Membrane Curvature Protein Exhibits Interdomain Flexibility and Binds a Small GTPase*

Background: APPL2 is an endosomal Rab effector forming part of a signaling pathway linking cell surface and nucleus. Results: Crystal and solution structures of APPL2 were solved, and Rab partners were identified. Conclusion: APPL2 interacts tightly with Rab31, and APPL2 structures reveal unexpected domain motion that could have functional implications. Significance: APPL2 dynamics and interactions may be crucial for its cell signaling role. The APPL1 and APPL2 proteins (APPL (adaptor protein, phosphotyrosine interaction, pleckstrin homology (PH) domain, and leucine zipper-containing protein)) are localized to their own endosomal subcompartment and interact with a wide range of proteins and small molecules at the cell surface and in the nucleus. They play important roles in signal transduction through their ability to act as Rab effectors. (Rabs are a family of Ras GTPases involved in membrane trafficking.) Both APPL1 and APPL2 comprise an N-terminal membrane-curving BAR (Bin-amphiphysin-Rvs) domain linked to a PH domain and a C-terminal phosphotyrosine-binding domain. The structure and interactions of APPL1 are well characterized, but little is known about APPL2. Here, we report the crystal structure and low resolution solution structure of the BARPH domains of APPL2. We identify a previously undetected hinge site for rotation between the two domains and speculate that this motion may regulate APPL2 functions. We also identified Rab binding partners of APPL2 and show that these differ from those of APPL1, suggesting that APPL-Rab interaction partners have co-evolved over time. Isothermal titration calorimetry data reveal the interaction between APPL2 and Rab31 has a Kd of 140 nm. Together with other biophysical data, we conclude the stoichiometry of the complex is 2:2.

an APPL-Rab complex, though a model of the APPL1-Rab5 complex has been proposed (12).
Here, we describe the first structural information on APPL2, specifically, the BARPH domains of human APPL2 (hAPPL2 BARPH ). We report both the crystal structure and SAXS solution structure of APPL2, showing unexpectedly that there is flexibility in the arrangement of the BAR and PH domains. In addition, we identify by yeast two-hybrid screen that Rab31 is a binding partner for APPL2, and we establish the thermodynamics and stoichiometry of the interaction between hAPPL2 BARPH and Rab31.

EXPERIMENTAL PROCEDURES
Cloning-For human APPL2 (Open Biosystems, accession no. BC033731; GenBank accession no. 55198) residues 2-384 were subcloned into the LIC vector pMCSG7 (15) encoding an N-terminal polyhistidine tag with a tobacco etch virus cleavage site. For Rab31, the same subcloning strategy was adopted using a codon-optimized construct consisting of residues 1-167 of hRab31 (GeneArt; GenBank accession no. U59877) subcloned into the LIC vector.
Protein Production-hAPPL2 BARPH and Rab 31 were expressed in 500-ml cultures of Escherichia coli BL21(DE3)pLysS using auto induction (16). Cultures were grown at 30°C and typically harvested after 17 h at an A 600 of ϳ8.0. Pellets were harvested by centrifugation, snap frozen in liquid nitrogen and stored at Ϫ80°C.
The pellet was thawed in buffer A (25 mM Tris, pH 7.5, 150 mM NaCl, 1 mM-mercaptoethanol), with 0.5% Triton X-100, 0.30 units/ml DNase (Roche Applied Science), and protease inhibitor mixture (100 M PMSF, 2.0 M bestatin, 0.3 M pepstatin A, 0.3 M E-64, 0.08 M aprotinin, and 1.0 M leupepti-Astral) at 4°C. The thawed pellet was sonicated on ice, and the cell debris was removed by centrifugation. The supernatant was then mixed with TALON TM resin (Clontech) pre-equilibrated in buffer A. This suspension was mixed with rotation for 1 h at 4°C and left overnight on ice, before transferring to a gravity column. The bound lysate was washed with three bed volumes of buffer A, followed by three bed volumes of buffer A containing 10 mM imidazole. Protein was then eluted in three bed volumes of buffer A containing 300 mM imidazole. The eluant was concentrated and exchanged into buffer B (25 mM Tris, pH 7.5, containing 25 mM NaCl and 1 mM ␤-mercaptoethanol) using a 10-kDa concentrator (Amicon) before loading onto a MonoQ 5/50 GL column equilibrated in buffer B on an ÄKTA FPLC (GE Life Sciences). The column was washed with 10 ml of buffer B, and protein was eluted using a 70-ml 25-500 mM NaCl gradient in buffer B.
