Myristoylation Restricts Orientation of the GRASP Domain on Membranes and Promotes Membrane Tethering

Background: An unknown mechanism promotes trans interactions by the GRASP homotypic membrane tethers rather than unproductive cis interactions.

Results: Neutron reflection shows that the myristoylated GRASP domain has a fixed, upright orientation on the membrane incompatible with cis interactions.

Conclusion: Myristoylation restricts the orientation of the protein on the membrane to favor interactions in trans.

Significance: Orientation of membrane proteins is functionally significant and may be regulated by myristoylation.

Abstract

The mammalian Golgi reassembly stacking protein (GRASP) proteins are Golgi-localized homotypic membrane tethers that organize Golgi stacks into a long, contiguous ribbon-like structure. It is unknown how GRASPs undergo trans pairing given that cis interactions between the proteins in the plane of the membrane are intrinsically favored. To test the hypothesis that myristoylation of the self-interacting GRASP domain restricts its orientation on the membrane to favor trans pairing, we established an in vitro assay that recapitulates GRASP-dependent membrane tethering and used neutron reflection under similar conditions to determine the orientation of the GRASP domain. In vivo, the membrane association of GRASP proteins is conferred by the simultaneous insertion of an N-terminal myristic acid and binding to a Golgi-associated binding partner. In our assay, the latter contact was replaced using a C-terminal hexa-His moiety, which bound to Ni²⁺-conjugated lipids incorporated into a substrate-supported bilayer lipid membrane. Nonmyristoylated protein lacked a fixed orientation on the membrane and inefficiently tethered liposomes. In contrast, myristoylated GRASP promoted tethering and exhibited a unique membrane complex. Thus, myristoylation restricts the membrane orientation of the GRASP domain favoring interactions in trans for membrane tethering.

Introduction

Proteins interact to bridge two adjacent membranes in many important cellular processes including cell-cell adhesion, virus-membrane attachment, vesicle docking, and organelle fusion (1–4). Such cross-bridging may be either homotypic or heterotypic depending on whether the process involves the pairing of self-interacting molecules or the pairing of distinct components on each of the apposed membranes. Uniquely, in these types of molecular interactions, the effectiveness of pairing is governed by the relative propensity of the proteins to form interactions in a trans configuration across the membranes, versus binding in cis in the same membrane. Particularly in the case of homotypic cross-bridging where cis interactions may predominate, a mechanism is needed to suppress such cis pairing.

An important example of homotypic tethering is the interaction of the GRASP⁴ proteins that underlies stacking Golgi cisternae and/or linking the stacks into the elongated, ribbon-like membrane network evident in mammalian cells (5–9). GRASP65 is localized to cis Golgi cisternae, and GRASP55 is localized to medial and trans cisternae. The immediate effect of inhibiting GRASP65 is a loss of lateral continuity of cis cisternae, whereas inhibiting GRASP55 first disrupts trans cisternal continuity (10). Each GRASP protein contains two noncanonical PDZ domains and an internal PDZ ligand that protrudes from the surface of the molecule (11). Self-association occurs via the intermolecular insertion of the ligand into the binding groove of the first PDZ domain (12).

The GRASP proteins are anchored to their respective cisternal membranes by insertion of an N-terminal myristic acid and interaction of the second PDZ domain, which is positioned near the C terminus of the GRASP domain, with a membrane-bound partner protein (13, 14). Such dual anchoring of GRASP is required not only for its stable membrane association but also for its membrane-tethering activity. Stable, singly anchored versions of GRASP proteins have been created using transmembrane domains but, when expressed in cells, they fail to tether, whereas similar dually anchored constructs retain full activity (12, 15). The singly anchored constructs are recovered in complexes, suggesting that they form cis interactions within the same membrane that are unproductive for tethering (15).

The relative positions of the internal ligand and the PDZ1 binding groove on the molecule (11) suggest that GRASP orientation on the membrane could provide a mechanism that favors homotypic interactions in trans over those in cis (15). However, for such a mechanism to work, the orientation of the protein with respect to the membrane needs to be tightly controlled, raising the possibility that N-myristoylation, as a second anchor point for the protein on the membrane, could set this orientation. Fig. 1 shows possible GRASP orientations on the membrane and their implications for forming trans or cis interactions. Many orientations will not prevent cis interactions from occurring, whereas some that do are fundamentally different but cannot be discriminated in functional assays.

FIGURE 1.

