Dimeric peptides with three different linkers self-assemble with phos- pholipids to form peptide nanodiscs that stabilize membrane proteins

Three dimers of the amphipathic α-helical peptide 18A have been synthesized with different interhelical linkers inserted between the two copies of 18A. The dimeric peptides were denoted ’beltides’ where Beltide-1 refers to the 18A-dimer without a linker, Beltide-2 is the 18A-dimer with proline (Pro) as a linker and Beltide-3 is the 18A-dimer linked by two glycines (Gly-Gly). The self-assembly of the beltides with the phospholipid DMPC were studied with and without the incorporated membrane protein bacteriorhodopsin (bR) through a combination of coarse-grained MD simulations, size-exclusion chromatography (SEC), circular dichroism (CD) spectroscopy, small-angle scattering (SAS), static light scattering (SLS) and UV-Vis spectroscopy. For all three beltides, MD and combined small-angle X-ray and -neutron scattering were consistent with a disc structure composed by a phospholipid bilayer surrounded by a belt of peptides and with a total disc diameter of approximately 10 nm. CD confirmed that all three the beltides were alpha-helical in free form and with DMPC. However, as shown by SEC the different interhelical linkers clearly led to different properties of the beltides. Beltide-3, with the Gly-Gly linker, was very adaptable such that peptide nanodiscs could be formed for a broad range of different peptide to lipid stoichiometries and therefore also possible disc-sizes. On the other hand, Beltide-2 with the Pro linker, and Beltide-1 without a linker, were both less adaptable and would only form discs of certain peptide to lipid stoichiometries. SLS revealed that the structural stability of the formed peptide nanodiscs was also highly affected by the linkers and it was found that Beltide-1 gave more stable discs than the two other beltides. With respect to membrane protein stabilization, each of the three beltides in combination with DMPC stabilize the seven-helix transmembrane protein bacteriorhodopsin significantly better than the detergent octyl glucoside, but no significant difference were observed between the three beltides. We conclude that adaptability, size, and structural stability can be tuned by changing the interhelical linker while maintaining the properties of the discs with respect to membrane protein stabilization.


Introduction
Membrane proteins have great pharmaceutical interest since they are the specific targets of more than 50% of all drugs 1 . However, only 1-2% of the known protein structures in the Protein Data Bank are from membrane proteins. One reason is that membrane proteins rely strongly on a native-like amphiphilic environment to be stable, functional and accessible by crystallization for high-resolution diffraction experiments. This has motivated many studies on different reconstitution systems, from detergent micelles and amphipols 2 to membrane mimicking systems such as liposomes, bicelles 3 , SMALPs 4 and nanodiscs 5 . In the present study we investigate a new type of nanodisc system based on dimers of the peptide 18A with different interhelical linkers. The conventional nanodisc system is composed of a lipid bilayer, with the hydrophobic side screened by a double belt consisting of two amphiphilic membrane scaffolding proteins (MSPs) derived from human apoA1 5,6 . MSP nanodiscs stabilize membrane proteins well and are highly homogeneous from a structural point of view 7 . This is essential for structural studies such as small-angle scattering and single particle electron microscopy. However, reconstitution of membrane proteins in MSP nanodiscs is challenging from a sample handling point of view, and the system has limited flexibility with respect to the size of the membrane proteins that can be incorporated due to the fixed size of the MSP belt. We have therefore previously investigated nanodiscs with repeated, unlinked peptide units as a possible alternative to MSP, and showed that the amphipathic peptide, 18A, self-assembles with DMPC to form well-defined peptide nanodiscs with a diameter of about 100Å, that can stabilize the seven-helix transmembrane protein bacteriorhodopsin well 8 . The 18A peptide was first introduced by Segrest et al. 9 and variants have been studied ex-tensively [10][11][12][13] due to their potential as a therapy to prevent atherosclerosis 14 . Also NMR studies of 18A with DMPC was previously performed to elucidate the