We created a model of the 36-residue transmembrane domain of fukutin (FK1TMD) by threading the FK1TMD sequence to an idealized α-helix using Modeller 9v7 (13). The resulting helix was evaluated for stereochemical integrity using Procheck (14). This model was used for all of the atomistic simulations and also for creating the coarse-grained version of FK1TMD.

Atomistic simulations were performed using GROMACS 4.0.7 (www.gromacs.org) (15−17). The simulations used an extended united atom version of the GROMOS96 force field (18). Berger parameters were used for phospholipids, as described in ref (19). Water molecules were treated explicitly using the SPC water model (20). Sodium and chloride ions were used to neutralize the charge of the simulation systems. All lipid and protein bonds were constrained using the LINCS algorithm (21), and water molecules were constrained using the SETTLE algorithm (22), allowing a time step of 2 fs to be used. The velocity rescaling thermostat, with a time constant for coupling of 0.1 ps, was used to maintain the system temperature (23). The Berendsen barostat, with a time constant of 1.0 ps, was used to maintain the system pressure at 1.0 bar (24). Electrostatic interactions used a cutoff of 1.0 nm, with interactions beyond this cutoff treated using the smooth particle mesh Ewald (PME) method (25). The van der Waals interactions also used a cutoff of 1.0 nm with a long-range dispersion correction applied for the energy and pressure. The neighbor list was updated every 10 steps. The system components of each simulation are summarized in Table I of the Supporting Information. All simulations were run for 50 ns. The conformational properties of FK1TMD were analyzed using GROMACS tools and DSSP (26). Visualization was conducted with VMD (27).

All CG simulations were performed using GROMACS 4.0.7 (www.gromacs.org) (15−17) with the MARTINI CG force fields. All simulations involved self-assembly of a lipid bilayer in the presence of one FK1TMD molecule from a random configuration of protein, lipids, ions, and water as described in refs (28−30). The CG parameters for DPPC, DLPC, POPC, ions, and water molecules are given in ref (28). The parameters for the 1,2-dipalmitoyl-sn-glycero-3-phosphatidylinositol 4,5-bisphosphate (PIP2) lipid headgroup were derived through matching the bond and angle distributions from a CG inositol 1,4,5-trisphosphate (IP3) (31) simulation to atomistic simulations of IP3 performed using the GROMOS carbohydrate parameters (32). The remainder of the PIP2 lipid parameters were the same as those used for the CG DPPC lipid. Parameters for amino acids are given in refs (29), (33), and (34). The integrity of the FK1TMD helix was retained by implementing an elastic network model as described in ref (33). For simulations of FK1TMD in the more complex Golgi Apparatus membrane, we used the MARTINI2.0 force field because of the greater range of lipid models available (35). The complex GA membrane was composed of 50% PC, 20% phosphatidylethanolamine (PE), 12% PIP2, 8% phosphatidylserine (PS), and 10% sphingomyelin (SM). All phospholipids had palmitoyl fatty acid tails in both the sn-1 and sn-2 positions of the glycerol backbone. The ratio of phospholipid to cholesterol was 16:1 (0.16). This composition is similar to that of the GA (36). As with the atomistic simulations, analyses of the CG simulations were performed using GROMACS tools and locally written code, and visualization was conducted with VMD (27).

The hydrophobic peptide FK1TMD (MQRINKNVVL ALLTLTSSAF LLFQLYYYKH YLSARN) was synthesized by Peptide Protein Research Ltd. to more than 50% purity. DHPC (C6:0), DDPC (C10:0), DLPC (C12:0), DMPC (C14:0), and DPPC (C16:0) were purchased from Avanti Polar Lipids. Sodium phosphate was obtained from Sigma.

For lipids with chain lengths of more than 10 carbons, we reconstituted the FK1TMD peptide into unilamellar vesicles by dissolving the peptide and lipid at a molar ratio of 100:1 in methanol. This was dried to a thin film by vacuum evaporation. Multilamellar vesicles (MLVs) were formed by hydration and agitation in 5 mM sodium phosphate buffer (pH 7.4). MLVs were sonicated to clarity to form small unilamellar vesicles (SUVs). The final concentration of FK1TMD in the CD samples was 0.2 mg/mL as determined by its absorbance at 280 nm (E₂₈₀ = 5960 M⁻¹ cm⁻¹). For DHPC (C6:0), the samples were prepared in an identical manner, resulting in the reconstitution of FK1TMD into DHPC micelles.

