Phosphatidylserine and Phosphatidylethanolamine Asymmetry Have a Negligible Effect on the Global Structure, Dynamics, and Interactions of the KRAS Lipid Anchor

Abstract

The intrinsically disordered C-terminus of the prominent oncogenic protein KRAS-4B (KRAS) selectively interacts and clusters with phosphatidylserine (PS) lipids in the plasma membrane (PM). This 11-residue segment, called tK, contains a polybasic domain (PBD) of six contiguous lysine residues and a farnesylated cysteine. Previous molecular dynamics (MD) simulation studies of tK in phosphatidylcholine (PC)/PS bilayers have suggested that backbone conformational dynamics modulate tK–PS interactions. These simulations have been conducted in symmetric membranes whereas the PM is compositionally asymmetric, with the inner leaflet, where KRAS is localized, being enriched with PS and phosphatidylethanolamine (PE) lipids. To examine if bilayer asymmetry affects tK conformational dynamics and interaction with lipids, we conducted two 10 μs long MD simulations of tK bound to a PC/PS and a PC/PS/PE bilayer in which the PS and PE lipids are distributed in one leaflet. We found that, first, these compositional asymmetries caused differences in acyl chain dynamics between leaflets, but the equilibrium structural and dynamic properties of the two asymmetric bilayers are similar; second, in both systems tK is highly dynamic and samples at least two distinct conformational states; third, PS–tK hydrogen-bonding interactions vary with peptide backbone conformations, and lysine side chains in the PBD predominantly interact with the serine oxygens of PS. These results are in good agreement with previous observations of tK in symmetric membranes. The effects of POPS asymmetry or the presence of POPE on tK are limited to modulating the relative contribution of individual side chains to interactions with lipids and redistributing conformational substates. Additional observations include the larger flexibility of tK in the current simulations, which we attribute to the longer duration of the simulations and the use of the CHARMM36m force field, which more accurately models intrinsically disordered peptides such as tK.

Graphical Abstract

INTRODUCTION

KRAS-4B (hereafter KRAS) is a plasma membrane (PM)-associated signaling protein that is found mutated in ~20% of human cancers.^1–3 PM binding is required for KRAS function, which is achieved by a C-terminal 11-residue flexible segment called tK. The intrinsically disordered tK, which contains a polybasic domain (PBD) and a farnesylated carboxymethylated cysteine (Figure 1A), autonomously interacts with the PM and forms clusters in a lipid composition dependent manner.^4–6 We previously studied the interactions of tK with bilayers made up of the zwitterionic 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and anionic 1-palmitoyl-2-oleoyl-snlycero-phosphatidylglycerol (POPG) lipids mixed at a 4:1 ratio.⁷ In addition to the segregation and clustering of POPG around tK, we found that not all of the eight lysine residues interact with lipids at a given time and that lipid recognition appeared to be dependent on the peptide’s conformation. Subsequent simulation and experimental studies found that the lipid sorting capacity of tK is more complex than the simple electrostatic interaction between anionic phospholipids and the positively charged PBD region.⁵ For example, tK variants with equivalent net charge differ in backbone conformational fluctuations and lipid sorting.⁶

Figure 1.

Simulation of tK in compositionally asymmetric model membranes. (A) Primary sequence of tK and chemical structure of the farnesyl chain attached to the terminal cysteine residue. Underlined and in blue are polybasic domain (PBD) residues. (B) Simulation setup, showing tK tethered to a POPC/POPS/POPE bilayer. Lipids are shown in a color-coded surface representation, tK as tube with basic residues in blue, polar residues in green, and farnesyl in black. Setup of the simulation in a POPC/POPS bilayer is similar. (C, D) Time evolution of area per lipid (A; C) and bilayer thickness (L; D) in the POPC/POPS (black) and POPC/POPS/POPE (red) bilayers. 50 ns running average and the raw data sampled every 100 ps are shown in bolder and fainter shades, respectively. (E) Time evolution of tK Cα atom root-mean-square deviation (RMSD) from the starting structure (black) and bilayer insertion depth (I; red) measured as the z-distance between the COM of the farnesyl chain from the bilayer COM. Data are from the binary POPC/POPS bilayer; the profiles for the ternary system are similar.

