Enzyme–substrate interface targeting by imidazole‐based γ‐secretase modulators activates γ‐secretase and stabilizes its interaction with APP

Abstract

Alzheimer's disease (AD) pathogenesis has been linked to the accumulation of longer, aggregation‐prone amyloid β (Aβ) peptides in the brain. Γ‐secretases generate Aβ peptides from the amyloid precursor protein (APP). Γ‐secretase modulators (GSMs) promote the generation of shorter, less‐amyloidogenic Aβs and have therapeutic potential. However, poorly defined drug–target interactions and mechanisms of action have hampered their therapeutic development. Here, we investigate the interactions between the imidazole‐based GSM and its target γ‐secretase—APP using experimental and in silico approaches. We map the GSM binding site to the enzyme–substrate interface, define a drug‐binding mode that is consistent with functional and structural data, and provide molecular insights into the underlying mechanisms of action. In this respect, our analyses show that occupancy of a γ‐secretase (sub)pocket, mediating binding of the modulator's imidazole moiety, is sufficient to trigger allosteric rearrangements in γ‐secretase as well as stabilize enzyme–substrate interactions. Together, these findings may facilitate the rational design of new modulators of γ‐secretase with improved pharmacological properties.

Insights into drug‐target interactions and mode of action of an imidazole‐based γ‐secretase modulator may facilitate rational design of next‐generation compounds for treatment of Alzheimer's disease.

Introduction

The γ‐secretase intramembrane proteases (GSECs) play multifaceted roles in physiology and disease. Most studied are their essential roles in embryonic development and cellular differentiation, exerted through cleavage of Notch receptors, and in the pathogenesis of Alzheimer's disease (AD), where the aberrant cleavage of the amyloid precursor protein (APP) by mutant GSECs causes early‐onset familial AD (FAD) (Chávez‐Gutiérrez & Szaruga, 2020). However, the GSEC‐mediated processing of a large number of type‐I membrane proteins implicates their activities in diverse signaling cascades and a broad spectrum of biological functions (Jurisch‐Yaksi et al, 2013; Güner et al, 2020).

GSECs are multimeric, noncovalent complexes with presenilin (PSEN, catalytic subunit), nicastrin (NCSTN), presenilin enhancer 2 (PEN‐2), and anterior pharynx defective 1 (APH1A or B) as essential subunits (Escamilla‐Ayala et al, 2020). They cleave the transmembrane domains (TMDs) of their substrates in a sequential process that releases—from the membrane—both the intracellular domain and the ectodomain (N‐terminal fragment) of the substrate. The soluble intracellular domain may translocate into the nucleus to regulate transcriptional cascades; this is best exemplified by the Notch signaling pathway (Kopan, 2012). The initial GSEC cleavage (endopeptidase activity) generates also a de novo substrate which, already bound to GSEC, is subjected to successive carboxypeptidase‐like γ‐cleavages (carboxypeptidase‐like activity). The efficiency of these γ‐cleavages determines the length of the N‐terminal products, which are released into the luminal or extracellular environment (Fig 1A). In the case of APP, de novo generated longer amyloid β (Aβ) peptides (48 or 49 amino acids (aa) in length) are sequentially shortened by the carboxypeptidase activity to generate Aβ peptides varying from 37 to 42 aa in length (Qi‐Takahara et al, 2005; Yagishita et al, 2008; Takami et al, 2009). Our previous studies have shown that the stability of the GSEC‐APP/Aβ (enzyme–substrate, E‐S) complexes progressively decreases with each γ‐cut, shifting the equilibrium to favor E‐S dissociation and Aβ release (Szaruga et al, 2017; Fig 1A).

Figure 1. Deline C‐APP substrate interface.

1. Schematic representation of the processing of APP_C99 by GSECs.
2. Superimposition of the GSEC‐GSM co‐structure (PDB: 7D8X; PSEN in sienna brown, GSM E2012 in purple) on the GSEC‐APP_C83 co‐structure (PDB: 6IYC; PSEN in pink, APP TMD in orange) shows a large overlap between the binding modes of the modulator and the substrate. GSM E2012 structure is shown.
3. GSEC co‐structure with APP_C83 (PDB: 6IYC) as determined by Zhou et al (2019) indicating the position of the identified cavity (blue square). A magnification of the PSEN1‐NCSTN‐APP interface and putative imidazole‐based GSM binding pocket is shown in panel (E). Color scheme: PSEN1 in tan, PEN‐2 in sienna brown, APH1A in gold, NCSTN in green and APP_C83 substrate fragment in orange.
4. Potentially druggable (sub‐)pockets (colored in magenta and violet) located at the interface between PSEN1 and APP found in PDB structure 6IYC. Please also see Appendix Fig S4. These (sub) pockets were identified using Fpocket (Le Guilloux et al, 2009). The GSM III was docked to the merged pocket (Appendix Fig S4A).
5. Residues selected for investigations, the color code matches the surfaces and residues shown in this and other panels in this Figure. *Positions referring to the human NCSTN sequence previously evaluated in (Petit et al, 2019). Note that they are the equivalent to positions 241–242 in mouse NCSTN. # Known FAD pathogenic mutation (p.L113‐I114InsT or ‘Ins113T’) used as a reference in the study.

Mutations in PSEN1 causing FAD consistently impair GSEC processivity (Chávez‐Gutiérrez et al, 2012; Fernandez et al, 2014) and thereby shift Aβ production toward the generation of longer and amyloidogenic Aβ peptides (Aβ42, Aβ43 and potentially longer peptides; Szaruga et al, 2015; Veugelen et al, 2016). The accumulation of longer Aβ42 and Aβ43 peptides in the brain—caused by altered production in FAD or impaired clearance in the late‐onset sporadic AD (SAD)—is proposed to initiate pathogenic molecular and cellular cascades decades before the onset of symptoms (Selkoe & Hardy, 2016).

The enhanced production of longer Aβ peptides, linked to pathogenic variants, arises from the mutation‐driven destabilization of GSEC‐Aβ_n interactions (Szaruga et al, 2017). Notably, the more the mutation undermines E‐S interactions, the earlier the disease manifests. In fact, the molecular composition of FAD‐linked Aβ profiles not only informs about pathogenicity, but also enables the estimation of age at disease onset (Petit et al, 2022). These findings strongly support the notion that the stabilization of GSEC‐APP/Aβ_n interactions, and consequent shift in Aβ production toward shorter and soluble Aβ peptides, may serve as a therapeutic strategy to tackle early, Aβ‐driven pathogenic cascades in FAD, and broadly in the most common, late‐onset SAD.

GSEC modulators (GSMs) have been shown to enhance GSEC processivity while preserving the overall endo‐proteolysis (Weggen et al, 2001; Golde et al, 2013). These small compounds increase the efficiency of the sequential Aβ processing (along one or both GSEC product lines, Fig 1A) by stabilizing GSEC‐Aβ_n complexes and preventing their “premature” dissociation (Okochi et al, 2013; Szaruga et al, 2017). As a result, GSMs enhance the production of short peptides (Aβ37 and Aβ38) from the longer ones (Aβ43 and Aβ42), respectively. These short Aβs have been shown to attenuate the toxicity of longer Aβ42 and Aβ43 peptides in vivo in the Drosophila eye (Moore et al, 2018). These observations posit that short Aβ peptides are nontoxic and potentially protective. In support of this, clinical studies have recently linked increased CSF Aβ38 levels to slower cognitive decline in AD patients (Cullen et al, 2022). Furthermore, the CSF Aβ37/42 ratio has been shown to differentiate between individuals with low and high Aβ burden, as well as separate amyloid‐positive subjects who are cognitively normal from those who develop AD (Liu et al, 2022). These findings support the therapeutic potential of GSMs, at least in AD prevention settings, as well as indicate that the mechanism of action of GSMs (increment in the short‐to‐long Aβ peptide ratio) is inherently safe. Despite the promising profiles of GSMs, the associated drug‐binding interactions and underlying mechanisms of action are poorly defined. In this regard, we note that a cryo‐electron microscopy (cryo‐EM) analysis of GSEC in complex with the imidazole‐based GSM E2012 has been recently reported (PDB:7D8X) (Yang et al, 2021, Data ref: Yang et al, 2021). However, the superposition of the GSEC‐GSM E2012 and GSEC‐APP co‐structures (PDB: 7D8X and 6IYC; Data ref: Yang et al, 2021 and Data ref: Zhou et al, 2019, respectively) shows the binding modes of the substrate and GSM largely overlapping (Fig 1B). The structural model is thus not compatible with the simultaneous binding of the substrate (APP/Aβ_n) and GSM to GSEC.

Here, we applied experimental and computational approaches to define the binding mode of a potent imidazole‐based GSM (GSM III) to the biologically relevant GSEC‐APP (E‐S) complex. Our analyses locate the GSM binding site and define a mode of binding that implicates APP in drug–target interactions. Furthermore, our studies provide molecular insights into the underlying modulatory mechanism. Specifically, our analyses of “pocket‐filling” mutations show that the occupancy of a (sub)pocket in PSEN1—where the imidazole moiety of the GSMs binds—is sufficient to: (i) trigger allosteric rearrangements in the protease and (ii) mediate the stabilization of GSEC‐Aβ_n interactions. These findings demonstrate a preponderant role for the imidazole moiety of GSMs in a dual mechanism of action: enzyme activation and E‐S stabilization.

Our analyses also indicate that GSM‐driven GSEC activation arises from the transition‐state facilitation, while the stabilization of GSEC‐Aβ_n complexes may be connected to increased hydrophobic contact surface between GSEC and the substrate. The mechanistic insights into the modes of action of imidazole‐based GSMs, provided by this work, may guide the development of next‐generation GSMs.