structure of the discs 10,15 . In our previous study 8 , the formed 18A discs were found to be unstable over time and reorganize into larger particles. In this study we hypothesize that the structural stability as well as other properties, such as overall structure, homogeneity and ability to stabilize membrane proteins can be tuned by using dimers of 18A with different interhelical linkers, such that the formed peptide nanodiscs can be optimized to specific purposes. Therefore, we have investigated how dimers of 18A with three different types of linkers self-assembled with DMPC to form peptide nanodiscs. The belt peptides, which we denoted "beltides", were Beltide-1, which contained two copies of 18A connected directly via a standard peptide-bond, i.e. with no additional linker, Beltide-2 with the 18A-dimer linked by proline (Pro), and Beltide-3 with the 18A-dimer linked by a double glycine (Gly-Gly) linker residue (see Fig.  1). The choice of the Pro and Gly-Gly motifs were motivated by their presence as linkers in the amino acid sequence of human apoA1, and because of their helix-breaking properties 16 . They were expected to provide 18A-dimers with different degrees of flexibility. The helical structure of Beltide-1 is not explicitly interrupted so this peptide was expected to be rigid. Pro is the most abundant linker in apoA1 and is found in seven out of ten helical segments. It induces a ∼ 30°kink and is helix-unwinding 17 , such that the α-helix does not twist 100°a s usual, but only ∼ 30°. Ramachandran plots 18 , show how Pro limits the core regions for the dihedral angles before and after the residue, meaning that Beltide-2 has a limited flexibility. A double Gly motif is found between the seventh and the eighth helix in apoA1. Gly has no side-chain, and, excluding Pro, it has the lowest propensity for helix formation 16 , so the Gly-Gly motif is expected to locally break the helical structure such that adjacent helical segments can move almost freely with respect to each other 18 , hence Beltide-3 was expected to be very flexible. Peptides similar to Beltide-1 and -2, but without the amidation and acetylation of the ends, have previously been studied by circular dichroism and revealed a helical content of respectively 89 and 73 % when associated with DMPC 19 , consistent with the helix-breaking properties of Pro in Beltide-2. Beltide-2 in complex with DMPC has furthermore been studied with negative stain electron microscopy (EM) and found to form discoidal particles with a diameter of ∼90Å 10 at a 1:1 weight ratio. Several analogous peptide:phospholid systems have been investigated with EM, and they all form discoidal particles 13,[20][21][22] . A recent study investigated an asymmetric version of Beltide-2 in complex with POPC and showed that these particles were more temperature resistant than 4F:POPC particles 23 , with 4F being 18A with the two leucines (L) replaced by phenylalanines (F). To our knowledge, no studies have been made on Beltide-3. Our experimental study of the self-assembly and final structure of the peptide nanodiscs were based on size exclusion chromatography (SEC), circular dichroism (CD) spectroscopy, and combined small-angle X-ray-and neutron scattering (SAXS and SANS). The SAXS and SANS combination is particularly useful for these types of particles due to their internal multiple scattering contrasts. Coarse-grained molecular dynamics (MD) simulations were used to obtain further insight into the self-assembly process of the peptide nanodiscs. The time dependent structural stability of the formed peptide nanodiscs was studied by time resolved static light scattering. Throughout the study, results on 18A nanodiscs from our previous work 8 were used as reference and the present experiments were performed under the same conditions, allowing for a direct comparison. As a part of the study, bacteriorhodopsin was reconstituted into the beltide nanodiscs to investigate their ability to stabilize a transmembrane protein. It was found that the different flexibilities provided by the linkers were directly reflected in their self-assembly and in the structure of the formed beltide nanodiscs, as well as in the structural stability of the discs. It was also found that both Beltide-1, 2 and 3 stabilize bacteriorhodopsin very well and significantly better than the detergent octyl glucoside (OG). No internal difference between the three beltides were observed with respect to membrane protein stabilization.

