Crystal Structure of the Human Fe65-PTB1 Domain*

The neuronal adaptor protein Fe65 is involved in brain development, Alzheimer disease amyloid precursor protein (APP) signaling, and proteolytic processing of APP. It contains three protein-protein interaction domains, one WW domain, and a unique tandem array of phosphotyrosine-binding (PTB) domains. The N-terminal PTB domain (Fe65-PTB1) was shown to interact with a variety of proteins, including the low density lipoprotein receptor-related protein (LRP-1), the ApoEr2 receptor, and the histone acetyltransferase Tip60. We have determined the crystal structures of human Fe65-PTB1 in its apo- and in a phosphate-bound form at 2.2 and 2.7Å resolution, respectively. The overall fold shows a PTB-typical pleckstrin homology domain superfold. Although Fe65-PTB1 has been classified on an evolutionary basis as a Dab-like PTB domain, it contains attributes of other PTB domain subfamilies. The phosphotyrosine-binding pocket resembles IRS-like PTB domains, and the bound phosphate occupies the binding site of the phosphotyrosine (Tyr(P)) within the canonical NPXpY recognition motif. In addition Fe65-PTB1 contains a loop insertion between helix α2 and strand β2(α2/β2 loop) similar to members of the Shc-like PTB domain subfamily. The structural comparison with the Dab1-PTB domain reveals a putative phospholipid-binding site opposite the peptide binding pocket. We suggest Fe65-PTB1 to interact with its target proteins involved in translocation and signaling of APP in a phosphorylation-dependent manner.

Fe65 is an adaptor protein predominantly expressed in the brain, where it plays a critical role in neuronal development and APP 2 signaling (1,2). Fe65 exhibits three protein-protein interaction domains with different binding specificities as follows: one N-terminal WW domain and two contiguous C-terminal phosphotyrosine-binding domains (PTB1 and PTB2). The Fe65-WW domain recognizes polyproline sequences in several proteins, including c-Abl tyrosine kinase (3) and the mammalian homolog of the Drosophila actin cytoskeleton regulatory protein enabled (Mena) (4). The Fe65-PTB1 domain has been shown to interact with the low density lipoprotein receptorrelated protein (LRP-1) (5), the ApoEr2 receptor (6), the histone acetyltransferase Tip60 (7), and the transcription factor CP2/LSF/LBP1 (8), whereas Fe65-PTB2 interacts with the APP intracellular domain (9). The Fe65-mediated multiprotein complexes are involved in diverse processes, including learning and memory (10), regulation of neuronal growth cone motility (11,12), APP translocation and processing (13)(14)(15), and transcriptional regulation of gene expression (7, 16 -20). Fe65-PTB1 is essential for many of these functions. In particular, the described interactions with Tip60 and CP2/LSF/LBP1 are important for the regulation of gene expression (7,8), whereas the interaction with the C-terminal domain of LRP-1 has been shown to play a critical role for APP localization and processing (5) and therefore for Alzheimer disease pathogenesis.
PTB domains are protein-protein interaction modules recognizing short peptide regions, including the canonical NPXpY consensus sequence (sometimes reduced to NXXpY; X indicates relaxed residue selectivity) (21). PTB domains differ in their specificities for phosphorylated or unphosphorylated tyrosines, and most of the analyzed PTB domains bind their substrates in a Tyr(P)-independent manner. Based on evolutionary and structural comparisons, the PTB domains have been classified in three different subfamilies as follows: the Tyr(P)-dependent IRS-like and Shc-like PTB domains, and the Tyr(P)-independent Dab-like PTB domains (22). All PTB domains include the pleckstrin homology (PH) domain superfold consisting of seven antiparallel ␤-strands forming two orthogonal ␤-sheets (␤-sandwich) capped by a C-terminal ␣-helix. The peptide binding pocket is mainly formed by the ␤5-strand and the C-terminal ␣-helix. In addition, many PTB domains contain a basic phospholipid binding "crown" located on the opposite site of the peptide binding pocket (22). Shc-and Dab-like PTB domains contain two additional ␣-helices, one located N-terminally and one inserted between the strands ␤1 and ␤2. Shc-like PTB domains contain also a large loop insertion between helix ␣2 and strand ␤2 (␣2/␤2 loop).
