Research ArticleSTRUCTURAL BIOLOGY

Membrane surface recognition by the ASAP1 PH domain and consequences for interactions with the small GTPase Arf1

See allHide authors and affiliations

Science Advances  30 Sep 2020:
Vol. 6, no. 40, eabd1882
DOI: 10.1126/sciadv.abd1882

Abstract

Adenosine diphosphate–ribosylation factor (Arf) guanosine triphosphatase–activating proteins (GAPs) are enzymes that need to bind to membranes to catalyze the hydrolysis of guanosine triphosphate (GTP) bound to the small GTP-binding protein Arf. Binding of the pleckstrin homology (PH) domain of the ArfGAP With SH3 domain, ankyrin repeat and PH domain 1 (ASAP1) to membranes containing phosphatidylinositol 4,5-bisphosphate [PI(4,5)P2] is key for maximum GTP hydrolysis but not fully understood. By combining nuclear magnetic resonance, neutron reflectometry, and molecular dynamics simulation, we show that binding of multiple PI(4,5)P2 molecules to the ASAP1 PH domain (i) triggers a functionally relevant allosteric conformational switch and (ii) maintains the PH domain in a well-defined orientation, allowing critical contacts with an Arf1 mimic to occur. Our model provides a framework to understand how binding of the ASAP1 PH domain to PI(4,5)P2 at the membrane may play a role in the regulation of ASAP1.

INTRODUCTION

Most aspects of cell membrane signaling and trafficking are regulated by the assembly of dynamic multicomponent platforms, which associate with cell membrane surfaces in a lipid-dependent fashion (1, 2). Phosphatidylinositol phosphates (PIPs) are a common target for lipid-binding domains. Although present in cell membranes at relatively low concentrations, PIPs may cluster to form nanoscale domains (3). Pleckstrin homology (PH) domains are among the most common membrane-binding domains and are known to bind PIPs (4, 5). Here, we focus on the PH domain of the Arf GTPase (guanosine triphosphatase)–activating protein (ArfGAP) ASAP1 (6). In cells, ASAP1 affects cell behaviors dependent on adhesions and actin, including proliferation, invasion, and metastasis of cancer cells, which depends on its ability to hydrolyze guanosine triphosphate (GTP) bound to the Ras superfamily protein adenosine diphosphate–ribosylation factor-1 (Arf1) [(7) and references therein].

ASAP1 has a core catalytic domain composed of PH, Arf GAP, and ankyrin repeat domains {[325-724]–ASAP1, referred to as PZA (for PH, zinc binding, which comprises the Arf GAP catalytic domain, and ankyrin repeat domains)}. The ASAP1–PH domain ([325-451]–ASAP1) is immediately N-terminal to the catalytic Arf GAP domain. The PH domain core is a seven-stranded β sandwich consisting of N-terminal four-stranded (β1 to β4) and C-terminal three-stranded (β5 to β7) β sheets and is terminated with a long α helix wedged between the two sheets of the β sandwich at one end (Fig. 1, A and B) (8).

Fig. 1 Sequence and structure of the ASAP1–PH domain.

(A) Residues truncated in ΔN14ASAP1–PH are provided in the dotted box. Structured residues in the crystal structure (PDB: 5C79) are shown in plain font ([334-437]–ASAP1, referred to as 5C79). (B) Ribbon representation of the structure ASAP1–PH in PDB: 5C79. For visual guidance, N and C termini are labeled along with all β strands, as well as loops linking the β strands. Approximate sites of PI(4,5)P2 of interaction are labeled CA (for canonical sites) and NCA (for noncanonical site). Isoleucine side chains are shown as stick models.

The ASAP1–PH domain binds to phosphatidylinositol 4,5-bisphosphate [PI(4,5)P2]–containing membranes. Recently, a crystal structure of the PH domain was solved in complex with dibutyryl-PI(4,5)P2, providing structural evidence for the presence of two PI(4,5)P2 headgroup-binding sites in a single PH domain. One is formed by positively charged residues of the β12 and β34 loops [canonical site (CA)]; the other is on the opposite side of the β12 loop in between the β12 and β56 loops [noncanonical site (NCA)]. In line with multiple binding sites, PI(4,5)P2 lipids cooperate to promote membrane association, potentially switching on and off GAP activity within a narrow range of PIP concentration (8).

PH domains can also function in capacities other than membrane-targeting modules. PH domains can autoinhibit or position other structural elements of a protein to inhibit intramolecular catalytic domains, as described for kinases and guanine nucleotide exchange factors. In Akt, the PH domain inhibits the associated catalytic domain, a serine/threonine kinase (9). In the case of Arf exchange factors of the cytohesin family, cooperative binding of PIP and Arf6•GTP or Arl4•GTP to the PH domain relieves PH domain–mediated autoinhibition (10, 11). PH domains can also form part of the substrate-binding site or occlude the enzymatic site, as observed in Rho exchange factor such as Dbs (12) and FARP2 (FERM, ARH/RhoGEF and pleckstrin domain protein 2) (13). In the case of ASAP1, it was recently shown that deletion of residues 325 to 338 or mutation of I423 within PZA (to form [339-724]–ASAP1 and [325-724]–I423AASAP1) led to reduced PZA GAP activity by 100-fold and 4-fold, respectively, without affecting the affinity of the mutant PH domains to PIP-containing bilayers (14). These data suggest that recruitment of the catalytical ZA domain to a membrane containing PI(4,5)P2 and Arf1•GTP is not sufficient for maximum GAP activity and that the PH domain might contribute to Arf1 specificity and, perhaps, GAP activity through a direct interaction with Arf1 at the membrane. It is tempting to speculate that simultaneous binding of multiple PI(4,5)P2 to the PH domain at the membrane interface may promote functionally critical contacts of the ASAP1 core catalytical domain with its substrate Arf1•GTP. However, how PI(4,5)P2 binding may trigger ASAP1–PH/Arf1•GTP interaction, in addition to its membrane recruitment, and how PI(4,5)P2 affects the docking geometry of the PH domain at the membrane are, as yet, unknown.

The present study combines nuclear magnetic resonance (NMR), neutron reflectometry (NR), and molecular dynamics (MD) simulations to elucidate the membrane docking geometry, the interactions between the ASAP1–PH domain and PI(4,5)P2 in a membrane context, and their consequences. We show that the two loops between β sheets 1, 2 and 3, 4 and residues belonging to the β6 strand are involved in the binding interface at the bilayer surface with multiple PI(4,5)P2 headgroups and that other anionic lipids can potentiate those interactions without binding specifically. The ASAP1–PH domain is docked at the membrane, with the long axis of its core β sandwich lying almost parallel to the bilayer normal. Binding of the PH domain to a PI(4,5)P2 headgroup at the membrane, but not to a soluble analog, promotes subtle conformational changes along the C-terminal α helix, distant from the membrane interface. Titration of the PH domain bound to nanodiscs (NDs) with an activated Arf1 mimic (L8KArf1•GTP) suggests that this α helix plays a critical role in the formation of the Arf1•GTP/PZA complex at the membrane. Our findings provide insight into the details of anchoring of the ASAP1–PH domain to membranes and shed light on the Arf1/ASAP1–PH:PI(4,5)P2 assembly, demonstrating that the orientation of the PH domain and specific ASAP1–PH:PI(4,5)P2 binding at the membrane interface trigger a conformational switch, which allows ASAP1–PH:PI(4,5)P2 and ASAP1–PH/Arf1 interactions to occur simultaneously through different parts of the protein.

