Abstract

The membrane protease γ-secretase cleaves the C99 fragment of the amyloid precursor protein, thus producing the Aβ peptides central to Alzheimer's disease. Cryo-electron microscopy has provided the topology but misses the membrane and loop parts that contribute to substrate binding. We report here an essentially complete atomic model of C99 within wild-type γ-secretase that respects all the experimental constraints and additionally describes loop, helix, and C99 substrate dynamics in a realistic all-atom membrane. Our model represents the matured auto-cleaved state required for catalysis. From two independent 500-ns molecular dynamic simulations, we identify two conformation states of C99 in equilibrium, a compact and a loose state. Our simulations provide a basis for C99 processing and Aβ formation and explain the production of longer and shorter Aβ, as the compact state retains C99 for longer and thus probably trims to shorter Aβ peptides. We expect pathogenic presenilin mutations to stabilize the loose over the compact state. The simulations detail the role of the Lys53–Lys54–Lys55 anchor for C99 binding, a loss of helicity of bound C99, and positioning of Thr48 and Leu49 leading to alternative trimming pathways on opposite sides of the C99 helix in three amino acid steps. The C99 binding topology resembles that of C83-bound γ-secretase without membrane but lacks a presenilin 1-C99 β-sheet, which could be induced by C83's stronger binding. The loose state should be selectively disfavored by γ-secretase modulators to increase C99 trimming and reduce the formation of longer Aβ, a strategy that is currently much explored but has lacked a structural basis.

Introduction

The 99-residue C-terminal fragment (C99) of the amyloid precursor protein (APP) is produced upon cleavage by β-secretase; C99 is subsequently cleaved by the enzyme-complex γ-secretase, releasing the infamous amyloid-β peptides (Aβ) central to Alzheimer's Disease (AD) [1,2]. Aβ isoforms of different lengths are produced during the slow [3] consecutive trimming of C99 within γ-secretase [46]. The longer isoforms, notably Aβ42, are particularly toxic and aggregation-prone and are enriched in the plaques of AD brains and thus considered central to the development of AD [79]. Understanding how Aβ isoforms of variable length are formed within the molecular architecture of the large γ-secretase complex is an outstanding challenge that will be central to efforts to modulate C99 processing and the associated Aβ profile by specific medicine, e.g. γ-secretase modulators [10,11].

During the last decade, various computational and experimental studies have explored the transmembrane domain (TMD) (C99-TMD) structure and dynamics of C99 [1215]. Some studies have identified specific residues of the extra-membrane segment of C99 that may affect Aβ production, and some of these may change upon mutations leading to AD [1618]. Because of the complications of studying disordered proteins in membranes, a structure of full-length membrane-bound C99 is unavailable, and just recently was a structure published of the shorter C83 analog bound to γ-secretase [19]. NMR-derived structures of the TMD of C99 (PDB ID: 2LP1) have been reported [15], but the full structure of C99 beyond the transmembrane (TM) region is yet to be unveiled.

γ-Secretase, a membrane-embedded aspartate protease, comprises four subunits: nicastrin (NCT), the catalytic presenilin 1 (PS1) or presenilin 2 (PS2), anterior pharynx defective-1 (APH1A or APH1B), and presenilin enhancer-2 (PEN2) [20]. Presenilin comprises nine TM helices (TM1-9) and contains the two catalytic aspartate residues in TM6 (Asp-257) and TM7 (Asp-385) that are instrumental to substrate peptide bond hydrolysis of a large variety of type I integral membrane proteins including APP and Notch [21]. γ-secretase is catalytically matured by autocleavage of PS1 to form the N-terminal fragment (NTF, TM1-6) and C-terminal fragment (CTF, TM7-9) [22]. PEN2 and APH1 interact with PS1 and have significant roles in the maturation and assembly of the γ-secretase complex [23]. Furthermore, NCT comprises a single TM in its C-terminal region and a large N-terminal extracellular domain (ECD), which can discriminate access of substrates to the active site by size-selective control [24].

Over the last 4 years, cryo-electron microscopic (cryo-EM) structures have described apo-γ-secretase at increasing resolution (PDB ID: 4UI2, 5A63, 5FN2, 5FN3, 5FN4, and 5FN5) [2528]. New structures of Notch-100 [29] and C83 [19], bound to γ-secretase were published recently. The protein changes TM2 and TM6 conformation during substrate binding as expected from their special flexibility defining the open and closed states of the protein [30]; a new feature of the recent cryo-EM structure is that substrate binding induces a small strand segment in C83 and PS1 and extends the C83 TM helix [19]. Although foundational, the specific atomic motions underlying the density maps, the functionally important loops, the membrane's role in the conformational packing, and the effects of ambient temperature are not captured by the new structure, and it is also of interest to understand whether the natural, more weakly binding substrate C99 differs in its conformation from that of C83.

The mechanism of C99 cleavage depends on an intriguing interplay between loop and helix dynamics, with only the later clearly visible in the cryo-EM data [31,32]. Molecular dynamics (MD) simulations efficiently equilibrate all modes with fast time scales (such as water–protein interactions, water dynamics, and side-chain rotamer states [33]) and provide qualitatively correct dynamics at high resolution for the important loop and helix motions, including the hinge bending critical to this protein complex [30], if their characteristic timescale of typically ∼10–100 ns is covered [34,35]. Accuracy is limited by the realism of the chemical model, the sampling efficiency, and the physical force field. However, the first restriction applies also to the context- and protocol-dependent experimental structures, and for the helix movements, the force field and sampling problem are small compared with e.g. protein folding [3638]. More importantly, atomic resolution dynamics are not achievable in any other way than from MD and thus MD has been instrumental in understanding proteins within realistic membrane environments [13,14,39,40].

As an example of the importance of accurate chemical models, maturation cleavage of the large loop between TM2 and TM6 is crucial for activity, but elusive in any cryo-EM studies since the loops are too disordered [4,41]. The direct role of loop maturation is well understood from MD simulations showing that maturation leads to the two new dynamic fragments that enable access to the catalytic site and modulate the TM2–TM6 gate doors controlling C99 packing [30,42]. The Fit–Stay–Trim (FIST) mechanism [30,43] implies that a closer fit of C99 increases residence time and extends trimming, leading to shorter Aβ; this state is hypothesized to be disfavored relative to an open state by many PS1 mutations leading to increased Aβ42/Aβ40 ratios which directly correlate with clinical severity of disease [44].

We report here the first study of a complete molecular model of C99 within the essentially complete (except the experimentally elusive N-terminal loop of PS1) all-atom structure of γ-secretase. Our model respects all the available structural co-ordinates of γ-secretase and the major dynamics known, as validated below. We employed MD simulations of both the C99–γ-secretase complex and the apoform of γ-secretase in a realistic membrane environment. We specifically accounted for the maturation cleavage of γ-secretase [45,46] which substantially affects the conformational flexibility of the protein complex and is required for activity [42]. We also analyzed the binding free energy of C99 to directly probe the stability of the different conformational states. We identify two major conformation states of C99, a loose and a compact state, which largely confirm the hypothesis of the FIST mechanism and rationalize the observation of long and short Aβ peptides that are centrally important to AD.

Computational methods

Model of C99 bound to γ-secretase

The initial full-length C99 structure was obtained from a recent study [14]. The structure of C99 is sensitive to membrane lipid composition, and thus we used the structure of fragment 23−55 sampled by Dominguez et al. [40], which is in agreement with the experimental data. The N-terminal 1–22 and the N-terminal 56–99 residues were built using dihedral angles predicted via TALOS+ [47], using the Cα, Cβ, C, N, and H chemical shifts reported for C99 in LMPG micelles [48]. The protonation state of each residue of the C99 model was defined using UCSF Chimera [49] and the model was optimized by modeling the missing side chains using WHAT-IF [50]. The complete C99 model was then docked with our previously established semi-open state of γ-secretase [30] to obtain the C99–γ-secretase complex, using the ClusPro server [51] with the two catalytic aspartate residues as restraints. This structure is complete except for the experimentally elusive and functionally unimportant N-terminal of PS1, which has no experimental template and thus cannot be modeled at acceptable accuracy. The obtained cluster with the maximum number of structures having low energy (Supplementary Table S1) was chosen for optimization in the pure lipid bilayer system.

MD simulation of γ-secretase and C99–γ-secretase in a lipid bilayer system

To construct the complete membrane model of the C99-bound γ-secretase, the initial co-ordinates of the membrane were obtained using the Position Protein in Membrane server [52]. Then, the membrane-aligned structures were embedded into a membrane−solvent bilayer system using the CHARMM-GUI web server [53]. A homogeneous POPC lipid bilayer was generated around the protein by maintaining a water thickness of 20 Å using Monte Carlo randomization, and the system was neutralized by the addition of Na+ and Cl ions to a total concentration of 0.15 M NaCl. The resulting protein-membrane structural model was subject to extensive quality control vs. experimental facts as detailed in the Results and Discussion section.