Our previous MD simulations of tK mostly focused on a symmetric bilayer made up of 80% POPC and 20% 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoserine (POPS) lipids. Key results from these simulations included the observation that bilayer-bound tK fluctuates between three distinct conformational states⁵ and undergoes different lateral dynamics in different lipid bilayers (e.g., POPC/DSPS vs POPC/POPS).⁶ Using primarily electron microscopic (EM) spatial mapping of intact plasma membrane sheets, we also showed that tK has preference for asymmetric over symmetric POPS lipid species.⁵ Together, these studies have suggested that the KRAS lipid anchor is a combinatorial code for lipid sorting and, more broadly, the lipid acyl chain structure and dynamics participate in the regulation of the spatial distribution of prenylated polybasic membrane anchors found in many membrane-bound small GTPases.^5,6

While the previous simulations provided a useful mechanistic explanation for cell-based experimental observations, they were relatively short (<2.5 μs) and used symmetric membranes. The PM, on the other hand, is compositionally asymmetric,⁸ with the inner leaflet being enriched with phosphatidylserine (PS) and phosphatidylethanolamine (PE) lipids. Moreover, we previously used the CHARMM36 (C36) force field instead of its updated version CHARMM36m (C36m).⁹ C36m more accurately models intrinsically disordered proteins/peptides (IDPs) such as tK¹⁰ by optimizing the CMAP potential and scaling down the strength of Arg-Asp/Glu side-chain interactions. The current work extended the simulation time to 10 μs, replaced C36 by C36m, and used an asymmetric POPC–POPC/POPS bilayer. We examined the impact of lipid composition on tK conformational dynamics and lipid binding by running a second simulation in a ternary POPC–POPC/POPS/POPE mixed lipid bilayer. On the basis of a detailed analysis of these two trajectories, we show that the conformational diversity and conformation-dependent interactions of tK with anionic lipids are largely independent of bilayer asymmetry or lipid composition.

MATERIALS AND METHODS

Simulation Setup.

CHARMM-GUI^11–14 was used to build initial bilayer models using default parameters such as a water thickness of 22.5 Å and a hydration number of 50 per lipid. One leaflet of the binary bilayer contained 33 POPS and 77 POPC lipids while the other monolayer contained 107 POPC lipids (Table 1). This gave rise to a 30% PS content in the mixed-lipid leaflet, which we refer as the inner monolayer, and a 15.2% PS content in the entire bilayer. Similarly, one leaflet of the ternary bilayer consisted of 55 POPC, 33 POPS, and 22 POPE lipids while the other leaflet contained 105 POPC lipids. Therefore, the 30 mol % fraction of POPS in the inner leaflet is the same in the two bilayers, with the 20 mol % POPE diluting only the POPC fraction from 70% to 50%. A fixed POPS content ensures a one-to-one comparison of tK–PS interactions in the two bilayers. An additional consideration was monolayer surface area matching after the peptide is partially inserted (i.e., before running the simulations), which is required to ensure that assumptions for symmetric bilayers remain applicable.¹⁵ This was done based on areas per lipid determined by Membrane Builder in CHARMM-GUI. Monolayer surface area matching reduces packing differences and curvature arising from differences in phospholipid shape, which can be expressed in terms of the acyl chain area-to-headgroup area ratio.¹⁶ This ratio is nearly one for POPC, and therefore the intrinsic curvature of the outer monolayer is nearly zero. The head groups of POPS and POPE are smaller than the area of their acyl chains, resulting in intrinsic negative curvature of the lipids and generating a stress in the bilayer. This stress is counterbalanced by area-driven positive stress caused by the larger number of lipids in the inner leaflet.

Table 1.

Structural Properties of Asymmetric PS- and PS/PE-Containing POPC Bilayers Studied in This Work

	PC:PS:PE			D (μm2/s)a				tK
system	inner	outer	L (Å)b	PCd	PS	PE	tK	RMSD (Å)c	Ibb (Å)	no. of HBs
tK-PC/PS	77:33:0	107:0:0	38.4 ± 0.4	10.7 ± 0.2	8.3 ± 0.4		0.66 ± 0.01	2.0 ± 0.5	−0.8 ± 1.8	3.8 ± 1.4
tK-PC/PS/PE	55:33:22	105:0:0	38.8 ± 0.5	9.5 ± 0.2	7.0 ± 0.2	9.5 ± 0.3	0.41 ± 0.01	2.1 ± 0.4	−1.0 ± 1.8	3.7 ± 1.4

^aThe most linear portion of the MSD versus time plots (100–150 ns) was used to extract D from linear regression.

^bWe obtained L ≈ 39.0 Å for both the binary and ternary bilayers from the electron density profile of the P atoms.

^cWe excluded the first residue and the last two residues for RMSD calculations.

^dD for POPC (D_POPC) calculated separately for the outer and inner leaflets equals 10.2 ± 0.4 (outer) and 7.8 ± 0.2 (inner) for tK-PC/PS and D_POPC = 10.0 ± 0.5 (outer) and 7.2 ± 0.5 (inner) for tK-PC/PS/PE. Errors were calculated by block averaging after dividing the last 9 μs data into blocks of 1 μs, except for the diffusion coefficient of tK (D_tK) where we used the entire trajectory for MSD calculations and different ranges of lag time for estimating errors (25–55, 50–100, and 100–150 ns).

