 2ϩ entry into cells (36). It is interesting that this motif is partly conserved by amino acids 32-51 (HX 5 KKKX 3 YX 6 E) within the 33-kDa amino-terminal domain of AP-3 (11), which is known to contain the polyphosphate binding site (9). Further experiments into the determinants of ligand specificity of AP-3 might profit from studying this possibility.
Some very interesting structure-function relationships also emerge from a comparison of the inositide binding domains of AP-3 and AP-2. The latter protein is a tetramer containing an ␣-adaptin subunit, which has been shown to contain the inositide binding site (17), and a clathrin triskelia binding site (40), although this subunit cannot assemble clathrin triskelia into clathrin cages. On the other hand, the ␤-adaptin subunit of AP-2, while unable to bind inositides, can bind clathrin triskelia and assemble them into cages (41). AP-3 is a monomer with two functional domains, a 33-kDa N-terminal domain that contains the inositide binding site (9) and a clathrin binding site that can bind to clathrin triskelia but cannot promote clathrin cage assembly (34,42), and a 58-kDa C-terminal domain that does not bind inositides but can efficiently promote clathrin cage assembly (33). In both proteins, inositide binding and clathrin assembly are separated into either distinct domains within a monomeric protein or distinct subunits within a multimeric protein, and in both cases inositide binding regulates clathrin assembly. It is also interesting that the rank order of both binding constants and inhibition constants of inositides follow an unusual but similar pattern. For example, for both proteins InsP 6 binds more tightly to the protein than DiC 8 PtdIns(3,4,5)P 3 , yet the lipid is the more potent inhibitor in the clathrin assembly assays (compare the data presented here on AP-3 with the AP-2 data presented in Gaidarov et al. (17)). Interestingly, in nerve terminals both AP-2 and AP-3 have been co-localized to the same budding clathrin coated vesicles (43), suggesting that even though each protein is able to independently promote clathrin cage assembly in vitro, they must co-operate in some manner in vivo.
Another important aspect of this study with AP-3 is that it provides new information concerning the determinants of specificity for a lipid-protein interaction. Our data showing that the deacylation of inositol lipids reduces their affinity for AP-3 consolidates the idea that the fatty-acyl chains contribute to the specificity of these interactions (17) rather than their solely reflecting an interaction with the charged headgroup (1). The finding that the fatty acids contribute to the specificity of the interaction also prompts the consideration of how AP-3 would have access to these lipids in vivo. When the inositides are incorporated into the lipid bilayer of a membrane, it seems likely that the fatty acids would not be very accessible for an interaction with a soluble protein. Therefore, unless AP-3 has the ability to insert a hydrophobic region into the bilayer, the possibility must be considered that AP-3 competes with the membrane for PtdInsP 3 . Since AP-3 is a relatively abundant protein (estimated to be 0.2% of cellular protein; Ref. 44) such an interaction could have the consequence of modulating the cellular availability and actions of this lipid.