Results

Identification of the binding pocket for imidazole‐based GSMs within the GSEC‐APP_C99 complex

To identify the binding pocket of potent imidazole‐based GSMs within the GSEC‐APP complex, we thoroughly analyzed the GSEC‐substrate co‐structures (PDB: 6IYC and 6IDF; Yang et al, 2019, Data ref: Yang et al, 2019; Zhou et al, 2019, Data ref: Zhou et al, 2019) for potential cavities that might be consistent with the following key observations: the involvement of position I242 in NCSTN ectodomain and the first extracellular loop in PSEN1 in GSM binding (Takeo et al, 2014; Petit et al, 2019); and the drastic impairment mediated by the FAD‐linked PSEN1 intron 4 variant (p.L113‐I114InsT, Ins113T) of the response of GSECs to potent imidazole‐based modulators (Szaruga et al, 2015). Visual inspection of the E‐S structures pointed to a relatively narrow cavity present at the extracellular side of the complex (Fig 1C). A pocket identification routine (Le Guilloux et al, 2009) delineated two adjacent cavities, separated by the PSEN1‐Y240 residue (Fig 1C,D) and, on this basis, we speculated that GSMs could bind, via an induced fit or conformational selection mechanism, to the merged cavities (binding pocket). This putative binding pocket, primarily shaped by residues in PSEN1 (Loop 1: aa 105–106 & 109–114, TMD 3: aa 177 & 180–181 and TMD5: aa 232, 235–237 & 239–240) and the ectodomain of NCSTN (aa 242–243), was then investigated in a series of functional and theoretical analyses (Fig 1E). We subjected 18 PSEN1 residues, directly contributing to the formation of the putative binding pocket, to Ala/Phe scanning mutagenesis. In addition, we included the pathogenic intron 4 (Ins113T) variant as a reference in our analysis, given its deleterious effects in the modulation of GSEC by the potent (investigated here) GSMs (Szaruga et al, 2015). As mentioned above, mutagenesis data for residues 242–243 of NCSTN were already available and supported the involvement of this region in the interaction with GSMs (Petit et al, 2019).

We generated stable wild‐type and mutant PSEN1 cell lines by rescuing PSEN1 expression in Psen1 ^−/− /Psen2 ^−/− mouse embryonic fibroblasts (MEFs). The presence of mature glycosylated NCSTN and PSEN1 (N‐ and C‐terminal) fragments confirmed the efficient reconstitution of active protease complexes in all mutant cell lines (Appendix Fig S1). We then determined the mutation‐driven effects on the processing of APP by GSECs. To this end, we analyzed the composition of Aβ profiles secreted by the wild‐type and mutant cell lines transiently overexpressing human APP_C99 by ELISA (Fig 2A). Of note, Aβ profile composition is determined by the degree of GSEC processivity as well as the position of the first endopeptidase‐cut (product line preference). Factors that enhance GSEC processivity promote the conversion of the Aβ40 and Aβ42 peptides into Aβ37 and Aβ38, respectively (highlighted arrows in Fig 1A). The effects of factors modulating GSEC processivity are best reflected by changes in the Aβ(37 + 38)/(40 + 42) ratio. We thus used this ratio as a surrogate measure of enzyme processivity and looked at the relationship between (Aβ37 + Aβ40) and (Aβ38 + Aβ42) to infer mutation‐driven alterations in the protease product line preference.

Figure 2. Ala/Phe mutagenic scanning demonstrates the involvement of the investigated pocket in the regulation of GSEC processivity.

1. The graphs depict the Aβ profiles generated by WT or mutant PSEN1 MEF cell lines transduced with adenovirus encoding APP_C99. ELISA quantified secreted Aβ37, Aβ38, Aβ40, and Aβ42 peptide levels are shown as % of total Aβ (sum of the four peptides). The blue, orange, green, and purple dotted lines indicate the levels of Aβ37, Aβ38, Aβ40, and Aβ42 peptides in the WT Aβ profile, respectively. Data are presented as mean ± SD, N ≥ 3 independent experiments. Mutagenic primers used for the generation of mutant cell lines are listed in Appendix Table S1.
2. GSEC processivity levels for the cleavage of APP_C99 determined in WT and mutant PSEN1/GSEC cell lines. The mutant Aβ(37 + 38)/(40 + 42) ratios are shown as % of the WT cell line. Data presented as mean ± SD, N ≥ 3 independent experiments. One‐way ANOVA followed by Dunnett's post hoc test with comparison to WT was used to determine statistical significance (P < 0.05); **P < 0.01, ***P < 0.001, ****P < 0.0001. (F(DFn, DFd): F(34, 124) = 87.56.

As expected, the FAD‐linked intron 4 (Ins113T) mutation substantially enhanced Aβ42 production (Fig 2A) and impaired GSEC processivity (Fig 2B). Similarly, the F105A, G111F, L113A, F177A, I180F, L232A/F, L235A/F, V236A, and K239A/F substitutions significantly lowered enzyme processivity; with some of them displaying processivity levels that were similar to the pathogenic intron 4 variant (Fig 2B). In contrast, the Y106A, D110A/F, L113F, I114A, V236F, and Y240A mutations significantly promoted processivity, as indicated by the short‐to‐long Aβ peptide ratio. The total production of Aβ peptides—a proxy for global GSEC activity levels—indicated that only two mutations in PSEN1 (F237A and Y240A) lowered the GSEC endopeptidase activity, while four of them (Y106A, Y106F, F177A, and V236F) increased it (Appendix Fig S2A). Of note, the Y106A and V236F substitutions had consistent activation effects on both endopeptidase (global) and carboxypeptidase‐like activities. The results demonstrate that substitutions in this pocket mainly affect the processive function of GSEC and that there is a disconnection between the mechanisms securing the endopeptidase (global) and processive function of GSEC.

Moreover, the (Aβ37 + Aβ40)/(Aβ38 + Aβ42) relationship revealed a marked effect of the F237A substitution on the product line preference of GSEC and significant alterations for several other mutations (Appendix Fig S2B). Altogether, the results demonstrated that the investigated cavity substantially shapes Aβ profiles by regulating both GSEC processivity and product line preference.

We next determined the effects of the tested mutations on the response of GSEC to the potent imidazole‐based GSM III (1 μM). GSM III not only increased GSEC processivity (Appendix Fig S3A) but also activated the global (endopeptidase) processing of APP_C99 at concentrations higher than 3 μM, as indicated by the total Aβ product levels (Appendix Fig S3B). Intriguingly, GSM III shifted the product line preference of GSEC to promote the production of Aβ42 and Aβ38 (Appendix Fig S3C).

As expected, GSM III treatment drastically enhanced GSEC processivity in the WT cell line, where Aβ37 and Aβ38 peptides became the most abundant products (Fig 3A). The GSM‐driven enrichments in the short (Aβ37 and Aβ38) vs. long (Aβ40 and Aβ42) peptides varied among the mutant cell lines. However, the Aβ(37 + 38)/(40 + 42) ratio revealed significant mutation‐driven impairments for all but one tested mutant lines. The K239A substitution facilitated the response of GSEC toward the modulator while the K239F mutation impaired it (Fig 3B). These effects are likely due to the creation of a more spacious and hydrophobic pocket in the PSEN1 K239A mutant (Fig 1D shows the position of K239 in the pocket).

Figure 3. Investigated pocket plays a key role in GSM III‐driven modulation.

1. Aβ profiles generated by WT or mutant PSEN1 MEFs lines treated with 1 μM GSM III. The chemical structure of the bicyclic heterocycle GSM III is depicted. Data are presented as mean ± SD, N ≥ 3 independent experiments. The blue, orange, green, and purple dotted lines indicate the levels of Aβ37, Aβ38, Aβ40, and Aβ42 peptides in the WT Aβ profile, respectively.
2. Fold change in GSEC processivity of APP_C99 estimated by the Aβ(37 + 38)/(40 + 42) ratio and presented as % of the WT response. The data are presented as mean ± SD, N ≥ 3. The symbol # marks mutations that fully abolished the GSEC response toward GSM III, as indicated by data presented in Appendix Fig S3. As reference, a dotted line was drawn to indicate 10%. One‐way ANOVA followed by Dunnett's post hoc test with comparison to WT was used to determine statistical significance (P < 0.05). **P < 0.0001, ****P < 0.0001 (F(DFn, DFd): F(32, 101) = 112.5.

Most importantly, the data highlighted the critical roles of PSEN1‐Y106, G111, L113, F177, V236, and Y240 as substitutions in these positions (nearly) abolished the GSM III‐mediated modulation of GSEC. This was indicated by the drastic reduction in the fold‐change GSM vs. DMSO (≤ 10% of the wild‐type response indicated by a dotted line in Fig 3B). Note that Ala substitutions do not pose any steric hindrance in the pocket and, therefore, the impaired responses of these GSEC mutants to GSM III are most likely caused by the loss of key interactions between the compound and the protease. The fact that Phe substitutions at positions PSEN1‐Y181 and Y240 were deleterious suggests that hydrogen‐bonding interactions at these specific positions may play a role in the recognition of GSM III. In this regard, the in silico data suggest that GSM III can establish a hydrogen bond with Y240 (see below). Finally and in line with previous findings (Szaruga et al, 2015), the pathogenic intron 4 mutation was strongly deleterious.

Collectively, the data demonstrate the involvement of the investigated pocket, located at the E‐S interface, in the regulation of enzyme proteolytic processivity of APP_C99 and in the modulation of GSEC by potent imidazole‐based GSMs.

