Presenilin homologues influence substrate binding and processing by γ - 1 secretase: a molecular simulation study.

14 Presenilin homologues in the γ -secretase complex play a pivotal role in substrate binding and 15 processing, impacting β - amyloid (Aβ) peptide generation in Alzheimer's disease. We 16 conducted a molecular simulation study to determine substrate preferences between presenilin-17 1 (PS1) and presenilin - 2 (PS2) γ - secretase enzymes for amyloid precursor protein (APP) and 18 Notch1 processing. Using homology modelling, we generated PS1-and PS2-γ -secretase 19 models bound to substrates in the Aβ40 and Aβ42 generation pathways and Notch1 S3 and S4 20 cleavage site substrates. Metadynamics simulations and binding free energy calculations were 21 used to explore conformational ensembles and substrate preferences. PS2-γ -secretase exhibited 22 increased conformational flexibility and prefere ntial binding energy for initiating the Aβ42 23 pathway compared to PS1-γ -secretase. Additionally, Notch1 exhibits a preference for binding 24 to PS2-γ -secretase over PS1-γ -secretase. This study provides valuable insights into the 25 conformational dynamics of γ -secretase bound to different substrates within a cleavage 26 pathway, improving our understanding of substrate processivity. The findings highlight the 27 importance of considering both PS1-and PS2-γ -secretase in structure-based drug design 28 efforts, with implicat ions for stabilizing or destabilizing specific states during APP processing.


INTRODUCTION
Alzheimer's disease (AD) is the most common form of dementia, affecting approximately 35 million people worldwide.With no effective drug interventions that prevent or markedly slow down disease progression, a greater understanding of disease pathogenesis is required. 1A key pathological hallmark of AD is the formation of cerebral amyloid plaques, which contain Amyloid-β (Aβ) peptides. 24][5] The most common Aβ forms generated are Aβ40, accounting for the majority of peptides produced, and Aβ42, accounting for a substantial minority of peptides produced, with the homeostatic Aβ42:Aβ40 ratio being approximately 1:9. 6 However, these product ratios often shift in AD, such that increased production of longer forms -particularly Aβ42 and Aβ43 -occurs, 3,7,8 leading to increased Aβ aggregation. 4,5 hus, modulation of the production of Aβ peptides represents a potential therapeutic strategy for the treatment of AD. 9,10 Aβ peptides are generated via the regulated intramembrane proteolysis (RIP) of Amyloid Precursor Protein (APP), a Type I single-pass transmembrane protein.Initial cleavage by β-APP cleaving enzyme-1 (BACE1) results in ectodomain shedding of APP, allowing the membrane-embedded C-terminal fragment to be cleaved by the integral membrane enzyme γsecretase, ultimately releasing the APP IntraCellular Domain of APP (AICD) and Aβ peptides. 113][14][15][16] There are two homologues of Presenilin (PS) -Presenilin-1 (PS1) and Presenilin-2 (PS2) -and three .CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is The copyright holder for this preprint this version posted May 19, 2023.; https://doi.org/10.1101/2023.05.17.541079 doi: bioRxiv preprint isoforms of APH1 -Aph1a S , Aph1a L , Aph1b -giving rise to six discrete forms of the γsecretase enzyme. 171]  .CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is The copyright holder for this preprint this version posted May 19, 2023.; https://doi.org/10.1101/2023.05.17.541079 doi: bioRxiv preprint γ-Secretase generates Aβ peptides via a series of sequential cleavages of the transmembrane domain of APP.These cleavages occur within the lipid bilayer and involve an initial cleavage, releasing the AICD, followed by 2-3 further cleavages spaced at approximately three amino acids between each cleavage (approximately one helical turn).ultimately resulting in the Aβ peptide to be released from the membrane. 22,23 he APP processing pathway is well defined, and although there are multiple product lines, two primary pathways have been determined (Fig 1B, C). 23 These pathways produce Aβ40 and Aβ42 as their major products.The initial substrate, APP-C99, is the same for both product lines, and is cleaved to release the AICD.22,23 However, the cleavage position varies by one residue between the two pathways, with this initial cleavage site being the primary determinant of the product pathway.Cleavage between residues Leu49 and Val50 typically leads to the production of the Aβ40 product, whereas cleavage between Thr48 and Leu49 leads to Aβ42 product generation (Fig 1B , C). 22,23 Initial efforts to target γ-secretase involved the development of inhibitors; however, this has been hampered as a therapeutic strategy by side-effects believed to be associated with concurrent inhibition of the processing of other substrates, including Notch-1, which has a diverse array of cell type dependent signaling functions.24 Notch-1 is processed similarly to APP, where, prior to γ-secretase processing, ectodomain shedding by ADAM10 occurs, 25,26 and the membrane retained product, referred to as the Notch Extracellular Truncation (NEXT), is the γ-secretase substrate.NEXT is initially cleaved by γ-secretase at the S3 site to release the Notch Intra Cellular Domain (NICD).27,28 Subsequent cleavage leads to the release of an Aβequivalent product, and the final cleavage site is the S4 site (Fig 1D).28,29 Multiple S3 and S4 sites are known, and it is likely that there are similar intermediate processing pathways to those identified for APP processing; however, this has not been definitively determined for Notch.
A key advance in improving the functional understanding of γ-secretase has been the determination of its three-dimensional atomic structure via cryo-electron microscopy (cryoEM).The first structure by Bai et al., 30 was a 3.4 Å resolution structure (PDB 5A63), followed closely by an additional four structures 31 ranging from 4.0 -4.3 Å resolution.These structures included an apo-state structure (PDB 5FN5), a DAPT-bound state (a γ-secretase inhibitor; although the structure of the inhibitor could not be resolved) (PDB 5FN2), and two structures containing a volume believed to represent a substrate helix (PDB 5FN3 and 5FN4).
While these cryoEM structures provide static snapshots, they can be utilized to inform our understanding of the structure and dynamics of γ-secretase.Considerable molecular dynamics work has been completed using these structures, in particular, those prior to the most recent substrate bound structures, investigating the dynamics of substrate docking and entry pathways [34][35][36][37][38] and the effect of lipids on γ-secretase conformation. 37,39 ore recently, the effect of presenilin mutations on γ-secretase conformation and substrate binding have been investigated. 40,41 rediction of the binding site of γ-secretase modulators and inhibitors 42,43 has been complemented by the most recent atomic structures of PS1-γ-secretase with inhibitory and modulatory small molecules bound. 44 typically considered the more important PS in the context of AD, as the vast majority of pathogenic mutations in the presenilin protein associated with AD occur in PS1 compared with PS2. 45 However, PS2 expression has been shown to increase over time with in vitro neuronal differentiation 46,47 and in the human and murine brain with age, 48,49 while in cell culture, PS2-γ-secretase has been shown to generate more intracellular Aβ 50,51 and increased Aβ42:Aβ40 ratio.17,47,52,53 These observations collectively suggest a role for PS2-γ-secretase in disease presentation and progression.Understanding the specific mechanisms and contributions of PS1-and PS2-γ-secretase to substrate binding and processing is therefore important to ensure that effective therapeutics are developed that target PS1 and PS2 complexes effectively.
In this study, we used enhanced sampling approaches, specifically, targeted molecular dynamics and well-tempered metadynamics (WTMetaD), to explore the conformational landscape of PS1-and PS2-γ-secretase in the context of APP and Notch1 processing to examine the similarities and differences and enable improved structural understanding for the future development of Aβ modulating therapies.These approaches were further complemented by binding energy calculations to suggest specific preferences for substrate binding to the different forms of γ-secretase.Traditional experimental methods are limited in their ability to readily provide conformational information about the various states of γ-secretase bound to the multiple Aβ peptides.Similarly, traditional simulation approaches may overly sample specific protein conformations and lead to an incomplete understanding of the potential states that may be adopted by a given protein or protein complex.Homology modelling and metadynamics provide a tool kit that can allow the investigation of multiple γ-secretase and substrate combinations to understand the conformational ensembles of PS1-vs PS2-γ-secretase processing of APP and Notch1.