RESULTS

Lipid-ASAP1 PH domain interactions at the bilayer surface

PI(4,5)P2 binding sites by NMR and MD. Previous investigations of the interaction of the ASAP1 PH domain with a water-soluble analog of PI(4,5)P2-containing dibutyryl (C4) chains [diC4-PI(4,5)P2] and a PH domain truncated at its N terminus ([339-451]–ASAP1; referred to as ΔN14ASAP1–PH) were conducted by NMR in our laboratory (8). However, since these studies omitted the context of membrane-embedded PI(4,5)P2 and used a PH domain truncated at its N terminus, we felt it appropriate to repeat these studies using the full-length ASAP1 PH domain ([325-451]–ASAP1, referred to as ASAP1–PH) interacting with a membrane mimetic doped with lipids containing a PI(4,5)P2 headgroup to capture the effect of the bilayer. We used NDs as a well-established membrane mimetic to facilitate solution-state NMR spectroscopy of the membrane surface complexes (15). To do so, we titrated the 15N and I,L,V,A,T-13C methyl–labeled protein with increasing concentration of NDs containing PI(4,5)P2 and followed changes in chemical shifts by [15N-1H] TROSY HSQC (heteronuclear single-quantum coherence) (16) or by [13C-1H]-methyl TROSY HMQC (heteronuclear multiple-quantum coherence) (17) NMR methods (fig. S1).

1H-13C chemical shift perturbations (CSPs) of the methyl groups of residues I371, A381, L383, and L402 were used to generate binding isotherms and calculate dissociation constants (Kd). The soluble diC4-PI(4,5)P2 analog binds to the free state of ASAP1–PH in the fast exchange regime with a Kd of 250 ± 50 μM. Similarly, NDs containing a single PI(4,5)P2 lipid on average [1.25 mole percent (mol %)] bind to the PH domain with a Kd of 180 ± 20 μM. This indicates that (i) ASAP1–PH binds to a single phosphoinositide headgroup with relatively low affinity, and (ii) the dominant energetic contribution to membrane binding is electrostatic and not hydrophobic [as shown by the near equivalence of PI(4,5)P2 binding free energy in solution and in a membrane environment]. Increasing PI(4,5)P2 mole fraction from 0.0125 to 0.05 in NDs [four PI(4,5)P2 per ND] resulted in tighter binding (Kd = 12 ± 3 μM), indicating that binding of a single PH domain to multiple PI(4,5)P2 molecules increases its affinity to the membrane (fig. S2, A to H).

Binding of the soluble analog induced large 13CO CSPs for residues in the β12 loop (residues K349, S350, D351, and I353) and β34 loop (residues H373, T375, and Q379) and residues L402 and I403 (β6 strand), as observed previously for ΔN14ASAP1–PH (Fig. 2A). Binding to PI(4,5)P2-doped NDs resulted in CSPs in the same regions of the protein, suggesting that contacts between the protein and the phosphoinositide are similar at the membrane surface and in solution (Fig. 2B). The near-perfect correlation between 1H-15N CSP values at PI(4,5)P2 mole fraction of 0.0125 and 0.05 in the membrane suggests that the CA and NCA have similar affinity when presented with a single PI(4,5)P2 at the membrane (fig. S2I). Mapping the changes onto the PH domain structure suggested two pockets where PI(4,5)P2 headgroups can bind, in good agreement with the position of the CA and NCA sites previously described (fig. S2, J to M) (8).

Fig. 2 ASAP1–PH:PIP2 binding interface and ASAP1–PH:PS interactions identified through CSP and MD analyses.

CSPs between free and bound PH domain are plotted against residue number. (A) ASAP1–PH was titrated by diC4-PI(4,5)P2. (B) ASAP1–PH was titrated by PI(4,5)P2-containing NDs. Lipid composition of the NDs was 16:0-18:1 PC:18:1-18:1 PI(4,5)P2 (95:5). A carbon-detected experiment was used to measure diC4-PI(4,5)P2–induced CSPs. (C) Normalized PI(4,5)P2:ASAP1–PH interaction counts plotted against residue number. Data were averaged over three MD trajectories with a cutoff of 3.5 Å to detect nonhydrogen proximities. (D) (Left axis, ●) Free energy of binding of ASAP1–PH to 16:0-18:1 PC NDs containing 1.25 mol % of 18:1-18:1 PI(4,5)P2 and increasing mole fraction of 18:1-18:1 PS. Dissociation constants were calculated by following 1H-13C chemical shift changes of I371, A381, and L383 methyls as a function of ND concentration. (Right axis, ○) The percentage of GTP bound to myr-Arf1 hydrolyzed in 3 min by PZA is plotted against the mole fraction of PS in large unilamellar vesicles (LUVs) containing PI(4,5)P2 (1 mol %) and increasing concentration of PS. (E) Plot of 1HN-15N CSPs of ASAP1–PH bound to 16:0-18:1 PC NDs containing 1.25 mol % of 18:1-18:1 PI(4,5)P2 (x axis) and 1.25 mol % 18:1-18:1 PI(4,5)P2 and 15 mol % of 18:1-18:1 PS (y axis). 1HN-15N CSPs were collected at 0.1, 0.2, and 0.6 of bound PH domain (as measured by the CSP of I371, A381, and L383 methyls by 1H-13C HMQC). The data presented are for the fully bound ASAP1–PH domain. (F) Density distributions of PI(4,5)P2 (left) and PS (right) headgroups bound to ASAP1–PH, calculated using the last 100 ns of one of the all-atom simulation. A diffused density for PS highlights the nonspecific interaction of PS with the PH domain. Peaks in the density distribution correspond to the CA and NCA.

To complement the NMR results, multiple (nine) independent membrane-binding MD simulations were performed for a system containing ASAP1–PH and a PI(4,5)P2-containing lipid bilayer [PC:PS:PI(4,5)P2, 75:20:5]. To enhance sampling for this complex system, we used a highly mobile membrane mimetic model (HMMM) (18, 19), which accelerates lipid diffusion and reorganization using an atomistic representation. In seven of nine simulated trajectories, ASAP1–PH spontaneously and stably bound to the membrane and remained membrane-bound for the remainder of the simulation, without any dissociation events occurring (fig. S3). Out of all the PI(4,5)P2-bound replicas, we observed the interaction of the ASAP1–PH domain with two or three PI(4,5)P2 molecules without evidence for further clustering. To test the stability of the captured PI(4,5)P2-bound conformations from the HMMM membrane-binding simulations, representative replicas were then converted to all-atom membrane representations and simulated for an additional 200 ns. Orientation, depth of insertion (see below), and lipid:PH domain interactions involving residues in the β12 and β34 loops and β6 strand corresponding to the CA and NCA sites remained stable. In the CA site, five side chains (R354, Q358, R360, H373, and R378) interact with phosphate groups of the bound PI(4,5)P2 lipid: Phosphate-5 interacts with R360, Q358, R378, and H373; phosphate-4 interacts with R378; and phosphate-1 interacts with the basic side chain of R354. In the NCA site, side chains from K349, R407, W357, Q412, and R407 interact preferentially with phosphate-5, while W357 and Q412 side chains also form transient interactions with phosphate-4. Contrary to the crystal structure, phosphate-1 does not form any direct interactions with the protein. All the captured PI(4,5)P2-specific interactions are in good agreement with the binding regions observed by NMR (Fig. 2C and fig. S4).

PS:ASAP1–PH domain interactions by NMR and MD. In addition to phosphoinositides, the presence of other anionic lipid species in a bilayer, such as phosphatidylserine (PS), can affect the affinity of a PH domain for bilayers (20). In the absence of PI(4,5)P2, no 1HN-15N or 1H-13C CSPs were observed for bilayers containing 15 mol % of PS, even at tens of millimolar lipids. These results indicate that ASAP1–PH domain binding is very low for bilayers containing only PS as an anionic lipid.