Possible orientations of the GRASP domain, composed of PDZ1 (blue) and PDZ2 (red), and their implications on the protein's predilection for trans membrane tethering or unproductive cis protein binding. The GRASP crystal structure shows that the groove on PDZ1, which accepts the internal ligand located on PDZ2, is on the opposite face of the GRASP domain from the internal ligand on PDZ2 (11). A, GRASP may lie flat on the membrane with the PDZ1 groove oriented coplanar while two membrane anchors keep the binding groove and the internal ligand rigidly in an orientation such that cis interactions become unlikely, as shown on the right of the lower membrane. B, a similar restriction to cis interactions would be achieved by an orientation in which GRASP sits upright on the membrane with one PDZ domain membrane bound and the other located distal from the membrane. The PDZ1 groove is oriented roughly coplanar, but cis interactions are unlikely because the PDZ groove and internal ligand are located at different distances from the membrane surface, and misoriented with respect to each other (right side on lower membrane). The membrane-membrane distance in this scenario is expected to be larger than for the GRASP orientations shown in A. C, if one of the two membrane anchors is compromised, the increased orientational flexibility of the GRASP domain leads to many other mutual orientations of the membrane-bound protein that will not discriminate between cis and trans binding, and hence result in many unproductive cis ligations. C-term, C terminus; N-term, N terminus.

Because orientation is difficult to determine in vivo, we developed an in vitro assay of GRASP-mediated tethering on sparsely tethered bilayer lipid membranes (stBLMs) where protein orientation can be determined with neutron scattering. The results show that the GRASP domain is sufficient for membrane tethering and that myristoylation is required for full activity. Neutron reflection revealed decisive differences in GRASP membrane orientation conferred by myristoylation, yielding a model of how orientation promotes trans-membrane tethering.

EXPERIMENTAL PROCEDURES

Materials

1,2-Dioleoyl-sn-glycero-3-phospho-choline (DOPC),1,2-dioleoyl-sn-glycero-3-phos-phoethanolamine-N-(lissamine rhodamine B sulfonyl) ammonium salt (LR-PE), 1,2-dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxy-pentyl)iminodi-acetic-acid)succinyl] nickel salt (Ni²⁺-NTA-DGS), and a liposome extruder were from Avanti Polar Lipids (Alabaster, AL) and used as prescribed. The tether lipid for supported membranes, Z 20-(Z-octadec-9-enyloxy)-3,6,9,12,15,18,22-heptaoxatetracont-31-ene-1-thiol (HC18) (17), was a generous gift from Dr. David Vanderah (Institute for Bioscience and Biotechnology Research, Rockville, MD). β-Mercaptoethanol was from Sigma-Aldrich. No. 1.5 glass coverslips were from Electron Microscopy Science (Hatfield, PA), and microscopy glass slides were from Fisher Scientific. All other chemicals were of HPLC grade. Ultrapure water was prepared in a Millipore (Billerica, MA) Milli-Q water purification device.

Constructs

GRASP55 was cloned into the pRSET-B vector by PCR amplification into the EcoRI site. Subsequently, the N-terminal His tag was deleted by a PCR loop-out procedure and a 10-amino acid linker, ASSRSGGSGA, after which a hexahistidine sequence was inserted after residue 208 followed by a stop codon to create G55, i.e. GRASP55(1–208)-His₆. The His₆-GRASP55(1–208) construct was made by PCR deletion of the His tag from GRASP55(1–208)-His₆ and reintroduction immediately before the starting methionine. Mutations of the pocket and ligand (L59A/I100S+Δ196–199) and of the phosphorylation site (S189D) were generated exactly as described previously (12, 18). The E157R mutation was subsequently introduced into the S189D construct using QuikChange according to the manufacturer's protocol (Stratagene, La Jolla, CA). All constructs were confirmed by sequencing. Protein was induced and myristoylated as described previously (12) to yield myrG55. Buffers used for protein incubation of interfaces in surface plasmon resonance (SPR), fluorescence microscopy, and neutron reflection (NR) contained 50 mmol/liter K₂HPO₄, adjusted to pH 7.45. The mitochondrial tethering assay was as described previously (12, 18).

Formation of stBLMs

stBLMs were formed on solid substrates as described previously (19, 20). The substrates, No. 1.5 glass coverslips or microscope glass slides for SPR and microscopy, respectively, and 3-inch (75-mm) silicon wafers (El-Cat Inc., Ridgefield Park, NJ) for NR, were thoroughly cleaned as described (21) and metal-coated by magnetron sputtering (ATC Orion, AJA International, Inc., North Scituate, MA). For fluorescence microscopy, ∼6 nm of gold was sputtered on top of an ∼2-nm thick chromium bonding layer, which leaves the coverslips optically transparent. For SPR, ∼45 nm of gold was sputtered on the chromium bonding layer. Gold layers on chromium for NR were 10–15 nm thick. Freshly gold-coated substrates were immersed in an ethanolic solution of 30% HC18 and 70% β-mercaptoethanol, at a total concentration of 0.2 mmol/liter, to form a mixed self-assembly monolayer. Rapid solvent exchange (19, 22) or vesicle fusion was used to complete the formation of stBLMs.