In silico binding mode for a potent imidazole‐based GSM implicates the GSEC–APP interface in drug–target interactions

In parallel, we applied an in silico approach to investigate the drug–target interactions underpinning the modulation of GSEC activity by GSM III (Velter et al, 2014). We used the GSEC‐APP_C83 co‐structure (PDB: 6IYC; Zhou et al, 2019, Data ref: Zhou et al, 2019) to build a wild‐type GSEC‐APP_C99 model by adding the missing N‐terminal residues of the substrate and modeling the hydrophilic intracellular loop of PSEN1, as these parts were not present in the near‐atomic E‐S complex structure (initial model presented in Appendix Fig S4A). Due to the narrowness of the putative modulator pockets and the steric hindrance posed by the conformation of Y240, flexible side chain/flexible ligand docking was employed to generate an initial binding mode for GSM III. Specifically, we performed a 1 μs long simulation of the GSEC‐APP_C99 and a structural snapshot with a maximum‐sized binding cavity was used as structure “receptor” to dock the compound (Appendix Fig S4B). In order to mimic an induced fit process, the predicted GSM‐GSEC‐APP_C99 complexes were then subjected to flexible structural relaxation by molecular dynamics. After 400 ns of MD simulations, GSM III was re‐docked to the resulting receptor conformation (Appendix Fig S4C). After two docking/MD iterations, this procedure resulted in stable GSM III binding in the presence of the APP_C99 substrate. Remarkably, the resulting GSM III binding mode (Fig 4A–C) revealed direct contacts with most of the residues found experimentally to critically modulate the GSM‐driven effects in Fig 3. Indeed, calculation of the contact frequencies between the GSM (ligand) and GSEC suggested the direct participation of the PSEN1‐F105, Y106, Q112, L113, I114, F177, I180, Y181, L232, V236, and Y240 residues in ligand–protease interactions (Table 1) and was consistent with the involvement of position NCSTN I242 in GSM III recognition (Petit et al, 2019). The most notable drug–target interactions involved hydrogen bonding and π–π sandwich stacking with PSEN1‐Y106 and Y240 positions, respectively (Table 2). Furthermore, our analysis revealed that overall GSM III binding to the GSEC‐APP_C99 complex relies mainly on hydrophobic interactions, with 67% of the total solvent‐accessible surface occupied by apolar residues (PSEN1 + APP_C99) (i.e., 210.44 Ų of 315.76 Ų total area, Table 3). This observation provides a molecular understanding to the highly hydrophobic nature of GSMs (De Strooper & Chávez Gutiérrez, 2015).

Figure 4. In silico and experimental data propose a binding mode for the imidazole‐based GSM III.

1. GSEC‐APP_C99 model with GSM III docked in the identified pocket, which is highlighted with a blue square. Color scheme: PSEN1 in tan, PEN‐2 in sienna brown, APH1A in gold, NCSTN in green, and APP_C83 substrate fragment in orange. Further information on the construction of this model is given in (Appendix Fig S4B–D).
2. Binding mode of GSM III at the GSEC‐APP_C99 interface. Color scheme: PSEN1 in tan, and APP_C83 substrate fragment in orange. Residues involved in drug–target interactions are colored according to their mode of interaction with the drug (H‐bonding: salmon pink, π stacking: green, general hydrophobic: violet).
3. Schematic interaction diagram of the GSM III within the GSEC‐APP_C99 complex: PSEN1, NCSTN, and APP residues are shown;  green and orange circle mark positions in NCT and APP, respectively. The hydrogen bond between PSEN1‐Y106 and GSM III is depicted in green and van der Waals contacts in red. The plot has been generated with LigPlot+ (Wallace et al, 1995).
4. Aβ profiles generated by WT or mutant PSEN1 cell lines overexpressing APP_C99 and treated with increasing concentrations of GSM III. Data are presented as mean ± SD, N = 2 independent experiments.
5. GSEC processivity estimated by the Aβ(37 + 38)/(40 + 42) ratio was determined for the dose–response analysis (presented in panel D) and shown as % of DMSO. Data for N = 2 independent experiments are shown. 

Table 1.

Contact frequencies between GSM III and PSEN1 residues given in % of sampled frames. Only residues with a contact frequency of at least 30% are shown. All residues with at least one atom within 3 Å of the ligand were counted as contacts.

PSEN1 residue	Interaction type	Contact freq. (%)
Y240	π–π stacking	99.3
I114	apolar	98.8
V236	apolar	97.8
F177	apolar	97.5
I180	apolar	92.0
Y106	H‐bond	89.8
Q112	apolar	88.8
L232	apolar	85.5
F105	apolar	81.0
Y181	apolar	68.5
L113	apolar	65.7

Table 2.

Frequency of key interaction between the GSM and PSEN1 in % of sampled frames.

Interaction	GSM III (%)
Total polar Y106 interactions	77.5
Hydrogen bond with Y106	67.3
Water bridge to Y106	10.2
Y240 π–π stacking	94.7

Table 3.

Pocket descriptors calculated from the GSM III binding mode as obtained by Fpocket (Le Guilloux et al, 2009). Legend: SASA = solvent‐accessible surface area.

Parameter	GSM pocket
Total SASA (Å2)	315.757
Polar SASA (Å2)	105.313
Apolar SASA (Å2)	210.444
Volume (Å3)	959.320

Dose–response analysis of residues delineating the pocket demonstrates their critical involvement in GSM recognition and binding

We then asked whether the detrimental effects of mutations at critical positions in PSEN1 (Y106 (hydrogen bonding), Y240 (π–π stacking) as well as L113‐114 (Intron 4), F177, L232, and V236 (hydrophobic interactions)) could be rescued by higher GSM III concentrations (Fig 4B shows drug–target interactions color‐coded by type). Dose–response analyses with concentrations ranging from 0.03 μM up to 10 μM demonstrated that all mutations drastically shifted the effective drug concentrations toward higher values (Fig 4D), relative to the wild‐type cell line. Moreover, analysis of the Aβ (37 + 38)/(40 + 42) ratio demonstrated that higher concentrations of the GSM III (only) partially promoted the Aβ40➔Aβ37 and Aβ42➔Aβ38 cleavages by mutant GSECs, relative to the wild type (Fig 4E), with the exception of the L232A substitution. In the latter case, a full response was observed at the highest GSM III concentration. In this regard, our model suggests that—while the L232F substitution is well tolerated—the L232A replacement induces the narrowing (partial collapse) of the imidazole‐binding pocket, supporting the notion that imidazole‐based GSMs bind to a flexible and dynamic pocket in GSEC. The deleterious effects of Ala mutations suggest that the investigated side chains are either establishing important interactions with the drug or play a crucial role in the architecture of the binding site.

Hydrogen‐bonding geometry between PSEN1‐Y106 and the GSM imidazole moiety is critical for drug recognition

We then used the in silico structural model to investigate drug–target interactions at the molecular and mechanistic levels. Data presented here, in conjunction with previous functional and structural studies (Takeo et al, 2014; Liu et al, 2021; Yang et al, 2021), support the high significance of this hydrogen bonding in the recognition of imidazole‐based GSMs. In addition, our in silico analyses suggested that the geometry of this interaction plays a critical role in drug–target interactions. In silico modeled mutant systems showed that the exchange of PSEN1‐Y106 for Gln or Trp (residues offering a hydrogen‐bonding donor group albeit at different distances to the protein backbone) would have strong detrimental effects on drug–target interactions. Generation of the PSEN1‐Y106Q and Y106W cell lines and functional assessment, regarding Aβ generation, showed normal reconstitution of mutant GSEC complexes (Appendix Fig S1) and processivity levels that were slightly reduced or enhanced by the PSEN1‐Y106Q and the PSEN1‐Y106W mutations, respectively, relative to the WT control (Fig 5A,B). Nevertheless, these mutations fully or drastically impaired the response of mutant GSEC responses to GSM III (Fig 5A: DMSO vs. GSM III and Appendix Fig S5A), indicating that both the distance and the orientation of the key “PSEN1 Y106—GSM” hydrogen bond play roles in imidazole‐based GSM recognition.

Figure 5. Experimental and in silico analyses implicate GSEC and APP in drug–target interactions.

• A–C
  (A) Aβ profiles generated by WT and indicated mutant MEF lines in the presence of vehicle (0.1% DMSO), 1 μM GSM III or (C) 1 μM GSM II. Data are presented as mean ± SD, N ≥ 3 independent experiments. (B) WT and mutant GSEC processivity estimated by the Aβ(37 + 38)/(40 + 42) ratio and presented as % of WT. Data presented as mean ± SD, N ≥ 3 independent experiments. One‐way ANOVA followed by Dunnett's post hoc test with comparison to WT was used to determine statistical significance (P < 0.05). *P < 0.01, **P < 0.0001, ***P < 0.001, ****P < 0.0001 (F(DFn, DFd): F(3, 45) = 56.9.
• D, E
  Analysis of mutant APP substrates with regard to their effects on the response of GSEC to the imidazole‐based GSM III. Aβ profiles secreted by HEK293T cells transiently expressing WT or mutant APP_C99 substrates in the presence of vehicle (DMSO) or 1 μM GSM III were analyzed by ELISA or mass spectrometry (MS). Mutagenic primers used for the generation of mutant APP_C99 substrates are listed in Appendix Table S1. (D) ELISA‐based Aβ profiles are presented as % of total Aβ (37 + 38 + 40 + 42) (mean ± SD, N ≥ 3); (E) APP mutation‐driven effects on GSEC processivity according to the Aβ(37 + 38)/(40 + 42) ratio are presented as % of the WT condition (mean ± SD). One‐way ANOVA followed by Dunnett's post hoc test with comparison to WT was used to determine statistical significance; ****P < 0.0001. F(DFn, DFd): F(4, 17) = 71.68.
• F
  MS‐based profiles show the proportion of individual Aβ peptides on the total immunoprecipitated pool. Data are shown as mean of three independent experiments. See also (Appendix Fig S6B).

