Molecular Basis for Association of PIPKIγ-p90 with Clathrin Adaptor AP-2*

Phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2) is an essential determinant in clathrin-mediated endocytosis (CME). In mammals three type I phosphatidylinositol-4-phosphate 5-kinase (PIPK) enzymes are expressed, with the Iγ-p90 isoform being highly expressed in the brain where it regulates synaptic vesicle (SV) exo-/endocytosis at nerve terminals. How precisely PI(4,5)P2 metabolism is controlled spatially and temporally is still uncertain, but recent data indicate that direct interactions between type I PIPK and components of the endocytic machinery, in particular the AP-2 adaptor complex, are involved. Here we demonstrated that PIPKIγ-p90 associates with both the μ and β2 subunits of AP-2 via multiple sites. Crystallographic data show that a peptide derived from the splice insert of the human PIPKIγ-p90 tail binds to a cognate recognition site on the sandwich subdomain of the β2 appendage. Partly overlapping aromatic and hydrophobic residues within the same peptide also can engage the C-terminal sorting signal binding domain of AP-2μ, thereby potentially competing with the sorting of conventional YXXØ motif-containing cargo. Biochemical and structure-based mutagenesis analysis revealed that association of the tail domain of PIPKIγ-p90 with AP-2 involves both of these sites. Accordingly the ability of overexpressed PIPKIγ tail to impair endocytosis of SVs in primary neurons largely depends on its association with AP-2β and AP-2μ. Our data also suggest that interactions between AP-2 and the tail domain of PIPKIγ-p90 may serve to regulate complex formation and enzymatic activity. We postulate a model according to which multiple interactions between PIPKIγ-p90 and AP-2 lead to spatiotemporally controlled PI(4,5)P2 synthesis during clathrin-mediated SV endocytosis.

is required for the initial targeting of AP-2 to the plasma membrane as well as for cargo recognition, which in turn serves to stabilize nascent coated pits during CME (12). Given that CME is a pathway active in most cell types, one might expect that PIPKI-mediated synthesis of PI(4,5)P 2 would be subject to common mechanisms of spatiotemporal regulation. In support of this idea, it has been reported that PIPKI isozymes associate with AP-2 via direct interaction between the kinase core and a site within AP-2 distinct from that involved in recognizing YXXØ-based sorting motifs. Moreover, cargo protein binding has been shown to stimulate PIPKI activity, suggesting that synthesis of an endocytic pool of PI(4,5)P 2 is under the control of a positive feedback loop (17). Such a mechanism might be important for sustaining endocytosis under conditions of intense competition for PI(4,5)P 2 by other pathways.
Synapses face special challenges with respect to membrane trafficking because of the spatial and temporal constraints of chemical neurotransmission (4). A number of independent lines of evidence indicate a central role for PI(4,5)P 2 metabolism in exo-/endocytic cycling of SVs. Perturbation of PI(4,5)P 2 breakdown by genetic or acute manipulation of the inositol phosphatase synaptojanin impairs SV endocytosis at the stage of vesicle uncoating, resulting in synaptic depression following intense stimulation (18,19). Most importantly, in mice the loss of PIPKI␥, the major synthesizing enzyme at synapses, produces defects in SV recycling including significantly reduced frequencies of spontaneous miniature postsynaptic currents and decreased rates of clathrin-mediated SV endocytosis (20). These data suggest a crucial role for PIPKI␥-mediated PI(4,5)P 2 generation in CME of SV proteins. Consistent with these results, it has been reported that the tail domain of PIPKI␥-p90, the major splice isoform in neurons, can associate directly with the ␤2 appendage domain (21) and/or the sorting signal binding domain of the 2 subunit of AP-2 (22). In agreement with this, overexpression of the AP-2-binding PIPKI␥ tail domain has been shown to impair depolarization-induced SV endocytosis (21). The precise structural basis of these effects, however, has remained elusive.
In this work we report on the crystal structures of a peptide derived from the 28-amino acid splice insert of the PIPKI␥-p90 tail in complex with the ␤2 appendage or the C-terminal sorting signal binding domain of AP-2. Our biochemical and structure-based mutagenesis analysis revealed that association of the natively unfolded tail domain of PIPKI␥-p90 with AP-2 involves both of these interaction sites. Moreover, the ability of the PIPKI␥ tail to impair endocytic retrieval of SV proteins in primary neurons is critically dependent on its association with both AP-2␤ and AP-2. Our data indicate that interactions between AP-2 and the tail domain of PIPKI␥-p90 may serve to regulate complex formation and enzyme activity, suggesting a model according to which multiple interactions between PIPKI␥-p90 and AP-2 lead to spatiotemporally controlled PI(4,5)P 2 synthesis during SV endocytosis.
Protein Purification-Constructs in pGEX4T-1 and pET28 were transformed into Escherichia coli BL21-Codon Plus TM (DE3)-RP competent cells (Stratagene). His 6 -or GST-tagged fusion proteins were expressed and purified using HIS-Select TM nickel affinity gel (Sigma) or GST Bind resin (Novagen) following the manufacturer's instructions. For isothermal titration calorimetry (ITC) experiments, the His 6 -C-2 or His 6 -␤2 appendage was purified using HisTrap HP columns (GE Healthcare) according to the manufacturer's instructions. The proteins were further purified by size exclusion chromatography (Superdex S200, Amersham Biosciences) in 20 mM Hepes, pH 7.4, containing 200 mM NaCl.
