Mechanism of Tripeptide Trimming of Amyloid β-Peptide 49 by γ-Secretase

Abstract

The membrane-embedded γ-secretase complex processively cleaves within the transmembrane domain of amyloid precursor protein (APP) to produce 37-to-43-residue amyloid β-peptides (Aβ) of Alzheimer’s disease (AD). Despite its importance in pathogenesis, the mechanism of processive proteolysis by γ-secretase remains poorly understood. Here, mass spectrometry and Western blotting were used to quantify the efficiency of tripeptide trimming of wild-type (WT) and familial AD (FAD) mutant Aβ49. In comparison to WT Aβ49, the efficiency of tripeptide trimming was similar for the I45F, A42T, and V46F Aβ49 FAD mutants but substantially diminished for the I45T and T48P mutants. In parallel with biochemical experiments, all-atom simulations using a novel peptide Gaussian accelerated molecular dynamics (Pep-GaMD) method were applied to investigate the tripeptide trimming of Aβ49 by γ-secretase. The starting structure was the active γ-secretase bound to Aβ49 and APP intracellular domain (AICD), as generated from our previous study that captured the activation of γ-secretase for the initial endoproteolytic cleavage of APP (Bhattarai, A, et al. ACS Cent. Sci. 2020, 6, 969–983). Pep-GaMD simulations captured remarkable structural rearrangements of both the enzyme and substrate, in which hydrogen-bonded catalytic aspartates and water became poised for tripeptide trimming of Aβ49 to Aβ46. These structural changes required a positively charged N-terminus of endoproteolytic coproduct AICD, which could dissociate during conformational rearrangements of the protease and Aβ49. The simulation findings were highly consistent with biochemical experimental data. Taken together, our complementary biochemical experiments and Pep-GaMD simulations have enabled elucidation of the mechanism of tripeptide trimming of Aβ49 by γ-secretase.

Graphical Abstract

INTRODUCTION

Alzheimer’s disease (AD) contributes to more than 80% of all dementia cases.¹ Deaths related to AD in the United States increased by 89% between 2000 and 2014, and more than 6.2 million Americans are affected by AD in 2021 (www.alz.org). AD is characterized by the deposition of longer amyloid β-peptides (Aβ) in the form of cerebral plaques. The amyloid β-protein precursor (APP) is successively processed by β-secretase and γ-secretase to produce Aβ peptides. β-Secretase first sheds the APP extracellular domain to produce transmembrane peptide C99, followed by processive proteolysis by γ-secretase to produce Aβ peptides of varying lengths.² Membrane-embedded γ-secretase is a multidomain aspartyl protease with presenilin as the catalytic subunit. γ-Secretase is considered “the proteasome of the membrane,” with more than 140 known substrates, including APP and the Notch family of cell-surface receptors.^3,4 The location of the proteolysis and the number of cleavages within the APP transmembrane domain by γ-secretase determines the length of final Aβ products and the likelihood of forming plaques.

Of the many transmembrane substrates, processive proteolysis of APP by γ-secretase is the most studied. γ-Secretase first carries out endoproteolytic (ε) cleavage of C99 peptide near the cytosolic end of the transmembrane domain, producing Aβ49 and Aβ48 peptides and their respective AICD coproducts (AICD50-99 and AICD49-99, respectively).⁵ These initially formed long Aβ peptides are then cut generally every three residues from their C-termini to release tripeptide (and one tetrapeptide) coproducts. The two general pathways of γ-secretase processive proteolysis are Aβ48 → Aβ45 → Aβ42 → Aβ38 and Aβ49 → Aβ46 → Aβ43 → Aβ40,^6,7 producing Aβ42 and Aβ40 as their dominant products, respectively. Among these two, the longer Aβ42 peptide is more prone to aggregate and forms plaques.⁸ Moreover, early-onset familial AD (FAD) APP mutants can bias the enzyme to produce longer Aβ peptides that are pathological and cause AD.⁹ The trimming of APP substrate by a γ-secretase enzyme is dictated by active-site S1′, S2′, and S3′ subpockets that, respectively, bind to P1′, P2′, and P3′ substrate residues.¹⁰

Critical gaps remain in understanding the mechanism of intramembrane processive proteolysis by γ-secretase. Recently reported cryo-EM structures of γ-secretase bound to Notch and APP substrates provided valuable insights into the structural basis of substrate recognition of the enzyme.^11,12 However, artificial structural constraints were included that could affect the enzyme–substrate interactions. Molecular dynamics (MD) simulations have proven useful in understanding the structural dynamics of γ-secretase, notably the enzyme–substrate interactions, including many previous studies.^13-28 Recently, we computationally restored the wild-type (WT) enzyme–substrate costructure and applied all-atom simulations using the Gaussian accelerated molecular dynamics (GaMD) method to build the first dynamic model of γ-secretase activation.²⁹ GaMD is an enhanced sampling technique that works by adding a harmonic boost potential to smooth the potential energy surface and reduce system energy barriers.³⁰ Our GaMD simulations captured the extremely slow motions underlying enzyme activation, with the two catalytic aspartates and a coordinated water molecule poised for proteolysis of APP at the ε cleavage site. We showed that the I45F and T48P FAD mutations in APP enhanced the ε cleavage of the amide bond between Leu49 and Val50 compared with the WT APP. In contrast, the M51F mutation in APP shifted the ε cleavage to the adjacent Thr48-Leu49 amide bond, changing the proteolysis from the Aβ49 to the Aβ48 pathway. Despite these advances, the detailed atomistic mechanism of processive proteolysis by γ-secretase remains elusive. This is consistent with γ-secretase being a well-known slow-acting enzyme (k_cat for APP ε proteolysis ~2–6 per hour),^31,32 making it difficult to capture the dynamic transitions comprising large energy barriers in MD simulations. Hence, despite its importance in the pathogenesis of AD, the mechanism of processive proteolysis (tripeptide trimming) by γ-secretase remains poorly understood.