Liposome Preparation

DOPC, together with 0.74 or 2.2 mol % LR-PE or 0.125 mol % Ni²⁺-NTA-DGS as needed, was dissolved in chloroform in a vial and placed in vacuum to remove the solvent. The dry lipid films were redissolved in pentane and dried under vacuum again and then rehydrated in 50 mmol/liter potassium phosphate buffer at pH ∼7.45 and sonicated. The resulting liposome suspensions were extruded through polycarbonate membranes with a pore size of 100 nm, and liposome sizes were checked by dynamic light scattering (Malvern Zetananosizer, Malvern Instruments Ltd., Malvern, UK). Alternate liposome sizes were also investigated (extrusion through 30- and 200-nm pore sizes; data not shown). 100 nm-extruded liposomes were used throughout the experiments shown in the study.

SPR

stBLMs formed on gold-coated microscope glass slides were mounted on a custom-made SPR instrument (SPR Biosystems, Germantown, MD) (21) in the Kretschmann configuration. A convergent fan of light (λ = 763.8 nm) was reflected from the backside of the sample-covered interface and recorded by a CCD chip, and the position of the intensity minimum was determined as a function of the reflection angle. Shifts in the resonance were used to quantify protein binding to the stBLM (21) after injection into the sample chamber. Resonance shifts as a function of protein concentration were fitted to a Langmuir isotherm to determine the dissociation constants, K_d.

Fluorescence Microscopy

stBLMs formed on gold-coated glass coverslip were assembled in a Sykes-Moore chamber. Subsequently, 1.2 μmol/liter protein and 0.01 mg/ml liposomes were injected into the chamber. The solution was occasionally mixed and left to incubate the stBLMs for 3 h before excess protein and liposomes were rinsed away. The chamber was then transferred to a Carl Zeiss Axiovert 200M microscope equipped with a Zeiss 63× Plan-Apochromat (NA = 1.4) objective lens, a 100-watt HBO 103 W/2 mercury lamp, a Zeiss filter set at BP546/12:FT560: BP665/20, and a Hamamatsu C9100-12 EM-CCD camera (Hamamatsu, Bridgewater, NJ) and controlled with IPLab (BioVision Technologies, Exton, PA). Images of LR-PE labeled liposomes adhering to the stBLMs were taken with ∼300-ms exposure time at low excitation light intensity with the aperture stop closed to its minimum size. These image acquisition parameters were chosen to minimize the number of saturated pixels in the images. For each sample, at least 12 images (size of field, 100 × 100 μm²) were taken at random positions in the sample. The fluorescence intensity of these images after background subtraction was used as a measure to quantify the tethering activity of the protein constructs.

Neutron Reflection Measurements and Data Modeling

NR measurements were performed on the NG7 and MAGIK reflectometers at the NIST Center for Neutron Research and evaluated with garefl (23) using the GRASP55 crystal structure (Protein Data Bank (PDB): 3RLE), similarly to studies of other membrane-associated proteins (21, 24). To determine the protein orientation on the membrane, the origins of the two relevant Euler angles (see Fig. 4C, inset) were chosen to coincide with the longest (ϑ₀) and second longest (ϕ₀) principal axes of the mass distribution in the crystal structure. Parameter space was efficiently searched with a Monte-Carlo Markov Chain algorithm, and provided quantitative measures of parameter and neutron scattering length density confidence limits, as well as parameter cross-correlations (25). The crystal structure used in the refinement was reported for a truncated protein, GRASP55(7–208), and also lacks the N-myristoylation and the C-terminal decapeptide and His tag. Therefore, this structure needed to be supplemented to represent the protein used in the experiments. Peptide conformations were obtained from molecular dynamics (MD) simulation. Because the protein orientation on the membrane was initially unknown, a first, free MD simulation of the full-length proteins in buffer sampled mostly unrealistic peptide conformations. However, this procedure produced an approximate full-length protein model that could be used to gain a first estimate of the protein orientation on the membrane through refinement to the NR data. Based on this orientation, we subsequently modified the model using a steered MD simulation to implement constraints of the His tag and the N-myristoylation reaching to the membrane surface. This refined model was then used to fit the NR data in a second round of modeling. In practical terms, both models yielded the same protein orientation and distance to the bilayer within confidence limits, and this structural protein model was used to determine a unique component volume occupancy profile as a parent of the neutron scattering length density profiles for different isotopic samples.

FIGURE 4.