Preparation of Rat Brain and Synaptosomal Extracts-Rat brain extract was prepared from frozen rat brains homogenized in 10 ml of homogenization buffer (4 mM Hepes, pH 7.4, 320 mM sucrose, 1 mM PMSF, and protease inhibitor mixture (Sigma)) by using a glass-Teflon homogenizer. The homogenate was centrifuged for 10 min at 1000 ϫ g. The supernatant was then adjusted to 20 mM Hepes, pH 7.4, 100 mM NaCl, and 2 mM MgCl 2 containing 1% Triton X-100. Following incubation for 10 min on ice, the extract was centrifuged for 15 min at 43,500 ϫ g. The resulting supernatant was precleared by ultracentrifugation for 15 min at 184,000 ϫ g and used for pulldown experiments. For preparation of rat brain synaptosomal extracts, homogenized rat brain was centrifuged for 10 min at 1000 ϫ g. Postnuclear supernatant was centrifuged for 20 min at 14,600 ϫ g. The crude synaptosomal pellet was resuspended in 10 ml of homogenization buffer containing 1 mM PMSF and protease inhibitor mixture followed by further centrifugation for 20 min at 14,600 ϫ g. The resulting synaptosomal pellet was resuspended in homogenization buffer containing 1 mM PMSF and protease inhibitor mixture and adjusted to 20 mM Hepes, pH 7.4, 100 mM NaCl, and 2 mM MgCl 2 containing 0.5% CHAPS. Synaptosomes were incubated for 40 min at 4°C. The resulting lysate was precleared by further centrifugation steps, first for 15 min at 35,000 ϫ g and then for 15 min at 184,000 ϫ g, and used for pulldown experiments.
Crystallography of the ␤2 Appendage/PIPKI␥-p90 Peptide Interaction-An E. coli construct containing the gene coding for the ear domain of the ␤2 subunit of AP-2 (His 6 -␤2 appendage, amino acids 705-937) was cultured in "double" LB medium supplemented with kanamycin (40 g/ml each). When the absorbance at 600 nm reached 0.7, overexpression was induced by supplementing the culture with 0.8 mM isopropyl 1-thio-␤-D-galactopyranoside, and cells were harvested 3 h later. His 6 -␤2ear was isolated by nickel-nitrilotriacetic acid affinity chromatography (Qiagen), and the His 6 tag cleaved off by thrombin and ␤2-ear was purified by size exclusion chromatography (Superdex S75, Amersham Biosciences). The purified protein was dialyzed against 20 mM Tris, pH 8, and 150 mM NaCl, concentrated to about 60 -80 mg/ml, and supplied with a 5-fold molar surplus of the PIPKI␥-p90 pentadecapeptide YFPTDER-SWVYSPLH. Crystals were grown at 18°C using the sitting drop vapor diffusion method in the presence of trimethylphenylammonium-exchanged hectorite (Ͻ2 m). The reservoir solution contained 18% polyethylene glycol 8000, 100 mM Hepes, pH 7.5, and 4 mM dithiothreitol. Drops were prepared by mixing 1 l of reservoir and 1 l of protein/peptide solution at 80 mg/ml. Crystal showers with poorly defined microcrystals not suitable for single crystal x-ray studies appeared within 10 -30 s. To slow the crystallization process, nucleation seeds were introduced in the form of trimethylphenylammoniumexchanged hectorite (Ͻ2 m). Thin plate-like crystals formed around these sites within 60 min and reached their final size in about 3 h at 18°C.
Single crystals of the AP-2 ␤2 ear co-crystallized with the above mentioned PIPKI␥ pentadecapeptide were soaked briefly in a cryoprotection medium of 18% polyethylene glycol 8000, 100 mM Hepes, pH 7.5, 4 mM dithiothreitol, and 15% v/v glycerol, mounted in a nylon loop, and then flash-cooled in liquid N 2 . X-ray data were collected at Beamline BL2 at BESSY-II, Berlin, and processed using HKL2000 (24) and Scalepack. The space group was determined as P2 1 2 1 2 1 with a solvent content of 51%, indicating 1 molecule/asymmetric unit. The unit-cell parameters were a ϭ 37.56, b ϭ 83.47, and c ϭ 91.60 Å. The crystals diffracted to 1.83 Å resolution. The data set was 95% complete with an overall R merge of 6.0%. The phase problem was solved by molecular replacement with the CCP4 (25) program MOLREP using the 1.60 Å structure of the ␤2 ear (Protein Data Bank code 2G30) as a model (26). All water molecules and ligand atoms were omitted from the starting model. Subsequent cycles of the isotropic B-value and positional refinement to 1.83 Å resolution were performed using Refmac5 (27). The peptide chain and missing residues were built manually using the model-building program Coot (28). Orientation of the amino acid side chains and bound water molecules was modeled based on electron density Ͻ 3 in F o Ϫ F c Fourier maps. Modeled water molecules that refined with electron density Ͼ 3 were deleted. The final R-factor for the resolution range of 60 to 1.83 Å was 19.3% (R free 23.6%). The Ramachandran plot performed with PROCHECK indicated that 93.6% of the residues fell within the sterically most favored region with an additional 6.4 and 0% within the allowed and generously allowed regions, respectively. The statistics of the resulting structure (structural data available under Protein Data Bank codes 3H85 and 3H1Z) are reported in Tables 1-3.