Derivation of path between APP-bound and Notch1-bound PS1-γ-secretase
Targeted molecular dynamics and analysis of the derived trajectories was initially performed to identify a potential path linking the APP-bound and Notch1-bound states of PS1-γ-secretase.
The path ultimately identified was 20 frames in length.Key regions of dynamic movement in the frames comprising the path, representing key regions of structural variation between the two complex structures, are the region between the second and third transmembrane helices of presenilin, the region between the sixth and seventh transmembrane helices of presenilin, and the region between the third and fourth transmembrane helices of Aph1a (Fig S1A).All of these regions are located on the lumenal side of the complex.The corresponding PS2-γsecretase path was derived via homology modelling to the component frames of the PS1-γsecretase path, on the assumption that PS2-γ-secretase is likely to display similar dynamics to PS1-γ-secretase.

WTMetaD of PS1-and PS2-γ-secretase bound to APP substrates
Using the cryoEM structure of PS1-γ-secretase bound to APP-CTF in position for the initial cleavage of the Aβ40 pathway (PDB:6IYC) 32 , homology models of PS1 and PS2-γ-secretase bound to APP-CTF (and intermediate fragments) in position for cleaving Aβ49, Aβ48, Aβ46, Aβ45, Aβ43, Aβ42, Aβ40, and Aβ38 were prepared.Using the path derived from targeted MD simulation, WTMetaD in the position along (denoted as s, where s = 1 represents γ-secretase bound to APP as per PDB 6IYC, herein referred to as the 6IYC-like conformation of γsecretase, and s = 20 represents γ-secretase bound to Notch1 as per PDB 6IDF, herein referred to as the 6IDF-like conformation of γ-secretase) and deviation from (denoted as z, with z = 0 being along the path, and z > 0 being a deviation from the path) the path was performed.Energetic minima derived from WTMetaD of PS1-and PS2-γ-secretases bound to APP-CTF in the cleavage position to initiate the Aβ42 pathway (i.e., to generate the intermediate Aβ48 product) revealed two energetic minima for the PS1:CTF(Aβ48) complex, both of which are 6IYC-like (Fig 3A : PS1:CTF(Aβ48) s ≈ 3, z ≈ 0.035 and s ≈ 1.5, z ≈ 0.08).The lowest energy minimum of the PS1:CTF(Aβ48) complex (s ≈ 3, z ≈ 0.035) is comparable to the PS1:CTF(Aβ49) complex minimum in contour breadth, suggesting a similar conformational flexibility in PS1-γ-secretase when binding either substrate.However, the PS2 complex has only one 6IYC-like energetic minimum (Fig 3E : PS2:CTF(Aβ48) s ≈ 1.5, z ≈ 0.04), and while similarly positioned with respect to the path compared to the PS2:CTF(Aβ49) complex, this minimum has broader contours, suggesting that this complex is likely more flexible than the equivalent complex with PS1.
WTMetaD of PS1-and PS2-γ-secretase in complex with all intermediate substrates of the Aβ40 and Aβ42 processing pathways was then performed (Fig 2 & 3).All subsequent γ-secretasesubstrate complexes, after the initial cleavage to release AICD, have one energetic minimum, with the exception of the PS1:Aβ48(Aβ45) complex, which has two energetic minima (Fig 3B).All minima are 6IYC-like (s < 5 in all cases).The contour breadth in the PS1-γ-secretase complexes with Aβ40 pathway substrates is considerably tighter in the shorter substrates