Materials and Methods
Peptide Synthesis. The amino acid sequence of the 18A amphipathic helical peptide is DWLKAFYDKVAEKLKEAF. For this study the beltides were acetylated at the N-terminus and amidated at the C-terminus 21 . The two-helix peptides were Beltide-1 with two copies of 18A with no linker, Beltide-2 with two copies of 18A linked by Pro, and Beltide-3 with two copies of 18A linked by Gly-Gly, see Fig. 1. The peptides were synthesized by solid-phase peptide synthesis (SPPS) on an automated peptide synthesizer (Syro II, Biotage) on a TentaGel S Rink Amide 0.24 mmol/g (Rapp Polymere GmbH) resin with 9-fluorenylmethyloxycarbonyl (Fmoc) for protection of N α -amino groups and purified using RP-HPLC (Dionex Ultimate 3000 system) with preparative C18 column (FeF Chemicals, 200Å 10 µm C18 particles, 2.1× 200 mm). The purity of the peptides was evaluated by analytical highperformance liquid chromatography (HPLC), and the identification was carried out by electrospray ionization mass spectrometry (ESI-MS) (MSQ Plus Mass Spectrometer, Thermo). The synthesis was as in our previous work 8 with exception of the modified amino acid sequences. and can be found from the measured intensity using the Siegert relation. The field autocorrelation function g (1) (τ) falls exponentially with a mean decay rate Γ and for monomodal, polydisperse samples this can be estimated via the cumulant expansion 36,37 , where terms up to second order was used in the present study The second cumulant κ 2 corresponds to the variance of the distribution around the mean decay rate. The mean diffusion constant D can be deduced by Γ = Dq 2 , with q = (4πn/λ 0 ) sin(θ ) where n is the refractive index, 2θ is the scattering angle and λ 0 is the laser wavelength in vacuum. The apparent radius of hydration R h can then be found by exploiting the Stokes-Einstein relation 38 T is temperature, and η is the viscosity. Evolution of the mass of the particles could be monitored by the SLS intensity, due to the direct relationship between intensity I(q = 0) and the weight averaged molecular weight M of the particles, I(q = 0) = KcM, where c is the concentration and K is an optical constant. This direct relation holds true when the Rayleigh-Debye-Gans (RDG) approximation is satisfied, meaning that the wavelength used should be much larger than the studied particles. In practice, the condition is fulfilled for particle sizes less than ∼300Å. Also, it is assumed that the refractive index increment was constant over time 39 .
Stability of Bacteriorhodopsin in peptide nanodiscs. The stability of bR in the peptide nanodiscs was monitored by measuring the chromophore absorption at 550 nm. The chromophore of bR is only active when bR is natively folded. The absorption was measured with a NanoDrop 1000 spectrophotometer (Thermo Scientific).
Coarse-grained MD simulations Coarse-grained MD simulations were performed with ESPResSo 40 and visualized using the VMD package 41 . Lipids were simulated with four beads, and each 18A unit of the dimer peptides were simulated with 42 beads (see Fig. 3) as described in detail in previous work 8 . The size of each bead was controlled by the Weeks-Chandler-Anderson potential where r c = 2 1/6 a is the distance at which V WCA = 0. By varying a, the self-assembly of the lipids without peptides could be controlled 42   No explicit solvent was simulated and the hydrophobic effect was taken into account by an additional attractive potential between the hydrophobic beads: Following previous studies 8, 43 , the energy ε was set to unity and the interactions was ω lipid-lipid = ω peptide-lipid = 1.6 and ω peptide-peptide = 0.8. The peptide beads and lipid tail beads had a diameter of 1 σ MD and the head bead diameter was 0.9σ MD , where σ MD is the unit of length in the simulations. 1 σ MD corresponds to about 5.3Å, since the 18A peptide is 32Å long and simulated as being 6 beads long. The interhelical linker was represented as a point through which an angle dependent harmonic bond connected the two 18A peptides into a dimer where the linker strength K was set to 10 ε for the Gly-Gly linker and 100 ε for the Pro linker, and the resting angle φ 0 was set to respectively 0 and 30°. The fixed twist induced by Pro was obtained by cross-linking the two 18A subunits with harmonic bonds. The bonds had different lengths, such that the 30°kink was sustained. Total energy of the system and the number of lipids per particle were monitored as function of time. Simulation time was converted into seconds following Illya and Deserno 43 , by comparing simulated phospholipid diffusion time with the experimental value. Peptides and lipids were placed at random in a (60σ MD ) 3 simulation box with periodic boundary conditions. This simple initial point of the simulations was equivalent to the experimental starting point, where the lipid-peptide film was mixed with buffer without detergent. The 4-bead phospholipids matched to the length of DMPC, namely 20Å, but the width was different, so a computational conversion factor was found by comparing the area per simulated phospholipid headgroup, 22Å 2 , with the experimentally