Crystallography of the 2/PIPKI␥-p90 Peptide Interaction-An E. coli construct containing the gene coding for the C-terminal domain of the 2 subunit of AP-2 (His 6 -C-2; amino acids 158 -435) was cultured in double LB medium supplemented with kanamycin (40 g/ml each). When A 600 nm reached 0.7, overexpression was induced by supplementing the culture with 0.8 mM isopropyl 1-thio-␤-D-galactopyranoside, and cells were harvested 4 h later. His 6 -C-2 was isolated by nickel-nitrilotriacetic acid affinity chromatography (Qiagen), and the His 6 tag cleaved off by thrombin and C-2 was purified by size exclusion chromatography (Superdex S200, Amersham Biosciences). The purified protein was dialyzed against 10 mM Hepes, pH 7.5, 150 mM NaCl, and 4 mM dithiothreitol, concentrated to about 22 mg/ml, and supplied with a 5-fold molar surplus of the human PIPKI␥ octapeptide SWVYSPLH. Crystals were grown at 18°C using the sitting drop vapor diffusion

PIPKI␥-p90⅐AP-2 Complex
performed using Refmac5 (27). The peptide chain and missing residues were built manually using the model building program Coot (28). The orientation of the amino acid side chains and bound water molecules were modeled based on electron den-  The entire sequence of the bound peptide participates in a ␤-sheet-like hydrogen bonding with ␤2. C, surface representation of the PIPKI␥-p90 peptide binding interface with the ␤2 appendage including an overlay with the previously characterized Eps15-derived peptide, SFGDGFADF (yellow), (23) to compare the binding of the two peptides.

PIPKI␥-p90⅐AP-2 Complex
Isothermal Titration Calorimetry-High sensitivity ITC was performed on a VP-ITC device (MicroCal) at 25°C using gelfiltrated protein at concentrations of 60 to 100 M in buffer (20 mM Hepes, pH 7.4, and 200 mM NaCl). The PIPKI␥-p90 peptide (YFPTDERSWVYSPLH) was dialyzed against this buffer and injected in 30 steps of 10 l each from 1.12 mM stock solutions at 5-min intervals. Both the peptide and protein samples were gently degassed under stirring prior to the experiment to avoid air bubbles. Peptide was also injected into dialysis buffer to measure the heat of dilution. Base-line subtraction and peak integration were accomplished using Origin 7.0 as described by the manufacturer (MicroCal Software). All reaction heats were normalized with respect to the molar amount of peptide injected and corrected for the heat of dilution. Dissociation constants, K D , were calculated in Origin 7.0 using nonlinear regression analysis according to the "one set of sites" model. Initial injections were always excluded from the evaluation because they usually suffer from sample loss due to mounting of the syringe and equilibration preceding the actual titration.
PIPK Activity Assays-HEK293 cell extracts were incubated with GST fusion proteins or assayed directly. Samples were incubated in kinase buffer (50 l) containing 10 g of phosphatidylinositol 4-phosphate, 200 M ATP, and 10 Ci of [␥-32 P]ATP for 20 min at 37°C. Lipids were extracted and analyzed as described previously (17).
Quantification of Synapto-pHluorin Exo-/Endocytic Cycling in Primary Hippocampal Neurons-Synapto-pHluorin (a fusion protein between pHluorin green fluorescent protein and synaptobrevin 2) partitioning between vesicular and surfacestranded plasmalemmal pools was assayed according to published procedures. Briefly, primary hippocampal neurons were prepared from Wistar rats within 24 h after birth and co-transfected with plasmids encoding for ecliptic pHluorin-tagged synaptobrevin 2 (synapto-pHluorin) and mRFP (control) or mRFP fused to the C-terminal tail of PIPKI␥-p90 comprising amino acids 451-668 (WT or indicated mutants thereof) by calcium phosphate-DNA co-precipitation on day 14 in vitro. Neurons were analyzed by live imaging analysis within 16 to 18 h after transfection using a charge-coupled device camera For each neuron, more than 50 boutons were quantified. The fraction (p) of surface-stranded synapto-pHluorin (spH) was calculated by the following equation:

Structural Basis for the Association of PIPKI␥-p90 with the ␤2
Appendage Domain of AP-2-PIPKI␥-p90 has been reported to associate with AP-2 via three distinct interaction sites: direct binding of the PIPK catalytic core to AP-2 at a site distinct from the YXXØ motif recognition site (17) (site 1) and association of a peptide rich in aromatic and hydrophobic amino acids contained within the 28-amino acid splice insert of the p90 tail domain with either AP-2 (22) (site 2) or the AP-2 ␤ appendage domain (21) (site 3). At present it is unclear to what extent these sites determine AP-2⅐PIPKI␥-p90 complex formation and what is the precise molecular basis of these interactions. Furthermore, physiologically it is possible that multiple interaction surfaces for PIPKI␥-p90 within AP-2 are involved in the regulation of CME, i.e. during recycling of SV proteins at presynaptic nerve terminals (32).
We therefore decided to characterize these interaction surfaces in more detail, focusing on the binding  ␤2 appendage and PIPKI␥-p90-(451-668) confirm the structural data. A, affinity chromatography of native endogenous PIPKI␥ from rat brain synaptosomal extracts using the GST-␤2-ear (appendage) domain. Aliquots of rat brain synaptosomal extract and affinity-purified material were analyzed by immunoblotting with antibodies against PIPKI␥ or actin as a control. std., 7.5% of the total amount of rat brain synaptosomal extract added to the assay as the standard. B, GST-PIPKI␥-p90-(451-668) WT or mutant proteins were incubated with purified His 6 -talin FERM domain or His 6 -␤2 appendage. The bound talin FERM domain or ␤2 appendage was detected by immunoblotting using His 6 tag-specific antibodies. std., 50% (talin-FERM) or 10% (␤2 appendage) of the total amount of His 6 tagged-protein added to the assay as the standard.