MD production simulations

All-atom MD simulations were performed with the TIP3P explicit water model [54] and the structure-balanced CHARMM36m force field [55] in the graphical processing unit version of GROMACS (2018.2) [56]. Non-bonded interactions were evaluated using a Verlet cutoff scheme, with both Coulomb and Lennard-Jones interactions having a cutoff of 12 Å. Particle Mesh Ewald summation was used for long-range electrostatic interactions, and hydrogen bond lengths were constrained using Linear Constraint Solver (LINCS) algorithm [57]. The complete protein−lipid−solvent system was minimized by steepest descent optimization, followed by six-step position-restrained equilibration for 500 ps. During the first two steps, the system was equilibrated in a canonical (NVT) ensemble using Berendsen temperature coupling, and in the next four steps, equilibration was achieved within the isothermal–isobaric (NPT) ensemble.

After equilibration, 500 ns production simulations were carried out with a 2-fs time step in an NPT ensemble using the Nose−Hoover thermostat and the Parrinello−Rahman barostat at 303.15 K and 1 atm, respectively. We have recently found [30] that it takes 200 ns to reach stable trajectories as estimated from the horizontal RMSD graphs vs. initial structures. In addition, the functionally important helix motions of PS1 are, as expected [33,35], well sampled on a time scale of 100 ns after the first 200 ns required for equilibration, and a 500-ns simulation of the full system thus samples the central dynamics qualitatively well. To ensure this, we performed two complete 500-ns simulations using distinct initial random velocity seeds. As seen from the analysis below, these simulations are in essential agreement and serve as mutual confirmation. Both simulations were performed during 2018 using the graphical processing units of the high-performance-computing facility at the Technical University of Denmark.

Trajectory analysis

The obtained trajectories were analyzed using the built-in analysis tools of GROMACS version 2018.2 [56]. The deuterium order parameters of the acyl chains, density of the membrane environment, area per lipid head group, and bilayer thickness were analyzed using GROMACS utility tools and FATSLiM [58] to ensure the optimal quality control of the protein-membrane system. The backbone root-mean-square deviation (RMSD), Cα root-mean-square fluctuation (RMSF), and solvent-accessible surface area (SASA) were analyzed. All analyses presented in this work reflect the last, stable 300 ns of the simulations, since, as explained above, the first 200 ns are required to reach an equilibrated stable state. Ensemble-average clustering was performed using the clustering algorithm of Daura et al. [59] implemented in the gmx cluster tool of GROMACS with a cutoff 0.2 nm, applied to the last 300 ns of each simulation. Principal component analysis (PCA) and the corresponding collective motions were analyzed using the last 300 ns information of each simulation. Visual Molecular Dynamics v.1.9.3 [60] was used for visualization of MD trajectories, and PyMOL version 2.0 was used for rendering of figures.

Estimation of binding free energy of C99 to γ-secretase

The free energy of binding C99 to γ-secretase was determined using the g_mmpbsa tool [61], which employs molecular mechanics-based Poisson–Boltzmann Surface Area (MM/PBSA) calculations [62]. These energies were calculated for each of 100 frames extracted equidistantly from the last 300 ns of each MD trajectory. The total binding energy (ΔGbind) is the average (with reported standard deviations from this average) of all 100 calculations. We used solute and solvent dielectric constants of 2 and 80, respectively. Energy decomposition was carried out to understand the contribution of individual amino acids to the total obtained binding energy.

Results and discussion

Structure and dynamics of C99–γ-secretase within a membrane

Despite significant advancements in the general structure of γ-secretase, many features of the C99 binding remain unknown [25,27,28,32]. A very recent cryo-EM structure of C83-bound γ-secretase in ice [19], published after we submitted our paper, reveals important changes during substrate binding. Our MD simulations supplement with the underlying atomic dynamics causing these changes for the full C99 substrate in a relevant membrane environment at an ambient temperature. The membrane cleavage of C99 leads to the formation of Aβ and is therefore a cornerstone of structure-based attempts to understand AD and to develop causative medicine against the disease.

Based on the initial docked conformations of C99 obtained from ClusPro, PS1 and NCT consistently interacted strongly with the substrate (Supplementary Figure S1 and Table S2). After MD simulation, the C99 binding pocket comprised TM2 (Gln127–Tyr156), TM3 (Val161–Tyr189), and TM5 (Leu219–Tyr240), where TM2, TM6 (Glu243–Val261), TM9 (Ala434–His463), and the PAL motif (Pro433–Ala434–Leu435) interact closely with the substrate. The positioning of the C99–TMD (Gly29–Leu52) within the membrane is of particular interest and is analyzed further in the next sections. The TMD is strongly fixated by the positively charged lysines Lys53–Lys54–Lys55 of C99. Cleavage assays in cells and alanine-scanning mutagenesis suggest a central role of this motif in controlling substrate cleavage [63,64]. Other non-bonded contacts with C99–TMD are shown in Supplementary Table S2. The details of each subunit interaction with C99 resulting from full MD equilibration are discussed further in the next section.

One of the novelties of our work is to provide the membrane context of the substrate binding. To test the quality and realism of our protein-membrane model, we evaluated the membrane order parameters (SCd) of the head groups and acyl chains of the lipid tails, the density of membrane, and properties of the POPC bilayer. The membrane order parameters (Supplementary Figure S2) estimate the configurational flexibility of the lipids as a function of position within the lipid chain. The order parameters of the lipid head groups in Supplementary Figure S2 and the calculated membrane area per lipid molecule and membrane thickness (Supplementary Figure S3) show excellent agreement with the typical experimental values of these properties and also agree well with previous MD studies [65,66]. Most importantly, they provide quality control that the membrane system retains its conformational integrity during the simulations.

The equilibrated membrane structures can be compared with experimental structures without the membrane. The RMSD for the PS1 Cα atoms between the most representative computed structure from cluster analysis and the experimental apo-structure 5FN2 [27] is 1.55 and 1.49 Å for simulation 1 and simulation 2 (Supplementary Figure S4), respectively. The corresponding simulated apo structures displayed RMSDs of 1.30 and 1.57 Å, respectively. Compared with the new cryo-EM structure of C83–γ-secretase (PDB ID 6IYC) [19], the RMSD for Cα atoms of PS1 is 1.81 and 1.88 Å for simulation 1 and 2, respectively. The MD-equilibrated membrane-bound topology is thus in very good agreement with the topology of the cryo-EM structure for the parts that are not missing in the experimental structure, but differs in some parts close to the membrane and C99, as discussed below.

Our complete structural model is shown in Figure 1A, with Na+ and Cl ions shown but water molecules removed for clarity. Figure 1B shows the RMSD from the start structure at 200–500 ns during the two independent 500-ns simulations of this system. The full complex displays stable states after 200 ns, implying that we can collect statistics from the last 300 ns, as done below. The simulations reveal the important dynamics of the individual subunits and C99 discussed in further detail below. The most important of these is arguably the distance between the two catalytic aspartate residues (measured as the Cα–Cα distance) which displays considerable dynamic variation during the simulations in both cases (Figure 1C). This and other dynamic variations are relevant to the C99 binding and processing and are analyzed further below.

Properties of the simulated protein–membrane system.

Figure 1.
Properties of the simulated protein–membrane system.

(A) Complete structural model of membrane-bound γ-secretase with the C99 fragment bound. (B) Stability of trajectories based on the root-mean-squared deviation (RMSD) of the backbone atoms. (C) Average distance between the catalytic Asp-257 and Asp-385 Cα atoms of C99–γ-secretase during the last 300 ns of all-atoms MD simulation.

Figure 1.
Properties of the simulated protein–membrane system.

(A) Complete structural model of membrane-bound γ-secretase with the C99 fragment bound. (B) Stability of trajectories based on the root-mean-squared deviation (RMSD) of the backbone atoms. (C) Average distance between the catalytic Asp-257 and Asp-385 Cα atoms of C99–γ-secretase during the last 300 ns of all-atoms MD simulation.

Individual subunit contributions to C99–γ-secretase dynamics

To understand the underlying dynamics of the system in more detail, we analyzed the structural changes (as measured by co-ordinate RMSD plots) and fluctuations in atomic positions of all four subunits and C99 separately with respect to the initial pre-equilibrated structure. Figure 2A–E shows the structural variation for NCT, PS1, APH1, PEN2, and C99 along the equilibrated part of the simulation trajectories for simulation 1 (black curves) and 2 (red curves). Both systems reach dynamically stable states as seen from the predominant horizontal plots. However, the RMSD plot of PS1 (Figure 2B) indicates two distinct conformations in the two simulations, despite minor differences in the other subunits (Figure 2A,C,D). Finally, C99 displays structural variation within the overall enzyme complex, indicating that C99 is flexible at the membrane interface when bound (and probably more flexible than C83, see below), which may be important to the generation of different types of Aβ peptides, as explored in more detail below.

The root-mean-squared deviation relative to the initial prepared model (RMSD) and averaged root-mean-squared fluctuations (RMSF) of individual residue Cα atoms in C99–γ-secretase.

Figure 2.
The root-mean-squared deviation relative to the initial prepared model (RMSD) and averaged root-mean-squared fluctuations (RMSF) of individual residue Cα atoms in C99–γ-secretase.

(A) NCT RMSD; (B) PS1 RMSD; (C) APH1 RMSD; (D) PEN2 RMSD; (E) C99 RMSD; (F) NCT RMSF; (G) PS1 RMSF; (H) APH1 RMSF; (I) PEN2 RMSF; (J) C99 RMSF. Black and red colors represent simulation 1 and simulation 2, respectively.

Figure 2.
The root-mean-squared deviation relative to the initial prepared model (RMSD) and averaged root-mean-squared fluctuations (RMSF) of individual residue Cα atoms in C99–γ-secretase.