determined area per DMPC estimated from the mean residueal ellipticity at 222 nm 26 . In another study, 18A was 38% helical in the free form and up to 92% helical in complex with DMPC depending on the peptide:lipid ratio. In our study, 18A was estimated to be 48% helical in the free form and 77% in complex with DMPC, which was in good accordance with the previous data. Interestingly, the high helix content of Beltide-1 was close to invariant between the free peptide (89%) and the complex with DMPC (86%). This indicated that the secondary structure of free Beltide-1 was far more stable than that of free 18A. As expected, the proline and glycine containing Beltide-2 and Beltide-3 were less helical than Beltide-1 due to the helix puncturing residues. The slightly lower helicity of Beltide-3 compared to Beltide-2 is in accordance with the presence of two glycines in the first and only one proline in the latter. 37pA, a model peptide with the same sequence as Beltide-2 but without chemical modification of the termini, was previously found to be 26-28% helical alone and 34% helical in complex with DMPC 27,28 . Here, Beltide-2 was estimated to be 68% helical alone and 58% in complex with DMPC. Nterminal acetylation and c-terminal amidation of 18A significantly stabilized the peptide, which resulted in an increase of helicity from 6 to 38% 27 . It is noteworthy that the previous experiments were carried out under different conditions, particularly at higher temperatures, but given a more than two-fold increase in helicity from 37pA to Beltide-2, our results indicated that the chemically modified termini provided significant stability. For Beltide-2, we found a significant loss of helicity when complexed to DMPC. In the work by Sethi et al. 28 , the presented CD spectra indicate a slight loss of helicity for 37pA when complexed with DMPC, although they report an increase in helicity. A decrease in helicity is in accordance with our results. This result could indicate that Beltide-2 must undergo a conformational change to adopt a adequate hydrophobic match with the lipids. Beltide-3 was estimated to be 61% helical alone and 65.5% in complex with DMPC. This indicates that the probably highly flexible glycine linker is slightly stabilized upon DMPC binding. Thus, the CD data indicated that Beltide-1 and Beltide-3 underwent no or subtle conformational rearrangements when binding DMPC, whereas 18A gained a substantial amount of helix, and Beltide-2 lost a small amount of helix. Our results confirmed previous findings that chemical modification of the termini is contributing favorably to the stability of alpha helical peptides.
Nanodisc structure. SAXS and SANS data were consistent with a model of slightly polydisperse peptide nanodiscs composed of a flat lipid bilayer, with a belt of peptides screening the hydrophobic side of the bilayer (Fig. 2), i.e. the same model that described the 18A nanodiscs 8 . Comparison of the derived model parameters showed that different linkers resulted in difference in size, with Beltide-1 nanodiscs being the largest and Beltide-2 nanodiscs being the smallest. Beltide-3 did not confine the size of the formed peptide nanodiscs due to its adaptability and the size could be controlled by changing the lipid-peptide stoichiometry of the sample. The largest particles from the coarse-grained MD simulations were qualitatively consistent with the SAS models but slightly larger. The SAS modelling suggested that the peptide nanodiscs all had a relative polydispersity of 0.2 (see table 4). The dimeric beltide nanodiscs were expected to be more polydisperse than MSP1D1 nanodiscs 6,34,45 , but interestingly the analysis suggested that they were less polydisperse than 18A peptide nanodiscs with a relative polydispersity of 0.4 8 . The optimal composition of the discs differed, which can be seen from the extracted model parameters in table 4. The nanodiscs with Beltide-1 and DMPC had fewer peptides per lipid than discs with Beltide-2 and -3. This is a natural consequence of the difference in size and the fact that the number of phospholipids scales with the area, and the number of peptides scales with the circumference. It was expected that larger nanodiscs would contain more beltides. The size order including the 18A nanodiscs was: 18A nanodiscs → Beltide-2 nanodiscs→ Beltide-3 nanodiscs → Beltide-1 nanodiscs. The number of beltides per nanodisc was (listed in the same order): 15 → 21 → 25 → 22. So Beltide-1 nanodiscs had fewer peptides per disc than expected. Furthermore, the width of Beltide-1 was only 7.3Å, as compared to 9.6 and 9.8Å for Beltide-2 and -3 respectively. This can be explained by Beltide-1 forming both single belts and double belts at the rim of the nanodisc, and not only double belts as assumed in the model. The discrepancy between model and sample was then compensated for by decreasing the number of peptides per disc and the thickness of the belt. Coarse-grained MD simulations also indicated the presence of single belt formation in addition to the double belt, as seen in figure 11, however not to a higher degree for Beltide-1 than for Beltide-2 or -3.