Complex structure of the GRASP55 GRASP domain with a membrane. A and B, fractional CVO profiles of stBLMs composed of DOPC and Ni²⁺-NTA-DGS (9:1) on HC18/β-mercaptoethanol (1:3) with 6.5 μmol/liter myrG55 (A) and 6 μmol/liter G55 (B). CVO profiles of the substrate (terminal gold film), oligo(ethylene oxide) tether, lipid bilayer components, and protein, indicated in the legend in A, are decomposed from multiple NR measurements. Protein CVO profiles determined from spline fits (red lines) and from atomistic models (black dashed lines) are shown in the same graphs for comparison. Lipid bilayer CVO profiles shown are those determined from the free-form fits and are very similar in the atomistic models. Dashed red lines indicate 68% confidence intervals for the free-form protein CVO profiles. C, probability distribution, determined from Monte Carlo data resampling (25), of myrG55 orientations (ϑ,ϕ) with respect to the bilayer normal. Only the half-sphere 0 ≤ ϑ ≤ 90° is shown. The best fit is observed at (ϑ,ϕ) = (26°,78°). The inset shows the definition of the Euler angles ϑ and ϕ. D, penetration depths of myrG55 into the bilayer at the orientations shown in panel C. The membrane surface is defined as the half-point in the decay of the lipid headgroup CVO toward bulk buffer (brown traces in panels A and B at d ∼50 Å). E, orientation of myrG55 on the bilayer that corresponds to the best fit, (ϑ,ϕ) = (26°,78°), in the probability distribution of panel C. The long principal axis of the GRASP55 GRASP domain, ϑ₀ = 0°, is shown as a blue column. The internal ligand on the PDZ2 domain is shown in volume-filling representation, and the PDZ1 binding groove is shown as a green surface. The orientations of the PDZ2 ligand and a putative ligand bound to the PDZ1 groove, indicated as yellow cylinders, are roughly collinear. Approximate positions of the C- and N-terminal membrane binding sites, i.e. the Ni²⁺-ligated His tag and the myristoyl group, are shown in purple and maroon, respectively. F, probability distribution of G55 orientations (ϑ,ϕ) with respect to the bilayer normal on the half-sphere 0 ≤ ϑ ≤ 90°.

Molecular Modeling

Missing amino acids on the N and C termini of GRASP were added to the crystal structure in an initial fully extended configuration. The patched protein structure was modeled using the Generalized Born Implicit Solvent model implemented in NAMD 2.8 (26). A solvent dielectric of 78.5 and an ionic concentration of 0.1 mol/liter salt were used in the Generalized Born Implicit Solvent model. The resulting GRASP structure was minimized, first with the core protein held fixed and only the newly constructed termini allowed to vary. Then, in a second step of minimization, the full structure was allowed to vary. Further equilibration was performed with 10 ns of MD simulations followed by a 10-ns production run at 300 K. During both equilibration and production simulations, the backbone of the core protein was held fixed while the side chains were unrestrained. Both the backbone and the side chains of the added terminal residues were unrestrained. CHARMM22 with CMAP dihedral correction (27) was used for the protein force field parameters, and the CHARMM36 parameter set (28) was used for the myristate group on the N terminus. Frames were collected every 100 ps from the production run and averaged to generate the volume and scattering length profile of the GRASP protein that was then used for reflectivity fitting. Steered MD was performed on the N and C termini using the SMD module in NAMD. An external force was applied to the N-terminal Gly residue and the center of mass of the C-terminal His₆ tag. The force was directed downward along the z axis toward the putative membrane interface, resulting in a constant migration velocity of 0.0025 Å/fs.

RESULTS

GRASP Domain-mediated Liposome Tethering to Supported Membranes

To quantify the tethering activity of the GRASP domain in vitro, we developed a fluorescence-based assay that reports the tethering of liposomes to a supported planar bilayer. The aforementioned stBLMs feature a nanometer-thick hydrated layer between the solid surface and the lipid bilayer (19, 25) that keeps the bilayer in a physiologically relevant in-plane fluid state (29). We used purified preparations of the N-myristoylated and unmyristoylated GRASP domain of GRASP55 (residues 1–208). The resulting proteins, myrG55 and G55, included a 10-residue linker and C-terminal hexahistidine tag so that each protein bound to nickel (Ni²⁺) associated with N-nitrilotriacetic acid (NTA)-functionalized lipids present in both the stBLMs and the liposomes in the assay. The C-terminal position of the His tag was meant to roughly approximate the position of the PDZ2 binding groove, which is near the C terminus of the GRASP domain and is normally responsible for the secondary site of membrane contact.