Pooled protein fractions were concentrated using a 10-kDa concentrator (Amicon) and applied to a size exclusion chromatography column (HiLoad 16/60 Superdex 200) equilibrated with 25 mM HEPES, pH 7.5, containing 150 mM NaCl and 1 mM DTT on an ÄKTA FPLC (GE Life Sciences). Fractions containing hAPPL2 BARPH were pooled and dialyzed against 25 mM HEPES, pH 7.5, containing 150 mM NaCl , 1 mM DTT and 1 mM EDTA. Purified samples of hAPPL2 BARPH were concentrated using a 10-kDa concentrator (Amicon) and stored in aliquots at Ϫ80°C at an A 280 of 80.0. Rab31 was purified in a similar manner. However, the buffers used after the lysis and throughout the purification were at pH 8.5 and contained 5 mM MgCl 2 . TALON TM eluant was concentrated and exchanged into buffer C (25 mM Tris, pH 8.5, containing 25 mM NaCl, 1 mM ␤-mercaptoethanol, and 5 mM MgCl 2 ) using a 10-kDa concentrator (Amicon) before loading onto a MonoQ 5/50 GL column equilibrated in buffer C on an ÄKTA FPLC (GE Life Sciences). The column was washed with 10 ml of buffer C, and protein was eluted using a 70-ml 25-500 mM NaCl gradient in buffer C. Size exclusion chromatography was performed with a HiLoad 16/60 Superdex 75 column (GE Life Sciences) equilibrated with 25 mM HEPES, pH 8.5, 150 mM NaCl, 1 mM DTT, and 5 mM MgCl 2 . Purified Rab 31 was stored frozen at Ϫ80°C typically at 3-4 mg/ml. MALDI-TOF mass spectrometry estimates of the masses of purified hAPPL2 BARPH and Rab31 were 46.3 and 21.3 kDa, respectively. These are consistent with the theoretical masses of 46.1 and 21.3 kDa, respectively. Chemical cross-linking performed using bis(sulfosuccinimidyl) suberate suggested that hAPPL2 BARPH is predominantly a dimer in solution (data not shown).
Crystal Structure Determination of hAPPL2 BARPH -Crystals were grown at 20°C in a 96-well hanging drop plate using View-Drop II seals (TTP Labtech) where the well conditions were 20% (v/v) glycerol, 40 mM hexammine cobalt (III) chloride, and 2-4% (w/v) PEG 3350, or 1.5-3.0% (w/v) PEG 8000. The hanging drops were prepared using TTP Labtech's Mosquito robot and were composed of 200 nl of 15.0 mg/ml hAPPL BARPH and 200 nl of the well solution. The plates were imaged using a RockImager (Formulatrix). hAPPL2 BARPH crystallized in the P2 1 2 1 2 1 space group with two dimers in the asymmetric unit. The crystals contain 79% solvent and have a Matthews coefficient of 5.9 A 3 /Da (estimated using the CCP4 suite (21)).
Diffraction data were measured from each of two crystals, both harvested within 2 days of setting drops. One crystal grew in 2.55% (w/v) PEG 3350 and had dimensions of ϳ100 ϫ 100 ϫ 100 m 3 . The second crystal grew from 2.9% (w/v) PEG 8000 and had dimensions of ϳ220 ϫ 140 ϫ 120 m 3 . The cryoprotectant used for both crystals contained 5.0% PEG 8000, 20% glycerol, 40 mM hexammine cobalt (III) chloride, and 2.3 M trimethylamine oxide. Data used for phasing were collected on the first crystal and measured on a Rigaku FR-E Superbright X-ray generator with Osmic HiRes 2 optics and a Rigaku R-Axis IVϩϩ image plate detector. Data used for refinement were measured from the second crystal on the MX2 beamline at the Australian Synchrotron (17). Data were processed using HKL2000 (18), and statistics are shown in Table 1.