Here, we report the first dynamic model of tripeptide trimming of Aβ49 to Aβ46 (ζ cleavage) by γ-secretase. Extensive all-atom simulations using a novel Peptide GaMD (Pep-GaMD) method³³ captured the slow dynamic molecular transition from the ε to ζ proteolytic cleavage step. In Pep-GaMD, a boost potential is applied selectively to the essential potential energy of the peptide to effectively model its high flexibility and accelerate its dynamic motions.³³ In addition, another boost potential is applied to the protein and solvent to enhance conformational sampling of the protein and facilitate peptide binding. Pep-GaMD has been demonstrated in the binding of model peptides to the SH3 protein domains. Independent 1 μs dual-boost Pep-GaMD simulations have captured repetitive peptide dissociation and binding events, which enable the calculation of peptide binding thermodynamics and kinetics. The calculated binding free energies and kinetic rate constants agreed very well with the available experimental data.³⁴

In this study, we have combined biochemical experiments, including matrix-assisted laser desorption/ionization–time-of-flight mass spectrometry (MALDI-TOF MS), liquid chromatography–tandem mass spectrometry (LC–MS/MS), and Western blotting, with Pep-GaMD enhanced sampling simulations to elucidate the mechanism of tripeptide trimming of Aβ49 by γ-secretase. Our findings from Pep-GaMD simulations of WT and five FAD mutants (I45F, A42T, V46F, I45T, and T48P) of Aβ49 bound to γ-secretase were highly consistent with quantitative biochemical analysis of their specific proteolytic products, providing important mechanistic insights into tripeptide trimming by the enzyme.

RESULTS

Probing ζ Cleavage of WT and FAD-Mutant Aβ49 by γ-Secretase in Biochemical Experiments.

To compare the ζ cleavage of the WT and FAD mutants of APP by γ-secretase, we performed in vitro cleavage assay experiments using purified γ-secretase and recombinant APP-based substrate C100-FLAG, which contained the C99 APP C-terminal fragment with an N-terminal start methionine and a C-terminal FLAG epitope tag.³⁵ Efficiency of the cleavage of substrate Aβ49 to products Aβ46 and tripeptide was calculated by measuring Aβ49 production and Aβ49 degradation. To quantify Aβ49 production by ε cleavage of APP substrate, levels of coproducts AICD50-99 were determined using a combination of matrix-assisted laser desorption/ionization–time-of-flight mass spectrometry (MALDI-TOF MS) and quantitative Western blotting.

First, AICD produced in the assay was immunoprecipitated with anti-FLAG antibodies and detected by MALDI-TOF MS (Figure 1A). For the WT, A42T, V46F, and I45T APP substrates, the signal intensities corresponding to AICD49-99 and AICD50-99 show higher level of AICD49-99 than AICD50-99. However, for mutants I45F and T48P APP substrate, signal intensities show a higher level of AICD50-99 than AICD49-99. This suggests that I45F and T48P favor the production of Aβ49 rather than the production of Aβ48, while A42T, V46F, and I45T favor the production of Aβ48 rather than the production of Aβ49.

Figure 1.

Tripeptide trimming of the wild-type (WT) and FAD mutants of Aβ49 bound γ-secretase characterized by MS, Western blotting, and Pep-GaMD simulations. (A) MALDI-TOF MS detection of AICD50-99 and AICD49-99 products. (B) Anti-FLAG immunoblot of total AICD-FLAG levels. Purified C100-FLAG at a range of known concentrations was used to generate a standard curve, (C) quantification of total AICD-FLAG levels from immunoblot by densitometry, (D) quantification of AICD50-99 using total AICD levels determined from immunoblot and intensity ratios determined from MALDI-TOF MS, (E) quantification of ITL tripeptides generated from trimming of WT and FAD mutants of Aβ49, (F) cleavage efficiency of the first trimming (ζ) step. Gray dotted line denotes cleavage efficiency from the WT APP substrate. (G–L) 2D free-energy profiles calculated from the Pep-GaMD simulations of (G) WT, (H) I45F, (I) A42T, (J) V46F, (K) I45T, and (L) T48P Aβ49 bound to γ-secretase. The distances between the Cγ atoms of Asp257 and Asp385 in PS1 and between the hydroxyl oxygen of PS1 Asp257 and the carbonyl oxygen of Aβ49 Val46 were selected as the reaction coordinates.

The same reaction mixtures were subjected to quantitative Western blotting with anti-FLAG antibodies (Figure 1B), where standards of known concentrations of C100-FLAG were also run to make a standard curve, plotting band intensity against the concentration of FLAG-tagged C100. From this standard curve, the concentration of total AICD-FLAG product obtained in the enzyme reaction was quantified (Figure 1C). Quantification of the total AICD revealed increased total AICD production for the V46F mutant substrate and decreased total AICD production for A42T, I45F, I45T, and T48P mutant substrates compared to the AICD production for the WT. The concentration of AICD50-99 was calculated using the total AICD level determined by quantitative Western blot, and the ratio of AICD49-99 to AICD50-99 was determined from MALDI-TOF MS (Figure 1D). The calculated concentrations of AICD50-99 thus provided the level of production of coproduct Aβ49. Aβ49 production was slightly increased for I45F mutant, while for all other mutants A42T, V46F, I45T, and T48P decreased Aβ49 production was observed compared to Aβ49 production of the WT.