(A) NCT RMSD; (B) PS1 RMSD; (C) APH1 RMSD; (D) PEN2 RMSD; (E) C99 RMSD; (F) NCT RMSF; (G) PS1 RMSF; (H) APH1 RMSF; (I) PEN2 RMSF; (J) C99 RMSF. Black and red colors represent simulation 1 and simulation 2, respectively.

Figure 2F–J shows the corresponding thermal fluctuations averaged over the last 300 ns, divided into residue. We identify high mobility of the NCT extracellular loop region and C-terminal and some, but comparatively little mobility of TMs (Figure 2F). The high thermal fluctuation of the C-terminal is a common feature also seen in APH1 (Figure 2H) and PEN2 (Figure 2J). TM5-7 of APH1 and some parts of PEN2 exhibit fluctuations due to their exposure to the membrane lipids and fewer protein–protein contacts and display very similar dynamic behavior in both simulations. The dynamics of PEN2 and APH1 compared with the dynamics of the other two subunits is intriguing: PS1 (Figure 2G) shows major fluctuations across its sequence, especially of TM2, TM9, the hydrophilic loop 1 (HL1), and the large maturation-cleaved cytosolic loop between TM6 and TM7 (HL2). The maturation is instrumental in producing the distinct conformational variability of PS1, as evident also from comparative PS1-alone simulations with and without maturation cleavage [42]. The TMs of PS1 that are close to PEN2 and APH1 γ-secretase display less mobility. Previous studies suggest that PEN2 and APH1 stabilize the catalytic site [67], and the compact dynamic states of these subunits (Figure 2H,I) probably cause this tendency. Specifically, we find that the major helix of NCT forms important contacts with TM3–TM4 (Tyr195–His214) of PS1, and the N-terminal region of PEN2 interacts with the PS1 near the catalytic site (Supplementary Figure S5). HL1 interacted with the TM of C99 and displayed high flexibility during the MD simulations indicating a role in substrate recognition. This corroborates earlier studies by Tomita et al. [68] and is also partly evident from the cryo-EM structure of PS1–γ-secretase with C83 [19], although the thermal motions are expected to be much smaller at cryo-temperature and thus possibly lead to only one bound conformation state.

The higher RMSF of TM2 and TM6 suggests that collective motions modulate the relative orientation of these helices within the membrane when C99 is bound. Tominaga et al. [69] found that changes in the distance between TM6 and TM7 (Val379–Thr399) relate to Aβ42 production, signifying the critical role of TM6 conformation in C99 processing. We have recently shown [30] that TM2 and TM6 form gate doors to the active site, controlled by the mature HL2 acting as a plug preventing access to this gate; however, these and all other structural findings reflect γ-secretase without C99, and thus our new results suggest that the most important TM motions are present both with and without substrate bound, although their magnitude changes. Cryo-EM structures also indicate substantial motion of these helices [27], and at ambient temperature, these tilt changes are major, as seen from the RMSF data in Figure 2G.

C99 within γ-secretase has a compact TM region but flexible membrane association

Our simulations also provide the first full-atom structural dynamics of C99 bound to γ-secretase in the membrane, as summarized by the RMSD and RMSF plots in Figure 2E and J, respectively. The two statistically independent simulations qualitatively display similar thermal fluctuations. The disordered N-terminal residues (1–14) of C99 display some flexibility, but these motions monotonically decrease along the sequence of C99 as the membrane dampens C99 motions. We find that the C99–TMD, specifically residues 29–52, is completely dampened, such that dynamic control is exerted by the PS1 states.

The hinge G37–G38 region of the C99–TMD, speculated to be vital for processing by γ-secretase [12] displayed small variations, slightly larger than the neighboring, very fixed membrane residues. Previous MD simulation studies have shown that the GG hinge defines a dimerization motif and is thus conformation-wise important [13,40,70]. Both NMR and MD simulation studies imply that G37–G38 of C99 can act as dynamic hinges [13,15,71], where the ‘hinge-bending’ presumably assists substrate binding within the γ-secretase active site [72]. The C-terminal part of C99 outside the membrane was very flexible in both simulations but with larger thermal fluctuation in simulation 1 (black, Figure 2J). Circular dichroism spectroscopy and amide exchange experiments have shown that mutations of Thr43 and Thr48 to hydrophobic valines modulate the extent and direction of helix bending without altering the local flexibility at the ε-sites [73], but affecting the generation of Aβ42 and Aβ40 [74]. Our results provide a plausible explanation of the altered cleavage as due to the large bending flexibility of the C99–TMD [75]. The bend motion profoundly affects the exact Asp–Asp distance dynamics within the catalytic site and thus probably the pattern of proteolytic processing. We conclude that there are minor fluctuations in C99–TMD once fit within the catalytic site, but major dynamics at the interface of the membrane interacting with the disordered N- and C-terminal parts of C99; the N-terminal part is not present in the new experimental cryo-EM structure of C83 [19]. We show below that these membrane-protein dynamics give rise to two conformational states of bound C99 with distinct catalytic Asp–Asp distances, which can explain the two C99 cleavage pathways leading to Aβ and the production of shorter and longer peptides, which is elusive in the one-state C83 cryo-EM structure.

Essential dynamics of membrane-bound C99–γ-secretase

To understand the atomic contributions to the subunit dynamics in Figure 2 in detail, we performed PCA to identify the most significant collective motions in the phase space, emphasizing the first two eigenvectors (EV1 and EV2). Figure 3A,B shows EV1 and EV2 for simulation 1, whereas Figure 3C,D shows EV1 and EV2 for simulation 2. Structural motions in NCT-ECD occur both in the part of NCT that orients toward APH1 and in the part that interacts with PS1 and the membrane (top, green subunit). The latter motions affect atoms associated with the PS1–C99 complex consistent with a role of these NCT motions in substrate recognition [24]. PCA confirmed that the NCT–ECD interacts with HL1 of PS1 in and open–close movement, suggesting that it can affect substrate entry into the PS1 active site. Compared with the PCA of the apoform of γ-secretase in this work (Supplementary Figure S6) and recent MD simulation studies [76], the large C99 substrate restricts this movement somewhat via packing interactions with HL1; thus, the bound state displays a somewhat dampened NCT motion relative to the apostate of the protein complex.

Major dynamics of γ-secretase bound to C99 in a membrane.

Figure 3.
Major dynamics of γ-secretase bound to C99 in a membrane.

(A) Porcupine plot of eigenvector 1 motions from simulation 1. (B) Porcupine plot of eigenvector 2 motions from simulation 1. (C) Porcupine plot of eigenvector 1 motions from simulation 2. (D) Porcupine plot of eigenvector 2 motions from simulation 2. Red needles show the direction and amplitude of motions. (EH) The structural superimposition of the representative top-ranked cluster structure (E,G) and top-two cluster (F,H) of C99–γ-secretase colored by different subunits (Green: Nicastrin, PS1-NTF: Cyan, PS1 CTF: Magenta, APH1: Yellow, PEN2: Wheat and C99: Light gray).

Figure 3.
Major dynamics of γ-secretase bound to C99 in a membrane.

(A) Porcupine plot of eigenvector 1 motions from simulation 1. (B) Porcupine plot of eigenvector 2 motions from simulation 1. (C) Porcupine plot of eigenvector 1 motions from simulation 2. (D) Porcupine plot of eigenvector 2 motions from simulation 2. Red needles show the direction and amplitude of motions. (EH) The structural superimposition of the representative top-ranked cluster structure (E,G) and top-two cluster (F,H) of C99–γ-secretase colored by different subunits (Green: Nicastrin, PS1-NTF: Cyan, PS1 CTF: Magenta, APH1: Yellow, PEN2: Wheat and C99: Light gray).

We also find that the large HL2 (as shown in Figure 3C,D), which has been cleaved in our modeled mature state of the protein, displays a characteristic ‘left-right’ and ‘in-out’ movement, which can affect not only the translocation of the C99 to the catalytic center but also water access, important because a water molecule is required as nucleophile in the active site, and membrane-channel-like morphology. As shown before, in the immature state, the HL2 is a plug in the active site, whereas autocleavage of the plug enables access to the active site in the mature protein complex [42]. Because of this maturation, the movements of HL2 are substantially more significant than those of HL1. Porcupine plots of PC1 and PC2 of the catalytic subunit PS1 (Supplementary Figure S7) show strong inward–outward motions of HL2, whereas negligible movements occur for HL1. The significant motions of HL2 contribute to the strong fluctuations in the RMSF plot (Figure 2).

The two most representative snapshot structures for each simulation are shown in Figure 3E–H. C99-bound γ-secretase is characterized by several conformational states, which in the case of PS1 even affects the secondary structure, which is unusual. The representative structures show different packing of the TM helices of PS1 around C99 (cyan) resulting in differential compactness. In one state, C99 shows large dynamic interaction with NCT (gray, Figure 3E), which is elusive in the cryo-EM structure because the N-terminal of C99 is missing in C83. The residue-wise thermal fluctuation parameters in Figure 2J document these structural differences as well as a major conformational flexibility of C99 in its out-of-membrane loop interactions, bottom of Figure 3, which will probably be smaller at cryo-temperature.