To build initial tK–bilayer complexes, an extended structure of the prenylated tK (sequence shown in Figure 1A) was taken from a previous work⁶ and attached to one side of the binary or ternary bilayer prepared as described above. (Note: our initial tK structure was already farnesylated, but CHARMM-GUI supports covalent attachment of prenyl groups.) Insertion was achieved by manually translating and rotating tK in VMD¹⁷ until about five terminal carbon atoms of the prenyl chain are inserted into the hydrophobic core of the mixed-lipid leaflet of the bilayers. This procedure yields subsequent spontaneous insertion of the rest of the lipid-modified moiety within short MD simulation times.^7,18–21 The resulting tK–bilayer complexes were solvated by TIP3P waters, and counterions were added to neutralize the systems. This resulted in two ~2700 atom systems. An example of the simulation setup is shown in Figure 1B.

Molecular Dynamics Simulation.

Simulations were conducted by using the C36m force field,⁹ following previously described protocols.^22,23 Simulation systems were prepared as follows with the NAMD program²⁴ for a subsequent run on the Anton 2 supercomputer. Each tK–bilayer system was first energy-minimized and equilibrated for 200000 steps with the lipid phosphate and protein heavy atoms restrained with a harmonic force constant k = 4 kcal/(mol Ų). The system was further equilibrated for 100000 steps each with k scaled by 0.75, 0.50. 0.25, and 0, followed by a production run of 20 ns on Stampede2 at the Texas Advanced Computing Center and local resources. We used a time step Δt = 2 fs, except for the restrained runs where Δt = 1 fs. The particle mesh Ewald²⁵ method was used to calculate long-range electrostatic interactions, restraining bonds involving hydrogens by using SHAKE.²⁶ A switching function with 10 and 12 Å distance cutoffs was used for nonbonded interactions and a 14 Å cutoff for pair-list updates. The constant number of particles, pressure, and temperature (NPT) ensemble was used, with the Nosé–Hoover Langevin piston and Langevin thermostat methods used to maintain P at 1 bar and T at 310 K. After a short simulation on Stampede2 under this condition, each system was transferred to Anton 2²⁷ and simulated with Desmond for 10 μs, with trajectories written out every 100 ps. Default parameters under the NPT ensemble were used, including the U-series cutoffs, semi-isotropic pressure control, and a 2.5 fs time step.

Trajectory Analysis.

We used in-house scripts and R²⁸ for analysis. The entire trajectory was used to examine time-dependent properties while equilibrium properties were calculated after excluding the first 1 μs data, which was considered an equilibration phase. Where applicable, trajectories were divided into nine 1 μs blocks to obtain mean and standard deviations.

Analysis of Bilayer Structural and Dynamic Properties.

Bilayer thickness and hydrophobic thickness, L and H, were measured as the average distance along the membrane normal between the centers-of-mass of the phosphorus atoms and between the centers-of-mass of the first CH₂ acyl chain carbon atoms of the two leaflets, respectively. The monolayer thickness, h, was defined as the average distance between the phosphate group and the terminal acyl chain carbon in a leaflet. The area per lipid, A, was estimated from the lateral dimensions of the simulation box (area of the box) normalized by the number of lipids per leaflet. Considering the area symmetry noted above, we used fluctuations in A to compute the bilayer isothermal compressibility modulus, K_A, at the simulation temperature (T = 310 K) as

$$K_A=\frac{2K_{\text{B}}T\langle A\rangle }{N{\sigma }^2}$$ (1)

where K_B is the Boltzmann constant, σ² is the variance of the area per lipid, N is the number of lipids, and the angle brackets denote time average. We realize that in asymmetric bilayers K_A needs to be computed separately for each leaflet if monolayer coupling is small.²⁹ In our case, there is a high degree of interleaflet coupling (see Results and Discussion).

Following previous work by others,^30,31 we analysed monolayer interdigitation using the overlap parameter q_oi(z) derived from ρ_o(z) and ρ_i(z), the partial number densities along the membrane normal of the terminal (CH₃) acyl chain carbon atoms of the outer and inner monolayers, respectively (eq 2).

$$q_{\text{oi}\left(z\right)}=4\frac{{\rho }_o\left(z\right){\rho }_i\left(z\right)}{{\left({\rho }_o\left(z\right)+{\rho }_i\left(z\right)\right)}^2}$$ (2)

When plotted against z (membrane normal), q_oi(z) = 1 indicates that half of the density at z is from the outer monolayer and half from the inner monolayer acyl chains, whereas q_oi(z) = 0 indicates that atoms from only one monolayer are found at that particular z value. Lipid chain order was examined by using the deuterium order parameter,³² S_CH, calculated from the trajectories using eq 3

$$S_{\text{CH}}=\frac{1}{2}\langle 3{\text{ cos}}^2\theta -1\rangle$$ (3)

where θ is the angle between the bilayer normal and the C–H bond of the methylene/methyl group in a given acyl chain. Lipid (and tK) lateral dynamics was examined with the aid of a VMD plugin called the Diffusion Coefficient Tool,³³ which we used to obtain mean-square displacement (MSD) as a function of lag time (τ). The 2-dimensional diffusion coefficient, D, can then be computed from the MSD by using the Einstein relation D(τ) = MSD(τ)/4τ. Here we used linear regression of the MSD vs τ curves because MSD was predicted by Einstein’s relation to grow linearly with τ.³⁴

Analysis of the KRAS Lipid Anchor Structure and Dynamics.