Binding free energies of γ-secretase -APP bound complexes
Molecular mechanics-generalized Born/surface area calculations (MM-GB/SA) were performed for all enzyme-substrate complexes to estimate the binding energies (ΔGbind) for γsecretase with each of the substrates (Table 1).Structures corresponding to low energy regions of the free energy surfaces derived from the WTMetaD simulations were used for the MM-GB/SA calculations.The PS1:CTF(Aβ49) complex, in the position to initiate the Aβ40 pathway, has a more favorable binding energy compared to the PS2:CTF(Aβ49) complex.
However, the opposite preference is observed for CTF(Aβ48), where PS2:CTF(Aβ48) has a marginally lower binding energy than PS1:CTF(Aβ48).While there are two minima evident from the metadynamics simulation for the PS1:CTF(Aβ48) complex, MM-GB/SA calculations at both minima afford a higher binding energy compared to the PS2:CTF(Aβ48) complex.
Thus, CTF(Aβ48) is predicted to have a preference for binding PS2-γ-secretase over PS1-γsecretase.Considering the binding energy results for all four complexes bound to the APP-CTF, there is a preference for both PS1-and PS2-γ-secretase to bind the APP-CTF in the Aβ48 position over the Aβ49 position.

Per residue decomposition of binding free energies for γ-secretase-APP bound complexes
Per-residue decompositions of the MM-GB/SA binding energies were performed to determine the individual contributions of residues from the enzyme and the substrate to the overall binding energy.The relative change in binding energy per-residue with respect to PS1-versus PS2-γ-secretase was also calculated (ΔΔGPS Pref, calculation described in Methods) to determine the precise contributors to selectivity (Figures 4 and 5).
The APP-CTF substrate bound in position to produce the Aβ49 product, thus initiating the Aβ40 pathway, shows preference for binding to PS1-γ-secretase over PS2-γ-secretase.This preference is driven by a series of contributions clustered at the N-terminus of the substrate, specifically, nicastrin residues Ile246 (ΔΔGPS Pref = +2.49kcal/mol) and Arg652 (ΔΔGPS Pref = +6.07kcal/mol), which form hydrogen bonds (mediated by the Ile246 backbone amide) and a salt bridge respectively with Glu22 (ΔGbind PS1 = +0.87kcal/mol) in the substrate.Substrate residue Leu17 (ΔΔGPS Pref = +4.26kcal/mol), which forms a hydrogen bond between its backbone carbonyl with the backbone amide of Trp653 (ΔGbind PS1 = -8.64kcal/mol) in nicastrin, and the PS1 residue Arg108 (ΔΔGPS Pref = +4.29 kcal/mol), which is likely forming cation-π interactions with the substrate residues Phe19 (ΔGbind PS1 = -4.60 kcal/mol) and Phe20 (ΔGbind PS1 = -2.17kcal/mol), also contribute to this preference.Arg108 also forms a salt bridge with Nct Glu245.Arg108 is not conserved in PS2 and is replaced by a glutamate at the analogous position (PS2 Glu114), with this residue unable to make the same interactions (Fig hydrophobically interacts with substrate residue Met51 (ΔΔGPS Pref = +0.77kcal/mol).These interactions likely contribute to the stabilization of the β-sheet structure between PS1 and the substrate that is a key feature of the γ-secretase bound substrate, 32,33 and do not readily occur in the PS2 complex.
The subsequent complex in the Aβ40 pathway, the Aβ49 substrate positioned to produce the Aβ46 product, demonstrates a preference for binding to PS2-γ-secretase over PS1-γ-secretase.
This preference is a result of two key regions of interactions.PS2 Gln118 (ΔΔGPS Pref = -    The Aβ46 substrate, processing of which leads to the generation of the Aβ43 product in the Aβ40 pathway, shows a preference for binding to PS1-γ-secretase over PS2-γ-secretase.Key contributors to the binding preference are the nicastrin residue Arg652 (ΔΔGPS Pref = +4.27kcal/mol), and the PS1 residue Arg108 (ΔΔGPS Pref = +4.24kcal/mol), which forms electrostatic interactions with Glu22 (ΔGbind PS1 = -0.85kcal/mol) in the substrate.As previously noted, the PS1 Arg108 residue is not conserved in PS2, with the analogous residue in PS2 being Glu114; consequently, these interactions cannot occur in the PS2-complex.Substrate residues at the Nterminal juxtamembrane region, Ser26 (ΔΔGPS Pref = +2.45kcal/mol) and Asn27 (ΔΔGPS Pref = +2.28kcal/mol), further contribute to the substrate preference for PS1 by forming a greater network of hydrogen bonds with residues in the loop between TM-1 and TM-2 (Ile114: ΔΔGPS notably, these appear to stabilize the β-sheet structure between the substrate C-terminus and PS1 previously noted. 32,33 hese interactions occur in tandem with the Trp165 (ΔGbind PS1 = -   The subsequent substrates in the Aβ42 pathway all demonstrate preference for binding to PS1γ-secretase (Table 1).The primary contributors to this preference in the γ-secretase complexes bound to the Aβ48 substrate in position to generate the Aβ45 product are the nicastrin residue ).These interactions do not form in the PS2complex; notably, PS1 residue Leu172 is not conserved (analogous PS2 residue is Met178), nor is the residue immediately adjacent to Phe177 (PS1 Phe176 analogous PS2 residue is Leu182), likely influencing the observed interactions between the PS1-and PS2-complex.The substrate helix itself in the PS2-complex is observably disrupted at its di-glycine motif, which has been identified as a point of flexibility in APP, 54 and also likely contributes to the reduced interactions with PS2 residues in the substrate binding pocket.Lastly, interactions involving The final complexes modelled in the Aβ42 pathway feature Aβ42 itself as the substrate, positioned for the generation of Aβ38 as the final product in the pathway.From the binding free energies determined by MM-GB/SA, this substrate exhibits a preference for PS1-γsecretase over PS2-γ-secretase.The preference is driven by the PS1 residue Phe176 (ΔΔGPS Pref = +2.44 kcal/mol), which can form NH-π interactions 55 with the substrate residue Asn27.An equivalent interaction is unable to form in the PS2 complex, as Phe176 is replaced by Leu182