To determine the degradation of Aβ49, we calculated and quantified trimming product tripeptide ITL. The mixtures from the cleavage assay were subjected to LC–MS/MS analysis to detect tripeptides. All substrate constructs studied produced ITL except for T48P mutant, which produced IPL due to the replacement of T with P. For quantification of the production of these tripeptides, standard curves of each peptide were generated by plotting the concentration of synthetic peptide against the integrated areas of the three most abundant ion fragments from MS/MS (Figure S1). The ITL and IPL peptides generated in the γ-secretase cleavage were monitored and quantified (Figure 1E). The quantification of the trimming product (ITL or IPL) or the Aβ49 degradation reveals a decrease in Aβ49 degradation for A42T, V46F, I45T, and T48P. For I45F, Aβ49 degradation is similar to that of the WT. The concentration of both Aβ49 production and Aβ49 degradation was used to calculate the percent efficiency (Figure 1F). For all constructs, cleavage efficiency was close to 100% except that for two mutants I45T and T48P, the cleavage efficiencies decreased substantially to 35 and 34%, respectively.

We selected these particular FAD mutations in APP substrate based on their different effects on the Aβ49 → Aβ46 trimming step in our recently reported study.³¹ In that study, we examined the effects of 14 different FAD mutations in APP substrate on all proteolytic steps carried out by γ-secretase. Moreover, we determined the Aβ42/40 ratios for these and other FAD mutations in APP substrate and found the relative effects of these mutations on this ratio compared to that seen with WT enzyme to be generally consistent with those reported from other groups.^36-38 To the best of our knowledge, the effects of the I45F, A42T, V46F, I45T, and T48P FAD mutations of the substrate on the Aβ49 → Aβ46 trimming step have not been reported by any other groups.

Activation of γ-Secretase for Tripeptide Trimming of Aβ49 Was Captured in Pep-GaMD Simulations.

In parallel with the biochemical experiments, Pep-GaMD simulations were carried out on the γ-secretase bound by the WT and the I45F, A42T, V46F, I45T, and T48P mutants of Aβ49 (Table 1). The active WT APP-bound γ-secretase was obtained from our previous study,²⁹ and the amide bond between Aβ49 and AICD50-99 was cleaved as the new simulation starting structure (Figure S2, see details in the Methods section (Supporting Information)). We initially performed dual-boost GaMD simulations on the γ-secretase bound to Aβ49 with AICD50-99 removed. However, even after running ~6 μs GaMD simulations, we could not effectively sample conformational transitions of the system for ζ cleavage of Aβ49 to Aβ46 (Figure S3). The distance between the enzyme Asp257 catalytic residue and substrate Val46–Ile47 amide bond presented a computational challenge for conformational sampling, with apparently high energy barriers to overcome. To address the challenge, we applied our recently developed Pep-GaMD³³ method, which selectively boosts the essential potential energy of the peptide to effectively model the peptide flexibility and further improve sampling. We built four Pep-GaMD simulation systems with γ-secretase bound to Aβ49 in the presence of AICD50-99 and a system in the absence of AICD50-99 (Figure S4). The C-terminus of Aβ49 and the N-terminus of AICD50-99 were either charged or neutral combined to form four different Pep-GaMD enzyme systems. Spontaneous activation of γ-secretase for ζ cleavage of Aβ49 was observed during 600 ns Pep-GaMD simulations with “charged C-terminal Aβ49 and charged N-terminal AICD50-99” (Figures S5 and S6, Movie S1, and Table S2). The enzyme activation for ζ cleavage was characterized by coordinated hydrogen bonding between the enzyme Asp257 and carbonyl oxygen of substrate Val46. The catalytic aspartates were at a distance of ~7–8 Å between their Cγ atoms, which could accommodate a water molecule for nucleophilic attack of the carbonyl carbon of the scissile amide bond (Figure S5). The water molecule formed hydrogen bonds with both catalytic aspartates and was at ~4 Å distance away from the carbonyl carbon of substrate Val46 residue. The activated γ-secretase conformation was well poised for cleavage of the amide bond between Val46 and Ile47 for ζ cleavage of the Aβ49. In the γ-secretase bound to WT charged C-terminal Aβ49 and charged N-terminal AICD50-99 system, we observed AICD50-99 dissociation in addition to enzyme activation for ζ cleavage (Figure S7 and Movie S2). The AICD50-99, initially located near the Aβ49, slowly moved downwards to the intracellular PS1 pocket and then dissociated completely from the enzyme. Meanwhile, the AICD50-99 transitioned from β-sheet to a loop/unstructured conformation during the Pep-GaMD simulations (Figure S7). Similarly, γ-secretase bound to “neutral C-terminal Aβ49 and charged N-terminal AICD50-99” was also observed to become activated for ζ cleavage of Aβ49. In comparison, the γ-secretase systems bound to “charged C-terminal Aβ49” (in the absence of AICD50-99), “charged C-terminal Aβ49 and neutral N-terminal AICD50-99,” and “neutral C-terminal Aβ49 and neutral N-terminal AICD50-99” could not sample enzyme activation for ζ cleavage of Aβ49 (Table S2 and Figures S6A,B,D and S8A-C). This showed that the presence of charged N-terminal AICD50-99 was crucial for the enzyme activation for ζ cleavage of Aβ49. Therefore, systems for γ-secretase bound by charged C-terminal Aβ49 and charged N-terminal AICD50-99 were set up for running Pep-GaMD simulations of the FAD mutants of Aβ49.

Table 1.

Summary of Pep-GaMD Simulations Performed on Different Systems of γ-Secretase Bound by Aβ49 and AICD50-99 Peptides

system	N atoms a	dimension (Å3)	simulation (ns)	ΔVavgb (kcal/mol)	σΔvc (kcal/mol)
WT Aβ49	254 233	152 × 123 × 146	600 × 3	106.9	14.0
WT neutral Aβ49–neutral AICD	254 337	152 × 123 × 146	600 × 3	155.9	11.8
WT neutral Aβ49–charged AICD	254 340	152 × 123 × 146	600 × 3	134.2	11.3
WT charged Aβ49–neutral AICD	254 334	152 × 123 × 146	600 × 3	133.9	11.1
WT charged Aβ49–charged AICD	254 377	152 × 123 × 146	600 × 3	134.0	10.9
I45F charged Aβ49–charged AICD	254 335	152 × 123 × 146	600 × 3	137.5	12.1
A42T charged Aβ49–charged AICD	254 341	152 × 123 × 146	600 × 3	148.9	11.4
V46F charged Aβ49–charged AICD	254 329	152 × 123 × 146	600 × 3	177.9	11.9
I45T charged Aβ49–charged AICD	254 323	152 × 123 × 146	600 × 3	149.2	11.4
T48P charged Aβ49–charged AICD	254 328	152 × 123 × 146	600 × 3	137.9	11.2

^aN_atoms is the total number of atoms in the simulation systems.