We computed the intermolecular contacts between C99 and γ-secretase (Supplementary Table S3 and S4) from the two simulations. The number of intermolecular hydrogen bonds formed between NCT and PS1 and C99 differed significantly between the two ensembles, such that one ensemble represents a more compact C99 state (simulation 2: Figure 4), and the other represents a distinctly looser C99 state (simulation 1) (Supplementary Figures S8 and S9). The compact conformation displayed the shortest average distance between interaction pairs and the most persistent C99–protein interactions, with the longest residence time of C99–γ-secretase interactions (Figure 4). Most residues of the C99–TMD (Gly29–Leu52), including the sites of ε-cleavage (Ala42), ζ-cleavage (Ile45), γ-cleavage (Thr48 and Leu49), and substrate recognition (Met51, Lys52) interact strongly with PS1 by hydrophobic interactions or hydrogen bonds. The triplet lysine anchor Lys53–Lys54–lys55 engages in direct salt bridges as well as persistent hydrogen bonding to PS1.

Dynamic persistency of C99 interactions with γ-secretase.

Figure 4.
Dynamic persistency of C99 interactions with γ-secretase.

The minimum inter-atomic distance profile of key residues interacting with C99 in the compact ensemble (simulation 2, 200-500 ns).

Figure 4.
Dynamic persistency of C99 interactions with γ-secretase.

The minimum inter-atomic distance profile of key residues interacting with C99 in the compact ensemble (simulation 2, 200-500 ns).

Assuming that longer residence time of C99–protein interactions correlates with the extent of cleavage of C99, our identified two states of C99 may explain the changes in the production of longer (loose state) vs. shorter (compact state) Aβ peptides in various forms of the protein. This possibility is substantiated below by calculations of the stability of the enzyme–substrate complexes and PS1 helix tilt angles relating directly to the Asp–Asp distance.

PS1 helix tilt angles define C99 residence time

Recent MD simulations of γ-secretase indicate that the compactness of the substrate-binding site and the distance between the catalytic aspartates (ddAsp) in PS1 depend on the orientation of different TM helices [76]. The functionally important ddAsp fluctuates around an average value of ∼1 nm. To understand the helix translocations, which are known to be well sampled on the applied timescale [34], we computed the distribution of TM tilt angles as shown in Figure 5. These angles effectively differentiate the various conformational states of PS1 in γ-secretase. TM2 displays a notable two-state behavior in simulation 1 (Figure 5 top), associated with a weak two-state behavior of TM3 and a broader TM6 tilt angle distribution. Intriguingly, these three helices (TM2, TM3, and TM6) also display correlated motions in the apo-complex and lead to open, semi-open, and closed conformational states [30]. Whereas most helices are more compact in simulation 2, TM8 is flexible in both simulations and displays a clearly visible two-state dynamics in simulation 2. Thus, from a protein perspective, the C99-bound states display flexibility that resembles the apo-complex, with simulation 1 showing most of the variation in helix tilt angles.

PS1 and C99 helix dynamics.

Figure 5.
PS1 and C99 helix dynamics.

Distribution of tilt angles of the PS1 and C99 TMs calculated from the last 300 ns trajectory of the two replicates.

Figure 5.
PS1 and C99 helix dynamics.

Distribution of tilt angles of the PS1 and C99 TMs calculated from the last 300 ns trajectory of the two replicates.

To understand the impact of these TM helix movements on the catalytically important ddAsp, which potentially defines residence time and processing of C99, the simulated structural ensembles were projected onto the ddAsp distance and tilt angles for all TMs in Figure 6. Supplementary Figures S10 and S11 display the corresponding distributions of the ddAsp distance and the tilt angles for the apoform of γ-secretase, to understand the impact of C99 binding. Most importantly, both the apo-γ-secretase (Figure S11) and C99-bound γ-secretase (Figure 6) ensembles feature a loose and a distinctly more compact conformation state. For the loose apostate, ddAsp reaches 1.3 nm, whereas the compact state features ddAsp in the range 0.9–1.1 nm. Once C99 is bound, ddAsp of the loose state varies from 0.98 to 1.15 nm, while for the compact state varies from 0.90 to 1.02 Å. These differences are highly significant and the opening correlates with changes mainly in TM2, TM6, and TM8 (Figure 6, top). Not only the variability but also the distribution of the tilt angles (horizontal axis in Figure 6) change. TM6 undergoes an important tilt angle modification, which strongly correlates with the modulation of ddAsp. In addition, changes in the tilt angles of TM2 and TM9 correlate with those in tilt angles of TM6 and TM7. Experimental studies have suggested that TM2 and TM9 contribute to the initial substrate-binding site in PS1 enabling C99 to enter either between TM6 and TM9 or between TM2 and TM6 [67,77]. The higher flexibility of TM2 confirms its recently suggested role in the gate opening [30]. The wide tilt angle distribution of TM2 and TM9 affirms the high mobility of these helices seen in previous studies [27].

The distribution of snapshots projected onto the distance between the catalytic Asp-263 and Asp-366 and the tilt angle distribution of TMs during the last 300 ns of the two MD simulations of C99–γ-secretase.

Figure 6.
The distribution of snapshots projected onto the distance between the catalytic Asp-263 and Asp-366 and the tilt angle distribution of TMs during the last 300 ns of the two MD simulations of C99–γ-secretase.

The colored scale defines the relative populations.

Figure 6.
The distribution of snapshots projected onto the distance between the catalytic Asp-263 and Asp-366 and the tilt angle distribution of TMs during the last 300 ns of the two MD simulations of C99–γ-secretase.

The colored scale defines the relative populations.

Distance measurement was performed between terminal Cα atoms as shown in Supplementary Figure S12. In the loose state of simulation 1, all the TMs except TM3, TM7, TM9, and TMD of C99 displayed higher fluctuation in distances. In contrast, the RMSD of TM3, TM7, and TM9 in the compact state demonstrated higher mobility (Supplementary Figure S13), signifying that the substrate binding induces cooperative changes in the structure.

Atomic resolution mechanism of C99 binding and processing within γ-secretase

After identifying two major conformation states of C99–γ-secretase, we explored the details and contributions of individual amino acids of both C99 and γ-secretase to these two states. To obtain the energy contribution of each residue to the total binding energy, MM/PBSA calculations were performed for 100 equidistant snapshots of the last 300 ns of each ensemble. This local binding energy obtained from such calculations relates to KM, which represents the lifetime of the active poses of C99 leading to a productive turnover. This is distinct from the total binding free energy (affinity) of C99 to the protein complex, which involves many other nonproductive conformation states. kcat of γ-secretase is very small, of the order of 0.0012 s−1 [3], possibly due to the rate-limiting access of nucleophile water to the buried active site of C99–γ-secretase. KM has been measured to be surprisingly large, 0.2 mM, for a substrate size of C99 [3], and thus the turnover number becomes extremely small, of the order of 6 s−1 M−1, and this suggests kinetic control of the cleavage products probably governed by a higher kcat of the more loose state but a lower KM of the more compact state.

The calculated energies of binding of C99 are summarized in Supplementary Table S4. The absolute binding energies are meaningless since the entropy is not included, whereas the relative values indicate a significant (vs. the standard deviations of the 100 calculations across the ensemble) difference in the binding energy of C99 for the two conformation states. We also computed the intermolecular contacts and contribution of each residue of γ-secretase to the C99 binding energy (Supplementary Figures S14 and S15). The two simulations revealed slightly different energies of contributing residues, but the major contributing residues were generally identical. The salt bridges dominate the energetics of binding, not surprisingly, but will be greatly compensated by entropy and thus again, the scale of energies is not physically meaningful, only the qualitative contribution.

We note that the Lys53–Lys54–Lys55 anchor interacts with both the CTF and NTF of PS1, which has implications for the two cleavage pathways initiating at Aβ49 and leading to Aβ40 in one case, and initiating with Aβ48 and leading to Aβ42 in the other case. The electrostatic interactions, which dominate the energetics of C99 binding, control the exact positioning of Thr48 and Leu49 in the two ensembles. An important observation is that Thr48 (belonging to the ε-cleavage site) is involved in hydrogen bonding with PS1–NTF, whereas Leu49 is hydrophobically packed with Ile387 of PS1–CTF; thus, maturation cleavage produces two fragments that interact with each of the sites required for initiation of cleavage in the two pathways leading to Aβ49 and Aβ48. With the catalytic aspartates positioned on each site of C99 with access to one of these two sites, our MD simulations directly provide an atomic resolution model of the processing of C99 along the two pathways. Since the C99 forms a TM helix in this part, the next accessible peptide bond after Thr48–Leu49 on one side or Leu49–Val50 on the other side occurs three amino acids down the sequence of C99, and the simple translocation of the C99 helix thus most probably occurs directly after the first cleavage, leading to Aβ42 and Aβ40, respectively.

We also computed the minimum distance between the interacting atom pairs of the enzyme and substrate. As shown in Figure 7, the interacting atom pairs of the compact C99-bound state displayed a strong and stable trend in distance, while the loose binding mode (simulation 1) displayed variable distances (Supplementary Figure S9). We suggest that these two states define the different processing of C99 leading to two trimming pathways and to longer or shorter Aβ peptides. The two fits of C99 may explain the experimental observation of two trimming pathways [6,78]. The loose conformation displays fewer interactions and is associated with shorter C99 residence time and by inference releases Aβ peptides of longer lengths.

Summary of important C99 interactions with γ-secretase.

