Roles of Amphipathic Helices and the Bin/Amphiphysin/Rvs (BAR) Domain of Endophilin in Membrane Curvature Generation^*

Preparation of Spin-labeled Rat Endophilin A1 Derivatives

Single and double cysteine mutants of rat endophilin A1 were expressed and purified as described previously (5). Proteins were reacted with 5× molar excess of the spin label MTSL (1-oxy-2,2,5,5-tetramethyl-d-pyrroline-3-methyl)-methanethiosulfonate, which generates the spin-labeled side chain R1. Unreacted spin label was separated using PD10 columns.

Liposome Preparation

Folch fraction I, type I, was purchased from Sigma-Aldrich. Lipids were dried with nitrogen gas and desiccated overnight. They were resuspended in 20 mm Hepes, pH 7.4, 150 mm NaCl, treated to brief bath sonication, and extruded through 400-nm pore filters using the Avanti mini-extruder.

Continuous Wave EPR Experiments

2 μm spin-labeled protein was combined with 1.4 mg/ml Folch liposomes and incubated at room temperature for 20 min. After incubation, unbound proteins were separated from membrane-bound proteins by high speed centrifugation at 152,800 × g for 20 min at 22 °C. EPR spectra were then obtained for the membrane-bound portion and also for samples in the absence of liposomes. Spectra were collected using a Bruker EMX spectrometer fitted with a dielectric resonator at 1.59-mW incident microwave power and a scan width of 100 gauss. Power saturation experiments to determine O₂ and NiEDDA accessibility (ΠO₂ and ΠNiEDDA, respectively) have previously been described (20). Accessibility data were obtained using oxygen from air in equilibrium with buffer, and 10 mm NiEDDA. The Φ parameter was calculated by the relationship Φ = ln(Π(O₂)/Π(NiEDDA)). The immersion depth was calibrated using 1-palmitoyl-2-stearoyl-(n-DOXYL)-sn-glycero-3-phosphocholine (Avanti Polar Lipids) as described previously (20). We obtained the following relation between immersion depth (d) and Φ: d[Å] = 6.3*Φ − 3.9. This immersion depth represents the depth of the nitroxide moiety.

Pulsed EPR and Distance Analysis

Samples were prepared by combining 4 μm protein and 1.4 mg/ml Folch liposomes. After incubation for 20 min at room temperature, samples were centrifuged at 152,800 × g for 20 min, and the pellet was recovered for measurement. The pellet was resuspended in buffer, and 20–25% sucrose was used as cryoprotectant before loading into quartz capillaries. Samples were flash frozen, and data were acquired at 78 K. Four-pulse double electron-electron resonance (DEER) experiments were performed using a Bruker Elexsys E580 X-band pulse EPR spectrometer fitted with a 3-mm split ring (MS-3) resonator, a continuous flow helium cryostat (CF935; Oxford Instruments), and a temperature controller (ITC503S; Oxford Instruments). Data were fit using gaussian distributions, as implemented in the DEERAnalysis2008 package (21).

Structural Refinement

The crystal structure of the rat endophilin dimer (Protein Data Bank code 2C08) (5) was used as a starting point for computational refinement based on distance and depth constraints derived from EPR data. A complete description of the details of this procedure is given in the supplemental material. An in-house algorithm (PRONOX) was used to add residues 64–86 to each subunit of the endophilin dimer: add spin labels at positions 63, 66, 70, 73, 74, 75, 77, and 83; locate the starting structure at the surface of an imaginary lipid vesicle; and produce a set of constraints for simulated annealing molecular dynamics (SAMD) calculations. Based on the secondary structure determined by EPR, residues 64–75 were built as helices and constrained using hydrogen-bonding constraints to maintain a helical structure. These constraints, interdimer distance constraints measured from continuous wave EPR and four-pulse DEER EPR, lipid depths for residues 63, 66, and 70, and other system constraints (supplemental Tables S1 and S2) were used in SAMD simulations in AMBER8 (22). During these calculations, the helix and loop were not covalently linked to each other or to the BAR domain to allow freedom of movement according to the distance and depth constraints. Sixty cycles of SAMD were performed starting from a conformation with the helical axes approximately orthogonal to the long axis of the BAR domain. The motion of the insert helices (residues 64–75 in each subunit) was monitored by calculation of the angle between the helical axis and the long axis of the BAR domain (see the supplemental material). The lipid vesicle was built as described previously (23).