TABLE 7 Interface statistics
The protein interfaces, surfaces, and assemblies (PISA) service at European Bioinformatics Institute (Krissinel and Henrick (43)) was used. iNat and iNres indicate the number of interfacing atoms and residues, respectively, in the corresponding structures. of the human PIPKI␥-p90 tail to the purified ␤2 appendage or 2-(158 -435) (see below). Truncation analysis (data not shown) narrowed the minimal binding site required for association of the PIPKI␥-p90 tail with the ␤2 appendage to a 15amino acid sequence within the 28amino acid splice insert of p90, in complete agreement with previous work (21). The corresponding peptide was synthesized and used for co-crystallization experiments with the purified ␤2 appendage. The crystals obtained diffracted to a maximal resolution of 1.83 Å, and the structure model was refined with excellent B-factors for protein and peptide ligand ( Table 1). The p90 peptide is seen to accommodate the sandwich subdomain of the ␤2 appendage (Fig. 1A) at a location similar to that reported previously for a short peptide fragment derived from Eps15 (23) (Fig. 1C and Tables 2 and 3). The peptide largely binds via aromatic amino acids Phe 640 (residues within human PIPKI␥ are indicated by superscripted numbering), Trp 647 , and Tyr 649 . Phe 640 , surrounded by Ala-806, Tyr-815, Gln-804, and Asn-758, exhibits the largest buried surface area (138.6 Å 2 ; see Table 2) within ␤2, Trp 647 projects into a shallow pocket formed by Lys-808, Gln-756, and Ala-754 (Fig. 1B). In comparison, Tyr 639 displays a buried surface area of only 53.8 Å 2 and does not form hydrogen bond interactions. This explains the findings by Thieman et al. (36) that mutation of Phe 640 to Ala impairs association with the AP-2␤ appendage, whereas Tyr 639 3 Ala does not. Additional major contacts involve Ser 650 , a site implicated previously in association with AP-2␤ by site-directed mutagenesis (21), as well as further hydrogen bonds along the entire sequence of the bound peptide resembling a ␤-sheet hydrogen bonding pattern (Fig. 1B and Table 2). Interestingly, Thieman et al. (36) also reported that mutation of Tyr 644 to Ala (Tyr 649 in human) completely eliminates ␤2 binding, whereas a Tyr 644 3 Phe mutation, which occurs naturally in PIPK␥-687, is tolerated, although it reduces the affinity of PIPKI␥ for ␤2. This observation can be explained by our structural data, which show that although Tyr 644 interacts directly (via hydrogen bonding) with Lys-808, it also exhibits a considerable buried surface area of 52.6 Å 2 . This buried surface area and the associated van der Waals interactions are not affected by the Tyr 644 3 Phe mutation.

Peptide
We also observed at least three major differences between the previously characterized association of an Eps15-derived peptide fragment with the ␤2 appendage (23) and our structure reported here for PIPKI␥-p90. First, in contrast to Eps15, which exhibits a turn around its glycine residue, the p90-derived peptide wraps around the sandwich subdomain much further and thus covers a considerably larger interface than the short Eps15 sequence (773.7 versus 395.2 Å 2 ; Table 7). Second, because of the substitution of Phe 6 for glutamate within the Eps15 peptide of p90, the hydrophobic contact with Val-813 is lost. Third, Asp 8 within Eps15, the side chain of which contacts Tyr-815 of ␤2, is exchanged for serine in p90 ( Fig. 1C and Tables 2 and 3).
In agreement with the critical role of ␤2 Tyr-815 and Lys-808 in complex formation with the p90 peptide, we observed a complete loss in the ability of GST-tagged ␤2 Y815A or K808A mutants to pull down PIPKI␥-p90 from rat brain extracts ( Fig.  2A). Conversely, mutation of either Phe 640 or Trp 647 to alanines completely eliminates the interaction between GST-PIPKI␥-p90-(451-668) with purified His 6 -␤2 (Fig. 2B). Somewhat surprisingly, a Trp 647 3 Phe substitution also reduces complex formation below the detection limit in these experiments. A possible explanation is the ability of the indol nitrogen of Trp 647 to function as a hydrogen donor to Gln-756. Although not detectable on the basis of current electron density, there is the possibility of a positional disorder of the amine and oxyl groups of Gln-756 resulting in formation of a hydrogen bond (distance ϭ 3.38 Å) between Trp 647 and Gln-756. Binding of GST-PIPKI␥-p90 to talin, another known ligand of the same peptide sequence (9, 10), is unaffected by the Phe 640 3 Ala mutation but is eliminated by a Trp 647 3 Ala substitution, in accordance with previous experiments. In summary, these data establish a firm structural basis for the interaction between PIPKI␥-p90 and the ␤2 appendage domain of AP-2 and identify Phe 640 3 Ala as a ␤2-adaptin interaction-defective mutant of PIPKI␥-p90.