Figure 7.
Summary of important C99 interactions with γ-secretase.

(A) Intermolecular contacts between NCT and PS1 (NTF and CTF) and C99 obtained from the top-ranked cluster of the compact ensemble of simulation 2. The catalytic Asp257 and Asp385 have been marked in bold. (B) Line ribbon representation showing interaction of C99 (magenta) with NTF and CTF of PS1. Residues of C99 and PS1 forming hydrogen bonds and salt bridges are visualized in stick model. Yellow and green dotted lines represent the electrostatic and hydrogen bond interactions. The interacting residues of NTF are blue, those of CTF dark brown, and those of C99 green.

Figure 7.
Summary of important C99 interactions with γ-secretase.

(A) Intermolecular contacts between NCT and PS1 (NTF and CTF) and C99 obtained from the top-ranked cluster of the compact ensemble of simulation 2. The catalytic Asp257 and Asp385 have been marked in bold. (B) Line ribbon representation showing interaction of C99 (magenta) with NTF and CTF of PS1. Residues of C99 and PS1 forming hydrogen bonds and salt bridges are visualized in stick model. Yellow and green dotted lines represent the electrostatic and hydrogen bond interactions. The interacting residues of NTF are blue, those of CTF dark brown, and those of C99 green.

Recent studies have shown two plausible binding sites of PS1: TM2, TM6, and TM9 form one site, and PS1-NTF, PEN2, and NCT-ECD form the other site [67,79]. In addition, coarse-grained MD simulations suggest that C99–TMD can bind to the catalytic subunit (PS1) via TMD2/6/9 [80]. We computed the intermolecular contacts of the top-ranked two structures obtained from clustering for both our simulations. We find that Met51 of C99–TMD forms crucial contacts with Ile149 and Met152 of TM2, while Val44 forms hydrophobic contacts with Leu166 (TM3). Furthermore, the key catalytic Asp257 (TM6) and Asp385 (TM7) form strong hydrogen bonds and salt bridges with the triplet Lysine anchor (Lys53–Lys54–Lys55 situated at the C-terminal end of C99-TMD), an indispensable element for substrate recognition [64] (Figure 7 and Supplementary Figure S9). Recent studies [79,81] suggest that Val44, Leu49, Met51, and Leu52 residues of the C99–TMD act as the γ-secretase interaction sites, an observation that is supported and explained by our atomic resolution simulations. Also supporting our obtained ensembles, Xu et al. [63] previously showed that the conserved positively charged residue Lys53 in the C99–TMD crucially contributes to substrate recognition and cleavage. Mutational studies have confirmed the requirement of Lys53 within the C99–TMD and Notch (another major γ-secretase substrate), indicating that substrate recognition by γ-secretase is highly conserved. Our results not only show this role clearly but also indicate that Lys53 along with Lys54 and Lys55 contributes substantially to the substrate recognition via electrostatic interactions (Figure 7 and Supplementary Figure S9). In total, NTF of PS1 forms substantially more intermolecular contacts with C99 compared with CTF.

Since γ-secretase controls Aβ production, it offers a primary therapeutic target for AD. The TM fragment of C83 is cleaved by γ-secretase to produce either Aβ48 or Aβ49; additional trimming releases Aβ45/42/38 or Aβ46/43/40 [6,82]. Comparison of our C99–γ-secretase structure in the membrane to the new structure of C83-bound γ-secretase in frozen ice [19] suggests that the overall binding topology is very similar for C99 and C83 and with and without the membrane: NCT and HL1 interact with the N-terminal of C99, and TM6 and TM7 of PS1 change conformation during binding, as compared both to previous apo-γ-secretase simulations in the membrane, and corresponding experimental cryo-EM data without the membrane.

The new structure of C83 [19] suggests that the TM helix of C83 is partially unwound once bound, which could aid intramembrane proteolysis, and the substrate forms both a helix and a β-strand segment (Supplementary Figure S16A). The unwinding plausibly exposes the peptide bonds to cleavage either at Thr719–Leu720 (resulting in Aβ48) or at Leu720–Val721 (yielding Aβ49). In our simulations, we observed loss of helicity of the C99–TMD (Supplementary Figure S17), which supports that a helix unwinding also occurs in the membrane. The susceptibility of C83 to sequential trimming may be due to the presence of several helix-destabilizing residues in its TM that promote helix unwinding, rationalized by our data in Supplementary Figure S17.

A specific feature of the new cryo-EM structure is a three-stranded hybrid β-sheet comprising two β-strands of PS1 (in the cytoplasmic HL2) and one from C83 (Met722-Lys725) (Supplementary Figures S16B,C). It is interesting to know whether the membrane and C83/C99 difference changes this tendency. In the cryo-EM structure, the induced β-strand of C83 interacts with the PAL motif, which in turn stabilizes the C83 β-strand. In our structures produced before publication of the new C83 structure, we observed two β-strands (Lys286–Ser290 and Thr327–Asn331) in HL2. The unusual propensity for strand formation in PS1 was first observed by MD simulation in 2016, although not discussed in the paper by Somavarapu and Kepp [42]. In our simulations, C99 did not form any β-strand (Supplementary Figure S16B) and thus we do not observe a substrate–PS1 β-sheet recognition motif. C83 binds substantially more strongly than C99 and may thus induce structure not seen for C99, or cryo-temperature could plausibly stabilize it. Alternatively, our MD simulations may be inaccurate on this detail. Regardless of the answer, it is evident from what is now known that HL2 has major secondary-structure variability that is important for both autocleavage (requiring helix in the loop when it becomes substrate) and substrate recognition.

We suggest that the extent of C99 trimming depends on the relative prevalence (as determined by the relative free energy) of the compact and loose conformation states identified in our work. This balance could easily be affected by binding small molecules or by mutation. The loose state is associated with less favorable interactions with the catalytic site and reduced residence time of important interactions, and this will plausibly reduce the extent of C99 trimming. These findings are in accordance with the FIST model (Fit–Stay–Trim) of C99 processing [42,43]. They are also supported by the observation that Aβ42/Aβ40 ratios of FAD-causing PSEN1 mutations correlate with disease severity [44], and both of these correlate with the tendency of the mutation to chemically stabilize the hydrophobic packing of the protein complex within the membrane, with PSEN1 mutations generally losing thermochemical stability and/or hydrophobicity [83].

Conclusions

γ-secretase, an important intramembrane-cleaving aspartyl protease, is responsible for cleaving C99 into the Aβ peptides central to AD, with longer Aβ peptides being more aggregation-prone and toxic and enriched in senile plaques of AD patients. The molecular processing of C99, well known from cell assays, has missed a structural basis until the new cryo-EM structure of C83 in frozen ice [19]. Translocation of the substrate to the active site is known from previous studies to involve major conformational changes of the γ-secretase complex [32,69,77]. The previous study [79] demonstrated that C99 binds first to NCT and PEN2 and then subsequently binds PS1–NTF and the active site.

Our all-atom MD simulations describe the probable modes of binding of full-length C99 to mature γ-secretase in a pure lipid bilayer at ambient temperature and thus substantially complement the new cryo-EM structure of C83–γ-secretase. The model obeys all relevant experimental constraints and was subject to extensive quality control, with its topology in excellent agreement with known direct (electron microscopy) and indirect (assaying and site-directed mutagenesis) data. Our study reveals that the weaker binding of C99 vs. C83 induces variability in the binding mode and produces two major binding modes of C99 which differ in compactness, stability, and retention time near the catalytic aspartates. The open state features longer catalytic Asp–Asp distance correlating with changes in TM2, TM6, and TM9. Both states are in equilibrium and suggest that the longer substrate retention and tighter intermolecular binding in the compact state lead to the maximum trimming of C99 and thus shorter Aβ peptides. We hypothesize that FAD PS1 mutations shift the equilibrium in favor of the loose ensemble to a variable extent.

We find that the GxxG motif of the N-terminal of C99 contributes consistently to the packing of C99 within γ-secretase. Supporting recent observations by Yan et al. [64], the triplet lysine anchor (Lys53–Lys54–Lys55) substantially contributes to C99 binding. We show that this occurs in a cavity formed primarily by TM2, TM3, and TM5, with movements in TM6 controlling precision and compactness of C99 binding, particularly near the hydrophobic cleavage sites at Thr48 and Leu49. Very interestingly, Thr48 interacts predominantly with PS1–NTF as most C99 residues do, whereas Leu-49 interacts mainly with the CTF of PS1, directly indicating how maturation cleavage enhances flexibility that affects C99 positioning. Since the two catalytic aspartates are positioned on each side of C99 close to either Thr48 or Leu49, the data suggest that the exact positioning of the aspartates relative to the Thr48–Leu49 and Leu49–Val50 peptide bonds defines the two cleavage pathways initiating with the production of Aβ49 and Aβ48. We show that the looser state (Figure 6, top) features a more dynamic C99, longer ddAsp and accommodates more of both cleavage pathways, whereas the compact state (Figure 6, bottom) favors cleavage at Thr48–Leu49 to form the innocent Aβ40. These two states cannot be directly seen in the cryo-EM data as they average out. The free energy balance between the compact and loose ensembles defines the extent of trimming, which is primarily determined by the reduced enzyme-complex stability as trimming proceeds [84]. We thus hope to have identified the molecular mechanism at atomic resolution that explains Aβ production.