Phasing was performed on the Rigaku-measured diffraction data set by performing molecular replacement with Phaser MR in the Phenix package (19). The APPL1 BARPH crystal structure (PDB code 2Q13, with waters removed) (12) was used as the search model. Initial attempts to produce a phase solution using molecular replacement were frustrated by the expectation that the asymmetric unit was composed of eight or ten molecules, corresponding to solvent contents of 58.4 and 48.0%, respectively. A partial solution was eventually obtained that included two monomers, which were well placed in the electron density. The two monomers represented one subunit each of two dimers, allowing all four molecules in the asymmetric unit to be built in by comparison with the dimer structure of APPL1. The structure was refined with several rounds of manual rebuilding in COOT (20) and maximum likelihood refinement including TLS refinement in Phenix.refine (19). As recommended by Phenix developers, hydrogens were included, and the riding hydrogen model was used in refinement.
The final structure includes the following: chains A and B, residues 2-378 and 5-378, respectively, except residues 75-78 in loop 1 of chain A were not modeled; chains C and D, residues 2-378 and 5-377, respectively, except residues 76 -78 in loop 1 of chain C were not modeled. No density was present for the polyhistidine tag at the N terminus, though for chains A and C the tobacco etch virus cleavage site (residues Asn-6 to Ala1) was modeled. The quality and geometry of the model were evaluated by Molprobity (21). Statistics for the final refined structure are provided in Table 1. Coordinates and structure factors have been deposited with the Protein Data Bank code 4H8S.
Yeast Two Hybrid Analysis-A Rab library, comprising constitutively active Rabs containing the catalytic glutamine to leucine mutation, was generated as described previously (22). Briefly, the cDNA encoding the Rab was cloned into the pGBKT7 vector (Clontech). hAPPL2 and hAPPL1 were cloned into the pGADT7 vector (Clontech). Plasmids for Rabs and APPLs were co-transformed into chemically competent Saccharomyces cerevisiae strain AH109 (Clontech) and selected for co-transformation by growth on ϪLeu/Trp plates. Co-transformants were spotted in serial dilution on ϪLeu/Trp plates (ϩHis) as a control and ϪLeu/Trp/His (ϪHis) plates to select for interaction.
SAXS Data Collection and Analysis-Data were collected on the SAXS-WAXS beamline at the Australian Synchrotron with a sample to detector distance set at 1567 mm and an X-ray wavelength of ϭ 1.033 Å, allowing access to a q-range spanning ϳ0.01-0.55 Å Ϫ1 (q ϭ (4sin)/, where is half of the scattering angle). Immediately prior to loading, all samples were centrifuged at 10,000 ϫ g to remove large particles from the solution. To minimize the effects of radiation damage, 50 l of each sample was flowed past the beam in 1.5-mm quartz capillaries (Hampton Research) at room temperature (22°C). Data were collected at several concentrations for hAPPL2 BARPH to assess the extent of concentration-dependent attractive or repulsive interactions. For each sample, five frames (each with a 2-s exposure time) were measured. Data reduction was carried out using SAXS15ID software (23), averaging all five measured frames and correcting for solvent scattering, sample transmission, detector sensitivity, and background radiation. Data were placed on an absolute scale by normalization against a water standard (24).
Data quality was assessed by inspection of the concentration dependence of the scattering data, linearity of the Guinier region of the data (qR g Ͻ 1.3), and the estimated molecular mass of the protein. At all concentrations, the Guinier plot (ln I(q) versus q 2 ) was linear, and yielded I(0) values normalized by concentration that showed no systematic trend. Estimated molecular masses were determined (25), where the program MULCh (26) was used to calculate the contrast (⌬) and partial specific volume (). Estimated molecular masses were close to the expected values. Taken together, these quality assessments indicate that the protein solutions are homogeneous, free of significant inter-particle interactions, and that the data are of high quality.
