Influence of membrane lipid composition on the structure and activity of γ-secretase

Model preparation

The cryo-EM structure of GS resolved at 4.2 Å resolution (PDB ID: 5FN2) was employed as the initial 3D coordinate model to perform our multiscale MD simulations in a variety of lipid compositions.¹⁴ The missing side-chains of the structure were completed using WHAT IF server.²⁶ The intracellular loop connecting TM6 and TM7 of the PS1 component (amino acids 264 to 377) was not modeled due to its extended length and the lack of available structural information to inform its structure. Modeling PS1 absent the loop allowed us to mimic the mature (auto-protolyzed) form of PS1 in the complex. Simulations employing a model of the complete mature and in-mature forms have been reviewed by Kepp’s group.²⁷ The arrangement of our GS models with respect to a lipid membrane was obtained from the Orientation of Proteins in Membranes (OPM) server.²⁸ The average hydropathy index of each amino acid was calculated with the hydropathy values reported by Kyte and Doolittle using a nine-residue window of the GS sequence.²⁹

Molecular dynamics simulations

We used a multiscale computational modeling approach, combining all-atom and coarse-grained representations, to characterize the structural ensemble and conformational behavior of GS in twelve different lipid bilayers (see Table 1). For the CG models, we simulated GS in the (1) unprotonated and (2) protonated states of Asp385 residue.³⁰

Coarse-grained simulations.

The unprotonated and protonated Asp385 CG models were embedded in the different lipid membranes using the CHARMM-GUI Martini bilayer maker³¹ and the MARTINI v2.2 forcefield.³² The pressure was set at 1.0 bar with a semi-isotropic Berendsen coupling and the temperature was set to 315 K using V-rescale coupling. This temperature was used for all simulated systems since it corresponds to the highest phase transition temperature within the lipids used in this work.³³ Each system was neutralized and, except for POPA system, brought to a concentration of 0.15 M with randomly placed sodium and chloride ions. We performed 20 NPT replicas of the protonated and unprotonated states of GS-membrane models for 1.5 μs, for a total of 60 μs of dynamics for each complex, using GROMACS 5.0.6 software.³⁴ The MD simulations were analyzed using the inbuilt GROMACS tools, MDAnalysis libraries^35,36 for python, g_lomepro³⁷, GridMAT-MD³⁸ and FATSLiM³⁹ tools. An extra simulation of 1.5 μs was performed for each GS model using the customized version of GROMACS 4.5 to calculate the 3D-resolved local pressure of the last 500 ns (10,000 frames).^40,41 Plots were generated with Gnuplot 5.0⁴² and Matplotlib v2.0⁴³ and protein figures with PyMOL v0.9⁴⁴.

All-atom simulations.

Five atomistic GS models were constructed with the CHARMM-GUI membrane builder⁴⁸ module in POPC, POPE, POPA, POPC(60):CHOL(40) and a lipid raft (Raft 1) bilayer compositions. The systems were energy minimized and equilibrated under standardized NVT and NPT conditions. Equilibration of each simulation was followed by 200 ns production run with a 2 fs time step. Temperature was set to 315 K using the Nosé–Hoover coupling thermostat algorithm. The pressure was set to 1.0 bar using the semi-isotropic Parrinello–Rahman barostat algorithm. The Lennard-Jones potential was set using a shift function between 0.9 and 1.2 nm, and the electrostatic interactions were calculated between 0 and 1.2 nm after which were calculated using the Particle Mesh Ewald (PME) approach. Neighbor lists were updated every 20 steps and bonds involving hydrogens were constrained using the Linear Constraint Solver (LINCS) algorithm.⁴⁹ The simulations were performed using GROMACS 5.0.6 with the CHARMM36 force field⁵⁰ and the flexible TIP3P water model. Each system was neutralized and, except for POPA system, brought to a concentration of 0.15 M with randomly placed sodium and chloride ions.

GS topology defines its orientation in the lipid membrane

To understand the effect of the membrane lipid composition on the structure and dynamics of GS, it is first necessary to characterize the amino acid properties and distribution of the complex.⁵³ Like most membrane proteins, the majority of surface exposed residues in the TM region of GS are hydrophobic, while the majority of hydrophilic residues are surface oriented in the extra-membrane domains (Figure 1B).²⁹ Using the hydropathic profile proposed by Kyte and Doolittle, GS regions were defined so as to optimize the number of protein-membrane interactions (Figure 1C). Except for TM3 of APH-1A, all GS TMs have a high index of hydropathy at the center region of the helices (Figure S1). Charged amino acids located close to the bilayer-water interface are usually involved in protein stability and conformational state transitions due to their interaction with surrounding water molecules and lipid headgroups.⁵⁹ A large number of charged amino acids can be found in these GS regions, where a majority of positively charged amino acids (Lys and Arg) in the TMs are exposed to the cytosol rather than the extracellular space (“positive-inside” rule) (Figure 1D).⁶⁰ This initial structural analysis was used to organize our observations of membrane and lipid-specific interactions with GS domains and residues.