The CD spectrum of the FK1TMD peptide reconstituted in unilamellar vesicles composed of saturated phospholipids of varying chain lengths was measured using a Jasco J720 spectropolarimeter fitted with a heater and a 1 mm path length quartz cuvette (Hellma) at 25 and 43 °C. The spectrum scan was performed from 300 to 190 nm using a spectral bandwidth of 2 nm, a scanning speed of 100 nm/min, and a response time of 4 s. Six spectra were recorded and averaged for subsequent analysis. CD spectra were analyzed between 195 and 240 nm to determine the secondary structure composition of FK1TMD in the different lipid environments using the CONTIN/LL analysis algorithm (37,38) provided on the Dichroweb CD analysis server (39,40). The analysis was performed using basis set 7 that has been widely used to study secondary structural elements of membrane peptides and surface-associated proteins.

Initially, the stability of the FK1TMD model was assessed in an aqueous environment. The protein lost much of its α-helical structure within the first 10 ns of simulation; by the end of the 50 ns simulation, only residues V8−L13 and L25−H30 were in an α-helical conformation (see Figures 1 and 2 of the Supporting Information).

Next, the stability of the FK1TMD model in the vicinity of a membrane-mimetic phospholipid bilayer was investigated. Simulations were initiated with the FK1TMD model in two positions relative to the phospholipid bilayer (i) in the water region just above the phospholipid headgroups of a DMPC bilayer and (ii) in a TM orientation, in which the principal axis of the α-helix was perpendicular to the bilayer plane. As expected given the in vivo environment of the ER/GA membranes, the secondary structure of the model was more stable near a phospholipid bilayer compared to that in just a water/ion environment. When initially positioned in the aqueous phase just above the lipid headgroups, FK1TMD retained its α-helical structure in two short regions at either end of the peptide (Figure 1). The region between these two α-helices (L13−F20) became unstructured (see Figure 3 of the Supporting Information). When manually positioned perpendicular to the bilayer plane, the helix generally remained in a stable α-helical conformation (between R3 and H30), although some unfolding was observed at the C-terminal end (Y31−N36) (see Figure 1 and Figures 4 and 5 of the Supporting Information). To assess the effect of the bilayer thickness on the stability of FK1TMD, we also simulated the helix in a DPPC lipid bilayer. DPPC has longer tails than DMPC, and thus a thicker hydrophobic core region. The pattern of unfolding at the C-terminal end observed in the DMPC simulations was reproduced in the DPPC simulations. Structural drift of the helix was measured by calculating the mean root-mean-square deviation (rmsd) of the backbone Cα atoms from the starting structure over the last 25 ns of each simulation. The rmsd values were similar in both lipid environments, 0.33 ± 0.04 nm in DMPC compared to 0.28 ± 0.02 nm in DPPC (see Figure 6 of the Supporting Information).

CGMD, in which four heavy atoms are replaced with a single, spherical particle, to reduce the complexity of the system enables longer time scales to be studied. This has permitted the study of bilayer self-assembly and the insertion of peptides into lipid bilayers (29,30,41). The preferred membrane localization and orientation of FK1TMD has been investigated using CGMD. The FK1TMD model was constrained to an idealized α-helix by using an elastic network. The 1 μs simulations were initiated from a random arrangement of lipids, peptide, water, and counterions to allow the self-assembly of the bilayer and unbiased insertion of the protein. The effect of the bilayer thickness on the orientation of FK1TMD was assessed by simulating the model in DLPC (three particles in both tails), DPPC (four particles in both tails), and POPC (four particles in one tail and five in one tail) phospholipids. In all the CG simulations, the bilayer self-assembly proceeded via the stalk formation mechanism as previously reported for DPPC lipids from comparable CGMD studies (28,29). FK1TMD adopted a TM orientation within the membrane, with the axis of the α-helix oriented approximately perpendicular to the plane of the bilayer (Figure 1). FK1TMD was occasionally oriented parallel to the plane of the bilayer at the start of the simulations, particularly in DLPC lipids. However, after a short period of time (∼75 ns), one end of the helix inserted into the bilayer followed by the rest of the helix, such that it was traversing the bilayer. It retained this TM orientation for the remainder of the simulation, although its precise tilt with respect to the bilayer plane was observed to fluctuate. The tilt angle of FK1TMD calculated during the final 100 ns of the CG simulations was 39 ± 11° in DLPC and 30 ± 9° in DPPC. Presumably, the greater average tilt angle in the shorter DLPC lipid tails is a result of matching the hydrophobic region of the protein with the lipid tails. To further investigate the possibility of the lipid tail length influencing the FK1TMD tilt angle, we ran an additional set of CG simulations, this time with POPC phospholipids. PO tails are longer than DP and DL tails (the POPC CG lipid model implemented within MARTINI has five particles for the oleoyl tail and four particles for the palmitoyl tail, compared to the four particles for each palmitoyl tail in DPPC) and also contain an unsaturated bond (oleoyl group), which gives the tails a characteristic “kink”. In agreement with the trend observed in our DPPC and DLPC simulations of a smaller tilt angle in longer lipid tails, the mean tilt angle of FK1TMD calculated during the final 100 ns of the CG simulations was 17.6 ± 7° (see Figure 7 of the Supporting Information).