First, we determined the affinity of myrG55 to DOPC stBLMs containing 1 mol % Ni²⁺-NTA-DGS using SPR. The reaction yielded a dissociation constant, K_d, of 0.32 ± 0.09 μmol/liter (data not shown) comparable with binding constants measured for other His-tagged proteins to Ni²⁺-NTA-DGS (30, 31). The extrapolated (c_protein → ∞) equilibrium SPR signal corresponded to a protein mass density of ∼68 ng/cm², as calibrated with neutron reflection.⁵ The background of nonspecific protein binding, quantified with myrG55 on DOPC stBLMs devoid of Ni²⁺-NTA-DGS, was <10% of the signal with Ni²⁺-NTA-DGS. This is consistent with the relatively weak affinity of myristate for the membrane, K_d ∼100 μmol/liter (32), and with the inability of myristoylation alone to localize the GRASP proteins to membranes in cells (13, 15).

Next, we tested whether the GRASP domain construct was sufficient to tether vesicles to stBLMs. Liposomes, labeled with 0.125 mol % fluorescent LR-PE, and stBLMs were prepared from DOPC and varying amounts of Ni²⁺-NTA-DGS. Liposomes (0.01 mg/ml) and myrG55 were incubated with stBLMs for 3 h after which excess protein and liposomes were rinsed off and the fluorescence intensity was determined at the interface (Fig. 2A). The background signal of the assay was established with stBLMs and liposomes (both with 2.2 mol % of Ni²⁺-NTA-DGS) incubated without protein (Fig. 2A, left). With 1.2 μmol/liter myrG55, tethered liposomes (Fig. 2A, right) were observed with an intensity of >10× background that increased with Ni²⁺-NTA-DGS concentration (Fig. 2B). A low level signal, ∼4× background, was also observed in the presence of myrG55 when only liposomes, but not the stBLMs, contained Ni²⁺-NTA-DGS. We attribute this increase in background over that measured without protein to GRASP-mediated tethering of liposomes in solution, resulting in higher fluorescence intensity upon nonspecific adsorption to the stBLMs. Indeed, large liposome aggregates, tens of μm across, were evident in the liposome suspension containing myrG55 after 3 h of incubation. Flushing tethered liposomes with 200 mmol/liter imidazole decreased the fluorescence intensity by a factor of ∼3 (Fig. 2B), demonstrating further that liposome tethering is contingent on membrane-associated GRASP. Under the conditions of the experiment, imidazole may have had incomplete access to the preformed complexes accounting for the residual activity. Overall, the assay showed conclusively that the isolated GRASP domain is capable of tethering membranes in the absence of any other factors.

FIGURE 2.

Quantification of the GRASP55 GRASP domain tethering activity in an in vitro assay. A, tethering of DOPC liposomes, doped with 0.125 mol % LR-DOPE, to HC18/DOPC stBLMs. Both liposomes and stBLMs contained 2.2 mol % Ni²⁺-NTA-DGS. Right, liposomes tethered by myrG55 in a solution that contained 1.2 μmol/liter protein. Control (left) shows only a few vesicles nonspecifically absorbed to an stBLM in the absence of protein under otherwise identical conditions. B, interfacial fluorescence intensities of GRASP-tethered liposomes (1.2 μmol/liter myrG55), read out from the assay shown in A, at various concentrations of Ni²⁺-NTA-DGS as indicated (central four columns). The first and the last columns show background (no protein) and remaining intensity after an imidazole (200 mmol/liter) wash, respectively. AU, arbitrary units. C, interfacial fluorescence intensity of tethered liposomes as in B, incubated in solutions that contained wild-type myrG55 or various mutants of this construct (see “Results”). D, relative tether efficiencies of the GRASP55 GRASP domain as a function of membrane anchor placement. Data in panels A–D are mean ± S.E. for n ≥ 3 independent samples. Asterisks indicate one-tailed p values < 0.01.

Structural Requirements of GRASP Domain-mediated Tethering

To explore the mechanism of GRASP-mediated tethering, we used the in vitro assay to test three specific hypotheses that emerged from previous work carried out in cells. The first is that GRASP proteins tether Golgi membranes through an intermolecular self-interaction in which the internal ligand (residues 196–199) binds to the PDZ1 binding groove (11). Therefore, we mutated the GRASP domain construct by introducing known blocking substitutions into the PDZ1 binding groove (L59A/I100S) and deleting the residues of the internal ligand (Δ196–199). This construct was strongly blocked in its tethering activity (Fig. 2C, L59A/I100S+Δ196–199), suggesting that the previously identified binding interface mediates tethering in an assay devoid of other components.