Lipid headgroup charge modulates the structure and activity of GS

We built three different protein-membrane systems, each consisting of GS in a lipid bilayer composed of (1) zwitterion phosphatidylcholine (PC) lipids, (2) hydrogen-bond donor phosphatidylethanolamine (PE) lipids, and (3) negatively charged phosphatidic acid (PA) lipids, leaving the acyl chains 1-palmitoyl-2-oleoyl (PO) constant. The local stress profile of our reference POPC bilayer system presents a homogeneous lateral pressure profile in the transmembrane region, with lower pressure values characteristic of the interfacial region (z ≈ 1.8 nm and z ≈ −1.8 nm) and slightly increased pressure values in the lipid acyl tail (1.5 > z > −1.5 nm) and headgroup regions (z ≈ 2.2 nm and z ≈ −2.2 nm) due to repulsive interactions (Figure 2A). Pressure fluctuations observed in the upper part of the lateral pressure plots (z > 2.0 nm) are derived from the collision of water molecules with the ECD of NCT. Figure 2B shows the probability distribution of our previously described order parameters: dd_Asp distance and T_TM6 angle obtain from our unprotonated and protonated Asp385 GS systems. As in our previous study, we found that the protonated Asp385 form facilitates the transition of PS1 to the active state (state 2) conformation (characterized by short dd_Asp and proper T_TM6 angles), while the unprotonated systems were found predominantly in the inactive state (state 1) conformation (long dd_Asp and proper T_TM6 angle). It is worth mentioning that the experimental GS structures were transformed to a CG resolution and minimized using backbone-restrains in order to compare our CG model simulations.

The local stress profile obtained for our POPE bilayer simulations showed a slight increment in the transmembrane pressure profile values with respect to our POPC system, in agreement with previously reported values from similar CG bilayer systems.⁴⁸ Interestingly, the higher lateral pressure values favoured the active state (state 2) PS1 conformations in our CG models with unprotonated Asp385. In order to corroborate the impact of higher lateral pressures in our structural ensemble of GS systems, we performed 1.5 μs CG MD simulations for three model systems and replicas at higher pressures with the unprotonated Asp385 CG model embedded in a POPC bilayer (Figure S8). The resulting probability distribution of these simulations confirmed that an increased lateral pressure of the system promotes the active state (state 2) conformation. Despite the behavior of the high-pressure profile of the system with POPE lipids, no significant changes were observed in the protonated Asp385 CG models with respect to GS in the POPC bilayer.

The local stress profile in our POPA bilayer system showed higher interfacial pressure values in the upper leaflet, suggesting structural modifications in the lipid bilayer due to its interaction with the dynamic GS complex. Additionally, the decreased pressure in the lower leaflet headgroup region confirms the presence of attractive electrostatic interactions between the lipid headgroups and the positively charged amino acids located in the intracellular region of GS. The repulsive interactions between lipids and the GS complex in the upper leaflet and relatively attractive lipid-GS interactions in the lower leaflet are consistent with the small lipid displacement observed in the POPA bilayer relative to the POPC and POPE bilayers (Figure 2C). Nevertheless, we observed no differences in the structural ensemble of GS obtained with the protonated Asp385 embedded in a POPA bilayer compared to the equivalent system simulated in a POPC or POPE bilayer.