Structural stability and ability to stabilize membrane proteins. SLS showed structural change of the sample over time, and the structural stability could be assessed by the slope of the SLS curves. A correlation between size and stability of the formed nanodiscs was observed: Beltide-2 nanodiscs were smallest and least stable, whereas the Beltide-1 nanodiscs were largest and most stable. The stability of the beltide nanodiscs was reflected in the energy of the system, as monitored in the MD simulations, see Fig. 10. The simulations converged to different total energies and these energies correlated well with the stability of the systems, such that Beltide-1 nanodiscs were most stable and reached the lowest energy, and Beltide-2 nanodiscs was most unstable and converged to the highest energy. However, this was not true for the 18A nanodiscs, that had the same total energy as the Beltide-2 nanodiscs, but where more stable, as judged by the slope of the SLS curves. In our previous work 8 , we showed that 18A nanodiscs are structurally less stable than apoA1-based MSP1D1 nanodiscs, and a motivation for investigating the two-helical beltides was to find a system that was structurally more stable than the 18A nanodiscs. By having two linked helical segments, the dimer peptides bridges part of the gap between the 18A peptides with one helical segment and MSP with ten helical segments. The Beltide-1 nanodiscs were shown to be structurally more stable than the 18A nanodiscs, when bR was embedded in the nanodiscs, indicating that higher structural stability can be obtained by increasing the subsequent helical units per beltide. All three peptide nanodiscs stabilized the 7 transmembrane protein bR equally well and significantly better than OG at 20°C , meaning that the nanodiscs have great potential as systems for handling of membrane proteins.
The role of interhelical linkers in apoA1 and its effect on nascent HDL. Due to the similarity between the beltides and apoA1, the study gave insight into the role of the interhelical linkers in apoA1, the main protein constituent of nascent discoidal high density lipoproteins (HDL). The Pro linker, which is present in seven out of ten helices in apoA1, had a 30°kink that confined the formed Beltide-2 nanodiscs to a specific size, corresponding to a SEC peak around 13 ml on the Superdex 200 column. The size of HDL is shown to be important for its function 48 , so one main function of Pro is possible to confine the HDL particles in size. The Gly-Gly linker induced more adaptability and we hypothesize that it provides structural flexibility in apoA1, meaning that it can change dynamically between different states, such as those suggested by Mei and Atkinson 49 . Hence, Pro and Gly-Gly has opposite effects as linkers, i.e. Pro linkers confine the formed HDL particle to a specific size, whereas Gly-Gly linkers gives structural variability.

Conclusion
We studied three dimer peptides: Beltide-1, which contained two copies of 18A with no linker, Beltide-2 with the two copies of 18A linked by Pro, and Beltide-3 with the two copies of 18A linked by Gly-Gly. CD spectroscopy showed that all peptides were helical in the free form and when bound to DMPC. It was found that all three self-assembled with DMPC to slightly polydisperse, circular peptide nanodiscs, as revealed by combined SAXS and SANS and coarse-grained MD simulations. The nanodiscs were more polydisperse than MSP nanodiscs but less than 18A nanodiscs. Interestingly, all peptides had a diameter between 100 and 150Å, as seen for similar systems, such as 18A nanodiscs, MSP nanodiscs, polymer SMALP discs 4 and other apoA1 mimicking peptide nanodisc systems with different peptide architecture 22 . Even though the overall structure was the same for all the beltides, the linkers affected the size of the formed peptide nanodiscs with Beltide-1 nanodiscs being the largest and Beltide-2 nanodiscs being the smallest. Coarse-grained MD simulations were consistent with the SAS data and gave insight into the timescale of the self-assembly, showing how small, intermediate particles merged to form peptide nanodiscs within few ms. The presence of an interhelical linker affected the adaptability of the peptide dimers, so the peptide belt could adapt to different lipid-peptide stoichiometries. Beltide-3 was very adaptable and several peptide-lipid stoichiometries gave single, symmetric SEC peaks, while Beltide-1 and -2 were not very adaptable. The linkers also affected the structural stability of the formed peptide nanodiscs, meaning that the nanodiscs with the rigid Beltide-1 were more stable than the two other peptide nanodiscs. It is noteworthy, that when bR was incorporated, Beltide-1 nanodiscs were more stable than 18A nanodiscs, showing that is was possible to increase structural stability by increasing the length of the peptide, and thereby bridging the gap between the dynamic 18A nanodiscs with one helical unit and the stable MSP nanodiscs with 10 helical units. This structural stability could have an effect on the ability to stabilize membrane proteins. It was shown that the peptide nanodiscs could stabilize bacteriorhodopsin much better than OG micelles at 20°C, so the peptide nanodiscs have great potential as a system for easy handling of membrane proteins. The Pro linker and the double Gly linker are both present in the apoA1 sequence. The Pro linkers confine the HDL particles to a specific size by inducing a 30°kink between the helices, and the Gly-Gly linker provides the flexibility that is needed in order to change between different states. This way, the two linkers balance each other. We conclude that it is possible to tune stability, adaptability, size and polydispersity of peptide nanodiscs by using beltides of different lengths and with different interhelical linkers.