The second hypothesis we tested is that mitotic phosphorylation of Ser-189 inhibits tethering by causing a conformational change in the internal ligand (33). Because the conformational change is evident in a crystal structure of the GRASP domain containing a phosphomimetic mutation, S189D, we tested this mutation in the assay and found that it significantly impaired tethering (Fig. 2C, S189D). Phosphorylation at Ser-189 is thought to disrupt an electrostatic contact between Ser-189 and Glu-157, thereby causing a rotation of the α2 helix that propagates to the ligand (33). Thus, in the double mutation S189D/E157R, the placement of Arg should re-establish, in the S189D background, the contact between residues 189 and 157 in wild-type GRASP, thereby reversing the conformational change and rescuing tethering activity. Indeed, the tethering activity of this construct was nearly restored to that of the wild-type protein (Fig. 2C, S189D/E157R). To confirm that these mutations similarly affect tethering activity in vivo, we used a previously characterized assay in which GRASP is targeted to mitochondria and the resulting tethering of the mitochondria is measured by the cellular fluorescence distribution (12). Indeed, despite being stably targeted to mitochondria, a version of full-length GRASP55 containing the S189D substitution was significantly impaired when compared with the corresponding wild-type version, and the double substitution S189D/E157R reversed this inhibition (Fig. 3).

FIGURE 3.

GRASP-mediated membrane tethering of mitochondria. A, HeLa cells expressing GFP-labeled proteins containing the ActA mitochondrial targeting sequence were analyzed 24 h after transfection. Cells were imaged using GFP fluorescence to identify transfected proteins. Representative GFP fluorescence patterns are shown for wild-type GRASP55, GRASP55 containing the S189D or S189D/E157R substitutions, and a control construct containing only the targeting signal (GFP-ActA). B, fluorescence distribution was quantified by radial profile analysis, which reports averages corresponding to the fraction of total fluorescence present in each concentric circle drawn from the centroid. Note that tethering activity is indicated by the degree of fluorescence near the centroid (n = 3, ± S.E., ≥15 cells per experiment).

The final test concerned N-myristoylation of GRASP, which is thought to be important not only for membrane targeting but also for membrane tethering (12, 15). Therefore, we compared the tethering activity of myrG55 with that of G55. Consistent with previous work in cell-based assays (12, 15), tethering activity was significantly lower in the absence of myristoylation (Fig. 2D, His tag at the C terminus). Because myristoylation is not essential for membrane anchoring of the His-tagged construct, this result suggests that N-myristoylation is required for efficient tethering. It is noteworthy that membrane association of the N terminus alone was ineffective as tethering activity remained at a reduced level if the His tag was moved from the C terminus to the N terminus (Fig. 2D, His tag at the N terminus). Thus, full tethering activity depends critically on simultaneous membrane anchoring of both termini of the GRASP domain, consistent with previous in vivo work.

Structure of the GRASP Domain Complex with the Membrane

To correlate the functional requirements for membrane tethering with structural details of the GRASP domain complex with a membrane, we studied the latter with NR, a technique that can determine the orientation of membrane-bound proteins of sufficient shape anisotropy within tight limits (34) and their localization along the membrane normal with Angstrom precision (24). The NR experiment involved the measurement of the as prepared and the protein-decorated stBLM, each under buffers of distinct isotopic compositions that contained ¹H₂O and ²H₂O in different proportions. Because ¹H₂O and ²H₂O differ in neutron scattering lengths, regions of the surface structure that contain water contribute differently to the scattering in isotopically distinct buffers, thereby facilitating the characterization of such regions within the sample. Neutron scattering is nondestructive, which allows all measurements to be conducted on the same sample area under identical conditions (19, 35). These measurements are then evaluated in reference to each other (24, 25) to obtain a unique, highly resolved model of the protein-membrane complex. Two complementing approaches to data modeling were employed. In both methods, the lipid membrane was described as continuous distributions of submolecular fragments (36). The contribution of the membrane-bound protein to the neutron scattering length density at the interface was modeled either as a sequence of spline functions (“model-free description”) (21, 37) or by using the crystal structure of the GRASP domain (11) by rotating the structural model about two Euler angles and varying its distance from the membrane surface while optimizing the fit to the experimental data (24). Both models allowed for protein penetration into the bilayer, which was found to be small (<5 Å) in all cases.