The CG simulations allowed us to obtain a general overview of the GS behavior when varying the headgroup charge of the bilayer. In order to obtain an atomistic description of the GS-lipid interactions and explore the origin of the low observed GS activity in the presence of POPE or POPA lipid bilayers,¹⁸ we performed 200 ns all-atom MD simulations (Figure 3) of the GS complex employing the CHARMM36 force field in three single component lipid bilayers (i.e., POPC, POPE and POPA). We evaluated the residence times of lipids located close to GS in order to identify those lipids in contact with the protein more than 70% of the time during the last 100 ns in our all-atom MD simulations. For those identified lipids, we further evaluated the spatial distribution and occupancies in upper and lower leaflets. Figure 3 (A and B) show occupancy plots for POPC, POPE and POPA lipids, located in the upper and lower leaflets, in contact with GS. A greater population of lipids was found in the lower leaflets due to non-bonded interactions of the headgroups (phosphate or phosphodiester groups) with the inner positive amino acids of the complex (Figure S9).⁵⁴ These interactions impact the membrane thickness of the charged headgroup systems (Figure 3C). In addition, we found that none of the lipid headgroups interact with the catalytic aspartates, suggesting that the inhibitory mechanism of POPE and POPA does not require direct contact between the lipids and the catalytic residues as was previously proposed.⁶¹ However, the high POPA lipid occupancy values obtained in the lower leaflet in our all-atom simulations, together with the high lateral pressure values in the upper leaflet and the significant lipid displacement alterations, suggest an alternative mechanism in which GS activity is inhibited in a POPA lipid bilayer due to high PS1 structural restrictions, possibly affecting the substrate recruitment and/or entry into the GS active site.

The lipid residence time analysis of our all-atom MD simulations led to the identification of four key lipid binding sites (LBS) with high occupancy values consistent across the three GS-bilayer systems (Figure 3A and B). LBS1 and LBS2 were assigned to density distributions located respectively in the TM2-TM3 and TM2-TM6 interfaces of PS1 component (Figure 3D). The sites LBS3 and LBS4 were assigned to density distributions located on the internal side of the GS horseshoe and near the juxtamembrane helix of APH-1A (Figure S10). Interestingly, two distearoyl-phosphatidylcholine lipids were also found at these LBS in the 5A63 PDB structure of GS, which confirms the high lipid occupancies in these regions.¹³ In addition to the aforementioned sites, a high prevalence of lower leaflet lipids was observed in regions previously identified as initial substrate-binding sites.^7,62 This observation suggests that displacement of the resident lipids is required in the process of ligand or substrate binding.

Hydrogen bond analysis of our all-atom MD simulations indicate that interactions between the lipid headgroups and the hydrophilic loop (HL1) that connects TM1 and TM2 are responsible for the longer lipid residence times observed at LBS1 and LBS2 (Figures S9 and S11). POPE lipids in our all-atom bilayer system showed the highest number of hydrogen bonds with HL1. The increase in membrane thickness facilitates the residence of POPE lipids in LBS1 and LBS2 (Figure 3D). Consequently, when comparing RMSF values of AA and CG simulations of GS in POPE and POPC lipid bilayers (Figures S4, S5 and S7) significant alterations in HL1 flexibility are observed. The region with the highest ΔRMSF_POPC-POPE values is related to hydrogen bond formation between POPE lipids and Tyr115. It is well known that tyrosine residues are suitable polar ligands for the interaction with phosphodiester groups or positively charged molecules,⁵⁴ and Tyr115 could play an important role in the GS cleavage mechanism due to its close proximity to the active site.^63,64 In addition, alterations in the HL1 movement could explain the experimentally observed POPE inhibition of GS activity.¹⁸

Bilayer thickness and degree and location of lipid unsaturation impacts the PS1 conformational ensemble

Selkoe and coworkers demonstrated that lipid membrane thickness affects the C99 cleavage pattern leading to the formation of different Aβ isoforms.¹⁸ The authors found that thinner membranes favor Aβ42 production, whereas thicker membranes favor the production of Aβ40. In a later study, the same research group⁶⁵ confirmed that increasing the number of carbons in the lipid chain also increase GS activity, reducing the Aβ42/Aβ40 ratio. Despite the membrane composition and PS1 conformational studies, the mechanism by which the different Aβ isoforms are produced remains controversial.⁶⁶ Herein, we focused this part of our study to identify the general behavior of three lipid types containing a phosphatidylcholine (PC) headgroup and different chain lengths: DLPC (12:0/12:0), DPPC (16:0/16:0) and DGPC (20:1/20:1). Figure 2 presents lateral pressure profiles, together with the PS1 structural ensemble, and the local membrane thickness analysis obtained for our CG simulated systems of GS embedded in DLPC (thinnest), DPPC and DGPC (thickest) lipid bilayers. Without considering the thickness distributions, the three systems show similar local stress profiles to that of POPC. The DPPC bilayer has the greatest similarities with POPC because both lipid acyl tails have the same length, differing only by the single unsaturation in the oleyl chain of POPC. Interestingly, the local membrane thickness analysis (Figure 2D) shows a thicker bilayer region close to the SBS1 of GS, in agreement with the lower leaflet density distribution observed in the POPC systems.