Rat Endophilin A1-BAR Is a Dimer upon Membrane Interaction

To investigate whether the BAR domain of the rat endophilin A1 dimer observed crystallographically is retained upon membrane interaction, we generated a series of spin-labeled endophilin derivatives (Fig. 1A) and incubated them with liposomes comprised of Folch lipids. The experimental conditions were chosen such that the Folch liposomes were predominantly vesiculated into highly curved vesicles with an average diameter of ∼ 230 Å, as determined by electron microscopy (supplemental Fig. S1A). We then measured intra- and intersubunit distances and compared them with the respective α-carbon distances of the crystal dimer. The distances measured by EPR are those between the nitroxide moieties of R1, and therefore, they would be expected to be similar but not necessarily identical to the α-carbon distances. Distances >20 Å were investigated using four-pulse DEER (24–27), whereas shorter distances were investigated using continuous wave EPR (28, 29) (see “Experimental Procedures”).

The DEER measurements are illustrated with the example of membrane-bound endophilin spin-labeled in helix 3 at position 216 (Fig. 1A). A pronounced periodic oscillation is observed for this 216R1 derivative (Fig. 1B), and frequency analysis of this oscillation yields a well defined distance distribution with a maximum at 37 Å (Fig. 1C). This distance is in good agreement with the respective intersubunit α-carbon distance of 34 Å in the crystal structure of endophilin (Table 1). Similarly, the intradimer distances of membrane-bound 96R1 (in helix 1b), 63R1 (in the helix 1 insert), and the previously reported (5) distance for 227R1 (in helix 3) are in good agreement with the crystal structure (Table 1 and supplemental Fig. S1, B and C). As a control, we also measured the intersubunit distance for one of the derivatives (96R1) in solution and obtained the same distance as that for the membrane-bound form (supplemental Fig. S1D).

The tip-to-tip distance of the endophilin dimer in the crystal is nearly 130 Å. Because this range is beyond the detection limit for intersubunit contacts, we employed a different approach and measured distances between two sites located toward the tip region in the same subunit. As shown in Table 1 and supplemental Fig. S1B, the distance of 31 Å between 178R1 and 200R1 (located directly at the tip and in helix 3, respectively) is also in good agreement with the α-carbon distances of the crystal structure (33 Å). It has been found previously that the structures of different families of BAR domain proteins overlay well in the central region but are much more divergent toward the tips or wings (15). To test for potential movement between the central region and the wing region, we measured the distance between 166R1 and 210R1 (located in helix 2 and 3, respectively). The distance between these residues for membrane-bound endophilin is in good agreement with the expected distance from the crystal structure (Table 1 and supplemental Fig. S1B). Moreover, the same distance was obtained when distances were measured between these residues in solution (supplemental Fig. S1D). Collectively, this distance analysis supports the notion that the overall structure of the BAR domain dimer is largely retained upon membrane interaction.

Concave Surface of the BAR Domain Does Not Penetrate Deeply into the Acyl Chain Interior of the Membrane

Next, we used mobility and accessibility analysis of selected spin-labeled derivatives to investigate the mechanisms by which the BAR domain interacts with the membrane. Spin labels were introduced at positions that were either located on the concave surface of the BAR domain (positions 96, 159, and 166) or at positions expected to face away from the membrane (positions 108, 172, and 247) (Fig. 1A).

The EPR spectra for these derivatives in the soluble (black trace, Fig. 2) and membrane-bound forms (red trace, Fig. 2) are qualitatively similar and consistent with the respective locations in the crystal structure. For example, the EPR spectra for 247R1 (Fig. 2B) are characterized by three sharp and narrowly spaced lines in both conditions. The high motion indicated by these spectra is consistent with the location of this site at the C terminus of a helix that is likely to be frayed in solution. However, some membrane binding-induced immobilization is detected for 159R1 and 166R1. The immobilization observed at those sites could be caused by a direct interaction with the lipids or other regions of the proteins. We note that both sites could be in proximity to the membrane-embedded N-terminal helix 0 or the linker that connects it to helix 1 (5).