NR was performed with myrG55 and G55, both at ∼6 μm/liter protein, on DOPC stBLMs with 10 mol % Ni²⁺-NTA-DGS. This protein concentration was sufficient to form a homogeneous protein layer at the interface in which the GRASP domain covered about 20% of the available area (Fig. 4A), necessary to resolve the structure of the interfacial protein with high confidence. Because the increase in Ni²⁺-NTA-DGS content in the membrane led to an increase in bound protein area coverage, one might be concerned that molecular crowding at the interface biases the structural results. The following observations suggest that this is not the case. 1) Structural investigations with NR of other proteins at membrane surfaces, conducted at similar protein surface concentrations (21, 34, 38), showed clearly that protein organization depended on other factors (electrostatic interaction; ligand binding) than surface crowding. 2) A model that places proteins randomly on the surface (39) predicts mean nearest neighbor distances of 1.6 or 2.25 protein diameters, respectively, for area coverages of 20% (myrG55) or 10% (G55). Although dynamic protein-protein contacts are indeed likely on the fluid lipid bilayer of an stBLM (29), which may result in cis pairing due to PDZ interactions if protein orientation is not appropriately confined, unspecific interactions are rare as each membrane-bound protein still has ample space for lateral diffusion.

The obtained NR data sets from different isotopic buffers were fitted with neutron scattering length density profiles, which, in turn, are based on unique volume occupancy profiles of molecular components that were independent of isotopic compositions (36). Component volume occupancy (CVO) profiles thus determined in the model-free and atomistic approaches are shown in Fig. 4A (myrG55) and 4B (G55). The protein concentration at the membrane surface is larger by a factor >2 for the myristoylated protein. In both cases, the main features of the atomistic models are well within the confidence intervals of the free-form fits. The myristoylated GRASP domain most likely forms a unique protein-membrane complex, characterized by a specific orientation and penetration depth into the bilayer (Fig. 4, C and D). Because NR is insensitive to rotations of the sample about the membrane normal, the modeled reflectivities depend only on two of the three Euler angles, ϑ and ϕ, defined in the inset in Fig. 4C where (ϑ₀,ϕ₀) = (0°,0°) is defined in a Cartesian reference frame with the longest principle axis of the protein crystal structure aligned along the z axis (membrane normal) and the second longest principle axis aligned along the x axis. Monte-Carlo Markov chain data refinement provided the probabilities of protein orientations and penetration depths and yielded a narrow range of orientation angles centered on (ϑ,ϕ) = (26°,78°) (Fig. 4C). A region with arcs of (Δϑ,Δϕ) ∼ (±15°,±20°) encloses the 95% (2σ) confidence intervals. This orientation places the myristoylated N terminus directly at the membrane surface and the C terminus into a position ∼10 Å away, from where the His tag can reach the membrane surface via the decapeptide linker. The associated CVO profiles and oriented protein structure are shown in Fig. 4, A and E, respectively. The protein is peripherally associated with the membrane, penetrating the membrane's lipid headgroup region by <5 Å (Fig. 4D). The protein attains an upright orientation, with its longest principle axis, indicated by the blue column in Fig. 4E, tilted ∼26° away from the surface normal, z. PDZ1 is close to the bilayer surface, and the center of mass of PDZ2 is further away by ∼28 Å, such that the internal ligand on PDZ2 cannot bind to the groove of PDZ1 of neighboring molecules in the same bilayer. The axes of the internal ligand and its target groove (yellow cylinders) are almost parallel, with an angular mismatch of ∼5°. Both point down toward the membrane surface (x-y plane, shown in blue in Fig. 4E) at shallow angles. Colored beads indicate locations of the N and C termini determined. Steered MD simulations in which only amino acid residues not resolved in the crystal structure as well as the added C-terminal decapeptide and His₆ tag were allowed to vary conformationally were used to bring both termini to the membrane interface. Overall, this model of the protein/membrane complex resembles the protein organization schematically shown in Fig. 1B and is distinct from our previous depiction (11) resembling that of Fig. 1A.

In contrast to myrG55, a unique organization of the nonmyristoylated G55 on the membrane cannot be distinguished from the NR data with confidence because multiple orientations lead to similar NR spectra. However, G55 shows a significantly smaller extension from the membrane surface into the bulk than myrG55 (∼40 Å versus ∼60 Å, compare Fig. 4, A and B). The CVO profiles for the various orientations that are compatible with the NR results are consistent with arrangements in which the GRASP domain rests on its side (ϑ >45°), with both PDZ domains near the membrane surface, as schematically shown in Fig. 1C. Because these ranges of orientational states give rise to similar CVO profiles, a particular protein orientation cannot be determined, nor can it be excluded that multiple orientational states, and therefore distinct structures of the protein/membrane complex, coexist. An alternate arrangement with a more upright orientation is also consistent with the NR results, but the protein would have to penetrate the membrane significantly (Fig. 4F). Unless further work unexpectedly indicates that GRASP penetrates the membrane, this orientation seems unlikely. A more complete account of the NR results and their evaluation is provided in supplemental Figs. S1–S13 and supplemental Tables S1–S4.