Modeling of hAPPL2 BARPH was performed using BUNCH (27), which is designed to model the structure of a single poly- c R free is R work calculated using 5.16% of reflections set aside from refinement. d Values were calculated using Molprobity (21).

hAPPL2 Structure and Its Interaction with Rab31
peptide chain, typically where flexible or unstructured regions link structured domains. The program models regions of known structure as rigid units, and the remainder as chains of dummy residues, to generate an envelope that includes side chains. The position of the BAR (2-263) domain was fixed, and the position and orientation of the PH domain (272-384) was optimized. The N-terminal portion of the protein incorporating the His 6 tag (24 residues) and linker region between the PH and BAR domains (264 -271) were modeled as flexible loops. C 2 symmetry of the dimer was enforced. The optimization was run 10 times, and the best model was chosen based on 2 . A summary of the SAXS data collection and analysis is given in Table 2.
ITC experiments were carried out at 10°C using a MicroCal AutoITC20 (GE Life Sciences). Origin software (version 7) was used to integrate the heat released and calculate the binding enthalpy (⌬H), equilibrium constant K a (1/K d ), and stoichiometry (N), based on a "one set of sites" model fit to the data. Several other more complex binding models were trialed, but there was no improvement in the fit to the data. The Gibbs free energy (⌬G) was calculated using the following equation: ⌬G ϭ ϪRTln (K a ); binding entropy (⌬S) was calculated using ⌬G ϭ ⌬H Ϫ T⌬S.
The titration was performed by adding hAPPL2 BARPH (160 M) in 19 injections (2 l) to 200 l Rab31-Gpp(NH)p (12 M). The sample was stirred at 1000 rpm for all measurements. The titration was performed in triplicate, and the reported parameters are the averaged values from the individual analyses.
Multiangle Laser Light Scattering (MALLS)-Rab31-Gpp-(NH)p was prepared as for the ITC experiment. MALLS was performed as described previously (28,29) using a miniDAWN Tristar laser light scattering photometer and Optilab DSP interferometric refractometer (Wyatt Technology). The size exclusion chromatography was performed using a Superdex TM 200 10/300 (Pharmacia Biotech) at 25°C in 50 mM Hepes, pH 8.5, containing 150 mM NaCl, 1 mM DTT, and 5 mM MgCl 2 . All samples were injected in a volume of 400 l, and M r estimates were determined using Debye fitting. Concentrations used were as follows: APPL2, 128 M; APPL2, 24.6 M and Rab31-Gpp(NH)p, 425.8 M for the APPL2-Rab31-Gpp(NH)p complex.

RESULTS
Structure of hAPPL2 BARPH -The crystal structure of hAPPL2 BARPH was refined to 3.5 Å resolution (Table 1), revealing two dimers in the asymmetric unit (Fig. 1). Each dimer is crescent-shaped and formed by an anti-parallel arrangement of the helical BAR domains of each hAPPL2 BARPH monomer. The apex of the crescent is shaped by a six-helix arrangement contributed by three helices (1, 2, and 3) from each of the BAR domains; this arrangement is typical of BAR domain dimers (2). Helix 4 is separated from helix 3 by two residues (residues Phe-218 and Ser-219) forming a kink that orients the long axes of the two helices ϳ120 o relative to each other. Helices 2 and 3 have minor kinks at residues 107-112 and 188 -192, respectively, which also contribute to the overall shape of the hAPPL2 BARPH dimer. The PH domain is a typical seven-stranded ␤-barrel capped by an ␣-helix at its C-terminal end (30). It is joined to helix 4 of the BAR domain by a linker region (Fig. 1A).