^bΔV_avg is the average of the Pep-GaMD boost potential.

^cσ_ΔV is the standard deviation of the Pep-GaMD boost potential.

Free-energy profiles were calculated from Pep-GaMD simulations to characterize the activation of γ-secretase for ζ cleavage of the Aβ49 substrate, for which the distance between the enzyme catalytic aspartates and the distance between the enzyme protonated Asp257 and substrate residue Val46 were selected as reaction coordinates (Figures 1G-L, S6, and S9 and Table S2). In the WT Aβ49, three low-energy conformational states were identified from the free-energy profile, including “Final,” “Intermediate,” and “Initial” (Figures 1G, S6C, and S10 and Table S2). In the “Final” conformational state, the aspartates were ~7–8 Å apart to accommodate the water molecule in between. The substrate Val46 maintained a distance of ~3 Å from the active-site Asp257 to form a hydrogen bond in the Final active state. In the “Initial” state, the substrate Val46 was distant (~8–9 Å) from the active-site Asp257, while the interaspartate distance was ~6–7 Å. The “Initial” state represented the active state for the ε cleavage of APP. In the “Intermediate” state, the aspartates remained ~6–7 Å apart, while the Aβ49 peptide (carbonyl oxygen of Val46) was at a distance of ~6 Å from the protonated Asp257 (Figure 1G).

In the I45F mutant system, two low-energy conformational states, “Initial” and “Final,” were identified from the free-energy profile of Pep-GaMD simulations (Figures 1H, S11A, and S12A and Movie S3). Two out of three Pep-GaMD simulations could capture the activation process, as the Asp257 could form a stable hydrogen bond with Val46 as reflected in the distance–time course plot (Figure S11A). The “Final” state in the free-energy profile represented the active conformation of the enzyme for ζ cleavage of the scissile amide bond between Val46 and Ile47 APP residues. In the “Final” conformational state, the two catalytic aspartates were ~7–8 Å apart, and APP Val46 was ~3 Å distance away from the protonated aspartate. In the “Initial” state, the substrate Val46 was further away from the catalytic aspartate (~6–7 Å), and the aspartates were ~7–8 Å distance away from each other (Figure 1H).

In the A42T-mutant APP system, four low-energy conformational states were identified from the free-energy profile (Figures 1I, S11B, and S12B and Movie S4). Mutation of Ala42 to Thr42 caused the enzyme–substrate complex to sample a larger conformational space. In addition to the “Initial” and “Final” states, two new “Inhibited-1” and “Inactive” conformational states were identified for the A42T-mutant system. The catalytic aspartates were ~4–5 Å (too close) apart in the “Inhibited-1” state and 13 Å away (too far) in the “Inactive” state. In the “Inhibited” state, the catalytic aspartates could not accommodate a water molecule between them and hence was inhibited from proteolytic activation. APP Val46 was ~4–5 Å from the protonated Asp257 in this “Inhibited-1” state. In the “Inactive” state, the aspartates were ~13 Å apart and thus too far to form the dual hydrogen bonds with the water in between them, even though the Asp257 could form a hydrogen bond with the Val46 carbonyl oxygen. This hindered activation required for ζ cleavage.

In the V46F mutant system, two low-energy conformational states were identified, including “Inhibited-2” and “Final” (Figures 1J, S11C, and S12C and Movie S5). Like other γ-secretase systems, the “Final” state corresponded to the active conformation of the enzyme poised for ζ cleavage of Aβ49. Moreover, the “Inhibited-2” state had the two aspartates at proximity (~4–5 Å) between the Cγ atoms and was unable to accommodate a water molecule in between for enzyme activation. APP substrate was ~10 Å away from active-site Asp257 in the “Inhibited-2” state.

Furthermore, Pep-GaMD simulations were carried out on I45T and T48P mutant Aβ49-bound γ-secretase (Figures 1K,L, S11D,E, and S12D,E). Both of these mutant systems were not able to activate the enzyme for ζ cleavage, being consistent with the experimental results where the ζ cleavage efficiency dropped to about one-third compared to that of the WT. In the Pep-GaMD free-energy profile of the I45T mutant system, only one “Intermediate” low-energy conformational state was identified. This “Intermediate” state was the same low-energy conformation as the one in the WT system. For the T48P system, the hydrogen bond between APP Val46 and the protonated Asp257 was formed for a certain time in one of the three Pep-GaMD production simulations (Figure S11E). However, in the free-energy profile, we could identify two low-energy conformational states, including “Initial” and “Inhibited-1,” but not the “Final” active state (Figure 1L). The “Inhibited-1” state resembled the one identified in the A42T-mutant system. The “Initial” conformational state was the same as the one identified in the WT, I45F, and A42T systems. These Pep-GaMD simulation findings were consistent with the biochemical experiments, verifying the I45T and T48P systems as negative controls.

Conformational Changes in Activation of γ-Secretase for Tripeptide Trimming of Aβ49.