Notch1 substrates
Given the functional implications of Notch1 inhibition, it is imperative that Notch1 processing by γ-secretase is considered in future therapeutic targeting of the enzyme.Consequently, we examined Notch1 bound to PS1-and PS2-γ-secretase enzymes positioned for processing at the two primary S3 sites -Val1754/Leu1755 and Gly1753/Val1754 -and the two primary S4 sites -Val1745/Leu1746 and Ala1741/Ala1742.The cryoEM structure of PS1-γ-secretase bound to Notch1 in the Val1754/Leu1755 (PDB: 6IDF) 33 was used to generate homology models of PS1-and PS2-γ-secretase enzymes bound to substrates in the S3 and S4 positions.
WTMetaD in the position along and deviation from the path was performed, using the path derived from targeted MD.WTMetaD of all four substrate/cleavage positions in complex with the PS1-or PS2-γ-secretase enzyme revealed only one energetic minimum for each complex, except for PS1-γ-secretase bound to the S3 substrate in position to cleave at the   The binding energies for the Notch complexes were determined by MM-GB/SA using the structures corresponding to the energetic minima derived from WTMetaD (Table 2).The data indicates that the NEXT substrate demonstrates a considerable preference for binding PS2-γsecretase in both the Val1754/Leu1755 and Gly1753/Val1754 positions compared to PS1-γsecretase.Similarly, the S3 substrate positioned to cleave at the Val1745/Leu1746 S4 site exhibits a preference for binding to PS2-γ-secretase.The S3 substrate positioned to cleave at the Ala1741/Ala1742 S4 site, however, preferentially binds PS1-γ-secretase.The binding energies for the enzyme bound to the S3 substrate in position for the final γsecretase cleavage at S4 site Val1745/Leu1746 suggest a substrate binding preference for PS2γ-secretase (Table 2).This preference is supported by three key interacting regions.At the luminal juxtamembrane region, the substrate residue Val1726 (ΔΔGPS Pref = -2.21kcal/mol) forms hydrophobic interactions with nicastrin residues Phe240 and Ile242 (ΔΔGPS Pref = -2.08 kcal/mol), which do not occur in the PS1-complex.Two PS2 residues, Met152 (ΔΔGPS Pref = -   The S3 substrate in position to cleave at the Ala1741/Ala1742 S4 site for peptide release shows a preference for binding to PS1-γ-secretase over PS2-γ-secretase, unlike the other Notch complexes.This preference for binding to PS1-γ-secretase is predominantly driven by the interactions of the substrate C-terminal residues.Phe1748 (ΔΔGPS Pref = +5.23 kcal/mol) and