Hydrophobic mismatch between a protein and lipid membrane can induce conformational changes in the tilt and rotation of the helices in a way that enhances protein-lipid interactions.⁵³ The hydrophobic mismatches observed in the simulated GS systems appear to modulate the conformational states explored by PS1. From these three systems, the thinner (DLPC) bilayer is observed to increase the population of the active state (state 2) whereas the thickest (DGPC) bilayer exhibits a broader distribution of PS1 conformations with dd_Asp values greater than 1.1 nm. In general, GS with protonated Asp385 simulated in the three lipid bilayers were observed to have a smaller population of active state (state 2) conformations when compared to previous simulations of GS in monounsaturated lipid bilayer systems. Therefore, the absence of unsaturated chains in DLPC and DPPC lipids appears to alter the spatial arrangement and packing of the protein, hindering the transition between the inactive state (state 1) and active state (state 2) conformations.⁵⁷ This result is in agreement with the study of Holmes et al. in which they demonstrated that the trans isomer of a monounsaturated fatty acid chain in phosphatidylcholine lipids increases the activity of the enzyme.⁶⁵ Meanwhile, the protein conformational freedom allowed by the diunsaturated DGPC bilayer thickness results in a broad probability distribution for the inactive state (state 1), reducing the population of active state (state 2) conformations. Moreover, it has been observed that the position of the unsaturation is also crucial for GS activity, with the double bond in the omega-9 position of the oleic acid chain promoting the greatest activity.⁶⁵ These results suggest that GS activity may not be entirely associated with membrane thickness alone but also depends critically on the degree and location of unsaturation. The lipid composition and site-specific protein-lipid interactions may also play a critical role. It is therefore critical to study the effect of thickness variation for those lipid mixtures in which GS has shown increased cleavage activity.¹⁸

Site-specific interaction of cholesterol with GS modulates the PS1 structural ensemble

Cholesterol (CHOL) is an essential component of lipid rafts. Its high concentration in the brains of patients suffering from AD has been related to enhanced GS activity.^56,67 Osenkowski et. al. observed that Aβ production is directly proportional to the abundance of CHOL in various lipid membranes.¹⁸ The relative abundance of CHOL in lipid rafts has been observed to vary from 20 to 25% in the plasma membrane and from 40 to 55% in AD patient brain tissues and Chinese hamster ovary (CHO) cells.^18,23,45

To evaluate the influence of CHOL concentration on the structural ensemble of GS, we conducted CG MD simulations of three POPC:CHOL systems with 20, 40, and 60% CHOL as a function of the Asp385 protonation state. Figure 4 shows the lateral pressure profiles, the probability distributions of dd_Asp distance and T_TM6 angle, and the local bilayer thickness analysis of the POPC:CHOL bilayer mixtures. As expected, increasing CHOL concentration results in an increase in (1) bilayer lateral pressure and (2) bilayer thickness.^68–71 As mentioned earlier, higher lateral pressures promote the population of the PS1 active state (state 2), even in the unprotonated Asp385 CG models of POPC(40):CHOL(60) (Figure S12). This result suggests that lateral pressure at high cholesterol concentrations could facilitate the transition from an inactive state (state 1) to an active state (state 2) of GS. However, such high CHOL concentrations also reduce the flexibility of GS (Figures S4 and S5), which could hinder TM2 movement and blocking the substrate entry to the active site.⁶²