A conserved N-terminal motif, R/K-x-x-R/K, has been shown to be required for the correct localization of some glycotransferases within the GA (3). The R/K-x-x-R/K motif of FK1TMD (R-I-N-K) is located between residues R3 and K6. To investigate the role of this motif in the membrane localization and orientation of FK1TMD, we calculated the number of contacts (contact defined as an interatomic distance of ≤6 Å) between the protein and the phospholipid headgroup particles by sampling the last 100 ns of the 1 μs CGMD trajectories at 0.4 ns intervals. Full details of the protein−lipid contacts are provided in the Supporting Information. In all three lipid bilayers (DLPC, DPPC, and POPC), the N-terminal residues between M1 and N7 made regular contacts with the phospholipid headgroup particles. In particular, R3 and K6 had a strong propensity to interact with lipid headgroup particles; together, they accounted for >20% of all headgroup−protein contacts in simulations of all three lipid bilayers (see Tables 2−4 of the Supporting Information). Presumably, these charged residues serve to anchor the N-terminal end of the α-helix in the headgroup region of the zwitterionic lipid bilayer.

At the C-terminal end, residues Y28−N36 made regular contacts with the lipid headgroup particles. In particular, residues K29, R35, and N36 had a marked propensity to interact with the lipid headgroups; together, they make up ∼15% of the total headgroup−protein contacts in the DLPC simulations and ∼20% in the DPPC and POPC simulations. These observations suggest that the charged residues at both N- and C-terminal ends of the helix play a key role in anchoring the protein in a TM orientation in the lipid bilayer. The fluctuations in the tilt angles suggest that there may be a degree of mismatch between the width of the hydrophobic patch on FK1TMD and the hydrophobic region of the bilayer defined by the lipid tails. In addition to the protein−lipid contacts, we also measured the protein lateral motion by monitoring the position of the center of mass of the terminal residues in the x−y plane during the last 100 ns of the simulations (Figure 2). The helix exhibited a high degree of lateral mobility. In all three lipids, the lateral movement of the helix in the x−y plane was up to ∼7 nm in either the x or y dimension.

In Vivo Environment of FK1TMD. the GA membrane is composed of a complex mixture of lipids, including anionic lipid headgroups. Having established the influence of the lipid tail length on FK1TMD orientation and mobility, we then investigated any potential affinity of the protein for particular lipid headgroups, in particular the influence of the anionic lipids. First, we performed self-assembly simulations of FK1TMD in POPS lipids. Although the high charge density of a 100% POPS bilayer does not reflect the in vivo environment of FK1TMD, these simulations were useful in determining the effect of anionic lipid headgroups, as we could compare directly with our simulations of FK1TMD in POPC. Changing the headgroups in our simulations had the largest effect on the protein tilt angle. We observed an increase in the tilt angle from 17.6 ± 7° in POPC to 25.5 ± 9° in POPS. This is presumably driven by the interaction of R3 and K6 in the N-terminus and K29 in the C-terminus with the POPS headgroup atoms (see Table 5 of the Supporting Information). The lateral mobility of the protein in the two different lipid types was similar. To investigate the effect of the different headgroups in an environment more representative of the GA membrane, we simulated a more complex bilayer. This consisted of a mixture of DPPC, DPPE, PIP2, SM, and DPPS lipids in a 50:20:12:10:8 ratio and also cholesterol (CHOL) at a 16:1 phospholipid:cholesterol ratio (36).