Our molecular structures and dynamics will aid the development of new medicine for the treatment of AD with a structural causal basis. Specifically, new γ-secretase modulators should enhance the processing of C99 by favoring the ensemble of the compact state over the loose state that we have identified in this work. Such compounds would contribute their binding energy to the ternary complex to increase the residence time of C99, and accordingly increasing proteolytic trimming toward shorter non-toxic Aβ peptides. Further mechanistic and structural studies are needed to address the dynamics of the system in different biophysical environments, as our study only details the most important qualitative properties of the ensemble, with quantitative insights being most likely context-dependent.

Abbreviations

     
  • AD

    Alzheimer's Disease

  •  
  • APH1

    anterior pharynx defective-1

  •  
  • APP

    amyloid precursor protein

  •  
  • CTF

    C-terminal fragment

  •  
  • ECD

    extracellular domain

  •  
  • FIST

    Fit–Stay–Trim

  •  
  • HL1

    hydrophilic loop 1

  •  
  • LINCS

    linear constraint solver

  •  
  • MD

    molecular dynamics

  •  
  • MM

    molecular mechanics

  •  
  • NCT

    nicastrin

  •  
  • NTF

    N-terminal fragment

  •  
  • PBSA

    Poisson–Boltzmann surface area

  •  
  • PCA

    principal component analysis

  •  
  • PEN2

    presenilin enhancer-2

  •  
  • PS1

    presenilin 1

  •  
  • PS2

    presenilin 2

  •  
  • RMSD

    root-mean-square deviation

  •  
  • RMSF

    root-mean-square fluctuation

  •  
  • TM

    transmembrane

  •  
  • TMD

    transmembrane domain

  •  
  • cryo-EM

    cryo-electron microscopic.

Author contribution

B.D. performed most calculations and analysis and wrote the first draft of the paper. K.P.K. planned the project, analyzed results, and wrote the paper with B.D. N.T. contributed analysis of helix tilt angles and trajectory and statistical analysis.

Funding

The authors acknowledge the financial support received from the Novo Nordisk Foundation [grant NNF17OC0028860], and the Danish Council for Independent Research | Natural Sciences (DFF) [grant 7016-00079B].

Acknowledgements

The authors acknowledge computer time for this project at the DTU high-performance computing facility, Technical University of Denmark (DTU), Lyngby, Denmark.

Competing Interests

The Authors declare that there are no competing interests associated with the manuscript.