DISCUSSION

Membrane tethering that involves trans complex formation is widespread and important in cell biology, but an understanding of how it occurs in the face of competing cis pairing is incomplete. We found that the purified GRASP domain is sufficient to mediate membrane tethering. This tethering occurred via GRASP domain self-interactions, as GRASP needed to be present on both membranes and blocking mutations at the interaction interface inhibited tethering. In our assay, membrane anchoring of the GRASP domain was achieved via the N-terminal myristic acid and binding of a C-terminal His tag to Ni²⁺-conjugated lipids. Although the latter contact alone resulted in stable anchoring of the protein on each membrane surface, efficient tethering required both anchor points. Our structural studies showed that the dually anchored GRASP domain is complexed with the membrane in a unique, upright orientation that is likely compatible with the formation of interdigitated multimeric tethering complexes (Fig. 1B). In the absence of myristoylation, such a unique protein orientation could not be established, and the GRASP domain-membrane complex structure is more likely consistent with that depicted in Fig. 1C, which allows the formation of unproductive cis dimers.

As with the GRASP proteins, myristoylated proteins often depend on additional interactions for stable membrane association at their correct cellular localization. Therefore, N-myristoylation creates a second membrane attachment point, raising the possibility that a general purpose of N-myristoylation is to restrict protein orientation on the membrane, which could be beneficial to many protein functions. In the case of the GRASP domain, we show here that dual anchoring establishes an orientation that appears incompatible with homotypic interactions in cis. For membrane proteins that interact with the cytoskeleton, a similar mechanism could favor selective binding based on filament orientations. For protein-protein interactions within the plane of the membrane or for soluble components assembling on membrane proteins, myristoylation could control the final geometry of the complex.

Although protein orientation is likely a key determinant in many tethering reactions including cadherin-mediated cell-cell adhesion (1), this mechanism is ill-suited for flexible or unstructured binding partners. For example, the SNARE family of membrane-tethering/fusion proteins readily undergoes interactions in cis, forming stable complexes. ATP hydrolysis by the NSF ATPase is required to disassemble these complexes, enabling the SNARE proteins to participate in trans interactions. Thus, there exist at least two distinct modes of controlling trans interactions. Although an active mechanism, such as NSF, is needed to disassemble stable cis complexes of SNARES, membrane orientation may serve for well structured proteins.

Orientation-dependent binding could also impact the eventual dissociation of GRASP tethering complexes. This is because GRASP-mediated tethering precedes membrane fusion and the membrane rearrangement that occurs upon fusion transforms GRASP complexes from trans to cis. Our results here suggest that membrane-associated GRASP complexes in cis could only persist if the surrounding membrane forms a highly strained, i.e. high energy (40), configuration. Assuming that the GRASP membrane association in vivo is sufficiently strong, the relatively weak PDZ-ligand interaction (16) may provide a predetermined breaking point that releases once membrane fusion is completed. Thus, membrane fusion may provide the energy to dissociate GRASP trans complexes. This would constitute a novel type of regulation in which protein-protein interaction is controlled by membrane dynamics.

It is also conceivable that by providing a second membrane contact point, N-myristoylation confers a conformational change to the GRASP domain that affects its propensity for self-interaction. As peripheral membrane proteins, GRASP proteins are present in the cytosol, at least in small concentrations. Thus, both the membrane association of preformed soluble dimers and the binding of a soluble monomer to a membrane-associated monomer could render membrane tethering inefficient. Therefore, a reasonable hypothesis is that membrane binding imparts a conformational change, which activates the GRASP domains for self-interaction. Indeed, the purified GRASP55 GRASP domain self-interacts only weakly in solution (33).

The tethering assay described here allows manipulation of its core components, including lipid composition, liposome size, surface density of the proteins, and buffer composition. Although not physiological, the membrane anchoring of the GRASP domain using the His tag is a versatile aspect of the assay. Myristoylated GRASPs are recruited to Golgi membranes by binding golgins (14). Thus, further efforts to better recapitulate the biological tethering complex will be directed toward anchoring a His-tagged golgin such that the GRASP-binding domain of the golgin will recruit the myristoylated GRASP.

In conclusion, this study reveals an elegant structure-based biological mechanism that promotes homotypic protein-protein interactions in trans in which myristoylation stabilizes protein membrane orientation, thereby preventing interactions in cis. The work also shows that the GRASP domain is sufficient to mediate membrane tethering. An important area of further study concerns how additional cellular factors contribute to GRASP-mediated tethering. It will also be exciting to investigate whether myristoylation restricts the orientation of other proteins and to determine the ways in which orientation participates in the diverse functions of membrane proteins.