We then investigated the influence of background PS on the binding affinity of ASAP1–PH in the presence of PI(4,5)P2. It was previously proposed that background anionic lipids may (i) favor initiation of a complex through nonspecific electrostatic effects or (ii) compete for PI(4,5)P2 binding sites through specific interactions, potentially weakening the interaction of the PH domain with a PI(4,5)P2-containing bilayer (20). The binding affinity was monitored by titrating the ASAP1–PH domain with NDs containing, on average, one PI(4,5)P2 and increasing concentration of PS ranging from 0 to 30 mol %. Addition of 30 mol % PS resulted in a 10-fold increase in affinity for the membrane. Figure 2D shows that the free energy of binding, assessed from Kd measurements (closed symbols) and GAP activity (open symbols), increases linearly with the mole fraction of PS in the membrane. This indicates (i) that an increase of the background anionic charge increases the affinity of ASAP1–PH for the membrane and (ii) that those electrostatic interactions are nonspecific; i.e., binding of one or more anionic lipid headgroups to sites on the ASAP1–PH domain does not compete directly with PI(4,5)P2 at these sites. The latter was confirmed by comparing 1HN-15N combined CSPs in the presence of NDs containing only one PI(4,5)P2 with that of NDs containing one PI(4,5)P2 and 15 mol % of PS [1 PI(4,5)P2 and 11 PS molecules per ND, on average]. If PS was displacing a PI(4,5)P2 molecule from its CA or NCA binding pocket, then one should observe different CSP values for the two sites. In contrast, the near-perfect correlation between CSP values with and without PS demonstrates that even an excess of background PS does not displace PI(4,5)P2 from its binding pockets at the ASAP1–PH domain–membrane interface (Fig. 2E).

In addition, we analyzed the distributions of PI(4,5)P2 and PS in MD simulations and observed a diffuse distribution of PS lipids around the whole footprint of the PI(4,5)P2-bound ASAP1–PH domain (Fig. 2F). These results corroborate the NMR result that there are no specific binding sites for PS on the domain. Rather, PS accumulates around the protein through nonspecific, electrostatic interactions between the highly cationic PH domain and its anionic headgroup.

The increase in affinity in the presence of PS and the nonspecificity of PS:ASAP1–PH contacts strongly suggest that a background of membrane PS favors the initiation of the protein-membrane complex. Paramagnetic relaxation enhancement (PRE) NMR data obtained for the ASAP1–PH domain in the presence of bilayers containing 15 mol % of PS without PI(4,5)P2 suggest that ASAP1–PH is steered to the membrane in a correct orientation through electrostatic interactions (fig. S5).

Membrane docking geometry of ASAP1–PH by NMR, NR, and MD

Nuclear magnetic resonance. First, bilayer proximities of the I, L, V, A, and T methyl groups of ASAP1–PH were estimated by the time-averaged, r−6 distance-dependent increase in transverse relaxation rates (R2) upon interaction of the protein with spin-labeled bilayers and compared to spin label–free bilayers. Figure 3 (A and B) shows the resulting PREs plotted as ΔR2, the difference between methyl proton relaxation rates with and without 5- and 10-doxyl–substituted lipids. When bound to 5- or 10-doxyl–labeled NDs, methyl resonances of I353 and V356 (β12 loop) and A374 and T375 (β34 loop) are absent from the spectra, indicating that those methyl groups are within 20 Å or less of the spin label. In addition, methyl groups of L346 and L347 (β1 strand) have large PREs from the 5-doxyl phosphatidylcholine (PC) spin label, buried at a shallower depth in the membranes. This is not the case for methyl groups of I368 or I369 (β3β4 sheet, on the opposite flank of the ASAP1–PH domain β core), which would be located at a similar z position relative to the membrane surface assuming an upright orientation (where z is the direction perpendicular to the bilayer plane). This suggests that, at least part of the time, ASAP1–PH explores a tilted orientation, bringing the β1β2 sheet closer to the membrane surface than in the upright configuration.

Fig. 3 Membrane and solvent PRE data.

Bilayer proximities of I, V, L, T, and A side chains were measured to PC:PI(4,5)P2 (95:5) NDs doped with an average of two (A) 5-doxyl spin-labeled PC molecules or (B) 10-doxyl spin-labeled PC molecules per leaflet. The resulting PREs were plotted as ΔR2 = R2paramagR2diamag, the difference between methyl proton relaxation rates with and without doxyl-substituted dipalmitoylphosphatidylcholine (DPPC). The uncertainties indicated are those of the fits of the exponential decays and their differences. (C) Solvent PRE expressed as the change of relaxation rate normalized to the probe concentration (mM−1·s−1) as a function of the residue numbers for bound (dashed) or free (solid) ASAP1–PH domain. n = 2, *P < 0.0001, t test. (D) Membrane PRE data mapped on the crystal structure of the ASAP1–PH domain; I/L/V/A/T residues are represented with spheres corresponding to ΔR2 > 30 s−1 (red), 15 s−1 < ΔR2 < 30 s−1 (orange), and ΔR2 < 15 s−1 (green). (E) Solvent PRE data mapped on the crystal structure of the ASAP1–PH domain. Sites protected by more than 50% are marked on the crystal structure by red spheres. An orange sphere is used to label A374 (~30% protection). Deprotection is marked by green spheres. Spectral superimposition prevented the measurement of reliable data for residues marked by ice blue spheres (T375, T408, T444, and L449).

Second, methyl PREs were obtained in the presence of a soluble paramagnetic probe, a method referred to as co-solute or solvent PRE (sPRE) (21) that reports on the solvent accessibility of an atom. Figure 3C compares the sPRE rates measured for ASAP1–PH free in solution and bound to NDs [PC:PI(4,5)P2, 95:5], and the results are mapped onto the ASAP1–PH domain structure. For example, methyls of residues I353, V356, and, to a lesser extent, A374 show slower normalized relaxation rates when bound to NDs, validating the earlier finding that the β12 loop penetrates the bilayer and suggesting that its insertion depth is larger than that of the β34 loop. In particular, the side-chain methyl of residue I353 becomes almost completely solvent-protected, indicating its insertion below the phosphate plane of the bilayer.

Neutron reflectometry. Inherently, a low-resolution structural technique, NR from substrate-supported planar membranes—sparsely tethered bilayer lipid membranes (stBLMs) (22) or simple vesicle-fused bilayers on a Si wafer—can impose long-range restraints on structural models provided by NMR or crystallography (22, 23), thus providing highly resolved structures of protein-membrane complexes from rigid-body modeling (24). stBLMs have physical properties (25) similar to those of fully solvent-exposed membranes in vesicles or NDs and, in particular, are fully accessible to adsorbents from the adjacent buffer.

We first measured the neutron reflection of as-prepared (“neat”) bilayers, typically in two isotopically distinct solvents based on H2O and D2O, and then repeated the measurements, on the same physical sample, with proteins dissolved in the adjacent buffers. Because of the long measurement time required (hours), protein-membrane complexes thus characterized represent dynamic equilibria between dissolved and membrane-bound proteins. As a surface-sensitive scattering technique, NR is inherently sensitive to disordered protein segments but, on the other hand, insensitive to dissolved protein in the buffer adjacent to the supported membrane (23). We took advantage of the existing crystal structure [Protein Data Bank (PDB): 5C79] to interpret the NR spectra measured for ΔN14ASAP1–PH on a PIP2-containing stBLM and followed up with measurements of ASAP1–PH—the full-length PH domain—on a vesicle-fused, PI(4,5)P2-containing bilayer, seeking structural information on organization of the disordered segment ΔN14 on the membrane.

As an overview, Fig. 4A shows ΔN14ASAP1–PH on an stBLM [membrane composition: PC:PI(4,5)P2, 95:5] in which the protein profile was modeled as a Hermite spline. The figure shows the component volume occupancy (CVO) profiles for the entire bilayer, with bound protein to the right (positive z) of the gold film that covers the support. The bound protein imposes merely a disturbance on the structure of the as-prepared bilayer (see the Supplementary Materials for raw data of an exemplary set of measurements in fig. S6 and fitted model parameters in table S1). Because no assumptions on the protein structure are made, this representation provides a bias-free, albeit low-resolution, view of protein association with the membrane. Protein binding to the as-prepared membrane did not substantially alter bilayer structure (table S1) that was prepared, and remained, essentially defect free. The density of interfacially bound PH domain was ≈1 protein per 60 phospholipids, or per three PIP2, in the exposed membrane leaflet, adsorbed from a solution that contained 40 μM protein.

Fig. 4 Docking geometry of the ASAP1–PH domain on stBLMs [16:0-18:1 PC:18:1-18:1 PI(4,5)P2 (95:5)] by NR.