Stacking (π–π) and hydrophobic interactions at the GSEC‐APP interface contribute to imidazole‐based GSM recognition and binding

The in silico model also revealed a pronounced π–π sandwich stacking interaction between the aromatic PSEN1‐Y240 residue and the central bicyclic system in GSM III (colored in blue in Fig 3A). This interaction was present in 94.7% of the simulation snapshots (Table 2). Furthermore, modeling data suggested that the relatively larger hydrophobic area created by the bicyclic moiety of GSM III, relative to arylimidazole GSMs, contributes to drug recognition and potency by allowing a greater degree of flexibility in the establishment of π–π stacking interactions with GSEC. Accordingly, we reasoned that GSM III binding would be less affected by the size and the conformation of the PSEN1‐240 side chain than the binding of GSMs that do not present a bicyclic feature, such as the arylimidazole GSM II (depicted in Fig 5C). Note that GSM III is more potent in its modulatory effects than GSM II (Appendix Fig S5B). To challenge our hypothesis, we introduced a Trp residue at position 240 in PSEN1, generated the respective mutant cell line (Appendix Fig S1) and tested the responses of WT and mutant PSEN1‐Y240W GSECs to GSM II and GSM III (Fig 5C). Aβ profiles generated by the mutant PSEN1‐Y240W cell line showed a mild positive mutation‐driven effect on GSEC processivity (Fig 5B). In response to GSM III treatment, the PSEN1‐Y240W cell line still generated significantly higher levels of the short Aβ37 and Aβ38 products (Fig 5C, DMSO vs. GSM III), though the Aβ(37 + 38)/(40 + 42) ratio revealed that the mutant GSEC only partially responded to the modulator (Appendix Fig S5A). GSM II treatment significantly changed Aβ production by the PSEN1‐WT cell line but did not alter the Aβ profiles generated by the PSEN1‐Y240W mutant (Fig 5C, DMSO vs. GSM II). The lack of response toward GSM II is clearly shown by the overlapping 95% CI (Y240W, DMSO vs. GSM II in Appendix Fig S5C). The results indicate that the relatively larger hydrophobic area established by GSM III contributes to more stable interactions with GSEC, relative to imidazole‐based GSMs that do not present this feature (arylimidazole type). Furthermore, the data suggest that the recognition of imidazole‐based GSMs requires both proper hydrogen bonding with PSEN1‐Y106 and π–π stacking interactions with PSEN1‐Y240.

In addition, the in silico binding mode depicted hydrophobic contacts between the central bicyclic and terminal heterocyclic rings of the modulator (depicted in blue and yellow in Fig 3A, respectively) and both PSEN1 (loop 1 positions Q112‐I114) and APP (residues I32, V36, and V39) (Fig 4B,C). We speculated that the hydrophobic interactions between GSM III and the first extracellular loop of PSEN1 could provide a structural basis for the strong impairing effects of the pathogenic intron 4 variant (Ins113T) (Fig 2A). To further investigate this, we evaluated the effects of the analogous Ala and Phe intron 4‐like mutants (Ins113A and Ins113F, respectively) in our cellular model. These cell lines showed normal reconstitution of mutant GSECs (Appendix Fig S1) and their functional assessment revealed that the Ins113F insertion, but not the Ins113A mutant, rescued the impairing effects of the pathogenic Intron 4 (Ins113T) mutant on GSEC processivity (Fig 2B). The processivity data (Fig 2B) therefore attribute the pathogenic disturbances in Aβ profiles ‐exerted by the intron 4 variant—to the hydrophilic nature and length of the inserted side chain, rather than to the insertion per se. Moreover, our analysis showed that the impairing effects of the intron 4 mutation on GSEC processivity are—at least in part—mediated by the allosteric modulation of GSEC (see below). In addition, the intron 4‐like Ala/Phe variants showed slightly better responses to GSM III treatment than the pathogenic intron 4 (ins113T), as indicated by Aβ profiles shown in Fig 3A. The analysis of the Aβ(37 + 38)/(40 + 42) ratio however revealed that all insertions significantly impaired the response to the modulator.

To test the contribution of APP_C99‐ I32, V36, and M35 positions to GSM III binding (Fig 4B,C), we generated mutant APP constructs presenting Ala substitutions at the implicated positions and expressed them transiently in the HEK cell line. We then determined the secreted Aβ profiles in the presence of 0.3 μM GSM III or DMSO vehicle. As a control, we used the APP_C99 V24A mutation, which is situated in the vicinity but is not predicted to interact with GSM III. The control V24A mutation had no impact on GSEC processivity (DMSO) nor on the response to GSM III; while Ala substitutions at other tested positions had variable effects on enzyme processivity, but consistently all exerted significant impairments on the GSEC responses to GSM III (Fig 5D, DMSO vs. GSM III). Evidence of this is the relatively high levels of Aβ40 and/or Aβ42 generated in the presence of GSM III. Indeed, analysis of the Aβ(37 + 38)/(40 + 42) ratio demonstrated mutation‐driven impairments for all mutants, except the control V24A substitution (Fig 5E; Appendix Fig S6A). The more abundant presence of Aβ40 and/or Aβ42 peptides in the I32A, V36A, and M35A mutant Aβ profiles upon GSM treatment, relative to the wild‐type condition, was confirmed by mass spectrometry analysis (Fig 5F; Appendix Fig S6B). The M35A mutation had little impact on GSEC processivity (DMSO). Nevertheless, it nearly abolished Aβ37 generation in the presence of GSM III while producing substantial amounts of Aβ38 and Aβ40 peptides, implying a selective mutation‐driven impairing effect on the GSM III‐mediated modulation of the Aβ40➔Aβ37 cleavage. In conclusion, in silico and experimental data implicate the protease–substrate interface in the recognition and the binding of imidazole‐based GSMs.

Distinct drug–target interactions mediate GSEC modulation by imidazole‐ or acidic‐based GSMs

Previous findings suggested that acidic and imidazole‐based GSMs binding sites in GSEC may partially overlap (Takeo et al, 2014). We thus investigated whether residues critically involved in drug–target interactions with imidazole‐based GSMs contribute to the recognition of the acidic modulator GSM I. Cell lines bearing substitutions in PSEN1 that exerted drastic impacts on GSM III‐driven modulation (Y106A, L113A, V236W, Y240W, and Intron 4) were treated with 1 μM acidic GSM I (Appendix Fig S7). Importantly, the acidic GSM I selectively promotes the conversion of Aβ42 into Aβ38. As expected, analysis of Aβ profiles revealed high Aβ38 and decreased Aβ42 peptide levels for the GSM I‐treated wild‐type cell line (Appendix Fig S7A). Similarly, GSM I treatment shifted Aβ production in all tested mutants to increase Aβ38 levels at the expense of the Aβ42 peptide (Appendix Fig S7B). These findings support the specific involvement of the investigated pocket and drug–target interactions in the modulation of GSEC by imidazole‐based GSMs and imply that different mechanisms underlie the actions of imidazole‐ and acidic‐type GSMs.

Pocket‐filling mutations mimicking key GSM‐GSEC interactions allosterically activate PSEN1

Having demonstrated the specific involvement of the investigated pocket in imidazole‐based GSM binding, we sought to investigate the mechanism(s) underlying the GSM‐driven allosteric modulation of GSEC. We asked what direct interactions between the modulator and the GSEC‐APP complex could possibly mediate the allosteric effect. Given the critical role of the imidazole ring of the GSMs in the modulation of GSEC activity (Oehlrich et al, 2011), we tested whether the introduction of a bulky side chain in the sub‐pocket occupied by the imidazole moiety would not only sterically block (imidazole‐based) GSM binding, but could also allosterically activate PSEN1/GSEC. Inspection of the in silico model and the cryo‐EM structure featuring E2012 binding to GSEC pointed at the PSEN1‐V236W as a candidate mutation for this analysis (Fig 6A). The findings that the bulky PSEN1‐V236F substitution had a marked activating effect on GSEC processivity (~ 2.3 fold, Fig 2B) and fully blocked the GSM III‐driven modulation were encouraging.

Figure 6. Pocket‐filling PSEN1‐V236W mutation mimics key GSM‐GSEC interactions and allosterically activates GSEC.

1. In silico data show that the W236 side chain mimics the interactions of the GSM methylimidazole ring with the protease. The position of GSM III in the absence of substrate (identical with the position of GSM E2012 in the cryo‐EM structure PDB: 7D8X, Data ref: Yang et al, 2021) is shown in purple, while the binding mode of GSM III suggested by our work is illustrated in gray. The side chain of W236 (indicated in orange) occupies the same space as the imidazole moieties in both GSM binding modes.
2. Aβ profiles generated by WT or mutant PSEN1 cell lines transduced with APP_C99. Dotted lines indicate the levels of Aβ peptides (%) in WT profiles. Data presented as mean ± SD, N ≥ 3 independent experiments.
3. Pocket‐filling PSEN1‐V236W and Y106W‐V236W substitutions lead to increased Aβ(37 + 38)/(40 + 42) ratio, indicating the activation of the sequential GSEC‐mediated cleavage of APP_C99. Data presented as mean ± SD, N ≥ 3 independent experiments.
4. Aβ profile analysis in dose–response experiments for GSM III revealed no modulation of the PSEN1‐V236W mutant cell line. Data are presented as mean ± SD, N = 2 independent experiments. For comparison purposes, WT and V236W mutant Aβ profiles are presented in panel 6B. We note that dose–response analyses for GSM II showed similar results (Appendix Fig S8B).
5. GSEC activity assay using purified GSEC and an APP‐based fluorogenic peptide as substrate (Farmery et al, 2003). Upper panel: In the noncleaved state, the Dpn (2,4‐dinitrophenyl) quencher group suppresses fluorescent emission from the Nma (N‐methyl‐o‐aminobenzoic acid) fluorophore through Förster resonance energy transfer (FRET). Upon GSEC cleavage, the Nma group emits at λ = 430 nm a fluorescent signal upon excitation at λ = 355 nm. Lower panel: Specific activities for the purified WT or mutant GSECs (bearing the indicated substitutions in PSEN1) treated with DMSO (vehicle), 10 μM GSMIII or GSM II. Data are normalized to the WT GSEC activity in DMSO (vehicle). Data are shown as mean ± SD, N ≥ 3 independent experiments. Significance is shown for meaningful comparisons; two‐way ANOVA with comparison to WT was used to determine statistical significance; **P < 0.01, ***P < 0.001, ****P < 0.0001, ns: not significant.
6. Transition‐state analog inhibitor L‐685,458 (Inhibitor X) treatment of WT, V236W or Y240W mutant PSEN1 cell lines. Inhibitory profiles and derived IC50 values (table) indicate that the PSEN1‐V236W significantly increases the affinity for the inhibitor. Dotted line indicates 50% inhibition. Data are presented as mean ± SD, N = 3 independent experiments.