Standard definitions and tools were used to monitor common structural properties such as root-mean-square deviation (RMSD), radius of gyration (R_g), and hydrogen bonding (HB). For example, HB was defined by typical cutoffs of 3.1 Å and 30° for donor–acceptor distance and angle. The bilayer insertion depth, I, of the tK backbone (I_bb) or farnesyl chain (I_far) was defined as the difference between the average z-position of the tK backbone or farnesyl acyl chain atoms and the z-position of phosphate atoms in the tK-containing leaflet. We used previous experiences to define additional reaction coordinates (RCs) for analysis.⁷ These included a pseudo-dihedral angle involving virtual bonds between four consecutive Cα atoms. To identify a dihedral that most clearly classifies the simulated conformers, we scanned the entire backbone starting from the first four residues and stepping through every residue until the last four. We quantified the orientation of tK relative to the membrane normal by computing Euler angles that describe the 3-dimensional rotation of an object in space.³⁵ Given a fixed membrane reference frame x, y, z and mobile reference frame of tK X, Y, Z, Euler angles α, β, γ describe a rotation matrix that result in the mapping of (x, y, z) to (X, Y, Z). The protein (X, Y, Z) principal axis rotates about the x-axis by tilt angle β and about the z-axis by rotation angle α. These two angles were used to describe the orientational motion of tK with respect to the membrane plane (see ref 36 for a more detailed discussion of these angles). The lateral dynamics of tK on the bilayer surface was calculated by using 2D diffusion coefficients calculated from MSD as described above for lipids.

RESULTS AND DISCUSSION

There is a resurgence of interest in membrane compositional asymmetry.³⁷ However, the majority of MD simulation studies still focus on symmetric membranes. In this work, we simulated tK (an 11-residue polybasic farnesylated peptide from the KRAS C-terminus; Figure 1A) bound to asymmetric bilayers. We chose tK because it has been extensively studied in MD simulations using symmetric bilayers.^5,7 In the cell, tK binds to the inner leaflet of the PM that is enriched with PS and PE lipids. Therefore, we first simulated tK in a binary lipid bilayer and then in a ternary bilayer to test the impact of POPS and POPS/POPE asymmetry on its structure, dynamics, and interactions with lipids. During the simulations, the time evolution of the bilayer thickness (L) and area per POPC (A_POPC) show rapid equilibration and stabilization, indicating convergence (Figure 1C). Similarly, the peptide inserts and stabilizes in the bilayer relatively quickly, as shown for example for the binary system by using the time evolution of backbone RMSD and insertion depth of the farnesyl chain (Figure 1D). Below, we first briefly describe our observations on the structure and dynamics of the simulated membranes, followed by a comparative analysis of tK in the two bilayer systems.

Structure and Dynamics of Asymmetric PC/PS and PC/PS/PE Bilayers.

There is very little experimental or simulation data on the structure and dynamics of asymmetric bilayers involving POPS and POPE. Therefore, we first compared results from the POPC–POPC/POPS bilayer simulation with previous reports in symmetric bilayers of comparable lipid composition. Then we discuss similarities and differences between the binary and tertiary bilayer systems.

Global Structure and Dynamics of an Asymmetric PC–PC/PS Bilayer Are Similar to Its Symmetric Counterparts.