Structural Basis for the Association of PIPKI␥-p90 with AP-2-In addition to the association of the PIPKI␥-p90 tail with the ␤2 appendage of AP-2, it has also been reported that PIPKI␥-p90 binds to the medium chain () of both the AP-1B (30) and AP-2 complexes (22) via a YXXØ-based motif contained within the 28-amino acid splice insert of p90 that overlaps with the ␤2 appendage binding site. We synthesized the corresponding peptide (SWVYSPLH, with key residues impli-cated in complex formation being highlighted) and used it for co-crystallization experiments with purified 2-(158 -435) (C-2). Crystals were obtained under conditions similar to  and the sequence of the PIPKI␥-p90-derived peptide. B and C, differential heating power (⌬p) versus time (t) obtained by injecting PIPKI␥-p90-derived peptide into ␤2-ear (B) or 2 (C). Initial 5-l injections were excluded from the fitting procedure. All ITC data were recorded at 25°C. D, integrated, normalized, and dilutioncorrected heats of reaction (Q) versus molar peptide/protein ratio (R). Heats of reaction obtained by injecting peptide into ␤2-ear (squares) or 2 (circles) were fitted by a one-site binding model (solid lines) yielding the K D values shown.

PIPKI␥-p90⅐AP-2 Complex
those seen previously for other YXXØ motif-containing peptides complexed with C-2, and these diffracted to a maximal resolution of 2.60 Å (Table 4). Electron density was observed for the entire peptide sequence. As expected, the p90 peptide was found associated with the known YXXØ motif binding site within subdomain A of C-2 (Fig. 3A). We observed a threepin-plug interaction involving Trp 647 , Tyr 649 , and Leu 652 with Asp-176, Trp-421, and Arg-423 of C-2. Trp 647 is seen in a FIGURE 5. The binding sites for talin and the AP-2 subunits 2 and ␤2-ear overlap within the C-terminal splice insert of PIPKI␥-p90. A, affinity purification of native endogenous AP-2 from rat brain extract using PIPKI␥-p90-(451-668) WT or the indicated mutants as bait. Aliquots of rat brain extract and affinity-purified material were analyzed by immunoblotting with antibodies against AP-2 and actin as a control. std., 5% of the total amount of rat brain extract added to the assay as the standard. B, GST-PIPKI␥-p90-(451-668) WT or the indicated mutants were incubated with purified His 6 -␤2-ear (appendage). Bound ␤2 appendage was detected immunoblotting using a His 6 tag-specific antibody. std., 10% of the total amount of His 6 fusion protein added to the assay. Bottom, Ponceau S-stained membranes. Molecular mass standards are indicated on the left. Note that the His 6 -␤2 appendage is not visible in the Ponceau S-stained material due to the presence of GST-PIPKI␥-p90-(451-668) degradation products. C, GST-PIPKI␥-p90-(451-668) WT or the indicated mutants were incubated with purified His 6 -C-2. Bound C-2 was detected by immunoblotting using a His 6 tag-specific antibody. std., 50% of the total amount of His 6 -2 domain added to the assay as the standard. Bottom, Ponceau S-stained membranes. Molecular mass standards are indicated on the left. D, GST-PIPKI␥-p90-(451-668) WT or the indicated mutants were incubated with purified His 6 -talin FERM domain. Bound talin FERM domain was detected by immunoblotting using a His 6 tag-specific antibody. std., 50% of the total amount of His 6 -Talin FERM domain added to the assay as the standard. Bottom, Ponceau S-stained membranes. Molecular mass standards are indicated on the left.

PIPKI␥-p90⅐AP-2 Complex
hydrophobic pocket surrounded by Glu-391, Gln-318, and Pro-393 (Fig. 3B). These interactions very much resemble the complex between a peptide derived from the cytoplasmic loop of the GABA receptor ␥2 subunit and C-2 (25), where a tyrosine (Tyr 4 ) is seen to accommodate the hydrophobic pocket occupied by Trp 647 in the case of the p90 peptide (Fig. 3C). Hence, both of these complexes also share a similarly large interaction area between peptide and protein ligand (Table 7). Thus, the PIPKI␥-p90 tail domain interacts with AP-2 via a conventional YXXØ-based sorting motif normally found in transmembrane cargo proteins.
Interaction between the PIPKI␥-p90 Tail and ␤2 Appendage or C-2-Our structural data clearly indicate that binding of the PIPKI␥-p90 tail to the ␤2 appendage and to C-2 is mutually exclusive. We thus next asked which of these two interactions might be favored thermodynamically in vitro. To this aim we performed ITC experiments using a synthetic p90 tail peptide or a ␤2 binding-defective mutant thereof (Fig. 4A). We determined K D values of 6 M for 2 (Fig. 4, C and D) and 22 M for ␤2 appendage domain-containing (Fig. 4, B and D) complexes. As expected the mutation of Phe 640 to Ala reduced binding to ␤2 to nearly undetectable levels, whereas association with 2 was almost unaffected (data not shown). Thus, at least in vitro, the interaction between PIPKI␥-p90 tail and 2 is thermodynamically favored over that with the ␤2 appendage. This does not rule out the possibility that complex formation between PIPKI␥-p90 tail and the ␤2 appendage might be physiologically relevant, nor do these data exclude further regulatory mechanisms, i.e. phosphorylation events that could alter the availability or affinity of these binding sites under physio-logical conditions (2). For example, interaction of the PIPKI␥-p90 tail with AP-2 might require its prior phosphorylation by the clathrincoated vesicle-associated protein kinases AAK1 or GAK in order to induce a conformational opening of the AP-2 complex (2,3,16).