(A) CVO profiles of a lipid bilayer with the surface-bound ASAP1 PH domain (ΔN14ASAP1) parameterized as a Hermite spline. The protein distribution is indicated with its median (black dashed line) and 68.2% confidence region. (B) Data evaluation by rigid-body modeling with the crystal structure (PDB: 5C79) and comparison with the free-form model [same data as in (A)]. Inset: Comparison of protein CVO profiles for ΔN14ASAP1–PH (black line) and ASAP1–PH (dashed line with confidence region) from free-form fitting. (C) Orientation analysis of ΔN14ASAP1–PH at the membrane in terms of two Euler angles (β,γ), as indicated in the inset. Protein orientations are realized by first rotating the reference orientation of the high-resolution structure (see Materials and Methods) by an angle β about the x. Second, the structure is rotated by an angle γ about z′. (D) Visualization of the PH crystal structure on the stBLM in a configuration (orientation and insertion depth) on the membrane at the center of the 68.2% probability region shown in (C). Phospholipids and substrate are only schematically displayed—for example, atomistic corrugation of the interface has been neglected—but lipids are shown at the same scale as the protein. The CVO profile at the left shows the same data as in (B).

Subsequently, a rigid-body model, which optimizes the orientation and membrane penetration depth of crystal structure density projections on a grid of angular sections through the highly resolved protein structure, revealed more information. Figure 4B shows a close-up of the refined protein CVO profile at the bilayer surface, in terms of its median and confidence limits, in comparison with the free-form CVO (same data as in Fig. 4A). Both representations overlap fully within the resolution of the method, but the rigid-body model contains considerably more information, contingent on the assumption that the crystal structure is a valid representation—at the resolution of the experiment—of the structure of the membrane-bound protein. Figure 4C shows ranges of likelihood for the orientation of the PH domain on the bilayer in a reference system indicated in the inset, and Fig. 4D visualizes the penetration depth of the protein into the membrane. The orientation is well resolved, within ±10° about the Euler angles (β,γ) = (15°,90°), which points PH residue I353 and its surrounding β12 loop toward the membrane. The isoleucine residue snorkels into the membrane region down to the level of the lipid glycerols, as shown in Fig. 4D, substantially below the headgroup phosphates. The rigid-body model places its Cα position at Δz = −1.5 Å (+1.2/−1.9 Å) below the center plane of the lipid headgroups (which is approximately the center of the phosphate positions).

The inset in Fig. 4B shows results from a measurement of ASAP1–PH (protein concentration in the buffer: 20 μM) on a Si-supported bilayer of the same composition as for the truncated protein, in which the protein CVO from the free-form model is compared with that from ΔN14ASAP1–PH. No significant differences between the two sets of data are observed in the profiles, except for the protein density, which was higher for the full-length variant despite its lower solution concentration (table S1). This analysis indicates that the full-length ASAP1–PH domain binds to the membrane in a similar configuration to its truncated version, but it does not reveal further insight into the role of the ΔN14 segment.

MD simulations. NR and NMR analyses were complemented by performing extensive membrane-binding simulations of 5C79 near a lipid bilayer [PC:PS:PI(4,5)P2, 75:20:5] using HMMM. Figure 5A shows the Cα fingerprints for all nine replicas. In all simulations, the β12 loop contacts the lipid bilayer, suggesting a strong membrane-binding propensity in agreement with the PRE effects observed with NMR and the protein orientation/penetration depth derived from the NR work. We find that I353 is the residue that penetrates the membrane deepest, with an average location at Δz = −3.5 ± 2.7 Å below the phosphate plane, which is consistent with the NMR PRE data and the NR results. Meanwhile, the β34 loop shows a tendency to bind to the lipid bilayer in only some replicas, as seen in the sPRE NMR data. Figure 5B compares the regions that define the 68% probability orientation distributions of ΔN14ASAP1–PH at the membrane as determined by NR with that from the MD trajectories. Overall, there is good agreement between experiment and simulation, as both techniques show that the PH domain is sampling a fairly narrow range of orientations in the same region of Euler angles (β,γ).

Fig. 5 MD simulations reveal PH domain approach to the membrane surface and docking geometry.

(A) Ensemble-averaged Cα profile of 5C79 calculated during the last 50 ns of HMMM simulations in PI(4,5)P2-containing membranes for all nine independent simulations. Red dots correspond to the average z position of each residue, and blue bars represent SD over the averaged time, where z is the distance from the bilayer phosphate plane. (B) Comparison of the 68% probability distribution of ΔN14ASAP1–PH to average orientations obtained from the HMMM MD simulation trajectories.

ASAP1–PH interactions with PI(4,5)P2-containing bilayers—Structural consequences

A recent study reported that PI(4,5)P2 analogs increased PZA GAP activity as a function of increasing acyl chain length (14). This suggests that, in addition to membrane recruitment, PH binding to PI(4,5)P2 in the bilayer context may trigger functionally important structural rearrangements in the PH domain. To test that hypothesis, we compared 1H-13C CSPs of ASAP1–PH bound to soluble diC4-PI(4,5)P2 with those of the domain bound to PI(4,5)P2-containing NDs (Fig. 6, A and B). Differences, which report a change in the chemical environment of those nuclei, are observed not only for methyl groups of residues I353, T375, and I403—which are part of the PI(4,5)P2 binding pockets—but also for methyl groups of L386, T387, I420, I423, L434, and T435, which are distant from the bilayer interface but near or within the C-terminal α helix capping the PH domain. In addition, we measured higher solvent accessibility for the methyl groups of L386, V390, A421, and I423 when ASAP1–PH is membrane-bound. While a change in the chemical environment of methyl groups close to the bilayer interface is expected upon membrane binding, the changes distant from the bilayer interface indicate that PI(4,5)P2 interaction with ASAP1–PH also promotes allosteric changes distant from the bilayer interface. These experimental observations thus suggest that a subtle reorientation of residues with methyl side chains along the C-terminal α helix is taking place upon interaction of the PH domain with the membrane containing PI(4,5)P2 (Fig. 3C). We then interrogated the MD simulations by comparing the root mean square fluctuations (RMSFs) calculated for the Cα atoms of each residue along the sequence for the ASAP1–PH domain free in solution and bound to PI(4,5)P2-containing membranes (fig. S7, A and B, and table S2). Lower RMSFs are observed for the membrane-bound protein loops and part of the NCA site, as expected from restricted motions at the membrane-binding interface, as well as for the C-terminal helix and surrounding regions. Conversely, RMSF of the protein core backbone was independent of the bound state of the protein. These observations suggest that PI(4,5)P2 binding at the membrane changes the distribution of conformational states explored by the C-terminal helix, as suggested by the NMR experiments. Solvent-accessible surface areas, calculated for the membrane-bound and free ASAP1 PH domain, show that part of the C-terminal helix is more solvent-exposed when the PH domain is at the membrane, which is in general qualitative agreement with the NMR results showing increased solvent exposure of the C-terminal helix (fig. S7C).

Fig. 6 PI(4,5)P2 triggered conformational switch and interface with L8KArf1•GTP identified through CSP.

(A) 1H-13C CSPs observed when ASAP1–PH is bound to NDs containing 18:1-18:1 PI(4,5)P2 (filled bars) and when bound to diC4-PI(4,5)P2 (open bars). n = 3, *P < 0.0001, **P < 0.05, t test. (B) Methyl residues that show significant CSP differences and are not located in the CA or NCA sites are plotted in orange on the crystal structure. Residues in the CA or NCA sites are plotted in dark blue. (C) Difference between the 1H-13C CSPs observed for PH binding to NDs in the absence and in the presence of L8KArf1•GTP at a ratio ASAP1–PH:Arf1 1:1. In addition to CSPs on the α helix, small CSPs were observed for residues around the PIP2 binding sites, likely indicating tighter binding of the PH domain in the presence of Arf1 or altered PI(4,5)P2 interactions. n = 2, **P < 0.05, t test. (D) Methyl residues exhibiting significant CSPs upon L8KArf1•GTP binding are plotted in light blue on the crystal structure.