Two regions of the BAR domain interact with the PH domain of a dimer mate. Residues 148 -153 in loop 2 of the BAR domain monomer interact with residues 334 -339 of the PH domain in the loop between ␤-strand 5 and 6 of its monomer mate. On the other side of the same PH domain, residues 312-316 in the loop between ␤-strands 3 and 4 interact with amino acids 16 -21 in the N-terminal region of the BAR domain of its monomer mate. Two alternate monomer conformations are present in the asymmetric unit because in chain C the orientation of the PH domain relative to the BAR domain differs from that of the other three chains. This is discussed further below.
Electrostatic Surface-The electrostatic charge mapped to the surface of hAPPL2 BARPH and hAPPL1 BARPH dimers is presented in Fig. 2. The concave surface has prominent basic patches at either end of the dimer at the BAR/PH domain interface. These patches are symmetrically arranged and are contributed by residues at the BAR and PH domain interface, from the BAR domain residues Arg-20, Lys-146, Arg-149, and Arg-171 of one monomer and from the PH domain of its dimer mate residues Arg-299, Arg-314, and Arg-338. The less prominent two patches near the center of the dimer are again symmetri- The electrostatic surface of the convex face of the hAPPL2 BARPH and hAPPL1 BARPH dimers is also presented in Fig. 2. A prominent basic patch is apparent at the apex of the hAPPL2 BARPH dimer composed of residues Lys-212 and Lys-213 of one monomer and Lys-220 and Arg-221 of its dimer mate.
An end-on viw of the electrostatic surface of hAPPL2 BARPH is shown in Fig. 3. Two basic patches are apparent. The more prominent one is formed by residues Lys-152, Lys-153, and Lys-154 in loop 2 of one BAR domain monomer and by residues Arg-338, Arg-339, Arg-360, and Lys-361 of the PH domain from the dimer mate and is part of the basic groove that essentially runs around the molecule at the BAR and PH domain interface. The less prominent patch in Fig. 2C is contributed from residues Arg-287 and Lys-289 of the PH domain. Minor differences in the distribution of surface charge were observed between the AB chain and CD chain monomers.
The hAPPL2 BARPH dimer surface can be described as mostly electronegative with distinct electropositive ends and a central electropositive band that rings the BAR domain dimer. When the electrostatic surface of hAPPL2 BARPH is compared with that  Potential Phosphoinositol Binding Site-The PH domain of APPL2 is reported to interact with phosphoinositides (11,14). We investigated the potential binding site by comparing the structure of the hAPPL2 BARPH PH domain with the crystal structure of the ARHGAP9 PH domain bound to inositol (1,4,5)-trisphosphate (Ins(1,4,5)P 3 ) (PDB code 2P0D) (31). DaliLite (32) alignment of this structure with the PH domain of hAPPL PH gave a root mean square deviation (r.m.s.d.) of 2.0 -2.2 Å for 89 -92 C␣ atoms and sequence identity of 20%. An overlay of the two structures suggested that similar to ARH-GAP9, hAPPL2 BARPH may bind Ins(1,4,5)P3 in a spectrin-like noncanonical manner. In this binding mode, the phosphoinosi-  tol is bound between strands ␤1 and ␤2 and the loop joining ␤5 and ␤6 of the hAPPL2 BARPH PH domain, rather than between strands ␤1 and ␤2 and strands ␤3 and ␤4 of the PH domain as in the canonical binding mode (33). If the binding mode of Ins(1,4,5)P 3 is conserved between the two proteins, then Ins(1,4,5)P 3 would be located close to a complementary basic surface in hAPPL2 BARPH (Fig. 3). Furthermore, there is an interesting correspondence between the contact residues for Ins(1,4,5)P 3 in ARHGAP9 (31) and the hAPPL2 BARPH residues in this region: ARHGAP9 has Arg-342, Lys-343, and Trp-345; hAPPL2 BARPH has Lys-289, Arg-287, and Trp-297. Curiously, the equivalent residues of hAPPL2 are not related to those in ARHGAP9 by their position in the sequence, suggesting that if hAPPL2 does bind Ins (1,4,5)P3 in this manner, it may have evolved by convergent rather than divergent evolution.