DISCUSSION
The propensity for either PS1-or PS2-γ-secretase to generate a specific profile of Aβ species is a function of both the initial cleavage site and the subsequent likelihood of the successive tri/tetra-peptide cleavage events occurring.In this study, we undertook well-tempered metadynamics simulations of PS1-and PS2-γ-secretase enzymes complexes with the initial and intermediate APP substrates of the two major Aβ species, as well as Notch1-derived substrates.We analyzed these simulations to determine the likely low energy states for γsecretase-substrate complexes and the binding free energy for each substrate bound to γsecretase.All data was generated and compared for both PS1-and PS2-γ-secretase enzymes in order to assess enzyme preference for given substrates.
Our metadynamics results suggest a comparable ability for PS1-γ-secretase complexes to initiate either the Aβ40 or Aβ42 pathway (Fig 2A , 3A), while binding of subsequent substrates in the Aβ40 pathway (Fig 2B-D) involves a restricted conformational ensemble and may be less favorable.PS1-γ-secretase binding free energy results indicate a preference for binding APP in the position to initiate the Aβ42 pathway.Interestingly, subsequent substrates in both pathways are generally more efficiently bound, suggesting that PS1-γ-secretase processing of APP likely leads to the release of shorter Aβ peptides (Table 1).The metadynamics results for PS2-γ-secretase suggest a preference for PS2-γ-secretase to initiate the Aβ42 pathway over the Aβ40 pathway, marked by a less restricted verses a more restricted conformational ensemble for binding the respective substrates (Fig 2B , 3B).The subsequent substrates in each pathway, however, elicit broad conformational flexibility in PS2-γ-secretase, with the exception of the Aβ42(Aβ38) substrate, which yields a restricted conformational ensemble and may suggest reduced propensity for PS2-γ-secretase to generate Aβ38 products (Fig 3H).Binding free energy results for PS2-γ-secretase complexes support a considerable preference for binding APP-CTF to initiate the Aβ42 pathway.Interestingly, PS2-γ-secretase generally binds subsequent substrates in both pathways with lower binding energy (Table 1), suggesting that PS2-γ-secretase may generate longer Aβ products.
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is The copyright holder for this preprint this version posted May 19, 2023.; https://doi.org/10.1101/2023.05.17.541079 doi: bioRxiv preprint While it is likely that Notch undergoes similar tri-and tetra-peptide successive cleavage, this detail has not been elucidated; however, multiple initial cleavage (S3 sites) and final cleavage (S4 sites) γ-secretase sites have been identified. 28,29,56 Hre, we investigated PS1-and PS2-γsecretase binding to Notch1 substrates aligned with the primary S3 and S4 sites.While PS1and PS2-γ-secretase-complexes bound to the NEXT substrates in the S3 cleavage site positions elicit similar conformational flexibility in γ-secretase, the conformation of γ-secretase suggested by metadynamics simulations in the PS2 complexes is more akin to the cryoEM PS1-Notch complex (PDB 6IDF) than the PS1 complexes (Fig 6A -D).Furthermore, the PS2:NEXT(VL) and PS2:NEXT(GV) complexes have considerably more favorable binding energies over the equivalent complexes with PS1 (Table 2).Our data suggests that PS2-γsecretase would preferentially process Notch1 substrates to generate the NICD over PS1-γsecretase.
Generation of AICD and NICD products is indicative of the propensity for the initial cleavage by PS1-and PS2-γ-secretase, however, experimental data identifying PS1-or PS2-γ-secretaseICD generation levels is discordant.Some studies show that PS1-γ-secretase generates more AICD (or APP-CTF accumulation) or NICD, [57][58][59][60][61] while others show that PS1 and PS2 produce similar levels of ICD products. 17,50 otably, these studies are undertaken in different experimental conditions, use different cell lines, and importantly, do not account for differences in PS1 and PS2 expression that will likely influence total γ-secretase activity levels.
We have recently shown in HEK293 cells that PS1 expression is approximately 5-times that of PS2 expression and that this expression profile is retained in an exogenous expression system.Subsequently, we show that when PS expression is accounted for PS2-γ-secretase processes more APP and Notch substrate than PS1-γ-secretase in an exogenous system. 53In this study, we see that both PS1-and PS2-γ-secretase have comparable conformational flexibility when bound to NEXT substrates (Fig 6A -D).However, the substantial preference for PS2-γ-secretase binding of NEXT substrates (Table 2) shown in this study supports the notion that PS2-γsecretase would generate more NICD product at an individual enzymatic level.
With respect to APP processing, it is the initial cleavage position and the different propensities for PS1-vs PS2-γ-secretase to cleave these that is important in the context of AD and Aβ generation.However, any preference for the initial cleavage site between PS1-and PS2-γsecretase remains contentious; one study 62 , using PS1+/+ PS2+/+ genotype HEK293 and HeLa cells, shows that the AICD products aligning with Aβ42 pathway initiation predominates in endosomal fractions, where PS2 localises, 50,60 whereas AICD products aligning with Aβ40 pathway initiation predominate in plasma membrane fractions, where PS1 primarily localises. 50,60 nother study shows that both PS1-and PS2-γ-secretase generate similar ratios of the AICD product of both Aβ40 and Aβ42 pathways. 63While studies investigating the gamut of Aβ products by both PS1-and PS2-γ-secretase are not plentiful, PS2-γ-secretase has been shown to generate a higher Aβ42:Aβ40 ratio than PS1-γ-secretase. 47,52,53,64 Aditionally, PS1γ-secretase generates higher levels of Aβ38 than Aβ42, while the opposite is true of PS2-γsecretase, 57,59,63,65 The final Aβ profile generated by γ-secretase is affected not only by the initial cleavage site, but also by the likelihood of continued processing.The data presented in this study shows that PS2-γ-secretase not only has a preference for binding to CTF(Aβ48), but also a broader conformational ensemble when binding this substrate, compared with CTF(Aβ49) (Table 1, Fig 2A-B, 3A-B), supporting the view that PS2-γ-secretase is likely the predominant Aβ42-generating enzyme.This is further supported by comparably unfavorable free binding energies of the PS2-γ-secretase bound to subsequent substrates of the Aβ42  1), and the restricted conformational ensembles observed for the PS2:Aβ42(Aβ38) complex (Fig 3H).Combined, these data indicate that PS2-γ-secretase will preferentially initiate Aβ42 pathway and will likely release the substrate prior to Aβ38 product generation.
Improved understanding of PS1-and PS2-γ-secretase specific substrate processivity and the repertoire of enzyme-substrate conformations is critical for the future development of novel γsecretase targeting therapeutics.Following the failures of γ-secretase inhibitors to gain traction, attention has turned to the development of molecules that selectively inhibit the processing of APP -γ-secretase modulators (GSM).[67][68] The atomic structure of PS1-γ-secretase bound to the GSM E2012 was recently solved and from this, it has been proposed that the facilitation of substrate helix unwinding is a possible mechanism by which GSMs may increase the production of shorter Aβ products. 44Our study identifies conformational bottlenecks between PS1-and PS2-γ-secretase, in particular, with APP processing, presenting opportunities to stabilize or destabilize specific states.Additionally, we provide insight into γ-secretase targeting more broadly, i.e.APP vs Notch vs other substrates, as different types of substrates are shown to affect γ-secretase conformation differently, which may be harnessed for future structural based drug design. 69bly, we observe different conformations between PS1-and PS2-γ-secretase bound to NEXT substrates, which implies that PS1-and PS2-complexes could be targeted differently.This is supported by PS1 vs PS2 selectivity that is already evident in γ-secretase-targeting small molecules that have been developed through traditional medicinal chemistry pipelines. 51,59,65,70 This sudy provides insight into the conformational repertoire of γ-secretase bound to the various substrates within a cleavage pathway, improving our understanding of substrate processivity and highlights the importance of due consideration for both PS1-and PS2-γsecretase in structure-based drug design.