The NMR, NR, and MD results, which include binding and PRE data as well as an explicit docking orientation, consistently suggest that the β12 loop of the ASAP1 PH domain penetrates the bilayer, with burial of the hydrophobic I353 side chain into the membrane. To probe whether I353 membrane penetration triggers the conformation change, a series of mutants of I353 were generated in the PH domain and in the catalytic PZA construct [325-724]–ASAP1. The binding of the PH mutants to PI(4,5)P2-containing NDs clearly showed a correlation with the hydrophobicity of the mutated residue. However, the activity of the corresponding PZA mutants did not perfectly correlate with their bound fractions (fig. S8B). These results suggest that an isoleucine at position 353 may be critical for the conformational change of the PH domain α helix.

We then titrated NDs with and without bound ILV-labeled PH domain with unlabeled, unmyristoylated L8KArf1•GTP, a point mutant that is soluble in its GTP-bound form contrary to myr-Arf1•GTP (26). Consequently, L8KArf1•GTP binds to PH domain–free NDs with a Kd of 10 μM (to be compared to nanomolar range for myr-Arf1•GTP). Nevertheless, L8KArf1•GTP can be used as a proxy for probing the functional relevance of the conformational change observed on the PH domain. In the presence of an equimolar concentration of PH domain bound to NDs, CSPs and reduced signal intensity were observed for residues I423, L426, T427, T435, and A437 located along the C-terminal α helix (Fig. 6, C and D), suggesting that these residues are implicated in the interaction between L8KArf1•GTP and ASAP1–PH.

DISCUSSION

It was previously shown that PI(4,5)P2 is necessary for efficient ASAP1 GAP activity [(14) and references therein]. Here, we examined the docking geometry and lipid interactions of the ASAP1 PH domain on model membranes containing PI(4,5)P2 by combining data from NMR and NR experiments with MD simulations. We report that PI(4,5)P2 binding (i) maintains the PH domain in a defined orientation that allows critical contacts between the newly exposed protein segments and an Arf1 mimic to occur and (ii) triggers functionally relevant allosteric conformational changes that involve regions of the PH domain distant from the membrane. The results are consistent with a model where PI(4,5)P2 not only recruits but also activates the PH domain for GAP activity regulation.

The membrane docking geometry of the ASAP1–PH domain bound to a bilayer containing PI(4,5)P2 is well defined by and consistent between NR, MD, and NMR. All three methods support a model in which PH is membrane peripheral and oriented such that the long axis of the β sandwich is almost aligned with the bilayer normal, positioning the C-terminal α helix and unstructured regions within 25 ± 5 Å from the lipid bilayer surface. Since mutations along both of those segments decrease ASAP1 ArfGAP activity, it is tempting to suggest that the docking geometry described here allows critical contacts between those segments and Arf1•GTP.

The penetration of ASAP1–PH into the bilayer is unique for PH domains. It penetrates deeper than the GRP1 (general receptor for phosphoinositides 1) PH domain, which constitutes the only case for which partial penetration has previously been reported (27). The deepest backbone penetration occurs at I353, whose Cα resides 3.5 ± 2.7 Å below the plane of average lipid phosphate positions as observed by MD. In this location, it is then expected that the δ1 side-chain methyl of residue I353 penetrates deeply into the acyl chains, where it experiences a different chemical environment from the aqueous solution. This localization in the bilayer core is consistent with the NMR-observed CSPs and both membrane PRE and sPRE. In all MD trajectories, we observed that the β12 loop plays a key role in initiating and stabilizing ASAP1–PH membrane binding. This is in good agreement with experimental data on the GRP1 and cytohesin-2 PH domains, which suggest that the β12 loop region constitutes the primary PIP-binding site on PH domains. Hydrophobic interactions of I353 within the bilayer core combined with specific interactions of PIP2 with the CA and NCA binding sites establish an effective anchor point for binding. Furthermore, the position of I353 within the domain structure defines the orientation of the domain on the membrane, as experimentally established by NR and confirmed by MD simulations.

We observed allosteric effects within ASAP1–PH upon binding to PI(4,5)P2 headgroups at the bilayer surface that are not seen upon binding to soluble PI(4,5)P2 headgroups. CSP results show that PI(4,5)P2:ASAP1–PH interaction, in the context of a bilayer, alters the chemical environment of methyl side chains along the C-terminal α helix located ≈25 Å from the membrane, and sPRE data show that conformational changes in that region result in the exposure of buried side chains. MD simulations suggest that membrane binding induces a shift in the distribution of conformational states explored by the C-terminal helix. Collectively, experimental and in silico results suggest a conformational switch in ASAP1–PH. Having an isoleucine at position 353 appears to be important for function. Reports of conformational switches are numerous in proteins (28, 29). For example, agonist binding to G protein–coupled receptors results in conformational changes tens of angstroms away from the binding pocket to initiate G protein activation, thus propagating information from one side of the membrane to the other. In the context of Arf GAP activity, such effects and their dependence on the membrane-bound state of ASAP1–PH could be significant for Arf1 recognition and GAP activity. First, PI(4,5)P2 headgroups with short acyl chains do not stimulate GAP activity (14). Furthermore, we recently reported interactions between the N-terminal segment of Arf1 and the PH domain, which revealed that the I423A mutation, also in the α helix, had no effect on PI(4,5)P2 binding. However, it reduced binding to the N-terminal segment of Arf1, GAP activity in PZA, and activity of full-length ASAP1 in cells. Last, we showed here that the C-terminal α helix interacts with L8KArf•GTP, a mimic for Arf1•GTP. Those interactions, which occur in approximately the same region affected by binding to PI(4,5)P2-containing membranes, point to the functional relevance of that conformational change. It is possible that a shift in the distribution of conformational states of the C-terminal helix is part of the recognition of Arf1•GTP by ASAP1–PH and of its regulation. The molecular details of the allosteric network connecting the membrane and Arf binding regions, particularly the relationship between position 353 and loop structure and penetration, as well as the Arf1 residues at the Arf1/ASAP1–PH interface, are being pursued in other studies.

At the membrane, our NMR and MD data indicate that ASAP1–PH can bind simultaneously to multiple PI(4,5)P2 molecules, via the CA (positively charged residues of the β12 and β34 loops) and NCA sites (positively charged residues of the β5 strand, on the opposite face of the β12 loop in between the β12 and β56 loops). The two positively charged pockets correspond to the two phosphoinositide sites observed in the ASAP1 PH domain crystal structure, solved in complex with diC4-PI(4,5)P2. However, a more physiologically relevant orientation of NCA-bound PI(4,5)P2 with the acyl chains pointed in the direction toward the membrane, as noted previously by others (30), was observed in our simulations.

Binding isotherms derived from NMR data show that the affinity of the PI(4,5)P2 headgroup for the CA and NCA sites is relatively weak but that simultaneous occupancy of the two binding sites increases the overall avidity of the ASAP1–PH domain for a PI(4,5)P2-containing membrane. Recently, analysis of a large number of yeast PH domains revealed that only one of them bound phosphoinositides with high affinity, while six others bound with only moderate affinity and low specificity (31). The ASAP1–PH domain belongs to the latter group. It is likely that binding to the first PI(4,5)P2 places the second binding pocket of the molecule in an environment where the local concentration of PI(4,5)P2 is enhanced, thus facilitating binding with a second PI(4,5)P2 (32). Those reactions would appear to be cooperative, as recently observed experimentally for PZA (8) or the Arf Guanine Exchange Factor Brag2 (33).