In our setup of a two-component asymmetric bilayer, one monolayer was 100% POPC and the other contained 30% POPS by composition (Table 1), which is close to the overall PS content of the plasma membrane (~15%).³⁸ We have recently simulated symmetric pure POPC and mixed POPC/POPS (4:1 ratio) bilayers.³⁹ We therefore compared the current simulation results with those from the previous ones in terms of common measures of equilibrium membrane structure and dynamics. These included bilayer thickness (L), area per lipid (A), and lateral diffusion coefficient (D) of lipids. The average L = 38.4 ± 0.4 Å we obtained from the asymmetric POPC–POPC/POPS simulation compares reasonably well with the 38.9 and 39.4 Å thicknesses in a pure POPC and mixed POPC/POPS symmetric bilayer simulations, respectively.³⁹ An earlier simulation study arrived at a similar value of L = 39.1.⁷ Experimental values of L for a pure POPC bilayer range between 37.9 Å at temperature T = 323 K and 39.1 Å at T = 303 K.⁴⁰ Comparison of areas per lipid is not as straightforward because we have different numbers of lipids in the two leaflets of the asymmetric bilayer (110 vs 107). This resulted in a monolayer surface area symmetry but complicated the calculation of a single area per lipid from the area of the simulation box. Therefore, we calculated A_POPC and A_mix representing areas per lipid for the pure-POPC and POPC/POPS monolayers and obtained A_POPC = 67.9 ± 1.1 Ų and A_mix = 64.9 ± 1.0 Ų. These values are only slightly larger than the 64.8 and 63.4 Ų areas per lipid obtained from simulations of symmetric POPC and POPC/POPS bilayers.³⁹ Experimental values of A_POPC in a pure POPC membrane are 67.3 and 63.3 Ų at T = 323 and 303 K.⁴⁰ Similarly, the lateral diffusion coefficients averaged over the entire POPC (D_POPC = 10.7 ± 0.2 μm²/s) or POPS (D_POPS = 8.3 ± 0.4 μm²/s) lipids in the asymmetric bilayer are comparable to those in a symmetric POPC/POPS bilayer (8.9 ± 0.5 for both lipids).³⁹ The slightly slower lateral mobility of POPS in the current simulation can be explained by charge interactions with tK that are discussed later. We found that the diffusion of POPC in the mixed-lipid leaflet is slightly slower than in the pure-POPC leaflet (D_POPC = 7.8 ± 0.2 vs 10.2 ± 0.4 μm²/s), with the latter being closer to an experimental value of 14 μm²/s for a fully hydrated POPC bilayer at 313 K.⁴¹ POPC lipids in the mixed-lipid inner leaflet also exhibit larger acyl chain order (S_CH) than those in the pure-POPC outer leaflet (Figure 2A). As in the diffusion coefficients, S_CH of both the sn-1 and sn-2 chains in the pure-POPC monolayer are more similar to results from a previous simulation of pure POPC bilayer as well as experiments (ref 42 and references therein). Therefore, POPS did not alter acyl chain dynamics in the opposing leaflet but measurably enhanced chain ordering in the host monolayer. Taken together, these results show that incorporation of POPS (and tK) in one leaflet of a POPC bilayer alters lipid packing and to some extent lateral dynamics locally, but it has little impact on the global structural properties of the membrane as a whole.

Figure 2.

Acyl chain order parameters. (A) Absolute value of POPC acyl chain order parameter per carbon atom, S_CH, for the sn-1 (top) and sn-2 (bottom) chains in the outer (black and red) and inner (gray and dark red) monolayers from the binary (POPC–POPC/POPS; left) and ternary (POPC–POPC/POPS/POPE; right) mixed bilayer simulations. (B) S_CH profiles for the sn-1 (top) and sn-2 (bottom) acyl chains of POPC (left) and POPS (right) lipids from the binary (black) and ternary (red) bilayer simulations. The error was calculated by using block averaging by dividing the last 9 μs data into nine 1 μs blocks.

Addition of 11% PE to an Asymmetric PC/PS Bilayer Has a Negligible Effect on Membrane Structure or Dynamics.

A key difference between the two- and three-component asymmetric bilayers simulated in this work is the replacement of 22 POPC lipids by POPE in the mixed-lipid leaflet, resulting in a 55/33/22 ratio of POPC/POPS/POPE lipids. Table 1 shows that the resulting ternary bilayer is characterized by a numerically larger L (38.8 ± 0.5 Å) and a smaller A_POPC (66.2 ± 1.1 Ų) or A_mix (63.3 ± 1.1 Ų) than the binary bilayer described in the previous section. However, the values are within error and therefore statistically not significant. A similar conclusion could be drawn by comparing hydrophobic thicknesses: H = 27.4 ± 0.5 Å and 27.8 ± 0.5 Å for the binary and ternary systems, both of which are reasonably close to the experimental value of 26.8 Å for pure POPC bilayer.⁴³ We conclude that the two bilayer systems are similar in terms of global structural properties. Small differences are apparent when comparing the enhancement of POPC acyl chain ordering in the mixed-lipid monolayer, which is more pronounced in the ternary than the binary bilayer (Figure 2A). Similarly, comparison of 2D diffusion coefficients, D, shows that lipids (and tK) in the ternary system diffuse slightly slower than in the binary system (Table 1). Even though the reduction in diffusivity is less than 1.5-fold, it is statistically significant. We therefore wondered if this suggests more acyl chain packing and/or interlipid interaction in the PE-containing bilayer. Consistent with this interpretation, the calculated order parameters for the POPS acyl chains, but not the POPC acyl chains, are slightly higher in the ternary than the binary membrane (Figure 2B). Moreover, more than half of the POPE lipids form hydrogen bonds with other lipids, particularly with POPS (discussed in the next section).