We calculated root-mean-square fluctuations (RMSFs) of γ-secretase bound by the WT and FAD-mutant APP from Pep-GaMD simulations (Figure S13 and Movie S6). In the WT Aβ49-bound γ-secretase, the TM2, TM6, TM6a, and C-terminus of TM9 helix were flexible in the catalytic PS1 subunit. The Pen-2 subunit exhibited high fluctuations with ~3 Å RMSF. Helices 1, 2, 5, 12, 17, and TM domain of nicastrin were also flexible during the Pep-GaMD simulations. Structural clustering was performed on Pep-GaMD snapshots of the system using a hierarchical agglomerative algorithm in CPPTRAJ³⁹ (see the Methods section (Supporting Information)). The top-ranked cluster was selected as the representative “Final” active conformation for the ζ cleavage of Aβ49. The starting structure from ε cleavage of APP was obtained as the “Initial” active conformation. The catalytic PS1 of the “Final” conformation was compared to that of the “Initial” conformation in Figure 2A. Relative to the “Initial” conformation, the substrate helical domain tilted by ~50° in the “Final” conformation (Figure 2A,B). Residue Leu49 in the substrate C-terminus moved downwards by ~5 Å (Figures 2B, 4A, and S14). The last residue in a helical conformation in the “Final” state of Aβ49 was Thr43, whereas it was Ile45 in the “Initial” state. In the transition from the “Initial” to the “Final” conformational state, two substrate residues, Val44 and Ile45, unwound their helical conformation and changed to a turn/loop conformation. Residues Thr43 and Ile45 were in similar positions in the “Initial” and “Final” active conformations relative to the membrane perpendicular axis (Figure S14). In comparison, the substrate C-terminal Leu49 moved downwards by ~5 Å while straightening the C-terminal loop (Figures 2B and S14B).

Figure 2.

Conformational changes of the PS1 catalytic subunit and substrate during activation of γ-secretase for tripeptide trimming of Aβ49 in Pep-GaMD simulations. (A) Comparison of the initial (active for ε cleavage, blue) and final (active for ζ cleavage, red) conformations of the Aβ49-bound PS1. The enzyme activation for the tripeptide trimming was characterized by coordinated hydrogen bonding between the enzyme Asp257, carbonyl oxygen of Aβ49 Val46, and a water molecule accommodated between the two aspartates poised for cleavage of the amide bond between Val46 and Ile47 residues. (B–F) Conformational changes of (B) Aβ49 substrate, (C) catalytic aspartates, (D) TM3, (E) TM6 and TM6a, and (F) β2 strand from the initial to the final conformational state. The helical domain of Aβ49 tilted by ~50° and residue Leu49 at the C-terminus of Aβ49 moved downwards by ~5 Å. Protonated catalytic Asp257 moved ~3 Å toward the Aβ49 substrate. The enzyme TM3 moved outwards by ~2 Å and TM6a moved downwards by ~2 Å. The enzyme β2 strand (N-terminus of TM7) moved away from APP and closer toward the β1 strand (C-terminus of TM6a) by ~5 Å.

Figure 4.

Active-site conformations of γ-secretase for tripeptide trimming of Aβ49 observed in the Pep-GaMD simulations. (A–D) Conformations of the substrate P1′, P2′, and P3′ residues in the final active conformations of the (A) WT (red), (B) I45F (pink), (C) A42T (green), and (D) V46F (cyan) Aβ49 systems. (E) Comparison of the PS1 active-site S1′, S2′, and S3′ pockets that accommodate the WT and mutants of Aβ49.

At the enzyme active site, the catalytic Asp385 did not have significant movement during the adjustments for substrate peptide trimming (Figure 2C). In comparison, the protonated catalytic Asp257 moved by ~3 Å toward the substrate. Asp257 moved forward to form a hydrogen bond with the carbonyl oxygen of the scissile amide bond between the substrate residues Val46 and Ile47. Similarly, TM3 moved outwards by ~2 Å (Figure 2D), and TM6a moved downwards by ~2 Å (Figure 2E). Flexibility in these helices involved important FAD mutation sites including Tyr154, His163, Ala164, Leu166, Trp165, Ser169, Ile168, Tyr256, Ala260, Leu262, Cys263, Pro264, Pro267, Arg269, Val272, and Leu271 (www.alzforum.org). Trp165 and His163 from TM3 and Arg269 from TM6a showed significant movements in their side chains. With a major part of the C-terminus of APP absent (as AICD dissociates, see the next section), the β2 loop at the N-terminus of TM7 moved away from the APP by ~5 Å in the Final state as compared to the Initial state (Figure 2F). FAD mutation residues in the β2-TM7 region including Arg377, G378, L383, and G384 showed flexibility in the simulations. In particular, residue Arg377 reoriented its side chain in the “Final” conformational state.

Changes in the Secondary Structures of the WT and FAD-Mutant Aβ49 during Tripeptide Trimming.

The secondary structures of the WT and FAD-mutant Aβ49 bound to γ-secretase were recorded during the Pep-GaMD simulations and plotted in Figures 3 and S15. Changes in the secondary structures of Aβ49 during ζ cleavage were compared to that of APP substrate (“Initial” active conformation) during ε cleavage from our previous study²⁹ (Figure S16). Unwinding of the helix C-terminus in Aβ49 during ζ cleavage was observed in the secondary structure plot. During the ε cleavage, the C-terminus of the WT APP substrate could maintain helical conformation up to Ile45/Val46 (Figure S16). In comparison, WT Aβ49 was helical up to Thr43 in the C-terminal region (Figures 3A and 2B). About 2–3 residues unwound near the ζ cleavage site to expose the scissile amide bond between Val46 and Ile47 to the catalytic aspartates and the coordinated water for activation. A new helix was formed for residues Ser26 to Ala30 in the Aβ49 during the transition from ε to ζ cleavage in the WT system (Figures 3A and S17). With the 50° tilt of Aβ49 peptide in the space between TM2 and TM3, the N-terminus is exposed to the hydrophobic lipid bilayer (Figure S17). This helped the N-terminal loop to transition to an α-helical conformation. The effects of the mutations on the new helical conformation are mentioned and explained in the next paragraph. A turn/unstructured conformation at residues Ala30–Ile31 separated these two helices. In addition, the N-terminus of Aβ49 lost its interactions with the hydrophobic loop 1 (HL1) because of the tilting away from this loop.