Domain Flexibility-The APPL2 BAR and PH domains are linked by residues 258 -278. The hAPPL2 BARPH crystal structure reveals that these two domains move with respect to one another because the arrangement in chain C differs from the other three chains in the asymmetric unit (Fig. 4). Analysis of the PH domain rotation in the hAPPL2 BARPH dimers was assessed using DynDom (34). This identified loop 254 -263 254 LLSVDESVYT/P 264 as the bending region and found that the PH domain rotates by 13°with respect to the aligned BAR domains with minimal translation.
We SAXS Analysis of hAPPL BARPH -Structural modeling was performed against the solution scattering data using the program BUNCH. The PH domains sampled a range of dispositions relative to the BAR domains during the optimization, producing a model that provided an excellent representation of the data ( 2 ϭ 0.9, cf. the crystal structures: 2 (AB dimer), 2.6; 2 (CD dimer), 2.7). The SAXS-derived solution structure of the hAPPL2 BARPH dimer (Fig. 4a) is similar to the crystal structure, FIGURE 4. Solution x-ray scattering data and model for hAPPL2 BARPH . A, molecular surface of hAPPL2 BARPH (obtained by optimizing against the SAXS data using the program BUNCH (27)), overlaid with the AB dimer crystal structure (normalized spatial discrepancy, 1.11). The model derived from the solution scattering data suggests a small outward movement of the PH domain relative to the BAR domain in both the AB and CD dimers of the crystal structure. B, scattering data for hAPPL2 BARPH , and associated BUNCH model profile (solid black line, 2 ϭ 0.9). Inset, Guinier plot of the scattering data displaying a linear fit (R 2 ϭ 0.99), consistent with a monodisperse solution. C, pair-distance distribution function derived from the scattering data using GNOM (47) D, structural superimposition of the C chain monomer (green) with the D chain monomer (magenta) of the hAPPL2 BARPH crystal structure showing the altered conformation of the PH domain of chain C (13°rotation relative to the aligned BAR domains). The two residues Leu-254 and Thr-263 at either end of the Dyndom-defined bending region are indicated. (these are part of the linker between the BAR and PH domains.) A and D were prepared using PyMOL (47).

hAPPL2 Structure and Its Interaction with Rab31
but the PH domain is shifted outward by a small amount. This movement is consistent with the experimental pair-distance distribution function, which indicated that the maximum dimension of the molecule in solution was ϳ180 Å (cf. the maximum size of a dimer in the crystal structure is ϳ170 Å).
Identification of Rabs That Interact with APPL2-APPL1 and APPL2 have been reported to interact with Rab5 (6, 12). APPL1 has also been shown to interact with Rab21 (12). We hypothesized that APPL2 may also bind to other Rab partners. We therefore performed a yeast two-hybrid screen with APPL2 and a comprehensive Rab library, comprising 46 Rab GTPases containing at least one member of each subfamily (12). The Rabs used in this analysis were variants that stabilize their GTPbound form and hence are constitutively active. As a control, we also performed the screen using APPL1 as the bait and, in agreement with published literature (12), we identified Rabs 5 and 21 as APPL1-interacting proteins (data not shown). For APPL2, the yeast two-hybrid screen revealed interactions with Rabs 5, 22a, 24, and 31 (Fig. 5). APPL2 was previously shown to interact with Rab5 (6,14), but Rabs 22a, 24, and 31 are novel APPL2 interactors.
Analysis of hAPPL2 BARPH Interaction with Rab31-To follow up on the yeast two-hybrid results, we cloned and expressed Rab31 and generated the Rab31 complex with the non-hydrolysable GTP analog (Gpp(NH)p). We evaluated the interaction between hAPPL2 BARPH and Rab31-Gpp(NH)p using ITC and found that two Rab31 molecules bind one hAPPL2 BARPH dimer (n ϭ 0.98 Ϯ 0.01) with a binding affinity, K d , of 140 Ϯ 30 nM. Thermodynamic analysis of the interaction indicated that binding is entropically driven with ⌬H of 8.85 Ϯ 0.30 kcal/mol and ⌬S of 62.5 Ϯ 0.7 cal/mol/degree, suggesting that the binding of Rab31-Gpp(NH)p to hAPPL2 BARPH involves the displacement of structured water into bulk solvent.