Electrostatic interactions between the PI(4,5)P2 headgroup and ASAP1–PH provide the dominant energetic contribution to membrane-binding affinity. Eight basic side chains (K349, K355, R354, R360, K365, H373, R378, and R407) can stably contact the PI(4,5)P2 headgroup at the membrane. Contrary to PLCδ1, hydrophobic interactions at the membrane only add a minor energetic contribution to the binding affinity (34). This is seen in the near equivalence between the affinity of ASAP1–PH for diC4PI(4,5)P2 and the PI(4,5)P2 headgroup at the membrane interface, as well as by the relatively small effect of I353 mutations on binding affinities measured by NMR. In addition, our experiments show that the presence of charged lipid species other than PIPs (in particular PS) in the bilayer potentiates the affinity of the ASAP1–PH domain for PI(4,5)P2. The enhancement is in agreement with previous reports, which indicated that the affinity of the GRP1 PH domain for PIP3 is increased by one order of magnitude in the presence of PS (35). Similar effects have been reported for other PIP-binding proteins that target the inner plasma membrane. For example, the PTEN tumor suppressor has been shown experimentally (23), and in related MD simulations (36), to increase in its affinity to PIP-containing membranes in the presence of PS. For ASAP1–PH, it has been proposed that the NCA site may be occupied by any anionic lipid and PS in particular (8). Our NMR and MD data show that those interactions are not spatially well defined and that PS does not bind specifically to the NCA pocket but rather interacts in a nonspecific manner with the PH domain. The combined absence of specific PS interactions and altered net affinity are consistent with, and explanatory of, prior observations (8) of PS-dependent sigmoidal binding (fig. S9). Nevertheless, the effect of PS binding translates functionally. Combined, these data indicate that, while many PH domains may have a relatively low canonical binding site affinity for PIPs, interaction of multiple PIP molecules and/or other anionic lipid species enables physiologically relevant binding of a PH domain to a membrane. In addition, detection of local clusters of PIP molecules via coincidence sensing may tune the concentration of PH domain at the membrane interface in a cooperative manner.

The revelation of a change of the conformational space explored by a segment of the ASAP1–PH domain, remote from the membrane surface and PI(4,5)P2 binding sites, is perhaps essential to unraveling the complexities of the concerted actions in ASAP1 membrane association and GAP activity on Arf1. The spectroscopically accessible methyl group probes revealed small yet statistically significant differences in the CSPs when the ASAP1–PH domain was bound to a PI(4,5)P2 headgroup at the membrane and in solution, as well as differences in solvent accessibility. Eight of 22 residues of the C-terminal α helix contain labeled side-chain methyl reporters; however, the N-terminal segment (residues 325 to 339), which is important for GAP activity, provides only one reporter position (L332). These structural effects were revealed using L8KArf1•GTP as a mimic for Arf1•GTP. While it is known that L8KArf1•GTP is 10-fold less efficient as a substrate compared to wild-type Arf1•GTP, when the PI(4,5)P2 in the reaction is incorporated into vesicles, mutational studies for residues in the C-terminal α helix support our structural observations. The apparent shift in dynamics seen in the MD needs to be examined via experimental methods, such as NMR relaxation. The sPRE experiments are quite sensitive at detecting the presence of minor states, which would be consistent with the apparent dynamics, and CSPs are indicative of changes; however, more sophisticated and thorough studies are needed to reveal conformational details. Therefore, the present analyses identify the allosteric conformational change, while additional experiments and approaches will be required to probe the N-terminal segment and more accurately depict the conformational shifts within the protein.

Overall, our studies suggest that PI(4,5)P2 acts as a lipid-tethered ligand that binds and activates the ASAP1–PH domain in model membranes for interaction with its substrate. These data highlight the fact that PH domains can be more than membrane-targeting modules and may rationalize the 10,000-fold increase in activity of ASAP1 PZA when bound to PI(4,5)P2-containing membranes compared to solubilized Arf1/PZA complexes (8). Understanding the molecular details of the Arf1/ASAP1 interface will be essential to intervene therapeutically in a rational fashion to block ASAP1-driven cell proliferation in cancer.

MATERIALS AND METHODS

Chemicals

The phospholipids sn1-, sn2-perdeuterated 1-myristoyl-2-myristoyl-sn-glycero-3-phosphocholine 14:0d27-14:0d27-PC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine 16:0-18:1 PC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoserine 16:0-18:1 PS, 1-oleoyl-2-oleoyl-sn-phosphatidylinositol-4,5-bisphosphate [18:1-18:1 PI(4,5)P2], 1-oleoyl-2-oleoyl-sn-phosphatidylinositol (18:1-18:1 PI), 1-oleoyl-2-oleoyl-sn-phosphatidylethanolamine (18:1-18:1 PE), cholesterol, and 1-palmitoyl-2-stearoyl-(5 or 10)-sn-glycero-3-phosphocholine (n-doxyl PC) were purchased from Avanti Polar Lipids. The 10-[(1SR,2RS)-2,3-dihydroxy-1-hydroxymethylpropyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triacetic acid, gadolinium complex (Gadavist) was a gift from N. Tjandra.

Protein expression and purification

Preparation of membrane scaffolding protein belt proteins. The plasmid for MSPΔH5 was a gift from F. Hagn and G. Wagner (Harvard Medical School). The proteins were expressed and purified as previously described (37).

Preparation of PH domain. The sequence for both forms of ASAP1 PH domains, [325-451]–ASAP1 and [339-451]–ASAP1 (~14 kDa), was cloned between Nde I and Bam HI restriction sites of the pET3a vector, which was then transformed into Escherichia coli BL21 Star (DE3) cells (Invitrogen) for protein expression, plated on LB agar containing ampicillin, and incubated overnight (o/n). For the production of [U-2H], [U-15N]-methyl specifically labeled protein, NH4Cl is substituted by ammonium chloride (15N ≥ 99%), d-glucose is replaced by d-(2H, 12C)-glucose (2H ≥ 98%), and 13CH3-methyl specifically labeled precursors are added as described below. For a typical cell culture of 500 ml, a few freshly transformed colonies of BL21 (DE3) cells were picked to inoculate 5 ml of M9/H2O minimal media for o/n growth at 37°C in a shaking incubator (250 rpm). One microliter of the o/n culture [typical optical density at 600 nm (OD600) ~ 1.2] was then used to inoculate 4 ml of fresh M9/H2O medium to achieve a starting OD600 of 0.25. At OD600 ~ 0.5, 5 ml of M9/D2O minimal media was added and cell growth continued until an OD600 of ~0.5 is reached. Cells were diluted again by a factor of 2 and growth followed to OD600 ~ 0.5. This cycle was repeated until a D2O/H2O ratio of 3:1 (20 ml total) is reached. Cells were then harvested by centrifugation (3000g for 30 min) and resuspended in 25 ml of M9/D2O, and growth was continued in a 100-ml baffled flask until an OD600 of 0.5 is reached, before an additional 25 ml of M9/D2O was added for o/n growth at 37°C. When the o/n OD600 was between 1.3 and 1.5, the o/n cell expression (50 ml) was added to 500 ml of M9/D2O and growth followed at 37°C, up to OD600 ~ 0.6. For selective I-[13CH3]δ, L-[13CH3]proS, V-[13CH3]proS, A-[13CH3]β, and T-[13CH3]γ methyl labeling, the PLAM-AβIδ1LVproSTγ kit was used (NMR-Bio). After the addition of the precursor according to the manufacturer’s protocol, cell growth continued until an OD600 of approximately 0.8 at 20°C is reached, at which time protein expression was induced with the addition of 1 mM isopropyl β-d-1-thiogalactopyranoside. After induction, another 2 g/liter of d-(2H, 12C)-glucose was added, and the culture was grown o/n at 20°C. Cells were harvested by centrifugation (6000g for 30 min) and resuspended in buffer A [50 mM tris (pH 7.4), 150 mM NaCl, 0.25 mM tris(2-carboxyethyl)phosphine (TCEP), and one tablet of EDTA-free protease inhibitor (Roche)]. The cells were lysed with a model 110S Microfluidizer (Microfluidics) and centrifuged (50,000g for 45 min, JA25.50), and the soluble protein was purified over a 5-ml SP anion exchange column (GE Healthcare). The column was equilibrated and washed with buffer A, and the protein was eluted using a 20-column volume gradient to 50% buffer B (buffer A and 1 M NaCl). The ASAP1 PH domain was usually eluted at 40 to 45% buffer B. The sample identity was confirmed by mass spectrometry, and purity was assessed by SDS–polyacrylamide gel electrophoresis. Protein was then dialyzed o/n into 20 mM tris (pH 7.4) and 150 mM NaCl before concentration. Concentration of PH domain was determined by ultraviolet (UV) spectroscopy (ε280 = 16,960 M−1 cm−1).