To investigate the proximity of the labeled sites to the membrane bilayer, we used accessibility of the labels to the paramagnetic colliders, O₂ and NiEDDA. Although the concentration of the nonpolar O₂ increases in the membrane, the more polar NiEDDA preferentially partitions into the solvent. Thus, the deeper a spin label penetrates into the acyl chain region, the more it becomes accessible to O₂ and inaccessible to NiEDDA. The accessibilities to O₂ and NiEDDA (ΠO₂ and ΠNiEDDA) are conveniently summarized by the depth parameter Φ (Φ = ln(ΠO₂/ΠNiEDDA). Interestingly, the Φ values for positions 96, 159, and 166 (on the concave face) or 108, 172, and 247 (toward the convex face) were all in the range from −1.2 to −2.1 (Fig. 2). Thus, the depth parameter Φ indicates little membrane penetration by residues on the concave surface of the BAR because the values of Φ do not differ between labels at three such positions and those at solvent-exposed positions. Although these data clearly indicate the lack of significant acyl chain penetration of the nitroxide moiety of the spin-labeled sites, they do not exclude the possibility that the concave surface of endophilin interacts with the interfacial region of the bilayer (see below).

Central Insert Region Becomes an Amphipathic Helix upon Membrane Interaction

To investigate potential membrane-induced conformational changes in the helix insert region, we conducted a nitroxide scanning experiment. The EPR spectra of the soluble derivatives (Fig. 3, black) are generally of higher amplitude with sharper spectral lines than those from the respective membrane-bound forms (Fig. 3, red). Control experiments, in which sucrose was used to slow down the tumbling of the dimer in solution, showed little spectral changes, indicating that the membrane binding-induced change in tumbling rate did not play a significant spectral role (data not shown). The line shapes of 70R1 to 77R1 indicate a largely disordered and unfolded structure in solution, whereas a more ordered structure is induced upon membrane binding. Only residues 64–69 are ordered in solution, and pronounced immobilization can be seen for 64R1, 65R1, 68R1, and 69R1. Such immobilization could be caused by tertiary contacts. This region was largely disordered in the crystal structure of rat endophilin (5) but was resolved in the structure of human endophilin (8), in which residues 63–70 were found to form a short helical segment. Moreover, residues 64, 65, 68, and 69 are pointing directly toward the BAR domain in the latter structure. Thus, the EPR spectra in the absence of membranes are consistent with a structure akin to that in the crystal of human endophilin (also see below). Residues 64–69 are also the most ordered region of the membrane-bound form, but in the presence of membranes the ordered region becomes much more extended.

To obtain more detailed secondary structure information in the membrane-bound form, we performed O₂ and NiEDDA accessibility measurements for 63R1 to 77R1. As summarized with the depth parameter Φ (Fig. 4A), the accessibility data give rise to a periodic oscillation. The local Φ-maxima (residues 63, 66, 70, 73, and 74; magenta) represent preferentially O₂ accessible, lipid-exposed sites, and these positions cluster on the hydrophobic face of a helical wheel (Fig. 4B). Conversely, the local minima (green) correspond to solvent-exposed sites that fall onto the opposite face of the helical wheel. The formation of an amphipathic α-helical structure is further supported by a comparison of the Φ-data with a cosine curve that has the standard periodicity of an α-helix (3.6 amino acids/turn). This curve agrees well with the experimentally observed periodicity from residues 63 to 75, but it starts to diverge for residues 76 and 77. For these residues, an increase in the Φ-values would be predicted, but the opposite is observed. Thus, these residues may no longer be part of the helix.

Previous studies demonstrated that Φ-values are proportional to the membrane immersion depth of R1 side chains that are penetrating into the acyl chain region. This depth can be calibrated using spin-labeled derivatives of phospholipids (20, 30). Based on such a calibration (see “Experimental Procedures”), we obtain average immersion depths of the nitroxide moieties of 63R1, 66R1, and 70R1 that are on the order of 5–6 Å. Previous work has shown that an R1 side chain is typically at a distance of 7–10 Å from the center of an α-helix to which it is attached (31). Thus, we can estimate that the center of the helix is located about 1–4 Å above the phosphate layer. The Φ-values for residues 73 and 74 are somewhat lower, and the reduced immersion depth of these sites is consistent with a slight helix tilt. However, residues 73 and 74 are also located on the periphery of the hydrophobic face (Fig. 4B) where the side chains might project laterally rather than directly toward the bilayer interior.