PIPKI␥-p90 Associates with Native AP-2 via Multiple Interaction Determinants-To determine whether sequences contained within the 28-amino acid splice insert of PIPKI␥-p90 represent the only AP-2 binding sites with the PIPKI␥ tail domain, we analyzed the ability of GST-PIPKI␥-(451-668) mutant fusion proteins to affinity-purify native AP-2 from rat brain synaptosomal protein extracts. Surprisingly, alanine substitution of Phe-640 (eliminating association with the ␤2 ear) together with Tyr-649 (a key residue involved in association with AP-2) or Trp-647 in conjunction with Tyr-649 or Tyr-649 and Leu-652 significantly reduced but did not completely abrogate AP-2 binding (Fig. 5A). As expected from the crystal structure and the analysis of single point mutants (see Fig. 2), these mutations eliminated association with the ␤2 appendage (Fig. 5B). All mutants, however, retained the ability to bind to His 6 -C-2 (Fig. 5C), suggesting that the residual AP-2 binding activity seen in affinity chromatography experiments is because of the association of PIPKI␥-(451-668) with the 2 subunit. GST-PIPKI␥-p90-(451-668) F640A/Y649A also bound to talin, whereas the W647A/Y649A or the W647A/Y649A/L652A mutant did not (Fig. 5D). These results therefore suggest the existence of a second AP-2 binding site within the PIPKI␥-p90 tail.
Inspection of the primary sequence of PIPKI␥-p90-(451-668) revealed the presence of a second putative YXXØ-based sorting motif between amino acids 497 and 500 (Fig. 6A). To analyze the importance of this proximal YXXØ motif for complex formation between the PIPKI␥-p90 tail domain and AP-2, we mutated the key conserved residues Tyr-497 and Leu-500 to alanines. Whereas GST-PIPKI␥-p90-(451-668) Y497A/L500A bound to AP-2, the ␤2 appendage, or talin with an efficiency similar to the wild-type protein, a mutant in which both YXXØbased AP-2 binding motifs along with Trp-647 had been mutated to alanines (Y497A/L500A/W647A/Y649A/L652A) did not associate with any one of these ligands (Fig. 6B). To assess the relative contribution of the binding sites for AP-2␤ and AP-2 within the PIPKI␥-p90 tail domain, we next performed affinity chromatography experiments from rat brain extracts. Mutational inactivation of the binding site for the ␤2 appendage domain (F640A) alone or in combination with the distal YXXØ motif within the 28-amino acid splice insert of the

PIPKI␥-p90⅐AP-2 Complex
tail domain (W647A/Y649A/L652A) somewhat reduced but did not abolish association of GST-PIPKI␥-p90-(451-668) with native AP-2 complexes. Mutational inactivation of the proximal YXXØ motif (Y497A/ L500A) did not affect the efficiency with which either AP-2 or talin was pulled down from brain extracts. However, a mutant in which both YXXØ-based motifs along with the ␤2 appendage binding site had been inactivated (Y497A/L500A/ W647A/Y649A/L652A) did not associate with native AP-2 above background levels (Fig. 6C), corroborating our analysis using purified proteins. Hence, the PIPKI␥-p90 tail domain associates with AP-2 via two distinct AP-2-binding tyrosine-based sorting motifs, the distal of which overlaps with the binding site for the AP-2␤ appendage.
Finally, we wanted to assess the contribution of these various determinants within the kinase tail to the association of native full-length PIPKI␥-p90 with AP-2. To this aim, we performed affinity chromatography experiments from transfected COS-7 cells using GST-C-2 or GST-␤2 appendage domain as baits. As expected, mutational inactivation of the ␤2ear binding site within PIPKI␥-p90 (F640A; ⌬␤2 site) completely eliminated association with the ␤2 appendage, whereas binding to AP-2 was unaffected ( Fig. 6D and E, ⌬␤2 site). A PIPKI␥-p90 mutant in which both YXXØ-based motifs along with the ␤2 appendage binding site within the tail had been inactivated (Y497A/L500A/W647A/ Y649A/L652A; ⌬␤22 sites) did not bind to the GST-␤2 appendage but retained the ability to form a complex with AP-2, although with somewhat reduced efficiency (Fig. 6,  D and E, ⌬␤22 sites). These results indicate that complex formation between AP-2 and PIPKI␥-p90 is, at least in part, driven by determinants outside of the PIPK tail domain, presumably by the association of its catalytic core with AP-2 (17). This conclusion also agrees with the dissociation constants measured for these various sites (22 M for PIPKI␥-p90 tail-␤2 ear; 6 M for PIPKI␥-p90 tail-C-2 (compare Fig. 4); 0.5 M for PIPKI␥ core-C-2 (17)). Our data therefore argue in favor of a model according to which multiple binding sites contribute to complex formation between PIPKI␥-p90 and the AP-2 complex.