In asymmetric bilayers, monolayer interdigitation can be affected by differences in acyl chain length and unsaturation (e.g., ref 31). This is not the case here because the three lipid types considered in this work share the same acyl chain. Instead, we asked if the differences in chain order and headgroup interactions noted above could cause differences in monolayer coupling. To examine this, we computed the partial mass density profiles for the terminal carbon atoms along the bilayer normal (Figure 3). The density profiles, particularly those from the inner leaflet, are nearly symmetric about the membrane center (the shoulders in the oleoyl (sn-2) densities reflect the curling due to the double bond). The plots demonstrate the penetration of the terminal carbon atoms into the opposite monolayer, suggesting a large extent of lipid interdigitation and strong coupling of the monolayers. We quantified the extent of coupling by computing the overlap parameter q_oi(z) from the density profiles. The function approaches zero at |z| > 10 Å in both the binary and ternary systems, indicating that there is no overlap between the two leaflets at large distances (Figure 3). However, q_oi(z) reaches a value of 1 near the center of both bilayers. Thus, at the bilayer center half of the density is contributed from the upper monolayer and the other half from the inner leaflet, demonstrating strong monolayer coupling in both membranes. This level of monolayer coupling and the monolayer surface area matching in our setup allows us to use the isothermal area compressibility modulus of the bilayer (K_A), instead of the monolayers separately, to gain some insight into the elastic properties of the two asymmetric bilayers. We computed K_A from the A_POPC values discussed above and their fluctuations using eq 1. We obtained K_A ≈ 233 ± 16 and 229 ± 16 mN/m for the binary and ternary bilayers. These values fall in the range of values observed in previous simulations and experiments (see ref 42 and references therein). Combining these results, we conclude that the equilibrium structural and dynamic properties of the binary and ternary asymmetric bilayers studied in this work are very similar to each other. In other words, at least for the molar ratios we have tested, POPS or POPS/POPE asymmetry does not significantly alter the global structure of a POPC/POPS bilayer.

Figure 3.

Partial mass density profile of the terminal acyl chain carbon atoms and the overlap parameter q_oi derived from the density distributions. During the binary and ternary simulations, the bilayer fluctuated within only 0.2 Å centered at z = 1.9 and 1.4 Å, respectively; these values were used to recenter each plot by shifting to the left.

Structure, Dynamics, and Interactions of tK in the Binary and Ternary Asymmetric Bilayers.

IDRs such as tK are characterized by a rugged free energy surface with multiple minima separated by small barriers, making them highly sensitive to environmental changes.¹⁰ We therefore examined if tK conformational dynamics and interactions with lipids is modulated by the asymmetric distribution of POPS and the presence of POPE in a POPC/POPS bilayer.

Ensemble-Averaged Structural and Dynamic Features of tK Are Insensitive to Bilayer Asymmetry and Lipid Composition.

Visual inspection of the trajectories suggested that bilayer insertion and stabilization of tK is fast and follows a similar trend in the binary and ternary bilayer systems. The same conclusion could be drawn from the analyses of time dependent structural properties (Figure 1D); complete membrane insertion accompanied by conformational adaptation occurred within 1 μs. Analyses of equilibrium properties of tK using the remaining 9 μs data yielded very similar results for the two bilayer systems. These included time-average RMSD, I_bb, and I_far (Table 1). Similarly, the average number of tK-POPS HBs is essentially identical in the two bilayers (Table 1), and these values are similar to those in a POPC/POPG symmetric bilayer.⁷ Consistent with the generally slower lateral mobility of lipids in the ternary bilayer than in the binary, Table 1 shows that the calculated D for tK is also smaller in the former (D_tK = 0.4 vs 0.6 μm²/s). However, the difference is small, and both values fall between the 0.1 and 0.9 μm²/s experimentally measured D values for the hypervariable region of KRAS, a 20-residue segment that includes tK.⁴⁴ Furthermore, the membrane insertion depth and localization of individual residues do not qualitatively vary with membrane composition (Figure 4A). To examine this quantitatively, we compared the average z-position of each Cα atom and the center-of-mass of each side chain relative to that of the hydroxyl, phosphate, and glycerol oxygen atoms of lipids (Figure 4B). The backbone and side chains of tK populate a wide range of the bilayers’ transverse dimension, distributing roughly between z = −10 Å and z = 12.5 Å relative to the average z-position of the inner leaflet phosphorus atoms. As expected, the prenyl group is buried deep in the hydrophobic core, dragging with it the proximal Lys182, Thr183, and Lys184 toward the phosphate and glycerol groups. As a result, residues at the C-terminal half of the peptide are either buried or reside near the phosphate atoms while Lys residues in the PBD (the six contiguous Lys in the N-terminal half) are localized at the membrane–water interface near the hydroxyl oxygen of POPS. That on average several of the N-terminal Lys residues reside at the lipid–water interface is consistent with previous observations from shorter simulations of tK in symmetric POPC/POPG⁷ and POPC/POPS symmetric bilayers.^5,6

Figure 4.