Here we report the first structure of a Fe65-PTB domain, Fe65-PTB1, in its apo-and in a phosphate-bound form at 2.2 and 2.7 Å, respectively. Although Fe65-PTB1 has been clas-sified as a Dab-like PTB-domain, its structure reveals characteristics of different PTB domain subfamilies. The data presented here set the stage for an atomic understanding of a molecular crossroad of APP metabolism and Alzheimer disease pathogenesis.

EXPERIMENTAL PROCEDURES
Structure Determination-Expression, purification, and crystallization of Fe65-PTB1 (residues Gly-366 to Glu-505) as well as X-ray data collection were performed as described (23). The structure of Fe65-PTB1 was determined by single wavelength anomalous diffraction collected on a mercury derivative crystal (P2 1 2 1 2 1 space group) (23). Phasing was performed with the program suite PHENIX (Python-based Hierarchical Environment for Integrated Xtallography) (24,25). After density modification, phases were of sufficient quality for automatic model building. The PHENIX-AUTOBUILD routine was able to build 230 residues (and the C-␣ traces of 22 additional amino acids) of the two molecules (A and B) in the asymmetric unit. The model was completed using iterative cycles of model building in COOT (26) and refinement with REFMAC5 (27,28). Refinement statistics are given in Table 1. The refined model was used as a search model to solve the native structure (H3 space group) (23) by molecular replacement using program PHASER (29). Model building and refinement were performed as for the derivative structure (Table 1). Fe65-PTB1 complexed with phosphate was obtained by soaking native crystals (H3 space group) for 10 min in the crystallization buffer containing 50 mM Na 2 HPO 4 . Data collection statistics are presented in supplemental Table S1. The phosphate-bound structure was subsequently solved by difference Fourier technique using the native structure and refined as the other models ( Table 1). The quality of all models was validated with programs PROCHECK (30) and WHAT-IF (31). The secondary structure was analyzed with program DSSP (32). Surface potentials were calculated with program GRASP (33), and figures were generated with programs MOLSCRIPT (34), Raster3D (35), and PyMOL.
Isothermal Titration Calorimetry-All isothermal titration calorimetry (ITC) experiments were carried out in a buffer containing 10 mM HEPES, pH 7.5, and 150 mM NaCl. Fe65-PTB1 and the respective peptides were extensively dialyzed against the same buffer. Solutions were degassed immediately before measuring, and all binding experiments were performed using a VP-ITC microcalorimeter (MicroCal, Northampton, MA) equilibrated at the desired temperature (15 or 25°C). For all measurements the cell was filled with 20 M peptide or inositol 1,4,5-triphosphate (IP 3 ), and 200 M of purified Fe65-PTB1 was used as the titrant in the syringe. A typical titration consisted of injections of 12-l aliquots of the titrant into the solution in the cell under constant stirring at 300 rpm at time intervals of 5 min to ensure that the titration peak returned to the base line. For data analysis the Origin 7.0 software was used.
Immobilized Peptide Libraries-Immobilized peptide libraries representing residues Met-229 to Trp-513 of the histone acetyltransferase domain of human Tip60 (10-and 15-mer peptides both tyrosine-phosphorylated and unphosphorylated with either one or three residue shifts) were synthesized as described previously (36,37). After 2 h blocking with 4% (w/v) skim milk (Sigma) in the incubation buffer (20 mM K 2 HPO 4 , pH 6.8, 100 mM KCl, 50 mM NaCl, 2 mM MgCl 2 , 1 mM CaCl 2 , and 0.05% (v/v) Tween 20), the membranes were incubated with 0.25 M purified Fe65-PTB1 in the incubation buffer for 1 h at 20°C. Bound Fe65-PTB1 was detected with an anti-His antibody (Qiagen) and visualized using an enhanced chemiluminescence (ECL) anti-mouse IgG secondary antibody (GE Healthcare). Incubations with the antibodies were performed in the incubation buffer supplemented with 0.5% (w/v) skim milk. All incubation steps were followed by three consecutive wash steps of 10 min in the incubation buffer.