Structure preparation.
The structures of PS1-γ-secretase bound to APP (PDB 6IYC) and Notch1 (PDB 6IDF) were obtained from the Protein Data Bank and used for the PS1:CTF(Aβ49) and PS1:NEXT(VL) models.The subsequent PS1-γ-secretase models with either the APP or Notch substrate in different positions commensurate with the intermediate substrate and positioned for the expected product (Fig S5 ), and all PS2-γ-secretase models were generated using Advanced Homology Modelling within Schrodinger 2018-3.ProPKA 71 was used to assign protonation states within prepared structures, typically predicting the catalytic aspartate Asp385/Asp366 to be protonated/neutral in charge and the catalytic aspartate Asp257/Asp263 to be deprotonated/charged.Once built, the structure was aligned to its coordinates as deposited in the Orientations of Protein in Membranes (OPM) database, 72 to facilitate the subsequent system setup for molecular dynamics simulations.Acetyl caps at the N-terminus and N-methylamine caps at the C-terminus were added to the PS-NTF chain, while the PS-CTF chain was only capped at its N-terminus, and APH1a, NCT and the APP-CTF and NEXT substrates were only capped at the C-terminus.The PEN2 chain and subsequent substrates were not capped.

Simulation box preparation.
Built complexes were set up for simulation adapting procedures from the Amber lipid force field tutorial. 73Briefly, built complexes were submitted to the CHARMM-GUI web server, where they were embedded in POPC bilayers (120 x 120 lipids in size) and solvated (to create a box with a minimum distance of 15 Å between the edge of the protein and the box edge), with salt added for charge neutralization of the system and for simulating relevant physiological ionic strength (150mM NaCl).Aguayo-Ortiz et al. 39 have shown that γ-secretase in-silico conformational behavior is similar in different homogenous membrane environments, hence we have simulated γ-secretase in 100% POPC.Relevant caps were chosen during CHARMM-GUI preparation to preserve the caps used during structure preparation.
Conversion of the prepared system from CHARMM to AMBER format, as well as system parameterization, was performed using AmberTools. 75The protein was parameterized using the AMBER ff14SB force field. 76Lipids were parameterized using Lipid14. 77TIP3P water was used throughout. 78The Joung-Cheatham ion parameters were used for sodium and chloride. 79resulting topology was then ported to GROMACS format using acpype. 80

System Equilibration
Simulations were performed using GROMACS 2018.3 81 patched with PLUMED 2.5. 82librations of the system in the NVT and NPT ensembles were adapted from previously described procedures for equilibration of membrane proteins. 83Briefly, heavy atoms in the protein and lipids were position-restrained with harmonic restraints at 240 kJ/(mol nm 2 ), and the system gradually heated in the NVT ensemble from 0K to 100K over 0.1ns, followed by heating in the NPT ensemble from 100K to 300K for a further 0.1ns.Following this, further NPT simulations (0.1ns in duration each) were conducted with only the protein heavy atoms position-restrained, utilizing gradually decreasing force constants (120 kJ/(mol nm 2 ), 96 kJ/(mol nm 2 ), 72 kJ/(mol nm 2 ), 48 kJ/(mol nm 2 ), 24 kJ/(mol nm 2 )), with a further 0.1ns simulation performed without position restraints.