A YXXØ Motif-containing Peptide Derived from the p90 Tail Can Stimulate the PI(4,5)P 2 -synthesizing Activity of AP-2bound PIPKI␥-Multiple interactions of AP-2 with PIPKI␥-p90, in addition to recruiting the kinase to sites of endocytosis, may serve to regulate enzymatic activity. In fact, it has been reported that receptor-derived YXXØ motif-containing peptides potently stimulate the PI(4,5)P 2 -synthesizing activity of PIPKI␥ bound to AP-2 (17). The presence of YXXØ motifs within the PIPKI␥-p90 tail domain could thus serve as an intramolecular switch to regulate PI(4,5)P 2 formation by PIPKI␥-p90⅐AP-2 complexes. To test this possibility we prepared lysates from stably transfected HEK293 flip-in cells expressing HA-PIPKI␥-p87 (i.e. a variant of PIPKI␥ lacking the AP-2/␤-binding splice insert) and analyzed lipid kinase activity. PI(4,5)P 2 synthesis in untreated lysates was comparably weak, and this did not change upon adding His 6 -C-2 (Fig. 7A). The addition of the PIPKI␥-p90-derived YXXØ motif-containing tail peptide together with His 6 -C-2 stimulated the formation of radiolabeled PI(4,5)P 2 , similar to that seen with an AP-2-binding sorting signal peptide (FYRALM, with key residues being highlighted) derived from the epidermal growth factor (EGF) receptor. A mutated nonfunctional peptide lacking the ability to bind to C-2 was inactive (Fig. 7B). Peptides were without effect in the absence of C-2 (data not shown; see Ref. 17). Similar results were obtained when the kinase activity of HA-PIPKI␥-p87 was determined following affinity purification on a GST-C-2-based matrix in the presence or absence of peptides (supplemental Fig. 1). These results are consistent with a model according to which an AP-2⅐PIPKI␥-p90 complex provides a local pool of PI(4,5)P 2 at endocytic sites, i.e. during SV recycling.

Inhibition of SV Endocytosis by Expression of the PIPKI␥-p90 Tail Domain Depends on Its Ability to Associate with AP-2-
PIPKI␥ plays an important role in the exo-/endocytic cycling of SVs in primary neurons (20). Loss of PIPKI␥ expression in mice severely impairs clathrin-mediated endocytosis of SVs, resulting in impaired retrieval of SV proteins from the presynaptic plasmalemma. A similar phenotype is observed upon expression of the PIPKI␥-p90 tail domain in hippocampal neurons in culture (21). These latter findings enabled us to functionally analyze the contribution of the various AP-2 binding sites within the PIPKI␥-p90 tail domain to its ability to impair SV endocytosis in primary neurons. To this aim we made use of synapto-pHluorin, a widely used tracer for SV exo-/endocytosis. Fluorescence of synapto-pHluorin is critically dependent on the pH and can thus be used to quantitatively analyze partitioning of the protein between the plasmalemma and intracellular SV-localized pools. Fluorescence changes were monitored from active synaptic boutons (i.e. displaying appropriate responses to electrical stimulation; compare Fig. 8A) expressing synapto-pHluorin together with mRFP (as a control) or mRFPtagged variants of the PIPKI␥-p90 tail. Fluorescence quantification after acid quenching and ammonium dequenching then revealed the relative ratio of vesicular-to-surface-stranded pHluorin molecules (Fig. 8B). An increased fraction of surfacestranded synapto-pHluorin is indicative of impaired SV endocytosis at active boutons. In agreement with previous results using FM4-64 to monitor SV endocytosis (21), we observed that overexpression of a fusion protein comprising the C-terminal tail of PIPKI␥-p90 tagged with mRFP significantly increased the surface-stranded pool of synapto-pHluorin (Fig. 8C) by more than 50%. By contrast, overexpression of the same construct had no significant effect on transferrin internalization (supplemental Fig. 2), in agreement with the specific role of PIPKI␥-p90 in the regulation of SV exo-/endocytosis. Mutational inactivation (F640A) of the interaction site for the ␤2 appendage domain within the PIPKI␥-p90 tail only modestly affected its ability to impair SV endocytosis (Fig. 8C, ⌬␤2 site). A PIPKI␥-p90 tail mutant unable to interact either with the ␤2 appendage or with AP-2 had nearly completely lost its dominant-negative effect on synapto-pHluorin retrieval from the cell surface (Fig. 8C, ⌬␤2⌬2 sites). Whether the remaining small degree of inhibition of SV endocytosis seen upon expression of this mutant construct resulted from sequestration of other yet unidentified binding partners of the PIPKI␥ tail or reflected nonspecific effects remains to be determined. In any case, these data corroborate our structural and biochemical analysis and further support an important physiological role for complex formation between AP-2 and PIPKI␥ in SV endocytosis, a function that is, at least in part, regulated by AP-2-binding determinants within the kinase tail domain.