RESULTS AND DISCUSSION
Structure Determination of Fe65-PTB1-The N-terminal PTB domain of human Fe65 (Fe65-PTB1, residues Gly-366 to Glu-505) crystallized in the two space groups H3 (native crystals; four molecules per asymmetric unit) and P2 1 2 1 2 1 (derivative crystals; two molecules per asymmetric unit). The orthorhombic space group was metal-induced as the result of co-crystallization experiments with methyl mercury chloride as described previously (23). The crystal structure was determined at 2.2 Å resolution by single wavelength anomalous diffraction using a mercury derivative crystal. The native structure (H3 space group) was determined by molecular replacement at 2.8 Å resolution. Refinement statistics are given in Table 1.
In the P2 1 2 1 2 1 space group, the two molecules of the asymmetric unit are well ordered except two residues (Gly-460 and Arg-461) in a loop region. The root mean square deviation (r.m.s.d.) between the P2 1 2 1 2 1 and the H3 structures has an average of 0.74 Å for all C-␣ atoms. The main differences occur in a loop region comprising residues Gln-400 to Lys-420 (see below). In the H3 space group this region is mostly not visible in the electron density.
Overall Structure of Fe65-PTB1-The structure of the Fe65-PTB1 domain reveals the folding pattern of the pleckstrin homology (PH) domain superfold (Fig. 1A). The core of this fold consists of seven antiparallel ␤-strands (␤1-␤7) arranged in two orthogonal ␤-sheets. This ␤-sandwich is capped by a C-terminal ␣-helix (␣3). Additional structural features of the PH domain superfold are also present in Fe65-PTB1. These include the peptide binding pocket formed by strand ␤5, helix ␣3, and the ␤1/␣2 loop, and a putative phospholipid-binding site next to the ␣2/␤2 loop (see below).
On an evolutionary basis, Fe65-PTB1 has been grouped into the Dab-like PTB domain subfamily (22). Consistent with this classification, an analysis of the Fe65-PTB1 structure using the Dali server (38) identified Dab-like PTB domains as the closest structural relatives (for details see supplemental Table S2). However, sequence identities are relatively low (below 25%), and the r.m.s.d. values of about 2.0 Å indicate significant structural differences between Fe65-PTB1 and Dab-like PTB domains.
The ␣1-helix located at the N terminus of Shc-and Dab-like PTB domains ( Fig. 1B and Fig. 2) is not present in the Fe65-PTB1 structure and is not predicted on the sequence level. Like most PTB domains, Fe65-PTB1 includes one additional helix (␣2) located between the ␤1and ␤2-strands ( Fig. 1 and Fig. 2). Fe65-PTB1 reveals a unique ␤1/␣2 loop (residues Met-380 to Arg-389) with an inserted 3 10 helix (3 10 A, residues Glu-382 to Leu-385) as part of the phosphotyrosine-binding pocket. The ␣2/␤2 loop is extended (residues Gln-400 to Lys-420), which is a characteristic feature of the Shc subfamily of PTB domains where this loop has accordingly been termed as Shc loop (22). However, this extended loop is also present in the Dab-like X11-PTB domain (39). In the P2 1 2 1 2 1 space group this loop is involved in crystal contacts and is part of the model. However, in the structures derived from the H3 space group, the ␣2/␤2 loop is not present indicating a high conformational flexibility.
Phosphate Binding to Fe65-PTB1 Suggests Phosphotyrosinedependent Peptide Interactions-The overall mode of peptide binding is conserved in all PTB domain structures determined so far (Fig. 2). PTB-binding peptides fold in trans as an antiparallel ␤-strand on top of strand ␤5 and are aligned parallel to the C-terminal helix ␣3. The consensus sequence fingerprint of PTB-binding peptides consists of the NPXpY motif. NPXpY motifs are typical internalization signals for membrane-associated receptor proteins (40 -42). This sequence has a high propensity of forming a ␤-turn structure, which in all PTB domains is specifically read out by an "anchoring" binding pocket harboring most of the peptide binding energy especially in the Tyr(P)-dependent interactions (43).