(A) Final snapshot from the binary (left) and ternary (right) tK-bilayer simulations illustrating the organization of tK in the host monolayer. Phosphorus atoms are in gray (POPC), red (POPS), and blue (POPE) balls. (B) Time-averaged Cα-atom (orange) and side chain center-of-mass (cyan) positions relative to the average z-position of phosphorus atoms of the host leaflet. Also shown are the z-position histograms of oxygen atoms from the POPS hydroxyl in black, phosphate in green, and glycero-ester in dark red and blue. (C) Normalized density maps of tilt and rotation angles sampled by the tK backbone residues 2–10, with the tilt angle [0, 90] along the radial coordinate and rotation along the angular coordinate [−180, 180]. (D) Radial pair distribution functions for oxygen atoms of POPS (black) and POPE (red) around NZ atoms of Lys residues.

In addition to localization and insertion depth, the orientation of tK relative to the membrane normal may affect interactions with lipids. To check this, we computed the tK backbone rotation (α) and tilt (β) angles, which together describe the orientation fluctuations of tK (see the Methods section). Figure 4C shows that the binary and ternary systems differ in dynamics along α (i.e., along the angular coordinate in Figure 4C). In contrast, tK samples a similarly narrow range of β = 56 ± 15.3° and 60 ± 14.9° in the binary and ternary mixed bilayers (along the radial coordinate in Figure 4C). Note that β can impact tK–lipid interactions more significantly than α whose effect is minimal because lipids are laterally mobile. Combining the lack of effect on axial localization (Figure 4B) and tilt angle (Figure 4C) with the comparable number of hydrogen bonds (Table 1), we conclude that POPE has no significant effect on the average membrane interaction properties of tK.

Analyses of radial distribution (g(r)) functions (Figure 4D) show a strong affinity of tK for POPS in both the binary and ternary bilayer systems, consistent with previous reports based on experiments and simulations.^5,6,44,45 There are also some interactions with POPE (Figure 4D) but very little with POPC (not shown). While tK’s preference for POPS is qualitatively similar, a closer look at the g(r) peak heights suggests a slightly weaker preference in the ternary than the binary system. This appears to be compensated for by the weak interactions of tK with POPE (Figure 4D), suggesting that the affinity of tK for the two bilayers is likely similar. We initially thought that POPE will reduce tK–POPS interactions by engaging PS in hydrogen bonding and thereby diluting the PS pool available for interaction with tK. Indeed, POPE donates a hydrogen bond from its amine group to the POPS headgroup oxygen atoms (average number of HBs = 7.2 ± 2.6) and to a lesser extent to POPC (5.6 ± 2.3) and to itself (2.9 ± 1.7). Therefore, on average, more than half of the total number of POPE lipids in the system are engaged in HB interactions with other lipids, and ~50% of these interactions are with POPS. However, this turned out to be insufficient to significantly reduce tK–POPS interactions. As a result, the overall structure and lipid interactions of tK are not significantly affected by the presence of POPE.

Bilayer Asymmetry and Lipid Composition Subtly Affect Specific Residue–Lipid Interactions and Redistribute Conformational Substates of tK.

To investigate the role of asymmetry and PE content on the interactions of individual side chains, we calculated the g(r) of POPS oxygen atoms around the NZ atom of each Lys (Figure 5). In both bilayers, all eight Lys residues of tK engage POPS to a significant degree. While each lysine side chain interacts with both the phosphate and the hydroxyl oxygens of POPS, all prefer the latter with the exception of Lys184. Lys184 interacts more strongly with phosphate oxygens, especially in the binary bilayer system. Lys177 and Lys179 in the middle of the PBD as well as Lys180 contribute the most to interaction with POPS, with some variation between the binary and ternary systems. Along with Lys182, Lys178 in the binary and Lys176 in the ternary bilayer contribute the least. This nonequivalency of the lysines in salt-bridge formation with anionic lipids is consistent with previous findings.^5–7 While in an earlier simulation Lys175 and Lys179 remained solvated,⁷ in the current simulations all lysine residues interact with POPS to some degree. This may be one effect of the asymmetric accumulation of POPS in the inner (tK-interacting) bilayer, since the increased local concentration could enhance formation of salt bridges. However, the use of the C36m force field and the ~5-fold increase in simulation length can also explain the enhanced tK–POPS interactions because increased dynamics could facilitate increased side-chain interactions with POPS oxygen. The effect of POPE appears to be limited to the scrambling of the relative contribution of individual side chains to interactions with POPS, as can be deduced from the peak heights highlighted as insets in Figure 5. Another effect is the weakening of some side chains’ interactions, such as that of Lys184 (Figure 5), with POPS hydroxyl oxygen atoms in favor of POPE phosphates.