Preparation of PZA domain and L8KArf1. Bacterial expression vectors for His10-[325–724] ASAP1 (i.e., ASAP1-PZA), myristoylated Arf1 (myrArf1), and L8KArf1•GTP were described in (6, 3840). Mutations in ASAP1 PH and PZA were generated using the QuikChange II site-directed mutagenesis kit (Agilent Technologies).

Preparation of NDs

All lipids in chloroform solutions were air-dried with nitrogen flow and resolubilized with cholate in aqueous buffer [20 mM tris-HCl (pH 7.4), 150 mM NaCl, and 75 mM sodium cholate]. NDs were assembled by mixing MSPΔH5 with solubilized lipids at a ratio of 1:45, followed by the removal of sodium cholate from the mixture with Bio-Beads SM2 resin (Bio-Rad), under o/n rocking at 22° and 4°C for NDs containing 14:0d27-14:0d27 PC and 1-palmitoyl-2-oleoyl-sn-glycero3-phosphocholine 16:0-18:1 PC, respectively. Assembled NDs were then purified via a Superdex-200 size exclusion column (GE Healthcare) and concentrated on a centrifugal concentrator [10 kDa molecular weight cutoff (MWCO), Thermo Fisher Scientific]. The concentration of NDs was determined by UV spectroscopy (ε280 = 18,450 M−1 cm−1).

NMR measurements and analysis

All NMR experiments were collected at 25°C on Bruker Avance III 700-, 800-, or 850-MHz spectrometers that were equipped with TCI triple resonance cryoprobes. All spectra were processed using TopSpin 3.6 and NMRPipe (41) and analyzed by NMRFAM-Sparky (42) for the membrane PRE or NMRPipe (41) for the sPRE.

Titration experiments

For titration experiments mapping the PIP binding sites, 500 μM (or 50 μM) ASAP1–PH domain was titrated by diC4-PI(4,5)P2 (or NDs) at a concentration ranging from 500 to 4000 μM (or 10 to 200 μM), and CSPs were followed using a carbon-detected CON experiment ([15N-1H] TROSY HSQC and [13C-1H] HMQC) (43).

For the CON experiment, CSPs were calculated using the following equationCSPCO=(Δv13CO)2

For proton-detected 1H-X titrations, CSPs were calculated usingCSPHX=(Δν H1)2+(A·ΔνX)2where Δν13CO, Δν1H, and ΔνX are changes in the observed backbone 13CO, 1H, or heteroatom chemical shift, respectively. A is a scale factor equal to 0.17 (or 0.185) when X is 15N (or 13C).

The 1H-X combined CSPs for individual residues were then fit to the equation to obtain the dissociation constant (Kd) according toCSPHX=CSPmax{(n[PH]+[ND]+Kd)(n[PH]+[ND]+Kd)2(4n[PH][ND])}(2n[PH])where n is the number of equivalent sites, CSPH-X is the change in the observed combined CSPs, and CSPmax is the maximum shift change at saturation [PH] and [ND], the concentrations of PH domain and NDs, respectively. For curve fitting, CSPmax, n, and Kd were set as free parameters.

Membrane PREs

All samples for PRE experiments were performed with 50 μM U-[15N,2H],13CH3-methyl-ILVAT-labeled ASAP1–PH in the presence of 200 μM ND in D2O buffer, 20 mM tris-D11, and 150 mM NaCl. Methyl-R2 was measured using an HMQC pulse scheme, with pulse field gradients selecting slow-delayed single-quantum component. The experiments were carried out with 2-s recycle delay, 64 scans, 200 complex points in the 13C dimension [20 parts per million (ppm)] and 2048 complex points in the 1H dimension (14 ppm), and a total of five relaxation delays (2.5, 5, 7.5, 10, and 20 ms). For analysis, peak intensities were extracted using NMRPipe and fitted to a mono-exponential decay curve such thatI(t)=A.eR 2twhere I(t) is the peak intensity of the relaxation experiment, t is the recovery delay, A is the amplitude of the magnetization at t = 0, and R2 is the transversal relaxation rate. Duplicate delays were used to determine the error for the fitted rates R2.

Solvent PRE

Methyl R1 was measured using a saturation recovery HMQC experiment with delays between 50 ms and 10 s. A 7.5-ms 1H trim pulse followed by a gradient was applied for proton saturation. Typically, experiments were carried out with 64 scans, 256 complex points in the 13C dimension (20 ppm), and 2048 complex points in the 1H dimension (14 ppm). For analysis, peak intensities or peak volumes were extracted using NMRPipe and fitted to a mono-exponential buildup curve such thatI(t)=A.eR 1t+Pwhere I(t) is the peak intensity of the saturation recovery experiment, t is the recovery delay, A is the amplitude of the z-magnetization buildup, P is the plateau of the curve, and R1 is the longitudinal relaxation rate. Duplicate recovery delays were used to determine the error for the fitted rates R1, and the set of experiments was repeated twice with two different samples (n = 2). The sPRE is obtained by performing a weighted linear regression using the equationR1[Gd3+]=A.[Gd3+]+R1[0]where [Gd3+] is the concentration of Gadavist, R1[Gd3+] is the fitted R1 rate in the presence of Gadavist [Gd3+], and R1[0] is the R1 in the absence of Gadavist. A is the slope of the plot of R1[Gd3+] versus [Gd3+] and represents the desired sPRE value. For the weighted regression, the error determined previously was used, and error on the concentration was neglected.

1H R1 relaxation rates were first measured for the PH domain in solution in the presence of increasing amounts of the paramagnetic probe Gadavist up to 2 mM concentration. There were no detectable 1H and 13C chemical shift changes between the paramagnetic and reference spectra. This indicated that the paramagnetic probe does not interact specifically with the protein and is likely homogeneously distributed around it. This was confirmed by noting that the sPRE values measured at two concentrations of the paramagnetic probe were very well correlated (correlation coefficient r = 0.95; fig. S10), and consequently, the increase in relaxation rates with increasing probe concentration was linear.

GAP assays

PZA-induced conversion of myr-Arf1•GTP to myr-Arf1•GDP was determined as previously described. GTP hydrolysis was plotted against protein concentration, from which the concentration of protein necessary to achieve 50% hydrolysis was estimated. The concentration dependence of GAP activity on PS (Fig. 2D, open circles) was probed by titrating wild-type PZA into a reaction containing 0.1 μM myrArf1•[α32P]GTP and large unilamellar vesicles composed of 1 mol % PIP2, PS between 0 and 30 mol %, and PC, at a total lipid concentration of 500 μM. The effects of I353X mutation on GAP activity (fig. S8B) were probed by titrating the wild-type or the indicated PZA mutants into a reaction containing 0.1 μM myrArf1•[α32P]GTP and large unilamellar vesicles composed of 40% PC, 25% phosphatidylethanolamine, 15% PS, 7.5% phosphatidylinositol, 2.5% PI(4,5)P2, and 10% cholesterol, at a total lipid concentration of 500 μM. Estimates from three experiments with SD are provided.

Neutron reflectometry

Preparation of supported bilayer membranes. For NR, 3″-diameter, 5-mm-thick n-type Si:P[100] wafers (El-Cat Inc., Ridgefield Park, NJ) were used. Substrates for stBLMs were coated with Cr (~20 Å) and Au (~150 Å) by magnetron sputtering (ATC Orion, AJA International, Scituate, MA). They were then immersed in a 7:3 (mol/mol) ethanol solution of HC18 (44) and β-mercaptoethanol at a total concentration of 0.2 mM to form a self-assembled monolayer (SAM). Substrates were assembled in a dedicated fluid cell (45), and vesicle solutions (5 mg/ml of lipid of the desired compositions) were allowed to incubate the dry SAMs for ~2 hours. Afterward, the systems were flushed with pure water to complete stBLM formation. Solid-supported BLMs for the measurement of the membrane-bound [325-451]–ASAP1 structure were directly prepared on clean, bare silicon wafers without the coating and surface passivation steps.