The three PTB domain subfamilies differ significantly in (phospho)-tyrosine recognition (Fig. 2). In Dab-like PTB domains, the anchoring pocket has only a minor role in peptide recognition, and tyrosine phosphorylation often decreases the binding affinity (39,44,45). On the contrary, IRS-like and Shclike PTB domains require phosphorylated tyrosines for high affinity binding (46 -48). The Shc-like PTB domains use three basic residues (two arginines and one lysine) to accommodate the Tyr(P) residue (46). The IRS-1 structure reveals two other arginine residues at different locations, which create a basic pocket for the negatively charged phosphate moiety (48). Fe65-PTB1 contains two equivalent arginine residues (Arg-451 and Arg-470) at the same position as IRS-1. Arg-451 is involved in a tight salt bridge with Glu-382, which is located in the unique 3 10 helix (3 10 A) of the ␤1/␣2 loop (Fig. 3A).
To test for a phosphorylation dependence of peptide binding to Fe65-PTB1, we soaked native Fe65-PTB1 crystals with phosphate and determined the structure at 2.7 Å resolution (see supplemental Table S1 for data collection statistics and Table 1 for refinement statistics). The binding of phosphate into the pocket could be confirmed by prominent difference densities in the unbiased difference density maps (Fig. 2). The occupancy for the phosphate is 75%, and together with the phosphate concentration of 50 mM used in the soaking experiments a K d value in the range of 20 mM can be estimated. Phosphate binding in the Fe65-PTB1 structure is exclusively performed by the two arginine residues Arg-451 and Arg-470 (Fig. 3A). Although the terminal nitrogen N-⑀2 of Arg-451 forms a single hydrogen bond to phosphate oxygen, Arg-470 orients the phosphate FIGURE 2. Comparison of Fe65-PTB1 with representative structures of the three PTB subfamilies bound to peptides. Top row, electrostatic surface potentials. Positive and negative patches on the protein surfaces are given in blue and red, respectively. Bound ligands are given as stick models. All proteins have the same orientation. Bottom row, structure models of the respective PTB domains in a color ramping from blue (N terminus) to red (C terminus). The unbiased F o Ϫ F c difference electron density for the phosphate in the Fe65-PTB1 structure is given at a 2 level. by two hydrogen bonds with its guanidinium group. Although the charge of Arg-451 is partially neutralized by the interaction with Glu-382, it will contribute to phosphate binding due to the nondirected character of Coulomb interactions. In addition to phosphate binding, the conserved Arg-451 is perfectly located to stabilize a putative phosphotyrosine binding by a -cation stacking interaction of the aromatic ring with the planar guanidinium group. When phosphate-bound Fe65-PTB1 is superposed on the IRS-1 PTB domain complexed with a phosphorylated interleukin-4 receptor peptide (48), the phosphate groups are found to bind in the same position and orientation (Fig. 3B). Therefore, we assume the bound phosphate group in Fe65-PTB1 to mimic the phosphate moiety of a phosphorylated peptide indicating a phosphorylation dependence of peptide recognition. The binding mode of the two arginines to the phosphate moiety of Tyr(P) is, however, slightly different. In IRS-1 the arginine residue (Arg-212) equivalent to Arg-451 in Fe65-PTB1 is not stabilized by a salt bridge, and the whole binding pocket is more open compared with Fe65-PTB1. Interestingly, the corresponding arginine to Arg-470 (Arg-227) is not involved in direct hydrogen bonding with the phosphate in the IRS-1 structure. The tighter binding of the phosphate moiety suggests at least an equal preference of Fe65-PTB1 for phosphorylated substrates.
Similarly, when the Dab1 structure in complex with an unphosphorylated APP peptide (49) is overlaid on Fe65-PTB1, the hydroxyl of the bound APP tyrosine matches with the phosphate oxygen that is hydrogen-bonded to Arg-451 (Fig. 3C). In Dab1, however, the binding pocket lacks the two arginine residues, and the binding of a nonphosphorylated tyrosine is favored (49,50). The ␤1/␣2 loop in Dab1 also contributes to the binding pocket but via an alanine residue. The Fe65-PTB1 residue Glu-382 located on helix 3 10 A is not conserved in Dab1.
In summary, peptide binding to Fe65-PTB1 is likely to be Tyr(P)-dependent as deduced from the conservation of the binding pocket and the binding mode of the bound phosphate moiety. The conservation of the binding pocket reveals a relationship in peptide binding to IRS-like PTB domains.