We also performed MALLS experiments to confirm the stoichiometry of the complex. MALLS data yielded molecular mass estimates of 129.5 kDa Ϯ 0.8% for the complex (theoretical mass of 114 kDa for 2:1 complex; 135.4 kDa of for 2:2 complex), 95.1 kDa Ϯ 2.0% for hAPPL2 BARPH (theoretical mass of 92.1 for dimer), and 24.9 kDa Ϯ 8.0% (theoretical monomer mass, 21.9 kDa) for Rab31-Gpp(NH)p. These results are consistent with the formation of a complex comprising an hAPPL2 BARPH dimer and two Rab31-Gpp(NH)p monomers in agreement with the ITC results.

DISCUSSION
BAR domain-containing proteins are reported to possess a charge dipole across the BAR domain dimer with one face, usually the concave face, thought to be important for their ability to bind to negatively charged membrane lipids (1). The electrostatic surface of the hAPPL2 BARPH dimer does reveal a concave surface that is less electronegative than its convex surface with patches of electropositive residues at the center and the ends of the banana-shaped molecule. A striking difference between the electrostatic surfaces of hAPPL1 and hAPPL2 dimers is that APPL2 has an additional basic patch on the apex of the convex surface formed by residues Lys-220 and Arg-221 of one monomer and Lys-212 and Lys-213 of the second monomer: the equivalent residues in hAPPL1 are Lys-Met and Glu-Gln, respectively (Fig. 6).
The crystal structure of hAPPL2 BARPH incorporates two dimers, whereas APPL1 crystal structures all have one dimer in the asymmetric unit. The APPL2 structure reveals that the PH domain is able to rotate with respect to the BAR domain, and that this can occur unilaterally. SAXS studies have shown the BAR-PX domain of SNX9 is flexible (35), however, while our SAXS data from hAPPL2 BARPH are consistent with a small movement of the PH domain (Fig. 4), it does not conclusively demonstrate flexibility between the BAR and PH domains of APPL2.
Many PH domains bind phosphoinositides (36), and although debate exists about the phospholipid binding ability of the PH domain of APPL1 (11,12), the isolated PH domains of APPL1 and APPL2 have both been shown to bind phosphoinositides (14). Both APPL PH domains lack the high affinity and high specificity binding motif for phosphoinositides, KX n (K/ R)XR (37). However, we found that the APPL2 PH domain residues Arg-287, Lys-289, and Trp-297 superimpose with the noncanonical inositol (1,4,5)-trisphosphate binding residues of ARHGAP9 (PDB code 2P0D) (31), suggesting that this alternate binding mode may be used. Equivalent residues are also present in hAPPL1, although are not all modeled in the available crystal structures. Nevertheless, it seems possible that both APPL proteins may bind phosphoinositides using this noncanonical binding site.
BAR domain-containing and PH domain-containing proteins are binding partners for small GTPases (10,38), and both APPLs are Rab effectors (6). In addition, APPL1 has been shown to bind directly to Rab5 and Rab21 (12). However, relatively little is known of the interacting partners for APPL2. We therefore undertook a global screen for its Rab partners and found interactions with Rabs 5, 22a, 24, and 31. APPL2 did not bind the APPL1 partners Rab21 and none of the novel APPL2-interacting Rabs were found to interact with APPL1. Thus, our yeast two-hybrid results show that APPL1 and APPL2 differ in their Rab binding spectrum. We confirmed by ITC, MALLS in conjunction with size exclusion chromatography that Rab31 does indeed interact with hAPPL2 BARPH , with both ITC and MALLS indicating the formation of a 2:2 complex.