Data acquisition. NR measurements were performed on the CGD-Magik reflectometer at the NIST (National Institute of Standards and Technology) Center for Neutron Research (NCNR). Reflectivity curves were recorded at room temperature for momentum transfer values 0.01 Å−1qz ≤ 0.25 Å−1. The fluid cell allows in situ buffer exchange, and using this capability, series of measurements on the same bilayer under different isotopic buffers (20 mM tris-HCl, 150 mM NaCl, and 5 mM dithiothreitol) in the absence and presence of proteins were performed on the same sample area. After preparation of the stBLM, NR data were sequentially collected with D2O- and H2O-based buffer. Buffer exchange was accomplished by flushing ~10 ml of buffer through the cell (volume ~1.3 ml) using a syringe. Protein dissolved in D2O- and H2O-based buffer at the desired concentration was introduced into the NR cell after characterization of the as-prepared bilayer, and measurements were carried out with protein in contact with the bilayer.

Data analysis. One-dimensional (1D) structural profiles along z of the substrate and the lipid bilayer were parameterized with a model that determines continuous distributions of the molecular components within the available (3D) space (46). Each reflectometry curve measured with a sample of a particular composition and isotopic makeup derives from the scattering length density distribution of that configuration. In our analysis, we process different distributions for a system that are related to each other—for example, isomorphic structures with distinct isotopic makeup—simultaneously and express the results in terms of CVO profiles, which represent space-filling, smooth material density profiles along z (24). This procedure increases the information content of the jointly evaluated measurements significantly compared to that of each individual measurement. Optimization of model parameters was performed using the ga_refl and Refl1D software packages developed at the NCNR (45). A Monte Carlo Markov Chain–based global optimizer was used to determine best-fit parameters and their confidence limits (45).

Protein modeling on the bilayer. The free-form CVO of a protein, defined by a Hermite spline function with control points, on average, 15 Å apart, is seamlessly merged with a representation of the lipid bilayer whose hydrophobic and hydrophilic submolecular components are suitably parsed (46). The extension of the protein along z is determined by the number of spline control points, which were iteratively refined. This procedure determines the protein contribution to the interfacial structure at low resolution, and it may provide the only information if the atomistic structure is unknown or the protein contains large disordered segments.

However, if the atomistic structure of the protein is known, the spline function is replaced with oriented 1D projections of the protein structure, and the most likely orientation and localization along z is inferred by rigid-body modeling (24). After hydrogens were added with MolProbity (47), the crystal structure of the PH domain (PDB: 5C79) was thus used to determine the proteins’ most likely orientation on, and its penetration depth into, the bilayer. The orientation (β,γ) = (0,0) of the 5C79 crystal structure was defined by having the major principal axis of the capping α helix (residues 415 to 435) aligned with the x axis, while the C terminus points toward the positive x direction. Keeping the capping α helix aligned with x, the central axis of the β barrel is aligned with z, the membrane normal, while the capping α helix is further from the membrane than the β barrel.

MD simulations

System setup. Starting from the crystal structure of the ASAP1 PH domain (PDB: 5C79), two constructs were designed: (i) [334-437]–ASAP1 (corresponding to PDB: 5C79) and (ii) [325-451]–ASAP1 (ASAP1–PH) domain, in which we modeled the missing N- and C-terminal regions. PSFGEN plugin of VMD (Visual Molecular Dynamics) (48) was used to add a C-terminal carboxylate capping group, an N-terminal ammonium capping group, and hydrogen atoms. Next, ASAP1 PH domain was placed in a water box using the SOLVATE plugin of VMD. The solvated PH domain was then neutralized with Na+ and Cl ions (0.15 M NaCl) using the AUTOIONIZE plugin of VMD, and the system was energy-minimized for 2000 steps and equilibrated for 1 ns. The final equilibrated ASAP1 PH domain was used for all the consequent membrane-binding simulations.

All membrane-binding simulations were performed using the HMMM model. Multiple independent HMMM membranes were constructed using HMMM BUILDER in CHARMM-GUI (49). In HMMM models, the presence of short-tailed lipids and the use of an organic liquid, 1,1-dichloroethane (DCLE), which mimics the bilayer interior, enhance lipid diffusion and membrane reorganization, thereby allowing spontaneous peripheral protein insertion. This approach has been extensively used to study a variety of peripheral membrane proteins (18, 50, 51). With the aid of this membrane model, we were able to perform multiple membrane-binding simulations of both forms of the PH domain in the presence of mixed-lipid membranes containing PC, PS, and PI(4,5)P2 in varying compositions (table S2).

The membrane-binding simulations started by placing the PH domain in the aqueous solution at least 12 Å away from the cis-leaflet phosphate plane. To ensure that the final membrane-bound conformation of the PH domain is independent of the initial placement, we generated different initial configurations by varying the angle between a vector passing through the β sandwich core and the membrane normal, z. Each replica was simulated for a total of 150 ns. Furthermore, to test the stability of the membrane-bound configuration obtained from HMMM simulations, the resulting membrane-bound replicas were converted to the full membranes and simulated for another 200 ns.

Simulation details. All the simulations were performed in NAMD2 (52) using CHARMM36m protein and lipid force fields and TIP3P water (53, 54). Short-tailed HMMM lipids are best simulated in a fixed area ensemble. All the HMMM simulations were performed with the time step of 2 fs in NPnAT ensembles at 1 atm and 310 K. Constant temperature was maintained by Langevin dynamics with a damping coefficient of 0.5 ps−1 applied to all the atoms, and constant pressure was maintained using the Nosé-Hoover Langevin piston method (55). Nonbonded interactions were calculated with a 12-Å cutoff and a switching distance of 10 Å. Long-range electrostatic interactions were calculated using the particle mesh Ewald method (56).

Analysis. All analysis was performed in VMD. The membrane-binding configuration and depth of PH domain insertion into the lipid bilayer was measured by monitoring the ensemble-averaged z positions of all Cα atoms of the protein with respect to the cis-leaflet phosphate plane over the last 50 ns of HMMM simulations. The PH domain was considered to be membrane-bound if the β12 loop (I353) penetrated below the phosphate plane. A heavy-atom cutoff of 3.5 Å was used to define specific lipid-protein contacts. The distributions of PS and PI(4,5)P2 around the membrane-bound PH domain were calculated using the last 100 ns of the full-membrane simulations.

Statistical analysis. Data are expressed as means ± SD. Representative data from at least two independent experiments were analyzed. Statistical significance was assessed by unpaired Student’s t test.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/6/40/eabd1882/DC1

https://creativecommons.org/licenses/by-nc/4.0/

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

REFERENCES AND NOTES

Acknowledgments: We acknowledge the use of the Biophysics Resource, Structural Biophysics Laboratory, and the assistance of S. Tarasov and M. Dyba. Funding: O.S., Y.Z., J.L., and R.A.B. were supported by the Intramural Research Program of the National Cancer Institute, Projects ZIA BC 011419, ZIA BC 011131, and ZIA BC 011132. N.S.R., X.J., M.E.Y., and P.A.R. were supported by the Intramural Research Program of the National Cancer Institute, Project BC007365. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. F.H. and M.L. were supported by the U.S. Department of Commerce, Award 70NANB17H299. Research was performed, in part, at the National Institute of Standards and Technology (NIST) Center for Nanoscale Science and Technology. Certain commercial materials, equipment, and instruments are identified in this work to describe the experimental procedure as completely as possible. In no case does such an identification imply a recommendation or endorsement by NIST, nor does it imply that the materials, equipment, or instrument identified is necessarily the best available for the purpose. Portions of the research reported in this publication was supported by the NIH under award numbers P41-GM104601 (to E.T.) and R01-GM123455 (to E.T.). We also acknowledge computing resources provided by Blue Waters at the National Center for Super-computing Applications and Extreme Science and Engineering Discovery Environment (grant MCA06N060 to E.T.). S.P. would like to thank the Beckman Institute Graduate Fellowship for funding. Author contributions: O.S., S.P., F.H., Y.Z., J.L., N.S.R., and X.J. performed experiments or simulations and analysis. The research reported emerged from discussions between all authors, and all authors contributed to the writing of the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.
View Abstract

Stay Connected to Science Advances

Navigate This Article