Figure 3.

Time-dependent secondary structures of Aβ49 bound to γ-secretase calculated from the Pep-GaMD simulations. (A) WT, (B) I45F, (C) A42T, and (D) V46F systems of Aβ49. Results from other simulations are plotted in Figures S12 and S15.

Similarly, secondary structural changes were recorded for the I45F, A42T, and V46F Aβ49 mutant systems (Figures 3B-D and S15). Like the WT, the I45F and A42T Aβ49 mutants maintained a helical conformation up to Thr43 at the N-terminus during the Pep-GaMD simulations. C-terminal residues after the Thr43, which included the ζ cleavage site bond between Val46 and Ile47, were observed mostly in a turn/unstructured conformation. This allowed catalytic aspartates and water to approach the scissile amide bond for forming coordinated hydrogen bonds required for this cleavage. Likewise, bands of new helix formation were observed in the secondary structure plots from Asn27 to Ile31 and from Asp23 to Lys28 for I45F- and A42T-mutant Aβ49 systems, respectively (Figure 3B,C). The new helix formed was due to its exposure to the hydrophobic lipid membrane. V46F Aβ49 was observed to be the most dynamic in terms of secondary structure changes (Figure 3D). A band of the helix was observed between Gly29 and Thr43, with a turn conformation formed between Leu34 and Val36. Thr43 to Ile47 transitioned between the helix and turn conformations during the Pep-GaMD simulations of the V46F Aβ49. Like the WT and other mutant systems, new N-terminal helix formation was observed at residues Phe20 to Gly25 in the V46F mutant APP (Figure 3D). The hydrophobic lipid environment helped these residues transition from turn to a helical conformation in the V46F mutant APP.

Active-Site Subpockets Formed in γ-Secretase for Tripeptide Trimming.

The “Final” active conformational state of Aβ49-bound γ-secretase was further analyzed for the P1′, P2′, and P3′ substrate residues at the ζ cleavage active site and the respective S1′, S2′, and S3′ subpockets in which they reside.¹⁰ The S1′ subpocket accommodating the P1′ residue in the WT Aβ49 was formed by residues from PS1 TM6a helix, β1 loop, TM3 helix, and TM7 N-terminal region (Figures 2 and 4A). The residues that formed the subpockets are listed in Table S1. The S2′ subpocket occupied by the P2′ substrate residue consisted of residues from PS1 TM6 helix, TM6a helix, PAL motif of TM9 helix, and β1 and β2 loop region. Moreover, the S3′ subpocket accommodating the P3′ residue was formed by residues from the PS1 TM6 helix, TM6a helix, and β1 loop. In reference to Aβ49, S1′ and S3′ pockets were located on the same side (TM6a and TM3 helices), whereas the S2′ pocket was located on the opposite side (TM6 and TM9 helices).

Similarly, the “Final” active conformational states of the I45F and A42T-mutant Aβ49-bound γ-secretase systems had the same subpockets formed at the active site during the ζ cleavage as that of the WT system (Figure 4B,C and Table S1). In the I45F and A42T “Final” active conformation, the S1′ and S3′ subpockets occupied by the respective P1′ and P3′ substrate residues consisted of residues from PS1 TM6 helix, TM6a helix, TM7 helix, and β1 and β2 loop region. In comparison, the S2′ subpocket was located on the opposite side of Aβ49 and consisted of residues from PS1 TM6 helix, TM6a helix, PAL motif of TM9 helix, and β1 and β2 loop. Furthermore, in the V46F “Final” active conformation, the locations of the S1′ and S2′ subpockets accommodating P1′ and P2′ Aβ49 substrate residues, respectively, were different as compared to that of the WT system (Figure 4C,D). The S1′ pocket occupied by the P1′ residue of Aβ49 consisted of residues from TM6 helix, TM6a helix, and TM2 helix (Table S1). The S2′ subpocket occupied by the P2′ residue of the V46F mutant in the “Final” active state was the same as the S1′ subpocket in the “Final” active state of the WT, I45F, and A42T systems (Figure 4C,D). Moreover, the S3′ subpocket accommodating the P3′ substrate residue in the V46F mutant was the same as the one in the WT, I45F, and A42T systems (Figure 4E).

DISCUSSION

Current AD treatments ease symptoms, but none has been clearly demonstrated to slow or halt disease progression. While the molecular cause of AD remains poorly understood, the hallmark pathological criterion for AD diagnosis is the deposition of amyloid-β (Aβ) plaques in the brain.⁴⁰ Aβ peptides are products of processive proteolysis by γ-secretase. Dominant missense mutations in the substrate (APP) and the enzyme (presenilin component of γ-secretase) cause early-onset FAD, and these mutations result in deficient carboxypeptidase trimming of initially formed long Aβ peptides to shorter secreted forms.^31,41,42 Yet, the mechanism of processive proteolysis of APP by γ-secretase is unknown. Recent reports of cryo-EM structures of γ-secretase bound to APP and Notch substrates as well as to γ-secretase inhibitors and modulators revealed details of the structural basis of substrate recognition as well as enzyme inhibition and modulation.^11,12,43 Regardless, static conformations of the enzyme cannot explain the underlying mechanism of enzyme activation and substrate processing. Essentially, nothing is known about the dynamic mechanism of processive proteolysis by γ-secretase.