Identification of a path between APP-bound and Notch-bound states
Targeted molecular dynamics was used to derive a path between the APP-bound and Notch1bound states of PS1-γ-secretase.10 simulations of 5ns duration were conducted, each starting from the APP-bound state of γ-secretase biasing towards the Notch1-bound states.During these simulations, a restraint of 50 kcal/mol was employed on the RMSD of all heavy atoms to the Notch1-bound state, as well as a concurrent added restraint of 100 kcal/mol on the RMSD of heavy atoms of presenilin transmembrane regions 2, 3 and 4, which are key interactors with the substrate.Frames from each simulation were clustered to 0.1nm, with the simulation giving rise to the largest number of clusters used to give the frames of the path.A vehicle routing solver 84 on the RMSD matrix of the clusters was used to identify the order of clusters giving rise to the smallest distance between adjacent frames.With the exclusion of the nicastrin ectodomain, all heavy atoms were used in the final determined path.
The frames comprising the PS2-γ-secretase path were generated by homology modelling against each corresponding frame of the PS1-γ-secretase path determined by targeted molecular dynamics.

Metadynamics Simulations of γ-Secretase complexes
After equilibration, the systems underwent well-tempered metadynamics (WTMetaD) to explore the conformational ensembles of PS1-and PS2-γ-secretase enzymes bound to APP and Notch derived substrates, similarly to that previously reported. 85Briefly, the collective variables for the WTMetaD bias were the position along the targeted MD-generated path (s) and the distance from this path (z).σ for s and z were set as 0.5 and 0.001 respectively, as determined by approximately half of the standard deviation in these variables at the conclusion .CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is The copyright holder for this preprint this version posted May 19, 2023.; https://doi.org/10.1101/2023.05.17.541079 doi: bioRxiv preprint of a 5ns unbiased simulation of PS1-γ-secretase bound to APP (Fig S6).Simulations were performed at 310 K for 500 ns, with Gaussian hills 1 kJ/mol in size added every 1 ps and employing a bias factor of 40.Calculation of the metadynamics reweighting factor 86 was enabled and the bias was stored on a grid for computational efficiency (updated every 10 ps using grid bin widths of 0.2 in s and 0.0005 in z).Atomic coordinates, velocities and energies were saved every 10 ps.To limit the exploration of deviations from the defined path, an upper wall in z at 0.1 was used, with a force constant of 100 and a rescaling factor of 0.001.

Identification of Low-Energy States of γ-Secretase complexes
Free energy surfaces (FESs) in terms of s and z were calculated from the WTMetaD simulations and reported relative to the lowest energy value in the determined FES.Convergence of FESs was assessed by monitoring the difference between free energy surfaces at 1ns intervals, as well as the Gaussian hill height and the collective variable space sampled over the duration of the simulations (Fig S7 -S10).Structural ensembles for each minimum in the FES within 2.5 kJ/mol of the global minimum were extracted from the WTMetaD simulations and clustered using the GROMACS gmx cluster utility, employing the GROMOS algorithm 87 and a 0.2nm threshold.

Binding free energy calculations
The structural ensembles extracted for each energetic minimum were used to calculate the substrate-enzyme binding energies, which were determined using the molecular mechanics generalized Born/surface area (MM-GB/SA) approach, facilitated by the MMPBSA.pytool of AmberTools. 88The single-trajectory protocol was utilized, 89 with the following equation: . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is The copyright holder for this preprint this version posted May 19, 2023.; https://doi.org/10.1101/2023.05.17.541079 doi: bioRxiv preprint ΔGbind = Gcomplex -Gprotein -Gligand = ΔH + ΔGsolvation -TΔS = ΔGvdw + ΔGele + ΔGGB + ΔGSA -TΔS where ΔGvdw is the molecular mechanics van der Waals interaction energy, ΔGele is the molecular mechanics electrostatic interaction energy, ΔGGB is the change in polar desolvation energy upon complex formation and ΔGSA is the change in nonpolar desolvation energy upon complex formation.The entropic term (-TΔS) was not calculated, due to high computational cost and poor accuracy.The GBneck2 model (igb = 5) was used to calculate the polar desolvation energy. 90The non-polar component of the desolvation energy was calculated via solvent accessible surface areas calculated with the LCPO method. 91MM-GB/SA energies were also decomposed per-residue (idecomp = 1).
To determine the preference between PS1-and PS2-complexes bound to the same substrate ΔΔGPS Pref = ΔGbind PS2 -ΔGbind PS1 was calculated, where positive values indicate preference for PS1, while negative values indicate preference for PS2.
further along the pathway, in comparison to the initial CTF(Aβ49) substrate (Fig 2B-D), compared to the broader contours in the equivalent complexes with PS2-γ-secretase (Fig 2F-H).In contrast, PS2-γ-secretase displays a restricted conformational ensemble when binding to the initial CTF(Aβ48) substrate in the Aβ42 pathway (Fig 3E), while subsequent enzymesubstrate complexes in the pathway (with the exception of the PS2:Aβ42(Aβ38) complex) display relatively broader free energy surface contouring, suggesting increased conformational flexibility (Fig 3F-H).