Indeed, MD simulations of the PSEN1‐V236W and V236F GSEC mutant suggested that the W236 and F236 side chains would occupy the same space as the critical methylimidazole ring of the modulator (Appendix Fig S8A). To test whether the occupancy of the imidazole‐binding sub‐pocket would allosterically activate PSEN1/GSEC, we mutated PSEN1‐V236 alone or in combination with PSEN1‐Y106 to Trp, generated the single and double mutant PSEN1 cell lines, and confirmed the normal reconstitution of GSEC complexes in these lines (Appendix Fig S1). Analysis of secreted Aβ profiles revealed remarkable increments in Aβ38 peptide levels for both single and double mutations (Fig 6B). In fact, the single and double PSEN1 mutants substantially enhanced GSEC processivity and caused ~ fourfold and ~ fivefold increments in the Aβ(37 + 38)/(40 + 42) ratio, respectively, and relative to the WT enzyme (Fig 6C). In addition, both cases, the GSEC product line preference was shifted to favor Aβ42 and Aβ38 generation (Appendix Fig S8B). Remarkably, dose–response analyses with GSM III or GSM II at concentrations up to 10 μM showed no change in the Aβ profiles generated by the mutant PSEN1‐V236W cell line (Fig 6D; Appendix Fig S8C, respectively). The data thus demonstrate that the introduction of bulkier side chain in the imidazole‐binding (sub)pocket activates the processive function of GSEC and sterically blocks the binding of imidazole‐based GSMs.

We then investigated whether the mutation‐mediated increment in GSEC processivity arises from allosteric effects. To investigate this point, we used the cleavage of an active‐site‐directed, fluorogenic substrate as a reporter of PSEN1 catalytic status. Of note, this short APP‐based substrate presents neither an ectodomain nor a TMD (Fig 6E; Farmery et al, 2003). The cleavage of this short peptide by GSEC leads to loss of Förster resonance energy transfer (FRET) between the fluorophore (Nma) and a quencher (Dnp) groups and fluorescent emission at 430 nM (upon excitation at 355 nM) (Takeo et al, 2014). Kinetic analyses using purified, wild‐type GSEC showed that both imidazole‐based GSM III and GSM II allosterically activate GSEC proteolysis in a dose‐dependent manner (Appendix Fig S9A,B). Furthermore, the degree of activation was consistent with the GSM potencies (Fig 6E; Appendix Fig S5B). We then performed similar in vitro GSEC activity assays using purified (wild‐type or PSEN‐V236W mutant) proteases and the short fluorogenic substrate (Farmery et al, 2003). In addition, we analyzed the pathogenic intron 4 mutation, which impairs both GSEC processivity and response to imidazole‐based GSMs (Figs 2B and 3B).

Fluorescence analysis demonstrated marked increment and decrement in the specific activities of the purified GSECs bearing the PSEN1‐V236W or intron 4 mutations, respectively (Fig 6E). These well‐controlled assays demonstrated that mutations in the investigated pocket allosterically modulate GSEC and that the occupancy of the GSM's imidazole‐pocket (by the PSEN1‐V236W mutation) is sufficient to trigger activating effects in PSEN1/GSEC, which mimic those elicited by imidazole‐based GSMs. Importantly, in silico modeling of the short peptide substrate within GSEC showed the W236 side chain placed at a minimum distance of 12.5 Å from the substrate (average distance: 15.6 Å in 280 ns simulation) (Appendix Fig S9C). This is much larger than a contact distance (< 5 Å), ruling out the possibility that the mutation‐driven activation is mediated by a direct W236‐subtrate interaction. Finally, as shown for the APP substrate in Fig 6D and Appendix Fig S8C, GSM III and GSM II failed to activate the processing of the fluorogenic peptide by the PSEN1‐V236W mutant GSEC (Fig 6E). This finding further strengthened the idea that the “pocket‐filling” PSEN1‐V236W mutation sterically blocks imidazole‐GSM binding.

In conclusion, the data show that occupancy of a (sub)pocket in PSEN1—where the GSM's imidazole moiety binds—is sufficient to trigger the allosteric activation of GSEC.

Transition‐state facilitation mediates the GSM‐driven allosteric activation of GSEC

E‐S complexes can either undergo catalysis, via a transition‐state intermediate (E‐S*), or decompose back to free enzyme and substrate. Our previous findings indicate that a less efficient stabilization of the transition state in GSEC complexes bearing FAD‐linked PSEN1 variants promotes the nonproductive dissociation of E‐S complexes, thereby favoring the generation of partially digested Aβ peptides (Szaruga et al, 2017). Given that pathogenic mutations and GSMs trigger opposite effects on GSEC‐mediated proteolysis (Okochi et al, 2013), we investigated whether the activating allosteric effects elicited by the PSEN1‐V236W mutation relate to the stabilization of the transition state. We used the transition‐state analog (TSA) inhibitor L‐685,458 (Inhibitor X) to determine whether a higher affinity for the transition state drives catalysis in the PSEN1‐V236W mutant protease. As a control, we analyzed the PSEN1‐Y240W substitution (Fig 5C), which also introduces a bulky side chain in the pocket, and, according to our in silico data, does not mimic drug–target interactions. While wild‐type and Y240W mutant complexes showed nearly identical inhibitory profiles in cell‐based assays, the V236W GSEC demonstrated higher affinity for the TSA inhibitor, as indicated by the shift in the IC₅₀ value (Fig 6F). Therefore, the data indicate that the occupancy of a narrow (sub)pocket in PSEN1 (colored in magenta in Fig 1C,D) by the GSM's imidazole moiety or the PSEN1‐V236W substitution is sufficient to trigger rearrangements that facilitate proteolysis by stabilizing the transition state.

Occupancy of the imidazole (sub) pocket in GSEC stabilizes E‐S complexes

We then investigated whether the pocket‐filling PSEN1‐V236W mutation also mimicked the stabilizing effects of GSM III (Szaruga et al, 2017). To evaluate this possibility, we analyzed GSEC‐APP/Aβ interactions in thermoactivity assays using purified APP_C99‐3xFLAG (substrate) and purified wild‐type or mutant (PSEN1‐V236W) GSEC complexes (Fig 7A). Similar thermoactivity assays have proven to provide valuable insights into the stability of GSEC‐APP/Aβ complexes (Szaruga et al, 2017). We quantified Aβ37, Aβ38, Aβ40, and Aβ42 generation by wild‐type or mutant GSEC complexes over a temperature gradient (37–57.3°C) (Fig 7B–E, respectively). Consistent with previous findings (Szaruga et al, 2017), GSEC‐Aβ_n thermostability decayed with each carboxypeptidase cleavage (Tm values for wild type: 57.2°C for Aβ42, 50.5°C for Aβ40, 47.2°C for Aβ38, and 44.1°C for Aβ37). Notably, a marked shift in the product profiles generated by the mutant GSEC (PSEN1‐V236W) toward higher temperatures demonstrated a mutation‐driven stabilization of E‐S interactions (Tm values for V236W: ND for Aβ42, 57.7°C for Aβ40, 57.3°C for Aβ38 and 47.1°C for Aβ37) (See Appendix Table S2 for 95% CIs).

Figure 7. Pocket‐filling PSEN1‐V236W mutation stabilizes GSEC‐Aβ (E‐S) interactions.

• A
  E‐S complex stability decreases during the sequential GSEC‐mediated cleavage mechanism, progressively shifting the equilibrium toward product release. In vitro GSEC activity assays using purified WT or mutant (PSEN1‐V236W) GSEC complexes and purified APP_C99 substrate were incubated at the indicated temperatures. Thermal destabilization promotes the release of partially digested Aβ peptides, while (mutation‐driven) stabilization increases processivity and production of shorter peptides.
• B–E
  De novo generation of Aβ37, Aβ38, Aβ40, and Aβ42 peptides over a temperature gradient was quantified by ELISA and plotted as a percentage of the respective peptide levels at 37°C. The data are shown as mean ± SD, N ≥ 3 independent experiments.

We then reasoned that a hydrophobic contact between the APP substrate and the GSM's imidazole moiety (or the  pocket filling mutation) could contribute to the E‐S stabilizing effects. We thus investigated in silico the impact of GSM binding on E‐S complex stabilities. To ensure that the encountered differences mainly result from the presence or the absence of the GSM and are not due to sampling of different conformational ensembles, we performed simulations on two structurally closely coupled pairs: GSEC‐APP_C99 ± GSM III. In silico calculated E‐S complex stabilities revealed that removal of the modulator destabilizes the E‐S complex by 13.2 kcal/mol (c.f. Appendix Table S3). Moreover, inspection of the trajectories indicated that direct interactions between a methyl moiety on the GSM and the substrate likely contribute to the increased E‐S stability. Notably, root‐mean‐square fluctuations (RMSFs) for the substrate residues in these simulations pointed at increased fluctuations of the N‐terminal part of the substrate (APP_C99 positions 34–36) in the absence of GSM III (Appendix Fig S10). These findings suggest that GSM binding has a profound effect on substrate dynamics and support the contribution of GSM‐APP interactions to the stabilizing effects exerted by the compound. Collectively, these observations propose a model where imidazole‐based GSMs promote the processive function of GSEC by a dual mechanism: allosteric activation of PSEN (by occupying the imidazole sub‐pocket) and stabilization of E‐S complexes (via pronounced interactions with the substrate), with the latter increasing the life span of the naturally less stable complexes containing shorter Aβ peptides.