Interaction with Putative Binding Partners-The histone acetyltransferase Tip60 was identified as a Fe65-PTB1 binding partner in a study performed by Cao and Sudhof (7) using a gene reporter assay and pulldown experiments with an N-terminal truncated Tip60␤ construct. This construct does not include the typical NPXY motif but two NXXY motifs within the histone acetyltransferase domain. The mutation of the 257 NKSY motif into 257 NASA was shown to abolish the interaction between Fe65-PTB1 and Tip60 suggesting this motif to be a putative binding site. However, structural modeling of Fe65-PTB1 onto the recently solved Tip60HAT structure (PDB code 2ou2), assuming the conserved PTB/NPXY binding mode, is not possible because of the different structural environment of the 257 NKSY motif. This motif does not form the characteristic ␤-turn structure but is completely embedded into an ␣-helix with the tyrosine being buried in the interior of the protein. Therefore, to investigate further the interaction and to discriminate between the two NXXY-binding motifs, we performed ITC experiments at different temperatures using synthesized 11-mer peptides, including either the 242 NEIY or the 257 NKSY sequence (phosphorylated and unphosphorylated; Table 2). However, none of the tested peptides revealed a significant binding to Fe65-PTB1 indicating a K D Ͼ300 M (data not shown). To test if binding of Fe65-PTB1 occurs at another site, we sampled the complete Tip60HAT sequence in an immobilized peptide library approach using a library of overlapping 10or 15-mer peptides (with phosphorylated and unphosphorylated tyrosines). Although we found some positive hits for surface patches (including both NXXY motifs) predominantly located opposite of the acetyl-CoA binding cleft, both NXXY motifs showed up in the phosphorylated and nonphosphorylated state, which neither allowed for a discrimination of the two sites nor for a clarification of the phosphorylation dependence (data not shown).
In addition to Tip60, we also tested the proposed interactions of FE65-PTB1 with the intracellular domains of the lipoprotein receptors LRP-1 and ApoEr2 (5,6). Again, we used synthesized peptides, including the NPXY motifs ( 4470 NPTY, 4503 NPVY for LRP-1, and 863 NPVY for ApoEr2) in both the phosphorylated and unphosphorylated forms ( Table 2) in ITC experiments. To our surprise and like for Tip60, none of the peptides showed binding to Fe65-PTB1 (data not shown).
In summary, the proposed interactions with Fe65-PTB1 could not be validated. As Fe65-PTB1 is stable in solution and experimental parameters like concentrations and temperature in the ITC experiments were carefully checked and modified, we assume either a low affinity binding or the requirement for a specific cellular environment or more complete protein fragments of the interaction partners.
Membrane Attachment-In addition to direct protein-protein interactions, many PTB domains were shown to be involved in membrane attachment (for review see Ref. 22). Direct membrane recruitment of PTB domains allows for a reduction of dimensionality, local concentration within specific membrane compartments like lipid rafts, and pre-orients the domain for favorable interactions with their membrane-bound peptide substrates. Similar to the pleckstrin homology (PH) domains, binding occurs to headgroups of phosphoinositides. Phosphoinositides show a unique subcellular distribution dependent on the degree and position of headgroup phosphorylation (51). They play a fundamental role in the spatial and temporal control of many signaling processes at membranecytosol interfaces. Of special interest is phosphatidylinositol 4,5-biphosphate, which has an important function in endocytosis events at the plasma membrane (49,52). The structural basis for phosphatidylinositol 4,5-biphosphate binding to PTB domains has been revealed by crystal structures of the ternary complexes of Dab1-PTB with the phospholipid headgroup IP 3 and either an ApoEr2 (50) or APP peptide (49). In the Dab1 protein the PTB domain is essential for membrane recruitment and reveals high affinity binding for IP 3 (K D ϳ 0.7 M) (53) and a highly positively charged surface patch opposite to the peptide binding pocket and oriented toward the membrane (52). The pocket is located at the C-terminal end of helix ␣2 and involves the adjacent ␣2/␤2 loop as well as a set of basic residues from or next to