A recent experimental study reported that only 8% of CHOL molecules bind to the GS complex in a membrane with approximately 43% relative abundance of CHOL.⁴⁵ To characterize CHOL binding sites in GS, we performed 200 ns all-atom MD simulations of GS in a POPC(60):CHOL(40) bilayer system (R1). The CHOL molecules with higher lifetimes (residence time > 70%) located close to GS during the last 100 ns of the all-atom MD simulations display seven potential CHOL binding sites (CBS) (Figure 4D). CHOL molecules have longer residence times in the upper leaflet binding sites than the other simulated lipids. Key residues involved in the GS-CHOL contacts were identified using per amino acid occupancy analysis of the all-atom simulation (Figure 4E). To corroborate the occupancy analysis results for the all-atom MD simulation, a second 200 ns simulation (R2) was performed with a different initial CHOL spatial distribution. These AA simulations were analyzed and compared with the CG simulations (Figure S13). In general, we observed high CHOL preference for amino acids located in TM2, TM3, TM5, TM6 and the PAL motif (TM8-TM9) of PS1; TM1 of NCT; TM3, TM6, TM7 and the juxtamembrane helix of APH-1A; and TM3 of PEN2 (Figure 4F). Table S1 of the supplementary material lists the amino acids that constitute the lipids and cholesterol binding sites. Interestingly, the cholesterol binding sites identified as CBS1, CBS3 and CBS5 involve the same amino acids as the lipid binding sites LBS1, LBS4 and SBS1. Moreover, the local membrane thickness of AA simulations shows a thicker region close to the CBS3 site, as observed in the DLPC, DPPC and DGPC CG simulated systems (Figure 4C). Nevertheless, the presence of a CHOL molecule in CBS3 was not observed in the all-atom MD simulation replica (R2), suggesting a low turnover rate with other lipids at this site. Similarly, low occupancy values in amino acids comprising CBS2 were also observed in our R2 atomistic simulation. However, increased CHOL lifetimes in the N-terminal regions of TM2 and TM6 suggest that this molecule prefers the TM interfaces rather than the interaction with the lipid-exposed surface of these TMs observed in the R1 simulation. Note that the separation between TM2 and TM6 facilitating interaction with CHOL could be related to the possible “gate-open” movement essential for the substrate entry mechanism. The CBS2 in our second replica (R2) agrees with the POPC binding site LBS2. The binding of CHOL to this site and to the CBS1 would explain the reduction of HL1 flexibility observed in these simulations (Figures S4 to S7). This suggests the existence of a specific site for CHOL binding that results in the modulation of Aβ production.⁷² Finally, the proximity of CBS4 and CBS7 to the proposed substrate binding sites (SBSs) suggests that these sites could intervene in the initial recruitment of C99 through the recognition of a bound CHOL molecule.^73,74

To validate the observations drawn from simulations of the POPC(40):CHOL(60) CG models, we performed CG MD simulations of GS in three different lipid mixtures consistent with experimental and theoretical abundances in lipid rafts (see Figure 5 and Table 1).^23,45,46 Raft 1 was constructed from the information derived from the lipidome associated with GS, reported by Ayciriex and coworkers.⁴⁵ For the Raft 2 system, the composition of the lipid rafts studied by Risselada and Marrink with CG simulations was used.⁴⁶ Finally, the lipidomic analysis performed by Di Paolo and coworkers on the tissue derived from the brain of an Alzheimer’s disease patient was used for the construction of Raft 3, following Dal Peraro and coworkers.^23,47 The lateral pressure profile analysis obtained for Raft 1 (Figure 5A) results in a similar profile to that observed in a pure-POPC lipid bilayer, while Rafts 2 and 3 displayed alterations in headgroup and interfacial region pressures, while no significant pressure changes were observed in the transmembrane region. The latter suggests that the use of a more complex lipid mixture reduces the effect of transmembrane lateral pressures compared with the POPC:CHOL systems. In addition, Raft 1 (with protonated Asp385) exhibited dd_Asp distance and T_TM6 angle probability distributions similar to those observed in the POPC system with a uniform membrane thickness and a high flexibility of the PS1 TMs (Figure S14A).

NCT ECD adopts a compact conformation in the presence of lipid rafts

In our previous study, we characterized three unique conformational states of NCT ECD: compact, intermediate and extended in agreement with experimental single-particle EM results reported by Chávez-Gutiérrez and coworkers.^30,75 As in the previous study, the major axis length of the GS structures was measured using the distance between the furthest amino acids located at the complex. Figure 5D shows the major axis length distribution of ten GS-membrane CG models with a gray shadow that divides the compact (< 10.3 nm), intermediate (10.3 to 11.3 nm) and extended (> 11.3 nm) conformations. In agreement with our previous study, we observed a dominant preference for the intermediate conformation. This analysis also showed that as the thickness of the membrane increases, the preference for a more extended GS form is preferred. However, Raft 1 and Raft 3 display a wide up/down movement of NCT ECD with a high population of structures close to the compact conformation. This proximity of the NCT ECD to the outer monolayer was also observed in the CG MD simulations of GS performed by Audagnotto et. al. in a lipid raft of similar composition to Raft 3.⁴⁷ The latter behavior of GS embedded in lipid rafts may be associated with the bilayer thickness that facilitates a close interaction between NCT ECD and the lipid headgroups surrounding GS (Figure S14B). Further studies are needed to evaluate the impact of glycosylated NCT ECD in the distribution of compact, intermediate and extended GS conformational states.^76,77