References

References
1
Kimberly
,
W.T.
,
LaVoie
,
M.J.
,
Ostaszewski
,
B.L.
,
Ye
,
W.
,
Wolfe
,
M.S.
and
Selkoe
,
D.J.
(
2003
)
Gamma-secretase is a membrane protein complex comprised of presenilin, nicastrin, Aph-1, and Pen-2
.
Proc. Natl Acad. Sci. U.S.A.
100
,
6382
6387
2
Haass
,
C.
,
Kaether
,
C.
,
Thinakaran
,
G.
and
Sisodia
,
S.
(
2012
)
Trafficking and proteolytic processing of APP
.
Cold Spring Harb. Perspect. Med.
2
,
a006270
3
Kamp
,
F.
,
Winkler
,
E.
,
Trambauer
,
J.
,
Ebke
,
A.
,
Fluhrer
,
R.
and
Steiner
,
H.
(
2015
)
Intramembrane proteolysis of β-amyloid precursor protein by γ-secretase is an unusually slow process
.
Biophys. J.
108
,
1229
1237
4
Tomita
,
T.
(
2014
)
Molecular mechanism of intramembrane proteolysis by γ-secretase
.
J. Biochem.
156
,
195
201
5
Zhang
,
H.
,
Ma
,
Q.
,
Zhang
,
Y.-W.
and
Xu
,
H.
(
2012
)
Proteolytic processing of Alzheimer's β-amyloid precursor protein
.
J. Neurochem.
120
,
9
21
6
Fukumori
,
A.
,
Fluhrer
,
R.
,
Steiner
,
H.
and
Haass
,
C.
(
2010
)
Three-amino acid spacing of presenilin endoproteolysis suggests a general stepwise cleavage of gamma-secretase-mediated intramembrane proteolysis
.
J. Neurosci.
30
,
7853
7862
7
Zoltowska
,
K.M.
,
Maesako
,
M.
and
Berezovska
,
O.
(
2016
)
Interrelationship between changes in the amyloid β 42/40 ratio and presenilin 1 conformation
.
Mol. Med.
22
,
329
337
8
Hsiao
,
K.
,
Chapman
,
P.
,
Nilsen
,
S.
,
Eckman
,
C.
,
Harigaya
,
Y.
,
Younkin
,
S.
et al.  (
1996
)
Correlative memory deficits, Aβ elevation, and amyloid plaques in transgenic mice
.
Science
274
,
99
103
9
Duering
,
M.
,
Grimm
,
M.O.W.
,
Grimm
,
H.S.
,
Schröder
,
J.
and
Hartmann
,
T.
(
2005
)
Mean age of onset in familial Alzheimer's disease is determined by amyloid beta 42
.
Neurobiol. Aging
26
,
785
788
10
Oehlrich
,
D.
,
Berthelot
,
D.J.-C.
and
Gijsen
,
H.J.M.
(
2011
)
γ-Secretase modulators as potential disease modifying anti-Alzheimer's drugs
.
J. Med. Chem.
54
,
669
698
11
Golde
,
T.E.
,
Koo
,
E.H.
,
Felsenstein
,
K.M.
,
Osborne
,
B.A.
and
Miele
,
L.
(
2013
)
γ-Secretase inhibitors and modulators
.
Biochim. Biophys. Acta
1828
,
2898
2907
12
Pester
,
O.
,
Barrett
,
P.J.
,
Hornburg
,
D.
,
Hornburg
,
P.
,
Pröbstle
,
R.
,
Widmaier
,
S.
, et al.  (
2013
)
The backbone dynamics of the amyloid precursor protein transmembrane helix provides a rationale for the sequential cleavage mechanism of γ-secretase
.
J. Am. Chem. Soc.
135
,
1317
1329
13
Dominguez
,
L.
,
Meredith
,
S.C.
,
Straub
,
J.E.
and
Thirumalai
,
D.
(
2014
)
Transmembrane fragment structures of amyloid precursor protein depend on membrane surface curvature
.
J. Am. Chem. Soc.
136
,
854
857
14
Pantelopulos
,
G.A.
,
Straub
,
J.E.
,
Thirumalai
,
D.
and
Sugita
,
Y.
(
2018
)
Structure of APP-C991–99 and implications for role of extra-membrane domains in function and oligomerization
.
Biochim. Biophys. Acta
1860
,
1698
1708
PMID:
[PubMed]
15
Barrett
,
P.J.
,
Song
,
Y.
,
Van Horn
,
W.D.
,
Hustedt
,
E.J.
,
Schafer
,
J.M.
,
Hadziselimovic
,
A.
et al.  (
2012
)
The amyloid precursor protein has a flexible transmembrane domain and binds cholesterol
.
Science
336
,
1168
1171
16
Kukar
,
T.L.
,
Ladd
,
T.B.
,
Robertson
,
P.
,
Pintchovski
,
S.A.
,
Moore
,
B.
,
Bann
,
M.A.
, et al.  (
2011
)
Lysine 624 of the amyloid precursor protein (APP) is a critical determinant of amyloid β peptide length: support for a sequential model of γ-secretase intramembrane proteolysis and regulation by the amyloid β precursor protein (APP) juxtamembrane region
.
J. Biol. Chem.
286
,
39804
39812
17
Zhou
,
L.
,
Brouwers
,
N.
,
Benilova
,
I.
,
Vandersteen
,
A.
,
Mercken
,
M.
,
Van Laere
,
K.
, et al.  (
2011
)
Amyloid precursor protein mutation E682 K at the alternative β-secretase cleavage β’-site increases Aβ generation
.
EMBO Mol. Med.
3
,
291
302
18
Dimitrov
,
M.
,
Alattia
,
J.-R.
,
Lemmin
,
T.
,
Lehal
,
R.
,
Fligier
,
A.
,
Houacine
,
J.
, et al.  (
2013
)
Alzheimer's disease mutations in APP but not γ-secretase modulators affect epsilon-cleavage-dependent AICD production
.
Nat. Commun.
4
,
2246
19
Zhou
,
R.
,
Yang
,
G.
,
Guo
,
X.
,
Zhou
,
Q.
,
Lei
,
J.
and
Shi
,
Y.
(
2019
)
Recognition of the amyloid precursor protein by human γ-secretase
.
Science
363
,
eaaw0930
20
De Strooper
,
B.
(
2003
)
Aph-1, Pen-2, and nicastrin with presenilin generate an active γ-Secretase complex
.
Neuron
38
,
9
12
21
Haapasalo
,
A.
and
Kovacs
,
D.M.
(
2011
)
The many substrates of presenilin/γ-secretase
.
J. Alzheimer's Dis.
25
,
3
28
22
De Strooper
,
B.
,
Iwatsubo
,
T.
and
Wolfe
,
M.S.
(
2012
)
Presenilins and γ-secretase: structure, function, and role in Alzheimer disease
.
Cold Spring Harb. Perspect. Med.
2
,
a006304
23
Francis
,
R.
,
McGrath
,
G.
,
Zhang
,
J.
,
Ruddy
,
D.A.
,
Sym
,
M.
,
Apfeld
,
J.
, et al.  (
2002
)
aph-1 and pen-2 are required for notch pathway signaling, γ-secretase cleavage of βAPP, and presenilin protein accumulation
.
Dev. Cell
3
,
85
97
24
Bolduc
,
D.M.
,
Montagna
,
D.R.
,
Gu
,
Y.
,
Selkoe
,
D.J.
and
Wolfe
,
M.S.
(
2016
)
Nicastrin functions to sterically hinder γ-secretase–substrate interactions driven by substrate transmembrane domain
.
Proc. Natl Acad. Sci. U.S.A.
113
,
E509
E518
25
Lu
,
P.
,
Bai
,
X.-C.
,
Ma
,
D.
,
Xie
,
T.
,
Yan
,
C.
,
Sun
,
L.
, et al.  (
2014
)
Three-dimensional structure of human γ-secretase
.
Nature
512
,
166
170
26
Sun
,
L.
,
Zhao
,
L.
,
Yang
,
G.
,
Yan
,
C.
,
Zhou
,
R.
,
Zhou
,
X.
, et al.  (
2015
)
Structural basis of human γ-secretase assembly
.
Proc. Natl Acad. Sci. U.S.A..
112
,
6003
6008
27
Bai
,
X.-C.
,
Yan
,
C.
,
Yang
,
G.
,
Lu
,
P.
,
Sun
,
L.
,
Zhou
,
R.
et al.  (
2015
)
An atomic structure of human γ-secretase
.
Nature
525
,
212
217
28
Bai
,
X.
,
Rajendra
,
E.
,
Yang
,
G.
,
Shi
,
Y.
and
Scheres
,
S.H.
(
2015
)
Sampling the conformational space of the catalytic subunit of human γ-secretase
.
eLife
4
,
551
560
29
Yang
,
G.
,
Zhou
,
R.
,
Zhou
,
Q.
,
Guo
,
X.
,
Yan
,
C.
,
Ke
,
M.
et al.  (
2019
)
Structural basis of notch recognition by human γ-secretase
.
Nature
565
,
192
197
30
Somavarapu
,
A.K.
and
Kepp
,
K.P.
(
2017
)
Membrane dynamics of γ-Secretase provides a molecular basis for β-Amyloid binding and processing
.
ACS Chem. Neurosci.
8
,
2424
2436
31
Tolia
,
A.
,
Horré
,
K.
and
De Strooper
,
B.
(
2008
)
Transmembrane domain 9 of presenilin determines the dynamic conformation of the catalytic site of γ-secretase
.
J. Biol. Chem.
283
,
19793
19803
32
Elad
,
N.
,
De Strooper
,
B.
,
Lismont
,
S.
,
Hagen
,
W.
,
Veugelen
,
S.
,
Arimon
,
M.
et al.  (
2015
)
The dynamic conformational landscape of γ-secretase
.
J. Cell Sci.
128
,
589
598
33
Karplus
,
M.
and
McCammon
,
J.A.
(
2002
)
Molecular dynamics simulations of biomolecules
.
Nat. Struct. Biol.
9
,
646
652
34
Adcock
,
S.A.
and
McCammon
,
J.A.
(
2006
)
Molecular dynamics: survey of methods for simulating the activity of proteins
.
Chem. Rev.
106
,
1589
1615
35
Zwier
,
M.C.
and
Chong
,
L.T.
(
2010
)
Reaching biological timescales with all-atom molecular dynamics simulations
.
Curr. Opin. Pharmacol.
10
,
745
752
36
Best
,
R.B.
,
Buchete
,
N.-V.
and
Hummer
,
G.
(
2008
)
Are current molecular dynamics force fields too helical?
Biophys. J.
95
,
L07
L09
37
Somavarapu
,
A.K.
and
Kepp
,
K.P.
(
2015
)
The dependence of amyloid-β dynamics on protein force fields and water models
.
ChemPhysChem
16
,
3278
3289
38
Lindorff-Larsen
,
K.
,
Maragakis
,
P.
,
Piana
,
S.
,
Eastwood
,
M.P.
,
Dror
,
R.O.
and
Shaw
,
D.E.
(
2012
)
Systematic validation of protein force fields against experimental data
.
PLoS ONE
7
,
e32131
39
Miyashita
,
N.
,
Straub
,
J.E.
,
Thirumalai
,
D.
and
Sugita
,
Y.
(
2009
)
Transmembrane structures of amyloid precursor protein dimer predicted by replica-exchange molecular dynamics simulations
.
J. Am. Chem. Soc.
131
,
3438
3439
40
Dominguez
,
L.
,
Foster
,
L.
,
Straub
,
J.E.
and
Thirumalai
,
D.
(
2016
)
Impact of membrane lipid composition on the structure and stability of the transmembrane domain of amyloid precursor protein
.
Proc. Natl Acad. Sci. U.S.A.
113
,
E5281
E5287
41
Meckler
,
X.
and
Checler
,
F.
(
2016
)
Presenilin 1 and presenilin 2 target γ-secretase complexes to distinct cellular compartments
.
J. Biol. Chem.
291
,
12821
12837
42
Somavarapu
,
A.K.
and
Kepp
,
K.P.
(
2016
)
The dynamic mechanism of presenilin-1 function: sensitive gate dynamics and loop unplugging control protein access
.
Neurobiol. Dis.
89
,
147
156
43
Tang
,
N.
,
Somavarapu
,
A.K.
and
Kepp
,
K.P.
(
2018
)
Molecular recipe for γ-Secretase modulation from computational analysis of 60 active compounds
.
ACS Omega
3
,
18078
18088
44
Tang
,
N.
and
Kepp
,
K.P.
(
2018
)
Aβ42/Aβ40 ratios of presenilin 1 mutations correlate with clinical onset of Alzheimer's disease
.
J. Alzheimers Dis.
66
,
939
945
45
Knappenberger
,
K.S.
,
Tian