Rab31 is also called Rab22b due to its close similarity with Rab22a. Intriguingly, all three novel APPL2-binding Rabs are members of the same branch of the Rab superfamily, with Rabs 22a and 31 more closely related than Rabs 5 and 21. Rab24 is less closely related but still resides in the same branch (39). Our results imply that during evolution the Rabs and their cognate effector proteins APPL1 and APPL2 co-evolved.
Our structural studies shed light on two potential sequence motifs of BARPH proteins (Fig. 6). First, a candidate amphipathic helix has been identified at the N terminus of the BAR domains of the BARPH proteins centaurin-␤2 and oligophrenin-1 and have been suggested as possible membrane insertion motifs (40). The sequence 14 SPRFRAALEEVEGD-VAEL 31 in centaurin-␤2 aligns with residues 16 SPQTRSLLS-VFEEDAGTL 33 of hAPPL2. These residues form an N-terminal helical region in the crystal structure of hAPPL2 BARPH that interacts with a PH domain loop located between strands ␤3 and ␤4. Second, a putative nuclear localization sequence has been identified in the APPL2 sequence 151 PKKKENE 157 (6). Again, this motif is part of the BAR domain (loop 2), which interacts with a PH domain loop located between strands ␤5 and ␤6 in the hAPPL2 BARPH crystal structure. Taken together, these data support a model where the PH domain may mask interactions with these sequence motifs. We speculate that dynamic motion of the PH domain in solution or in response to specific stimuli may alter the accessibility of these motifs, switching on or off their specific functions. B-factor analysis of the linker region between the two domains (residues 258 -278) supports the notion that this region is flexible. Thus, for chain A, the average B value (all atoms) for this linker is 126 Å 2 compared with 109 Å 2 for the entire chain; values for chain B are 159 Å 2 linker, 116 Å 2 overall; and for chain D are 161 Å 2 for linker, 119 Å 2 overall. Curiously, the increased B-value for the linker region does not hold true for chain C in which the PH domain rotation is present: the average B value for the linker region in chain C is 110 Å 2, which is similar to the overall value for the chain of 114 Å 2 . We note that rotation of the PH domain in chain C does not fully expose the NLS and putative membrane insertion regions, although the SAXS data suggest the PH domain can adopt multiple conformations in solution. The possibility that PH domain rotation might enable these regions to be revealed, and whether this occurs in response to specific signals, will therefore require further study.
The interaction between APPL2 and Rab31 is of considerable interest in view of the known roles of these proteins in the regulation of metabolism. Adiponectin is an adipokine, which is secreted from adipocytes to fine control insulin regulation of metabolism in other organs similar to the liver and muscle (41). APPL2 is a negative regulator of adiponectin receptor signaling, whereas APPL1 has been reported to act as a positive regulator of adiponectin action (42,43). For example, knockdown of APPL2 in C2C12 myocytes enhances adiponectin-stimulated glucose uptake and fatty acid oxidation (42). Moreover, overex-FIGURE 6. Pair-wise alignment of full-length hAPPL1 and hAPPL2. Equivalent residues are highlighted in gray. The four helices of the BAR domain are indicated by orange underline, the PH domain by green underline, and the linker region by purple underline. The four basic residues that form the basic patch at the apex of the convex surface in APPL2 BARPH are boxed and shown in blue (see Fig. 2 APPL2 (AB) convex surface). It is evident from this alignment that the PH domain is more highly conserved than the BAR domain and that helix 4 is the least conserved of the four BAR domain helices.

hAPPL2 Structure and Its Interaction with Rab31
pression of Rab31 or its GTP exchange factor, Gapex-5, blocks glucose uptake in adipocytes (44). This suggests that APPL2 and Rab31 may act as a complex to negatively regulate the adiponectin signaling pathway to control glucose uptake in fat and muscle cells. If this model is correct, this would suggest that different APPL isoforms may coordinate opposing trafficking steps by invoking the activity of different Rab proteins that are the master controllers of these pathways. Consistent with this, Rab5/APPL1 controls endocytosis, whereas Rab31 is thought to control TGN/endosomal trafficking (45).