Removal of the substrate leads to an (apo)GSEC‐GSM binding mode that is in perfect agreement with experimental structural data

The functional data presented in Fig 6E and Appendix Fig S9A,B show that imidazole‐based GSMs allosterically modulate the processing of small peptides that have neither TMD nor ectodomain. We thus assessed in silico whether the determined GSM III binding mode would remain stable in the absence of the APP_C99 substrate. We performed three (800 ns) simulations of the GSM III‐GSEC‐APP_C99 system (2.4 μs sampling in total), but prior to starting the simulations the entire substrate was removed from the model. Two out of the three simulations resulted in stable GSM III binding modes, highly similar to the observed in the experimental GSEC‐GSM E2012 structure (7D8X). The third simulation sampled an alternative GSM III orientation, caused by a breakdown of the hydrogen bond with the PSEN1‐Y106 during the simulation.

Alignment of PSEN1 TMDs 2, 3, 5, 6, and 7 from our simulation trajectories to the 7D8X structure showed the in silico predicted GSM binding mode with apo‐GSEC, featuring a hydrogen bond between the GSM and PSEN1‐Y106, in perfect agreement with the experimental cryo‐EM data (Fig 8A).

Figure 8. Superimposition of the experimental and in silico GSEC‐GSM binding modes (E2012 vs. GSM III, respectively) in the absence of the APP substrate show congruent GSM binding modes.

• A–C
  (A, B) Superposition of the GSEC‐GSM‐InhX structure (purple) (PDB: 7D8X, Data ref: Yang et al, 2021) and the in silico GSM III binding mode after the removal of the substrate and performance of 800 ns of MD (gray). (B, C) GSM binding modes with GSEC (PDB: 7D8X, Data ref: Yang et al, 2021) or the (in silico) GSEC‐APP_C99 co‐structure. PSEN1 is indicated as brown surface, while the substrate helix is colored orange.

Moreover, MD simulations suggested that GSM III is able to form a hydrogen bond with PSEN1‐Y240, in addition to the existing one with PSEN1‐Y106. This extra polar contact is enabled by the acceptor (oxygen) on the central ring structure of GSM III and is not occurring with other modulators presenting a methoxy group at this position, like GSM II and E2012. The formation of this additional hydrogen bond may explain the better performance of GSM III, relative to GSM II, in the activation of the GSEC‐mediated cleavage of the fluorogenic peptide lacking a TM helix (Fig 6E). Whether this extra hydrogen‐bonding interaction facilitates the formation of the activated GSEC‐GSM complex prior to the binding of APP requires further investigations. These results overall support the specificity and the robustness of our findings and support the derived structural models featuring GSMs interactions with the biologically relevant E‐S target. In addition, the data suggest that transitions between different GSM binding modes, triggered by binding or dissociation of substrate, may be possible.

Discussion

The development of safe and effective, disease‐modifying treatments for AD continues to be an urgent matter today, despite the recent FDA accelerated approval of Aducanumab (Aduhelm) (Alexander et al, 2021).

Pathogenic variants causing early‐onset, autosomal dominant AD shift Aβ production toward the generation of longer and aggregation‐prone peptides (Chávez‐Gutiérrez et al, 2012; Fernandez et al, 2014; Kretner et al, 2016; Veugelen et al, 2016; Liu et al, 2022), and the mutation‐driven increments in these longer Aβ forms (≥ 42 aa) show a linear correlation with clinical onset (Petit et al, 2022). This observation strongly supports the therapeutic value of strategies aimed at lowering the levels of longer Aβ peptides.

GSMs shift Aβ production to favor the generation of short and soluble Aβ peptides without affecting the global GSEC activity; the latter controls essential signaling events. Their mechanism of action thus mitigates toxicity phenotypes associated with general GSEC inhibitors (Luo & Li, 2022). GSMs have become a promising class of small compounds that could be used as a primary prevention in early‐onset FAD and Down's syndrome, or a secondary prevention in amyloid‐positive, asymptomatic individuals. However, a poor understanding of their binding modes and mechanisms of action, as well as the high complexity of the target, have limited their development.

Most importantly, GSMs have suffered from poor drug‐like properties and toxicity issues related to high lipophilicity (De Strooper & Chávez Gutiérrez, 2015). Nevertheless, costly drug discovery efforts, involving high‐throughput screening (HTS) of small compound libraries in human cell lines (following Aβ42 production as a readout) and iterative optimization of hit compounds, have led to the discovery of potent GSM scaffolds with improved pharmacodynamics (Mekala et al, 2020). Regardless of recent progress, the cost‐effective development of potent and CNS permeable GSMs displaying good safety profiles and suitable pharmacokinetic properties (that enable repeat‐dosing) remains desirable. In this respect, structure‐based drug discovery represents an alternative that may overcome shortcomings and risks in the generation and/or design of improved GSMs. This approach however requires prior knowledge about the drug‐binding pocket and drug–target interactions.

Here, we characterized the binding pocket of a potent imidazole‐based GSM within the GSEC‐APP_C99 substrate complex, defined the associated binding mode, and investigated the underlying mechanisms of action by integrating in silico and experimental approaches. Based on structural and functional data, we hypothesized that imidazole‐based GSMs bind to a cavity located at the extracellular interface formed by GSEC and the APP_C99 substrate (Fig 1D). Site‐directed mutagenesis of the residues lining the putative GSM binding pocket and assessment of Aβ profiles generated by the respective mutant GSECs, in the absence or the presence of the imidazole‐based GSM III, revealed mutation‐induced impairments that were consistent with the implication of the pocket in drug–target interactions. Importantly, mutations drastically impairing GSEC responses to GSM treatment, without affecting GSEC proteolysis, supported the hypothesis that the identified cavity indeed represents the binding pocket of imidazole‐based GSMs. As examples, the PSEN1‐Y106A, L113F, V236F, and Y240A substitutions enhanced processivity while severely impairing the GSM‐mediated modulation of GSEC (Figs 1E vs. 2A,B). These findings dissociate GSEC processivity from the protease's ability to respond to GSM treatment.

In silico docking of GSM III into the putative binding pocket required an iterative flexible approach that mimicked an “induced fit” mechanism. This observation highlighted the structural plasticity of the investigated pocket and suggested that such a mechanism (induced fit) may operate in the formation of the GSM‐GSEC‐APP_C99 complex. Molecular dynamics generated an in silico binding mode for GSM III, which encouragingly showed contacts between the compound and all residues highlighted as critical by our functional studies. Furthermore, its comparison with the experimental GSEC‐GSM E2012 co‐structure showed the interactions between the imidazole moieties of the GSMs (GSM III vs. GSM E2012) and GSEC in excellent agreement (Fig 8A). In both models, the imidazole ring is inserted into a narrow hydrophobic pocket in PSEN1 (defined by PSEN1‐F105, Y106, F177, I180, L232, V236, K239, and Y240), where it forms a hydrogen bond with PSEN1‐Y106 and establishes a few other hydrophobic contacts with GSEC.

However, the interactions established between the GSM scaffolds and the target (GSEC‐APP_C99 or GSEC, respectively) substantially differed in other regards (Fig 8B,C). In the presence of APP, the in silico binding mode featured a “shifted” GSM III structure, relative to the GSEC‐GSM E2012 co‐structure (Fig 8C), with its central (bicyclic) and heterocyclic tail (fluorinated) rings engaged in π–π stacking (PSEN1‐Y240) and establishing hydrophobic interactions with both PSEN1 and APP. This binding mode is consistent with the elegant analysis by Takeo et al (2014) demonstrating the involvement of PSEN1 first loop and Y240 in the response of GSEC toward the imidazole‐based tetracyclic GSM ST1120 and provides structural basis for the impairing effects of the FAD pathogenic intron 4 variant (Szaruga et al, 2015). Notably, the in silico and functional data indicate that the relatively larger hydrophobic area (π–π stacking) established by the bicyclic central ring of GSM III, relative to arylimidazole GSMs, contributes to GSM recognition and possibly to potency too. In addition, experimental data revealed that GSM III alters the position of the endopeptidase cleavage to favor the Aβ42‐ and Aβ38‐producing pathway (Fig S3C). We speculate that this effect arises from direct interactions between the compound and the APP ectodomain. Our work indeed implicated APP in drug–target interactions; however, further analyses are required to understand the mechanistic basis of the product line preference of GSEC.

To define what interactions underlie the modulatory actions of GSMs, we analyzed the effects of mutations mimicking drug–target interactions. Given the central role of the imidazole moiety, we searched in silico for mutations that could fill this specific (sub)pocket in PSEN1. MD simulations of the PSEN1‐V236W mutation suggested that the bulkier PSEN1‐W236 substitution would function as a “hydrophobic plug” filling the imidazole‐binding sub‐pocket and thus (partially) mimicking GSM binding (Fig 6A). Remarkably, functional analysis demonstrated that the single PSEN1‐V236W substitution substantially activated GSEC function (Fig 6B,C) and, in agreement with the mutation pocket‐filling nature, abolished the GSEC response toward imidazole‐based GSMs (Fig 6D; Appendix Fig S8C). Kinetic analysis using an active‐site‐directed probe showed that the mutation‐driven enhanced processivity arose from a long‐range allosteric effect, and it was similar to the GSM‐mediated allosteric modulation of the wild‐type GSEC (Fig 6E). The elucidation of the precise structural strategies underlying the allosteric activation requires further investigations. Nevertheless, the activating role of the PSEN1‐V236W mutation demonstrates that the occupancy of a small sub‐pocket in the protease, either by the W236 side chain or the methylimidazole ring in GSM III, is sufficient to trigger allosteric changes that result in the activation of the GSEC processive function.