3 ± 1 . 0 *
The PS1-γ-secretase complex bound to APP-CTF in position to generate the Aβ48 product has a second minima located at s = 0 to s = 2 and z = 0.07 to z = 0.09, this complex conformation has a MMGBSA free binding energy of -194.7 ± 3.0 kCal/mol #The PS1-γ-secretase complex bound to Aβ48 to generate the Aβ45 product has a second minima located at s = 0 to s = 3 and z = 0.06 to z = 0.08, this complex conformation has a MMGBSA free binding energy of -181.1 ± 3.8 kcal/mol §Values shown are in kcal/mol +/-standard error

Figure 4 .
Figure 4. Aβ40 pathway complexes per residue heatmaps of ΔΔGPS Pref.(A) Enzyme ΔΔGPS Pref values.Only residues where the ΔΔGPS Pref magnitude is greater than 1.5 kcal/mol in any complex are shown.(B) Substrate ΔΔGPS Pref values.Cleavage position denoted by arrow.Positive (red) ΔΔGPS Pref values indicate preference for PS1, negative (green) ΔΔGPS Pref values indicate preference for PS2.

4 .
28 kcal/mol) sidechain forming a CH-π bond with Gly38 (ΔΔGPS Pref = 0.54 kcal/mol) in the substrate.The analogous tryptophan in PS2, Trp171 (ΔGbind PS2 = -4.29 kcal/mol), is positioned such that it interacts with Val39 (ΔGbind PS2 = -4.61kcal/mol), forming a CH-π bond.The residue adjacent to Trp165/Trp171 is not conserved between PS1 and PS2, being an alanine in PS1 (Ala164) and a glycine in PS2 (Gly170); this likely influences the presentation of the tryptophan sidechain, as the backbone in this region is expected to exhibit greater flexibility in PS2-vs PS1-γ-secretase.APP-CTF bound in the position to generate the Aβ48 product, representing the initial substrate in the Aβ42 pathway, has a slight preference for binding to PS2-γ-secretase over PS1-γsecretase.The primary residue contributing to this preference is the PS2 residue Asp263(ΔΔGPS Pref = -9.28kcal/mol), which is the catalytic aspartate in the N-terminal fragment of PS2 (Fig5).Asp366, protonated in our models, is presented in PS2 in a manner that allows for multiple hydrogen bonds between its side chain with substrate residue Thr48 and the backbone amine of Leu49 (Fig 5 & S3A), unlike the equivalent residue in PS1, where these interactions are absent.Additionally, the N-terminal residues of APP-CTF (Val18 (ΔΔGPS Pref = -1.75kcal/mol) and Phe20 (ΔΔGPS Pref = -0.24kcal/mol) interact with a cluster of nicastrin residues (Asp249 (ΔΔGPS Pref = -3.80kcal/mol), Glu650 (ΔΔGPS Pref = -2.98 kcal/mol), and Trp653 (ΔΔGPS Pref = -3.18kcal/mol)) and residues in the TM1 to TM2 loop region of PS2 (Asn116 (ΔΔGPS Pref = -1.94kcal/mol)), forming hydrogen bonds and hydrophobic interactions that are not evident in the PS1-γ-secretase complex (Fig 5 & S3B).Notably, Asn116 in PS2 is not conserved in PS1, the analogous residue being Asp110; the negatively charged aspartate residue in PS1 forms a salt-bridge with the positively charged Arg108 in PS1, preventing interactions with the substrate.

Figure 5 .
Figure 5. Aβ42 pathway complexes per residue heatmaps of ΔΔGPS Pref.(A) Enzyme ΔΔG PS Pref values.Only residues where the ΔΔGPS Pref magnitude is greater than 1.5 kcal/mol in any complex are shown.(B) Substrate ΔΔGPS Pref values.Cleavage position denoted by arrow.Positive (red) ΔΔGPS Pref values indicate preference for PS1, negative (green) ΔΔGPS Pref values indicate preference for PS2.

Figure 7 .
Figure 7. Notch1 complexes per residue heatmaps of ΔΔGPS Pref.(A) (A) Enzyme ΔΔGPS Pref values.Only residues where the ΔΔGPS Pref magnitude is greater than 1.5 kcal/mol in any complex are shown.(B) Substrate ΔΔGPS Pref values.Cleavage position denoted by arrow.Positive (red) ΔΔGPS Pref values indicate preference for PS1, negative (green) ΔΔGPS Pref values indicate preference for PS2.
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is

Table 1 . MM-GB/SA free binding energy of γ-secretase -APP bound complexes
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is 3.97 kcal/mol) forms a hydrogen bond with substrate residue Val24 (ΔΔGPS Pref = -2.19kcal/mol) in (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is

Table 2 . MM-GB/SA free binding energy of γ-secretase -Notch1 bound complexes
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is