the rigid ␤-sheet core (␤1, ␤6, and ␤7) (Fig. 4A). It has been assigned by threedimensional electrostatic analysis to form a basic crown (22) (Fig. 4B). Strikingly, the superposition of the ternary complexes with the Fe65-PTB1 structure reveals a conservation of onehalf of the basic pocket, including three basic residues originating from the ␤-sheet core (Arg-373, Arg-463, and Lys-483 in Fe65-PTB1) (Fig. 4A). In Dab1, these residues are responsible for recognition of the phosphates at positions four and five of IP 3 . Despite the conservation of positively charged residues, however, a pronounced phospholipidbinding crown is not present within Fe65-PTB1 (Fig. 4B). As in Dab1-PTB, helix ␣2 is elongated by two turns and the ␣2/␤2 loop is much shorter, and the second half of the binding pocket recognizing the phosphates at positions one and five cannot be correlated on the sequence level (Fig. 1B). Although Fe65-PTB1 contains several basic residues in this region (Lys-405, His-409, and Lys-420), their spatial orientation in the Fe65-PTB1 structure is different as compared with the Dab1-PTB complexes. However, the ␣2/␤2 loop of Fe65-PTB1 is flexible, as deduced by comparison of the structures from the two different space groups, and therefore it might adopt a conformation favorable for phospholipid binding.
We investigated the putative binding of Fe65-PTB1 to phosphoinositides by immobilized lipid-protein binding assays (PIP-strips), ITC, and co-crystallization experiments. In the PIP-strip assays, we could not detect binding to any of the tested lipids or phospholipids, including different mono-, di-, and tri-phosphorylated phosphoinositides. Similarly, ITC and co-crystallization experiments with IP 3 under various conditions did not reveal binding to Fe65-PTB1 (data not shown). Taken together, although Fe65-PTB1 reveals distinct features for phospholipid binding, we could not detect a direct interaction. However, the structural and functional homology between Fe65-PTB1 and the Dab1-PTB domain as well as previous studies on other PTB domains (for review see Ref. 22) make an interaction under physiological conditions very likely.
Conclusions-The Fe65-PTB1 domain shares the pleckstrin homology domain superfold and is structurally closely related to the Dab-like PTB domain subfamily. The comparison of Fe65-PTB1 with Dab1 reveals distinct features of a phospholipid-binding site, although the interaction could not be validated experimentally. In Dab1, the interaction is strong and essential for membrane recruitment. However, Fe65 is mainly recruited to the membrane by its interaction with the APP intracellular domain. In the close proximity to the membrane even a weak phospholipid binding of Fe65-PTB1 might therefore be sufficient to attach and to orient the domain at the membrane surface. Color scheme is as in Fig. 2. C, Fe65-PTB1 oriented in the same way with superposed IP 3 taken from the Dab1-PTB structure.
The structure of Fe65-PTB1 reveals as well unique and Dabunlike characteristics. Most strikingly is the Tyr(P) binding pocket that resembles IRS-like PTB domains. Phosphate binding to Fe65-PTB1 indicates a Tyr(P) dependence of substrate interactions. Fe65-PTB1 would therefore be the first Tyr(P)-dependent and Dab-like PTB domain. A phosphorylation dependence of Fe65-PTB1 interactions has a significant impact on our current understanding of APP translocation and signaling events. Noteworthy, the intracellular domain of APP binds to Fe65-PTB2 in a Tyr(P)-independent manner (9). Fe65-PTB1 is the cross-point that links APP to the lipoprotein receptors LRP-1 and ApoEr2 and to the transcription regulators Tip60 and CP2/LSF/LBP1, although we were not able to validate interactions in our experiments and further studies are needed to define the exact molecular mechanisms. However, it seems that yet to be identified kinases are important for Fe65-mediated multiprotein complex formation.
Finally, the various Fe65-mediated interactions are well known to play an essential role for APP processing and amyloid-␤ generation (54). The accumulation of the amyloid-␤ peptide is a central event in the pathogenesis of Alzheimer disease, and the peptide is the main constituent of the senile plaques. Elucidating the molecular details of the pathways leading to amyloid-␤ formation is therefore both promising and necessary for a successful treatment of the devastating disease.