It would require quantum mechanics/molecular mechanics (QM/MM) calculations to fully understand the catalytic mechanism of proteolysis by γ-secretase. The catalytic step is likely the rate-limiting step of the enzyme proteolysis, being slower than the substrate–enzyme interaction dynamics. Nevertheless, the latter (dynamic motions of the substrate–enzyme interaction) has been suggested to take place over minutes.³² This is still considered as slow dynamics and extremely long timescales that is way beyond the reach of state-of-the-art conventional MD simulations but amenable to enhanced sampling simulations. We initially performed ~6 μs regular dual-boost GaMD simulations but could not sample the stable enzyme–substrate hydrogen bond that characterizes system conformation for ζ cleavage of Aβ49 (Figure S3). Then, we turned to our recently developed Pep-GaMD method, which selectively boosts the essential potential energy of the peptides. Pep-GaMD has been demonstrated to greatly accelerate protein–peptide binding simulations by orders of magnitude.³³ Compared with previous GaMD, Pep-GaMD is a more powerful method that can be applied for further improved enhanced sampling of protein–peptide interactions. The new Pep-GaMD simulations allowed us to capture the ζ cleavage activation in 600 ns. In this context, novel Pep-GaMD simulations have, for the first time, captured slow dynamic conformational transitions in both the enzyme and substrate for tripeptide trimming of the wild-type and FAD mutants of Aβ49, being consistent with MS and Western blotting biochemical experiments.

Here, we have applied the combination of novel Pep-GaMD enhanced sampling simulations and biochemical experiments to address the issue. Different systems of γ-secretase bound by the WT and FAD-mutant Aβ49 substrates were investigated to understand tripeptide trimming, ζ cleavage (Figure 5). Five γ-secretase systems—bound to the WT, I45F, A42T, and V46F charged C-terminal Aβ49 in the presence of charged N-terminal AICD50-99 and bound to WT neutral C-terminal Aβ49 and charged N-terminal AICD50-99—underwent activation for ζ cleavage during 600 ns Pep-GaMD simulations (Figure 5B). This was consistent with biochemical experiments, as these mutant systems showed similar efficiencies for the Aβ49 to Aβ46 proteolytic step (ζ cleavage). In comparison, γ-secretase bound by I45T and T48P Aβ49 showed little or no sample activation (Figure 5C). Furthermore, Aβ49-bound γ-secretase in the absence of AICD50-99 was not able to sample the “Final” active state for ζ cleavage of the substrate (Figure S8A), similarly for γ-secretase bound to WT charged C-terminal Aβ49 and neutral N-terminal AICD50-99 and γ-secretase bound to WT neutral C-terminal Aβ49 and neutral N-terminal AICD50-99 (Figure S8B,C). This highlighted the importance of AICD50-99 and its N-terminal charge in facilitating processive proteolysis by γ-secretase. Following ε cleavage, both the C-terminus of Aβ49 and N-terminus of AICD50-99 at the active site could be exposed to water molecules and thus charged at physiological pH 7 (as carboxylate and ammonium, respectively). The charged state likely aided movement toward the polar aqueous environment and away from the hydrophobic transmembrane interior of the PS1 active site. Indeed, the AICD50-99 with charged N-terminus could dissociate from PS1 in the Pep-GaMD simulations that helped prepare for the next cleavage during processive proteolysis by γ-secretase.

Figure 5.

Dynamic model of tripeptide trimming of Aβ49 by γ-secretase. (A) “Initial” conformational state of Aβ49 bound γ-secretase. (B) WT Aβ49 and its I45F, A42T, and V46F mutants were able to transition to the “Final” state with ~50° tilting of the helical domain and unwinding of the helix C-terminus (residues V44-I45) and became poised for ζ cleavage of the V46-I47 amide bond by γ-secretase. (C) In contrast, the I45T and T48P mutant Aβ49-bound γ-secretase were trapped in the “Intermediate” or “Inhibited-1” state, being inactive for ζ cleavage of the substrate.

During ζ cleavage activation, two residues unwound from the C-terminus of the Aβ49 helix, changing to a turn conformation (Figure 5B). This was observed in the time courses of the substrate secondary structures as well. Unlike the helical conformation, the loop/turn conformation facilitated exposure of the scissile amide bond to the catalytic aspartates and the coordinated water molecule. In parallel, positions of the Thr43 and Ile45 residues in the “Initial” and “Final” states relative to the membrane were similar, whereas the C-terminal residue Leu49 moved downwards by ~5 Å. Moreover, the helical domain of Aβ49 tilted by ~50° (Figure 5B). Thus, tilting of the helical domain and unwinding of the C-terminal helix in the substrate apparently facilitated the proteolytic progression from ε to ζ cleavage by γ-secretase. Helix unwinding was accompanied by straightening of the C-terminal loop/turn and downward movement of the terminal residue Leu49. Similarly, the β-sheet conformation between the APP C-terminus and the β1 loop was broken as ε cleavage product AICD50-99 dissociates. This caused the β1 loop to move away from the APP C-terminus by ~5 Å. This region has been suggested to be important for substrate recognition and proteolytic processing.¹² Similarly, γ-secretase inhibitors (GSIs) and transition-state analogs (TSAs) bind to this region.⁴³ The present study also shows an important role of this region in the activation of γ-secretase for ζ cleavage of Aβ49.

Relevant to this study, Hitzenberger et al.¹³ performed restraint MD simulations to produce γ-secretase complex structure bound to the Aβ49, Aβ46, and Aβ43 peptides. Simulations on these complexes showed that both helix unwinding and sliding of active-site aspartates toward the scissile amide bond are responsible for peptide repositioning during substrate processing by γ-secretase. During repositioning of the Aβ peptides, the N-terminus was anchored to maintain its interaction with PS1 subunit. However, these enzyme–substrate model complexes were generated by combining apo γ-secretase missing the Nicastrin subunit and C99 peptide using restrained MD. AICD50-99 peptide was not included in the γ-secretase study, and charges on the terminal ends of the Aβ49 and AICD50-99 peptides were not considered. In comparison, our model was based on the holo enzyme activated for ε cleavage from our previous study²⁹ generated using the APP-bound γ-secretase cryo-EM structure. To the best of our knowledge, we are unaware of previous studies on the molecular dynamics of tripeptide trimming, in particular, trimming of Aβ49 to Aβ46, by the γ-secretase complex.