Figure 5.

Radial distribution of phosphate (PO4) and hydroxyl oxygen (OH) atoms of POPS around NZ atoms of individual Lys side chains for the binary (left) and ternary (right) lipid mixture simulations. Insets highlight variations in peak height.

Bilayer-bound tK is characterized by a rugged free energy surface.⁵ We therefore hypothesized that changes in the local lipid environment caused by asymmetry or addition of POPE to a POPC/POPS bilayer may modulate the distribution of conformational ensembles. To test this hypothesis, we used the pseudo-dihedral angle and radius of gyration (R_g), defined in the Methods section, to quantitatively assess the conformational fluctuations of tK. Two major ensembles, termed S₁ and S₂, are apparent from the dihedral angle vs R_g density distributions (Figure 6A). We observed three distinct ensembles in a previous study of tK in a symmetric membrane,⁵ including a small cluster (6%) populated by helical conformations. This ensemble was not sampled in the current simulations, which we attribute to the use of the C36m force field. Notice that the ternary and binary bilayers yielded similar conformer distributions, with small differences in population size or structure as described by the two reaction coordinates (Figure 6). We therefore isolated conformers belonging to S₁ into one group and those in S₂ into another and compared them in terms of their HB interaction patterns with POPS, as we have done previously.⁵ The tK–POPS side chain–headgroup HB heat maps in Figure 6B show differential interactions of S₁ and S₂ with POPS, consistent with a previous conclusion on the relationship between backbone dynamics and lipid recognition.⁵ Note that a key difference between the conformers in S₁ and S₂ is in the degree of compactness and local structure (or twist) rather than an order-to-disorder transition observed previously.⁵ This, however, is likely a consequence of differences in force field and simulation length rather than impact of asymmetry. An obvious impact of the lipid composition is the distribution of substates in the S₁ ensemble, which is split into a larger number of substates in the presence of POPE (Figure 7). This could arise from the small perturbation in lipid packing discussed in previous sections. Taken together, our analysis suggests that POPS asymmetry and especially the presence of POPE modulate individual side chain–lipid interactions and the distribution of conformational substates but do not significantly affect tK’s global dynamics or interaction with lipids.

Figure 6.

(A) 2D-density distribution of simulated tK conformers based on a pseudo-dihedral angle defined by the Cα atoms of residues 181–184 and the radius of gyration (R_g) for the binary (top) and ternary (bottom) bilayers. Two major clusters, defined as state S₁ and state S₂, are highlighted. (B) Superposition of representative backbone conformations (top) and normalized frequency of HBs between each lysine residue of tK and POPS lipids, separately for S₁ and S₂ clusters. Representative structures were collected by using one-sigma of the Gaussian fit of the main peaks in (A).

Figure 7.

Same as in Figure 6A highlighting the population and structure of subensembles within the S₁ ensemble of tK conformers derived from the binary (top) and ternary (bottom) bilayer simulations.

CONCLUDING REMARKS

A major aim of this work was to examine how the structure and dynamics of the intrinsically disordered polycationic lipid anchor of KRAS (tK) might be modulated by membrane asymmetry and composition. To this end, we conducted two 10 μs long simulations of tK bound to a binary POPC–POCP/POPS and a ternary POPC–POPC/POPS/POPE bilayer. We first showed that in each case, apart from slightly decreasing lateral mobility and increasing acyl chain order in the mixed-lipid monolayer, the effect of asymmetry on the global structural properties of a POPC/POPS bilayer is negligible. For example, average bilayer thickness and diffusion coefficient values obtained in the current work are similar to those observed previously in symmetric bilayers. The two asymmetric bilayers we studied are also remarkably similar in terms of global bilayer structural properties. However, there were small yet potentially significant differences. Compared to the binary asymmetric mixture, both POPS and tK are slightly less diffusive in the ternary mixture. This is correlated to the acyl chain order parameters that suggest slightly more lipid packing in the ternary mixture likely due to hydrogen-bonding interactions of POPS with POPE. For tK, time-averaged structural properties such as RMSD and localization or insertion depth, as well as the extent of interactions with lipids via hydrogen bonding, are nearly identical between the two systems. The same conclusion could be reached from the analysis of radial pair distribution functions for POPS headgroup oxygens and amino nitrogen atom of lysine residues. Finally, despite some differences in the details, the previously observed ensemble-dependent POPS interaction of lysine residues in tK is reproduced by the ternary and binary asymmetric bilayer simulations. The effect of POPS asymmetry or the presence of POPE appears to be limited to modulating the relative contribution of individual side chains to interactions with lipids, and to redistributing conformational substates. These results in no way indicate that membrane asymmetry is not important for membrane protein function, but rather show that considerations of PS asymmetry or PE content are likely less critical than sampling and force field issues for simulating the KRAS lipid anchor and related systems.