To get a better understanding of the mechanistic bases of the activation, we interrogated the mutant GSEC bearing the PSEN1‐V236W with the transition‐state analog inhibitor X (Fig 6F). An increased affinity of the mutant GSEC (PSEN1‐V236W) for this active‐site inhibitor is consistent with a degree of cooperativity between the studied GSM binding pocket and the active site of GSEC, as reported earlier (Pozdnyakov et al, 2013; Yang et al, 2021). In addition, the data imply that the stabilization of the transition state underlies the proteolytic activation.

Our previous analyses have shown that imidazole‐based GSMs modulate GSEC by stabilizing E‐S interactions (Szaruga et al, 2017). We thus tested whether the PSEN1‐V236W mutation exerted similar stabilizing effects. Indeed, thermoactivity analyses of the purified GSEC (PSEN1‐V236W) mutant demonstrated marked mutation‐driven stabilization of GSEC‐Aβ interactions (Fig 7B–E). In this regard, in silico studies showed that GSM binding increases the hydrophobic contact surface between GSEC (+GSM) and the substrate, suggesting that the increased hydrophobic area contributes to the increased stability of GSEC‐Aβ_n complexes. Whether the stabilization arises from the elicited allosteric changes in the protease and/or involves the hydrophobic GSM‐APP interface warrants further investigations.

Do GSMs bind to a preestablished E‐S complex? Or do they interact with the apo‐GSEC, prior to the incorporation of the substrate? Previous investigations, using first‐generation GSMs, have shown that substrate docking to GSEC enables modulator binding to an allosteric site (Uemura et al, 2010). In the case of imidazole‐based GSMs, the fact that the formation of a stable GSEC‐GSM E2012 complex (for structural analyses) was facilitated in the presence of the TSA inhibitor X, which occupies the active site and partially mimics E‐S interactions, supports a strong interdependence between GSM and substrate binding to GSEC (Yang et al, 2021). Indeed, comparison of the GSEC‐APP and GSEC‐GSM E2012 co‐structures (PDB codes: 7D8X and 6IYC; Data ref: Yang et al, 2021 and Data ref: Zhou et al, 2019, respectively) revealed a substantial overlap between the binding modes of the modulator and the APP substrate (Fig 1B), implying that the formation of the biologically relevant complex (GSEC‐substrate‐GSM) is accompanied by substantial structural rearrangements in E‐S as well as drug–target interactions. Our in silico analyses showing the coherent conversion of the GSM‐GSEC‐APP into the GSM‐GSEC complex suggest the possibility that an equilibrium between these complexes exists and is regulated by the relative concentrations of the GSM and GSEC substrates. This strengthens the high pathophysiological relevance of protein dynamics in the regulation of GSEC function (Chávez‐Gutiérrez & Szaruga, 2020; Mehra & Kepp, 2021).

In conclusion, our combined experimental and computational studies delineate the binding pocket of imidazole‐based GSMs within the biologically relevant GSEC‐APP complex and provide the first molecular view of the underlying drug–target interactions. Our analyses demonstrate that the occupancy of a narrowed sub‐pocket in PSEN1—where the imidazole moiety of the GSMs binds—is sufficient to: (i) trigger allosteric rearrangements in the protease that result in its activation (via stabilization of the transition state) and (ii) stabilize E‐S interactions. They propose a model where imidazole‐based GSMs exert a dual (stabilizing and allosteric) effect on the GSEC‐APP complex that results in the increased active‐site occupancy of labile Aβ substrates and a faster turn over. We note that dose–response analysis of GSM III revealed that it potently activates the processive function of GSEC and, at concentrations higher than 3 μM, also increases the global (endopeptidase) processing of APP_C99. Whether the overall activation of GSEC by potent GSMs rescues pathophysiology in vivo has to be determined. However, evidence suggests that increased APP_C99 levels may contribute to neurodegeneration in AD via impairments in the endosomal‐lysosomal system (Lauritzen et al, 2021) and, therefore, treatments reducing both toxic Aβ peptides as well as APP_C99 levels may exert therapeutic effects. The observed activation of the global endopeptidase GSEC activity by GSM III may thus prevent or mitigate endosomal abnormalities and linked degenerative phenotypes associated with the accumulation of APP_C99 fragments within the membrane. However, this potential therapeutic path requires further investigations.

Finally, we speculate that the structural and mechanistic insights into the modes of action of imidazole‐based GSMs presented here may facilitate the rational design of next‐generation GSMs, displaying good safety profiles, suitable pharmacokinetic properties, and potentially including selective APP‐targeting GSMs. Compound selectivity toward APP, relative to other GSEC substrates, and the possibility that GSMs promote the processing of longer Aβ peptides as well as help to clear toxic APP‐derived C‐terminal fragments (APP_CTF) could be an added value in the mechanism of action of next‐generation GSMs.

Materials and Methods

Antibodies and compounds

To assess proper reconstitution of GSEC expression, the following antibodies were used: anti‐human PSEN1‐CTF (MAB5232) from Merck, anti‐human PSEN1‐NTF (MAB1563) from Millipore, horseradish peroxidase (HRP)‐conjugated anti‐mouse (#1721011) and anti‐rabbit IgG (#1721019) from Bio‐Rad, and HRP‐conjugated anti‐rat IgG (#P0450) from Agilent. NCSTN and PEN‐2 were detected using in‐house developed antibodies (9C3 for NCSTN and B126.2 for PEN‐2). ELISA capture antibodies (JRD/Aβ37/3 for Aβ37, JRF AB038 for Aβ38, JRF/cAb40/28 for Aβ40, and JRF/cAb42/26 for Aβ42), Aβ N‐terminal detection antibody (JRF/AbN/25) and imidazole‐based (GSM II/III) modulators (synthesis described in Velter et al, 2014) were obtained through collaboration with Janssen Pharmaceutica NV (Beerse, Belgium). Complete protease inhibitor (PI) tablets and fluorogenic peptide were purchased from Sigma‐Aldrich, and GSEC inhibitor X (#565771) from Calbiochem.

Generation of PSEN1 wild‐type and mutant MEF cell lines

Stable MEF cell lines expressing wild‐type (WT) or mutant PSEN1 GSEC complex were generated through retroviral transduction of Psen1 ^−/− /Psen2 ^−/− mouse embryonic fibroblasts (MEFs) (Herreman et al, 2000), as described before (Chávez‐Gutiérrez et al, 2012). Cell lines were cultured in Dulbecco's Modified Eagle's Medium (DMEM)/F‐12 (Life Technologies) supplemented with 10% fetal bovine serum (FBS) (Sigma‐Aldrich), and stably expressing clones were selected by addition of 5 μg/ml puromycin (Sigma‐Aldrich) to the culture medium 1 day after transduction. Reconstitution of active GSEC complex in the resultant cell lines was evaluated by SDS–PAGE/western blotting for the different GSEC subunits (as indicated in Appendix Fig S1). Briefly, cell pellets were collected for each cell line and total membranes were prepared and solubilized in 1% CHAPSO, 28 mM PIPES pH 7.4, 210 mM NaCl, 280 mM sucrose, 1.5 mM EGTA pH 8 and 1x PI. CHAPSO‐solubilized membranes were used for western blotting analyses.

MEF cell‐based activity assays with human APP_C99 in different conditions

To evaluate Aβ production in the generated PSEN1 WT/variant MEF cell lines, APP_C99 substrate was transiently expressed in the cell lines using a recombinant adenoviral expression system as described before (Chávez‐Gutiérrez et al, 2008; Petit et al, 2019) with the following modifications. Freshly trypsinized cells were plated at the density of 12,500 cells/well into 96‐well plates and transduced approximately 7 h later with Ad5/CMV‐APP_C99 adenovirus. 16 h later, the culture medium was replaced with low‐serum medium (DMEM/F‐12 medium containing 0.2% FBS) supplemented with DMSO (0.1% for regular cell assays and 0.3% in dose–response assays) or GSMs reconstituted in DMSO at 1 μM final concentration (regular screening assay) or a concentration range (for dose–response assays). The conditioned medium was collected for Aβ analysis following a 24‐h incubation period at 37°C.

Quantitative analysis of secreted Aβ peptides in conditioned medium for Aβ profile determination

Aβ37, Aβ38, Aβ40, and Aβ42 peptides levels in conditioned medium were determined by multiplex Meso Scale Discovery (MSD) ELISA. Briefly, 96‐well MSD ELISA plates precoated with anti‐Aβ37, Aβ38, Aβ40, and Aβ42 antibodies were blocked in phosphate buffer saline (PBS) supplemented with 0.1% casein. Freshly harvested conditioned medium or Aβ standards (synthetic human Aβ1‐37, Aβ1‐38, Aβ1‐40, and Aβ1‐42 peptides at known concentrations) were mixed (1:1 v/v) with SULFO‐TAG JRF/AbN/25 detection antibody (diluted in blocking buffer) and loaded on the plate (50 μl sample/well). Following overnight incubation at 4°C, ELISA plates were rinsed five times with washing buffer (PBS supplemented with 0.05% Tween 20) and signals were immediately read in MSD Gold read buffer (MSD) (150 μl/well) on a Sector Imager 6000 (MSD).