DISCUSSION
In the present study we have provided a firm structural basis for the association of the PIPKI␥-p90 tail domain with AP-2 and the ␤2 appendage domain, respectively. We show that these interactions with different subunits and domains of AP-2 partly involve overlapping determinants within the 28-amino acid splice insert of the PIPKI␥-p90 tail. Both complexes largely depend on hydrophobic/aromatic amino acid-in-groove interactions. Specifically, the PIPKI␥-p90 tail peptide is seen to wrap around the side site of the sandwich subdomain of the ␤2 appendage in a position similar but not identical to that occupied by a peptide derived from the accessory protein Eps15 (23). As this site is a minor protein binding interface within AP-2, this mechanism might allow for privileged access of PIPKI␥-p90 to the endocytic machinery. The same peptide can also accommodate the sorting signal binding domain of AP-2 via a three-pin-plug interaction involving Tyr-649, Leu-652, and Trp-647, the latter a residue known to be required for the association of PIPKI␥-p90 with the actin-associated scaffolding protein talin at sites of cell adhesion or synaptic contacts (9, 10). The use of partly overlapping binding sites for AP-2, the ␤2 FIGURE 9. Hypothetical model illustrating molecular determinants and possible mechanisms of association regulating complex formation between AP-2 and PIPKI␥-p90. A, PIPKI␥-p90 may initially be recruited to or stabilized at endocytic sites by association of its tail with the ␤2 appendage domain of AP-2. Binding is comparably weak (K D ϭ 22 M) and will require stabilization or modulation by further contacts. B, for example, YXXØ motif-containing cargo recognized by AP-2 at PI(4,5)P 2 -containing plasmalemmal hot spots might aid the formation of a heteromeric complex involving binding of the PIPKI core to AP-2, an association that facilitates the PI(4,5)P 2 -synthesizing activity of PIPKI␥-p90. C, PIPKI␥-p90 may also contribute to AP-2-dependent sorting of SV cargo lacking conventional YXXØ motifs at presynaptic sites. This might involve the association of YXXØ-based motifs within the PIPKI␥-p90 tail domain with AP-2, thereby facilitating local PI(4,5)P 2 production. appendage, and talin suggests that these various interactions may be subject to regulation during exo-/endocytic cycling of SVs. In agreement with this proposal, PIPKI␥-p90 has been shown to undergo activity-dependent Cdk5-mediated phosphorylation at Ser-650 (corresponding to Ser-645 of mouse PIPKI␥-p90) during SV exo-/endocytosis (33), a modification that has been shown to impair its association with talin (33) and with the ␤2 appendage domain of AP-2 (21). Our crystallographic data (Table 2) reveal the molecular basis for this regulation. Ser-650-OH participates in hydrogen bonding with the peptide backbone of Phe-753 and Leu-770, interactions that are likely to be compromised by phosphorylation of Ser-650. Phosphorylation of Tyr-649 by Src or other tyrosine kinases has been shown to facilitate association of PIPKI␥-p90 with talin (34). Our structural data presented here would argue that the same modification should inhibit binding of the PIPKI␥-p90 tail to AP-2 or the ␤2 appendage, suggesting the existence of a phosphorylation-dependent cycling of PIPKI␥-p90 between sites of cell adhesion and endocytosis. Mutagenesis paired with biochemical binding experiments also suggests that complex formation between AP-2 and PIPKI␥-p90 is dependent in part on determinants outside of the p90 tail domain, presumably involving the direct association of AP-2 with the central kinase core (17).
The clathrin-based endocytic machinery together with the p90 splice isoform of PIPKI␥ is concentrated within the presynaptic compartment where SVs undergo activity-dependent rapid cycles of exo-/endocytosis (4,32). It is therefore tempting to speculate that the presence of additional binding sites within the 28-amino acid splice insert of PIPKI␥-p90 for AP-2␤/ serves to concentrate the enzyme at endocytic hot spots near the active zone. According to a hypothetical model (Fig. 9), one might envision that the initial recruitment of PIPKI␥-p90 to sites of SV endocytosis involves binding of the YFPTDER-SWVYSPLH (with key residues being highlighted) sequence within the p90 tail to the AP-2 ␤ appendage (Fig. 9A), which acts as an autonomous platform for the enrichment of endocytic accessory factors (35). Complex formation might be aided by the association of the PIPKI␥ core with AP-2 and/or by phosphorylation of AP-2. Phosphorylation of 2 at Thr-156 facilitates a conformational opening of the AP-2 complex and the dislocation of the C-terminal sorting signal binding domain of 2 from the remainder of the AP-2 core (16). This conformational change would enable C-2 to bind to YXXØ-based sorting signals of transmembrane cargo proteins (Fig. 9B) or to the SWVYSPL and RSYPTLED peptide motifs within the PIPKI␥-p90 tail (Fig. 9C). Occupation of the YXXØ motif binding site on AP-2 likely results in a potent stimulation of PIPKI␥-mediated PI(4,5)P 2 synthesis (compare Fig. 7). Recent in vitro experiments suggest that clathrin assembly might facilitate displacement of the PIPKI␥-p90 tail from the ␤2 appendage, thereby potentially providing directionality to the process of complex assembly and disassembly (36). Clearly, further experiments are needed to put such a hypothetical scenario to the test.
One unresolved question concerning clathrin-mediated SV endocytosis pertains to the mechanism of SV cargo recognition by adaptor proteins including AP-2, endophilin, and stonin 2 (37). It has long been known that SV proteins and conventional cargo such as the transferrin receptor segregate during their endocytic itinerary in neurons or neuroendocrine cells (38). Moreover, accumulating evidence suggests that SV proteins are sorted into clathrin-coated vesicles by cargo-specific, partly YXXØ motif-independent mechanisms (39). These include the recognition of VGLUT1 by the endocytic SH3 domain protein, endophilin (40), or the internalization of synaptotagmin 1 by its dedicated sorting adaptor, stonin 2, via sorting determinants contained in its C2 domains (41). Association of AP-2 with YXXØ peptide motifs of presynaptically enriched PIPKI␥-p90 might thus provide a potential means to 1) exclude conventional YXXØ motif-containing cargo such as growth factor or nutrient receptors from presynaptic clathrin-coated pits formed during SV retrieval and 2) activate PIPKI␥-p90 to generate a pool of PI(4,5)P 2 that allows for efficient AP-2/clathrin coat formation during SV endocytosis. Further studies capitalizing on the molecular insights provided here will need to address this possibility.