,
G.
,
Ye
,
X.
,
Sobotka-Briner
,
C.
,
Ghanekar
,
S.V.
,
Greenberg
,
B.D.
et al.  (
2004
)
Mechanism of gamma-secretase cleavage activation: is gamma-secretase regulated through autoinhibition involving the presenilin-1 exon 9 loop?
Biochemistry
43
,
6208
6218
46
Thinakaran
,
G.
,
Borchelt
,
D.R.
,
Lee
,
M.K.
,
Slunt
,
H.H.
,
Spitzer
,
L.
,
Kim
,
G.
, et al.  (
1996
)
Endoproteolysis of presenilin 1 and accumulation of processed derivatives in vivo
.
Neuron
17
,
181
190
47
Shen
,
Y.
,
Delaglio
,
F.
,
Cornilescu
,
G.
and
Bax
,
A.
(
2009
)
TALOS+: a hybrid method for predicting protein backbone torsion angles from NMR chemical shifts
.
J. Biomol. NMR
44
,
213
223
48
Beel
,
A.J.
,
Mobley
,
C.K.
,
Kim
,
H.J.
,
Tian
,
F.
,
Hadziselimovic
,
A.
,
Jap
,
B.
et al.  (
2008
)
Structural studies of the transmembrane C-terminal domain of the amyloid precursor protein (APP): does APP function as a cholesterol sensor?
Biochemistry
47
,
9428
9446
49
Pettersen
,
E.F.
,
Goddard
,
T.D.
,
Huang
,
C.C.
,
Couch
,
G.S.
,
Greenblatt
,
D.M.
,
Meng
,
E.C.
et al.  (
2004
)
UCSF Chimera - A visualization system for exploratory research and analysis
.
J. Comput. Chem.
25
,
1605
1612
50
Hekkelman
,
M.L.
,
te Beek
,
T.A.H.
,
Pettifer
,
S.R.
,
Thorne
,
D.
,
Attwood
,
T.K.
and
Vriend
,
G.
(
2010
)
WIWS: a protein structure bioinformatics web service collection
.
Nucleic Acids Res.
38
,
W719
W723
51
Kozakov
,
D.
,
Hall
,
D.R.
,
Xia
,
B.
,
Porter
,
K.A.
,
Padhorny
,
D.
,
Yueh
,
C.
et al.  (
2017
)
The clusPro web server for protein-protein docking
.
Nat. Protoc.
12
,
255
278
52
Lomize
,
M.A.
,
Pogozheva
,
I.D.
,
Joo
,
H.
,
Mosberg
,
H.I.
and
Lomize
,
A.L.
(
2012
)
OPM database and PPM web server: resources for positioning of proteins in membranes
.
Nucleic Acids Res.
40
,
D370
D376
53
Jo
,
S.
,
Kim
,
T.
,
Iyer
,
V.G.
and
Im
,
W.
(
2008
)
CHARMM-GUI: a web-based grraphical user interface for CHARMM
.
J. Comput. Chem.
29
,
1859
1865
54
Jorgensen
,
W.L.
,
Chandrasekhar
,
J.
,
Madura
,
J.D.
,
Impey
,
R.W.
and
Klein
,
M.L.
(
1983
)
Comparison of simple potential functions for simulating liquid water
.
J. Chem. Phys.
79
,
926
55
Huang
,
J.
,
Rauscher
,
S.
,
Nawrocki
,
G.
,
Ran
,
T.
,
Feig
,
M.
,
De Groot
,
B.L.
et al.  (
2017
)
CHARMM36m: an improved force field for folded and intrinsically disordered proteins
.
Nat. Methods
14
,
71
73
56
Abraham
,
M. J.
,
Murtola
,
T.
,
Schulz
,
R.
,
Páll
,
S.
,
Smith
,
J. C.
,
Hess
,
B
et al.  et al.  (
2015
)
Gromacs: high performance molecular simulations through multi-level parallelism from laptops to supercomputers
.
SoftwareX
1–2
,
19
25
57
Hess
,
B.
,
Bekker
,
H.
,
Berendsen
,
H.J.C.
and
Fraaije
,
J.G.E.M.
(
1997
)
LINCS: a linear constraint solver for molecular simulations
.
J. Comput. Chem.
18
,
1463
1472
58
Buchoux
,
S.
(
2017
)
FATSLim: a fast and robust software to analyze MD simulations of membranes
.
Bioinformatics
33
,
133
134
59
Daura
,
X.
,
Gademann
,
K.
,
Jaun
,
B.
,
Seebach
,
D.
,
Van Gunsteren
,
W.F.
and
Mark
,
A.E.
(
1999
)
Peptide folding: when simulation meets experiment
.
Angew. Chemie Int. Ed.
38
,
236
240
60
Humphrey
,
W.
,
Dalke
,
A.
and
Schulten
,
K.
(
1996
)
VMD: visual molecular dynamics
.
J. Mol. Graph.
14
,
33
38
61
Kumari
,
R.
,
Kumar
,
R.
and
Lynn
,
A.
(
2014
)
G-mmpbsa—a GROMACS tool for high-throughput MM-PBSA calculations
.
J. Chem. Inf. Model.
54
,
1951
1962
62
Kollman
,
P.A.
,
Massova
,
I.
,
Reyes
,
C.
,
Kuhn
,
B.
,
Huo
,
S.
,
Chong
,
L.
, et al.  (
2000
)
Calculating structures and free energies of complex molecules: combining molecular mechanics and continuum models
.
Acc. Chem. Res.
33
,
889
897
63
Xu
,
T.H.
,
Yan
,
Y.
,
Kang
,
Y.
,
Jiang
,
Y.
,
Melcher
,
K.
and
Xu
,
H.E.
(
2016
)
Alzheimer's disease-associated mutations increase amyloid precursor protein resistance to γ-secretase cleavage and the Aβ42/Aβ40 ratio
.
Cell Discov.
2
,
16026
64
Yan
,
Y.
,
Xu
,
T.-H.
,
Melcher
,
K.
and
Xu
,
H.E.
(
2017
)
Defining the minimum substrate and charge recognition model of gamma-secretase
.
Acta Pharmacol. Sin.
38
,
1412
1424
65
Zhuang
,
X.
,
Dávila-Contreras
,
E.M.
,
Beaven
,
A.H.
,
Im
,
W.
and
Klauda
,
J.B.
(
2016
)
An extensive simulation study of lipid bilayer properties with different head groups, acyl chain lengths, and chain saturations
.
Biochim. Biophys. Acta
1858
,
3093
3104
PMID:
[PubMed]
66
Klauda
,
J.B.
,
Venable
,
R.M.
,
Freites
,
J.A.
,
O'Connor
,
J.W.
,
Tobias
,
D.J.
,
Mondragon-Ramirez
,
C.
et al.  (
2010
)
Update of the CHARMM all-atom additive force field for lipids: validation on six lipid types
.
J. Phys. Chem. B.
114
,
7830
7843
67
Watanabe
,
N.
,
Takagi
,
S.
,
Tominaga
,
A.
,
Tomita
,
T.
and
Iwatsubo
,
T.
(
2010
)
Functional analysis of the transmembrane domains of presenilin 1: participation of transmembrane domains 2 and 6 in the formation of initial substrate-binding site of γ-secretase
.
J. Biol. Chem.
285
,
19738
19746
68
Takagi-Niidome
,
S.
,
Sasaki
,
T.
,
Osawa
,
S.
,
Sato
,
T.
,
Morishima
,
K.
,
Cai
,
T.
et al.  (
2015
)
Cooperative roles of hydrophilic loop 1 and the c-terminus of presenilin 1 in the substrate-gating mechanism of γ-secretase
.
J. Neurosci.
35
,
2646
2656
69
Tominaga
,
A.
,
Cai
,
T.
,
Takagi-Niidome
,
S.
,
Iwatsubo
,
T.
and
Tomita
,
T.
(
2016
)
Conformational changes in transmembrane domain 4 of presenilin 1 are associated with altered amyloid- 42 production
.
J. Neurosci.
36
,
1362
1372
70
Li
,
C.-D.
,
Xu
,
Q.
,
Gu
,
R.-X.
,
Qu
,
J.
and
Wei
,
D.-Q.
(
2017
)
The dynamic binding of cholesterol to the multiple sites of C99: as revealed by coarse-grained and all-atom simulations
.
Phys. Chem. Chem. Phys.
19
,
3845
3856
71
Lemmin
,
T.
,
Dimitrov
,
M.
,
Fraering
,
P.C.
and
Dal Peraro
,
M.
(
2014
)
Perturbations of the straight transmembrane α-helical structure of the amyloid precursor protein affect its processing by γ-secretase
.
J. Biol. Chem.
289
,
6763
6774
72
Tian
,
G.
,
Sobotka-Briner
,
C.D.
,
Zysk
,
J.
,
Liu
,
X.
,
Birr
,
C.
,
Sylvester
,
M.A.
et al.  (
2002
)
Linear non-competitive inhibition of solubilized human γ-secretase by pepstatin A methylester, L685458, sulfonamides, and benzodiazepines
.
J. Biol. Chem.
277
,
31499
31505
73
Scharnagl
,
C.
,
Pester
,
O.
,
Hornburg
,
P.
,
Hornburg
,
D.
,
Götz
,
A.
and
Langosch
,
D.
(
2014
)
Side-chain to main-chain hydrogen bonding controls the intrinsic backbone dynamics of the amyloid precursor protein transmembrane helix
.
Biophys. J.
106
,
1318
1326
74
Oestereich
,
F.
,
Bittner
,
H.J.
,
Weise
,
C.
,
Grohmann
,
L.
,
Janke
,
L.-K.
,
Hildebrand
,
P.W.
et al.  (
2015
)
Impact of amyloid precursor protein hydrophilic transmembrane residues on amyloid-beta generation
.
Biochemistry
54
,
2777
2784
75
Langosch
,
D.
and
Steiner
,
H.
(
2017
)
Substrate processing in intramembrane proteolysis by γ-secretase – the role of protein dynamics
.
Biol. Chem.
398
,
441
453
76
Aguayo-Ortiz
,
R.
,
Chávez-García
,
C.
,
Straub
,
J.E.
and
Dominguez
,
L.
(
2017
)
Characterizing the structural ensemble of γ-secretase using a multiscale molecular dynamics approach
.
Chem. Sci.
8
,
5576
5584
77
Kornilova
,
A.Y.
,
Bihel
,
F.
,
Das
,
C.
and
Wolfe
,
M.S.
(
2005
)
The initial substrate-binding site of -secretase is located on presenilin near the active site
.
Proc. Natl. Acad. Sci. U.S.A.
102
,
3230
3235
78
Takami
,
M.
,
Nagashima
,
Y.
,
Sano
,
Y.
,
Ishihara
,
S.
,
Morishima-Kawashima
,
M.
,
Funamoto
,
S.
et al.  (
2009
)
γ-Secretase: successive tripeptide and tetrapeptide release from the transmembrane domain of β-carboxyl terminal fragment
.
J. Neurosci.
29
,
13042
13052
79
Fukumori
,
A.
and
Steiner
,
H.
(
2016
)
Substrate recruitment of γ-secretase and mechanism of clinical presenilin mutations revealed by photoaffinity mapping
.
EMBO J.
,
e201694151
80
Li
,
S.
,
Zhang
,
W.
and
Han
,
W.
(
2017
)
Initial substrate binding of γ-Secretase: the role of substrate flexibility
.
ACS Chem. Neurosci.
8
,
1279
1290
81
Steiner
,
H.
,
Fukumori
,
A.
,
Tagami
,
S.
and
Okochi
,
M.
(
2018
)
Making the final cut: pathogenic amyloid-β peptide generation by γ-secretase
.
Cell Stress
2
,
292
310
82
Quintero-Monzon
,
O.
,
Martin
,
M.M.
,
Fernandez
,
M.A.
,
Cappello
,
C.A.
,
Krzysiak
,
A.J.
,
Osenkowski
,
P.
et al.  (
2011
)
Dissociation between the processivity and total activity of γ-secretase: implications for the mechanism of Alzheimer's disease-causing presenilin mutations
.
Biochemistry
50
,
9023
9035
83
Somavarapu
,
A.K.
and
Kepp
,
K.P.
(
2016
)
Loss of stability and hydrophobicity of presenilin 1 mutations causing Alzheimer's disease
.
J. Neurochem.
137
,
101
111
84
Szaruga
,
M.
,
Munteanu
,
B.
,
Lismont
,
S.
,
Veugelen
,
S.
,
Horré
,
K.
,
Mercken
,
M.
, et al.  (
2017
)
Alzheimer's-causing mutations shift Aβ length by destabilizing γ-Secretase-Aβn interactions
.
Cell
170
,
443
456