Pep-GaMD captured the enzyme activation for ζ cleavage for γ-secretase systems bound to WT and three FAD-mutant (I45F, A42T, and V46F) APP substrates. The low-energy “Final” active conformation was identified in the Pep-GaMD free-energy profiles of all of these systems. However, the PMF profiles representing each enzyme system were different in terms of distinct low-energy states and the conformational space sampled by the enzyme during ζ cleavage. The I45F and V46F mutant systems sampled two low-energy conformations with the I45F system being the least conformationally dynamic (Figure 1H,J). Three and four low-energy states were identified from free-energy profiles of the WT and the A42T-mutant systems, respectively, with the A42T-mutant system being more dynamic (Figure 1G,I). Each system had its own set of conformations and a distinct activation pathway. This suggested that the enzyme is remarkably dynamic, consistent with its ability to cleave over 140 different substrates.⁴⁴

In the “Final” active state of γ-secretase poised for the ζ cleavage, subpockets were formed in the active site that were different from that formed for the ε cleavage (Figures 4 and S20). This finding was consistent with the observation that the C-terminus of Aβ49 during ζ cleavage did not form a β-sheet conformation with the PS1 TM6a β2 region, instead adopting a loop conformation (Figure 2F). The locations of the active-site subpockets formed for ζ cleavage were compared to those formed for ε cleavage (Figure S20). Moreover, the locations of S1′, S2′, and S3′ subpockets formed for ζ cleavage were the same for different γ-secretase systems bound to the WT and mutant Aβ49 except for the V46F mutant system. The S2′ subpocket accommodating the T48 P2′ residue formed for ζ cleavage was the same as the S1′ subpocket accommodating the V50 P1′ subpocket formed for ε cleavage. In contrast, the ζ cleavage S3′ subpocket for the L49 P3′ residue was the same as the ε cleavage S3′ subpocket for the L52 P3′ residue. The S1′ subpocket for the I47 P1′ residue for ζ cleavage and the S2′ subpocket for the M51 P2′ residue for ε cleavage had their own unique location in their respective Final active states. Regardless, S1′/S3′ and S2′ subpockets were located on opposite sides of the substrate in both of the “Initial” and “Final” active states.

During 600 ns of Pep-GaMD simulations, we did not observe the enzyme activation at other cleavage sites except for ζ cleavage at the amide bond between Val46–Ile47 of the Aβ49 peptide. This can be observed from the time course plots of the distance between the protonated Asp257 and the carbonyl oxygen atoms of residues Ile47 (for cleavage at the second position) and Ile45 (for cleavage at the fourth position) (Figure S21). Moreover, the pathway of tripeptide trimming for ζ activation is energetically more favorable compared to that of the second or fourth amino acid residue cleavage. In this context, even though Pep-GaMD was able to capture the slow dynamic transitions of the enzyme activation for ζ cleavage during 600 ns simulation time, the simulations appeared to still suffer from an insufficient sampling of the entire system conformational space and the calculated free-energy profiles remained unconverged. Hence, the free-energy profiles reflect a semiquantitative picture of the tripeptide trimming process rather than the exact correctness of the free-energy values.

Here, we have investigated both the wild-type and five FAD mutants (including I45F, A42T, V46F, I45T, and T48P) of the substrate. In a recent report,³¹ we determined the Aβ42/40 ratios for these and other FAD mutations in the APP substrate and found the relative effects of these mutations on this ratio compared to that seen with WT enzyme to be generally consistent with those reported from other groups.^36-38,45 To the best of our knowledge, there are no other studies comprehensively exploring APP FAD mutations and how they are processed by γ-secretase beyond our own.³¹ Even with respect to Aβ42/40 ratios, a standard measure in the field, the only other comprehensive study of the 14 APP TMD FAD mutations was by our group,¹⁰ which gave closely similar relative changes in Aβ42/40 ratios with all of these mutations, even though the systems were different (cellular transfection of APP¹⁰ vs C100-Flag and purified proteins³¹).

Other reports on the effects of APP mutations on Aβ42/40 have studied only very selected mutations. Among the five mutations studied in the current work: (1) Aβ42 levels cannot be determined by ELISA for A42T, as this changes the ELISA epitope; (2) no reported data can be found for T48P; similar Aβ42/40 changes are seen with I45F and I45T;^36-38 the only discrepancy is with V46F: Devkota et al.³¹ showed no change vs WT, while Bolduc et al.,¹⁰ Lichtenthaler et al.,³⁷ and Tamaoka et al.⁴⁵ report ~4-fold increases in Aβ42/40. The reason for this discrepancy is unclear; however, only Devkota et al.³¹ used purified proteins, while the other reports measured secreted peptides in transfected cells.

In summary, we have presented here the first dynamic model of tripeptide trimming′of Aβ49 to Aβ46′by γ-secretase, which was highly consistent with mass spectrometry (MS) and Western blotting biochemical experiments. Specifically, MS and Western blotting were used to quantify the efficiency of tripeptide trimming of WT and FAD-mutant Aβ49. In comparison to WT Aβ49, the efficiency of tripeptide trimming was similar for the I45F, A42T, and V46F Aβ49 FAD mutants but substantially diminished for the I45T and T48P mutants. All-atom simulations performed in parallel with the biochemical experiments captured remarkable structural rearrangements of both the enzyme and substrate, in which hydrogen-bonded catalytic aspartates and water became poised for tripeptide trimming of Aβ49 to Aβ46. Our complementary biochemical experiments and all-atom simulations have enabled elucidation of the mechanism of tripeptide trimming of γ-secretase. It will guide our future studies on subsequent cleavage steps of the APP substrate and processive cleavage of the other substrates of γ-secretase. A detailed mechanistic understanding of these processes is expected to greatly facilitate the rational drug design of this critical enzyme.