Dimeric Amphipathic Helices in the Central Insert Region Are Antiparallel

To analyze the orientation of the amphipathic helices with respect to each other, we measured intersubunit distances between the same sites in each subunit using DEER (supplemental Fig. S2). As shown in Table 2, the distance between these sites generally increases with increasing residue number, indicating that the helices must be facing away from each other. This feature is best illustrated with the lipid-exposed sites 63R1, 66R1, and 70R1. For these sites, we can assume that the nitroxide side chains are facing in the same direction (i.e. into the membrane) and that the interlabel distances are likely to be close to those between the respective α-carbon atoms. An interesting feature of the DEER distances is that they increase by about 10 Å for each helical turn, which is in good agreement with antiparallel helices because an individual helix is extended by ∼5 Å/turn. The other distances in Table 2 are also in agreement with a helical structure (also see below).

Since Masuda and colleagues (8) observed a short antiparallel helical structure for residues 63–71 in the crystal structure of human endophilin A1, we sought to determine whether the distances in solution are similar to those in the membrane-bound form. As shown in supplemental Table S3, the intersubunit distances for 63R1, 67R1, and 70R1 in the absence of membranes differ from those observed in the membrane-bound state. Compared with the α-carbon distances from the crystal structure, the measured distances in solution are in good agreement. These data suggest that the structure of the insert taken up in the crystal of human endophilin A1 is similar to that of rat endophilin A1 in solution and that the structure taken up in the membrane-bound state is different.

Structural Refinement of the Central Insert Suggests Staggered Helices with High Mobility

To generate atomistic models of the membrane-bound form of endophilin, we used the crystal structure of the soluble rat endophilin A1 dimer as a starting structure and built and refined the structure of the insert region (residues 63–86) using a modified version of our previously reported (23) SAMD refinement constrained by the EPR results. These calculations included intersubunit constraints based on DEER distances, depth constraints based on EPR accessibility data, constraints to maintain the helicity of the insert region as detected by EPR mobility and accessibility data, and symmetry constraints on the insert regions of the two subunits. The intersubunit constraints were implemented as the mean distance ± 5 Å to reflect the relatively large width of the EPR distance distributions.

Endophilin structures were collected at the end of each of 60 SAMD cycles. The insert helices clearly adopted one of three positions with an angle between the helical axis and the long axis of the BAR domain of about 140°, 90 °, and 40°, respectively (supplemental Fig. S3), with 18, 33, and 9 occurrences, respectively, over the 60 SAMD cycles. The three positions of the insert helices are interconvertible, and all may occur in a population of membrane-bound endophilin proteins. This may account for the width of the distance distribution for the intersubunit DEER data. However, in the SAMD cycles shown in supplemental Fig. S3A and in other calculations performed with minor changes in parameters (see the supplemental material), we consistently find that the most common conformation is that with the insert helices orthogonal to the BAR domain axis (green helices in supplemental Fig. S3B).

To examine the interaction of the endophilin dimer with a curved lipid surface, we used a representative structure with orthogonal insert helices and constructed part of a lipid vesicle around the protein (Fig. 5). We were able to do this because the predicted structure is positioned in space based on depth constraints relating the center of the lipid vesicle to the protein. Two main findings for the insert helices emerge from the SAMD calculations and subsequent building of the lipid surface. First, the helices (residues 63–75) are tilted such that the N terminus of each helix is located below the phosphate of the lipid head group, whereas the C terminus and the loop (residues 76–86) are positioned above the lipid surface (Fig. 5B). Second, the axes of the helices adopted a staggered position with respect to each other (Fig. 5A).

The detailed positions of amino acids 63–75 are shown in Fig. 5C. The amino acid positions obtained by computational refinement using EPR distance, depth and mobility data, and geometrical and symmetry constraints show remarkably good agreement with the predicted orientations based on accessibility data only (Fig. 4). Hence, Ala-63, Ala-66, Met-70, Thr-73, and Met-74 are oriented into or toward the hydrophobic region of the lipids. All of the remaining side chains are positioned above the phosphate head groups and may be considered in three categories. First, Arg-65 and Lys-67 are oriented sideways from the helix and can interact with negative charges on the lipid head groups. Second, three serine residues and Asn-72 are either solvent-exposed or hydrogen-bonded to head groups, and these residues produce an element of amphipathicity in the helix. Third, Leu-68 and Ile-71 are two hydrophobic residues that apparently break this amphipathicity. However, the side chain of Leu-68 forms hydrophobic contacts with Ile-55 and Leu-58 in the BAR domain. This interaction, as well as the interaction between Ile-71 and Tyr-86 may help to stabilize the orthogonal orientation of the insert helices (supplemental Fig. S4).