In the complex GA membrane, R3 and K6 at the N-terminus of FK1TMD made a large number of contacts with the headgroups of all lipid types. These two positively charged amino acids made the most contacts with the highly negatively charged PIP2. Indeed, the interactions of PIP2 with R3 and K6 accounted for 12% of the total contacts between FK1TMD and the phospholipid headgroups. In contrast, the interactions of PIP2 with K29 and R35, at the C-terminus of the FK1TMD, accounted for only 5% of the total headgroup contacts. The hydroxyl group of CHOL makes the most contacts with a discrete region between I4 and L10 at the N-terminus of FK1TMD. This region accounted for approximately 55% of the total contacts between CHOL and FK1TMD. Conversely, CHOL makes fewer and more diffuse contacts with the C-terminal amino acids. The interactions of SM with FK1TMD followed the same trends as those with DPPC. While the tilt angle of FK1TMD in the more complex GA membrane was 23.6 ± 8°, this is intermediate between the tilts observed for the DPPC/DLPC and POPC bilayers but similar to the tilt angle of 25.5° measured from our simulations of FK1TMD in pure POPS.

To evaluate the distribution of the system components, we calculated density profiles of FK1TMD and the various lipid headgroups (Figure 3). The distribution indicates similar positions of the DPPC, DPPE, and DPPS lipids with respect to the bilayer normal, which would be expected for lipids with the same tail groups. The SM lipids are placed slightly farther apart; i.e., they are displaced slightly toward the aqueous phase. This appears to be a result of lipid−lipid contacts within the bilayer. The shorter CHOL moieties are located closer to the hydrophobic core of the bilayer. The slight asymmetry in the CHOL density is a result of one molecule flipping into the opposite leaflet during the equilibration procedure. Visual inspection of the simulation trajectories revealed a clustering of PIP2 lipids around the N-terminus of the FK1TMD (Figure 3). To further investigate this clustering, the radial distribution was calculated for the glycerol-linked phosphates of PIP2 relative to the K6 and K29 residues of FK1TMD (Figure 8 of the Supporting Information). The distributions show that PIP2 is found more often in the proximity (within 1 nm) of K6 than in the proximity of K29. In comparison to PIP2, the radial distribution of the DPPE phosphates was similar in both leaflets of the bilayer. The density of the DPPE phosphates within 1 nm of FK1TMD was also lower than that of PIP2.

The secondary structure of FK1TMD reconstituted into saturated phospholipids with varying chain lengths was investigated using CD spectroscopy. FK1TMD was reconstituted at a lipid:protein ratio of 100:1 that resulted in the formation of bilayer structures for all lipids with the exception of DHPC, for which micellar structures were formed. To ensure CD spectra were acquired for all bilayer-forming lipids in the liquid crystalline phase, data were recorded at both 25 and 43 °C. For all bilayer-forming lipids, the CD spectra at 25 and 43 °C showed distinct minima at 208 and 222 nm with a maximum at 195 nm consistent with FK1TMD forming an α-helical structure as predicted (Figure 4). In contrast, the spectra of FK1TMD in DHPC micelles showed a less distinct minimum at 208 nm and reduced elipticity at 195 nm suggesting that additional secondary structures may contribute the observed CD spectra.

The quantitation of CD spectra of integral membrane proteins and peptides remains challenging because of the paucity of membrane protein structures in the basis sets employed during analysis, the scattering observed from the lipid vesicles (42,43), and other environmental factors (44). However, they still provide useful insights into changes in the secondary structure composition of proteins and peptides within (or associated with) the lipid bilayer. A quantitative analysis of the α-helical contribution to the CD spectra is given in Table 1. As expected, FK1TMD in bilayer-forming lipids is almost entirely α-helical. In the case of DPPC, this indicates that the secondary structure of FK1TMD remained invariant irrespective of whether the surrounding bilayer was in its liquid crystalline or gel phase. Interestingly, the α-helical contribution to the CD spectra in the bilayer-forming lipids appears not to vary as a function of bilayer thickness, suggesting that changes in bilayer thickness are accommodated by changes in the orientation of the TMD within the bilayer in agreement with the molecular dynamics studies. FK1TMD in DHPC micelles also revealed extensive α-helical structures, with more than 90% of the residues in a helical conformation at both temperatures. The analysis of FK1TMD in DHPC micelles at 43 °C indicated a small decrease in helical content compared to that observed at 25 °C, although the significance, if any, of this observation will require more detailed structural studies.