Expression and purification of GSEC complexes and substrates

Human WT or D385A (DA) GSEC (PSEN1, APH1B, PEN‐2, NCSTN‐PreScission protease cleaving site‐GFP) complexes was expressed in Hi5 insect cells (BTI‐TN‐5B1‐4) using a baculovirus expression system (FlashBAC Gold, Oxford Expression Technologies). Briefly, cells were infected with baculovirus expressing all different subunits of the GSEC complex at multiplicity of infection (MOI) 15. 72 hour postinfection, cells were harvested and homogenized with an Emulsiflex‐C3 (Avestin) in 25 mM PIPES, 300 mM NaCl, 5% glycerol and 1X Protease Inhibitors (PI, Roche Complete tablets). Membrane fractions obtained after ultracentrifugation (100,000 g for 1 h) were washed twice in 25 mM PIPES, 1 M NaCl, 10% glycerol, and solubilized overnight at 4°C (25 mM PIPES, 300 mM NaCl, 5% glycerol, 2% CHAPSO, PI). The next day, GFP‐tagged recombinant enzymes were purified by immunoaffinity chromatography using anti‐GFP beads (Acx et al, 2014). Untagged GSEC complexes were eluted by protease digestion with GST‐tagged PreScission protease in 25 mM PIPES, 150 mM NaCl, 0.5% CHAPSO, 5% glycerol, 1 mM DTT and 1 mM EDTA. Removal of the PreScission protease was achieved by immunoaffinity pulldown using Glutathione Sepharose 4B (GST) beads (GE Healthcare). Human APP_C99‐3XFLAG‐PreScission protease cleaving site‐GFP was expressed in Hi5 insect cells using the same baculovirus expression system and purified following a similar protocol. GSEC and substrate purities were assessed by SDS–PAGE and Coomassie staining (InstantBlue Protein Stain, Expedeon).

In vitro fluorogenic GSEC activity assay

In vitro activity assays with purified GSEC enzyme were conducted as follows: fluorogenic peptide was mixed (20 μM final concentration) with purified wild‐type or mutant GSECs in assay buffer (24 mM PIPES pH 7.4, 150 mM NaCl, 0.25% CHAPSO, 5 mM EDTA, 2.5% DMSO and 0.1% phosphatidylcholine (PC)) in the presence or the absence of 10 μM GSM (or Inh X) in a 384‐well plate. The reactions were incubated for 2 h at 37°C, and during this time fluorescence emission at 430 nM (upon excitation at 355 nM) was read on the CLARIOstar Plus (BMG Labtech). Densitometric western blot analysis of GSEC samples was used to set reaction samples to similar enzyme concentrations, and WT/mutant GSEC‐specific activities were calculated from slopes (Fluorescence/time).

In vitro GSEC activity and thermoactivity assays

Proteolytic reactions were performed using purified ~10 nM WT or mutant GSEC complexes (GSEC PSEN1/APH1A) and purified recombinant FLAG‐tagged substrates in 0.25% CHAPSO, 2.5% DMSO, 0.1% Phosphatidylcholine, 150 mM NaCl and 25 mM PIPES. Protease and substrate were incubated at 37°C or over a temperature gradient for 20 min. Enzyme mixes (containing all reaction components except the substrate) and substrate were pre‐incubated separately at the indicated temperatures for 5 min, mixed at time zero, and incubated for 20 min. Substrate concentration in the assays was 1.5 μM.

Immunoprecipitation of Aβ peptides from conditioned media

HEK293 cells were cultured in 10 cm² dishes and transfected with wild‐type or mutant APP_C99 expressing constructs. At day one posttransfection, cells were treated with GSM III (0.3 μM) or vehicle (DMSO) and 24–30‐h posttransfection conditioned media were collected. Aβ peptides were immunoprecipitated using the 4G8 antibody (4 μg/10 ml of conditioned media). Conditioned media and antibody were incubated overnight at 4°C, and then 40 μl Protein G agarose beads were added and incubation continued for 3 h. Beads were washed in PBS/0.05% Tween 20 pH 7.4 two times. The last wash in PBS/0.01% Tween 20 pH 7.4 was performed to reduce the detergent concentration. Dry beads were frozen at −20°C and subsequently subjected to MS analysis.

MALDI‐MS sample preparation and analysis Aβ peptides

Beads were resuspended in 15 μL of SA matrix solution (38 mg/mL in water/ACN/TFA 20/80/2.5 (v/v/v)), and 30 nM Aβ1‐28 was added (internal standard). The sample was vortexed for 1 min and centrifuged for 5 min at ~1,000 g. The supernatant (matrix‐analyte mix) was collected, and 1 μl (9 technical replicates) was applied on a MALDI AnchorChip Target (Bruker Daltonics, Billerica, MA, USA) using dried droplet preparation and air‐dried. All mass spectra were acquired on a rapifleX MALDI‐TOF/TOF mass spectrometer (Bruker Daltonics) equipped with a 10 kHz Smartbeam™ laser using the AutoXecute function of the FlexControl 4.2.

Molecular dynamics simulations

All MD simulations conducted in this study have been performed with the AMBER18 package (Case et al, 2018), utilizing the GPU‐accelerated version of pmemd (Salomon‐Ferrer et al, 2013). For amino acids, the AMBER14SB force field (Maier et al, 2015) was used, while lipids and water have been described by the LIPID14 (Dickson et al, 2014) and TIP3P (Jorgensen et al, 1983) models, respectively. Parameters for GSM III have been obtained with the Antechamber (Wang et al, 2006) module of AMBER18, using the GAFF2 force field (Wang et al, 2004). Prior to obtaining RESP (Fox & Kollman, 1998) charges (Hartree‐Fock at 6‐31G* level) for the ligand, the structures were optimized to a gas‐phase energy minimum at the B3LYP/TZVP level. The QM calculations have been performed with GAUSSIAN09 (Frisch et al, 2009). All simulations have been performed at 303.15 K using the Langevin thermostat (Goga et al, 2012), with a collision frequency of 1 ps⁻¹. Furthermore, the Berendsen barostat (Berendsen et al, 1984), with a relaxation time of 1 ps, was used to sample NpT ensembles at a pressure of 1 bar. Pairwise nonbonded interactions have been calculated up to a distance of 9 Å, while long‐range Coulombics were described by the particle mesh Ewald (PME) method. In the sampling phase, time steps of 4 fs were used. This was enabled by combining the SHAKE (Andersen, 1983) algorithm with the hydrogen mass repartitioning method (Hopkins et al, 2015). Periodic boundary conditions have been applied, and box dimensions were approx. 109Åx117Åx194Å. Molecular docking was performed by using Autodock Vina 1.1.2 (Trott & Olson, 2010) with standard settings. The starting structure of the apo‐state simulation was based on the PDB structure 6IYC (Zhou et al, 2019, Data ref: Zhou et al, 2019). The structure was prepared for simulation by assigning side‐chain protonation states at pH 6.5, using the PDB2PQR server (Dolinsky et al, 2004). The protonation state of the active site was assigned by hand, defining D257 as protonated and D385 as charged. The bilayer builder of the CHARMM‐GUI server (Jo et al, 2008; Wu et al, 2014; Lee et al, 2020) was used to place the GSEC‐APP_C83 complex in a POPC bilayer, consisting of 304 molecules and solvated in a 0.15 M KCl solution (60,306 water molecules). The missing residues in the HL2 loop of PS1, as well as missing N‐terminal residues of APP_C83, have been restored (automatically by CHARMM‐GUI). The N‐termini of NCSTN and PSEN1 have been terminated by ACE, while the C‐termini of APH1 and APP_C83 were capped by NME residues. The HL2 loop was cleaved between positions M298 and A299. Next, a 7‐step equilibration procedure was applied to the generated starting structure (see Appendix Table S4.1 for details). The same procedure was also applied to all following simulations prior to sampling.

For the investigation of the APP_C99 substrate, instead of APP_C83, the missing N‐terminal residues were added one by one with subsequent minimizations (200 minimization cycles per step). This was performed prior to the initial sampling run of the E‐S complex without modulator.

All other simulations have been performed using the same E‐S model and therefore feature the complete APP_C99 N‐terminus. Data related to mutant GSEC simulations are provided in Appendix Table S4.2. Only in case of the short 60 ns simulations performed for calculating RMSF on Appendix Fig S10, a slightly different simulation protocol was used: In order to compare dynamics and substrate affinities in the presence and the absence of GSM III sampling of closely coupled conformational pairs was necessary. This was ensured by starting all simulations of both, the GSM III‐GSEC‐APP_C99 as well as the GSEC‐APP_C99 complex from the same initial conformation. In each case, 10 independent 60 ns long simulation trajectories were produced of which the last 30 ns were subsequently used for data analysis. The amino acid sequence of the fluorogenic peptide is shown in Fig 6E and, instead of the Nma moiety, an additional Trp was added to the N‐terminus. The C‐terminal Dnp‐tail was omitted since interactions between the C‐terminus and W236 are impossible.

For the detection of the potential GSM binding pocket within the GSEC‐APP_C99 complex, we used Fpocket (Le Guilloux et al, 2009) which predicts cavities on input PDB structures, employing an algorithm based on Voronoi tessellation. Additional analysis of theoretical data was performed using VMD (Humphrey et al, 1996) and CPPTRAJ (Roe & Cheatham, 2013). Simulation snapshots were rendered using VMD and UCSF ChimeraX (Pettersen et al, 2021). Further information about the MD simulations is provided in the Appendix Supplementary Methods.

Statistical analysis

Statistical analysis of the data was accomplished using the GraphPad Prism 8 software. Student's t‐test or one‐way ANOVA with Dunnett's post hoc test were used to test the significance of the changes as indicated in the figure legends. P‐value < 0.05 was used as a predetermined threshold for statistical significance.

Author contributions

Manuel Hitzenberger: Formal analysis; investigation; writing – original draft; writing – review and editing. Matthias Koch: Formal analysis; investigation; writing – original draft; writing – review and editing. Lucía Chávez‐Gutiérrez: Conceptualization; formal analysis; supervision; funding acquisition; writing – original draft; project administration; writing – review and editing. Martin Zacharias: Formal analysis; supervision; funding acquisition; writing – original draft; writing – review and editing. Sam Lismont: Investigation. Thomas Enzlein: Formal analysis; investigation. Carsten Hopf: Formal analysis; investigation. Dieter Petit: Formal analysis; investigation; writing – original draft. Katarzyna Marta Zoltowska: Formal analysis; investigation.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Appendix

The EMBO Journal (2022) 41: e111084.

Data availability

This study includes no data deposited in external repositories.