Presenilin, γ-secretase and the search for pathogenic triggers of Alzheimer’s disease

Abstract

Cerebral plaques of the amyloid β-peptide (Aβ) are a defining pathology in Alzheimer’s disease (AD). The amyloid hypothesis of AD pathogenesis has dominated the field for over 30 years, ostensibly validated by rare AD-causing mutations in the substrate and enzyme that produce Aβ. The γ-secretase complex carries out intramembrane proteolysis of substrate derived from the amyloid precursor protein (APP). Mutations in APP and presenilins, the catalytic component of γ-secretase, typically increase the ratio of aggregation-prone 42-residue Aβ (Aβ42) over the more soluble 40-residue form (Aβ40). Nevertheless, the inability to clarify how Aβ aggregation leads to neurodegeneration, along with poor progress in developing effective AD therapeutics that target Aβ, raises concern about whether Aβ is the primary disease driver. γ-Secretase carries out processive proteolysis on APP substrate, producing long Aβ peptides that are generally trimmed in tripeptide intervals to shorter secreted peptides. Recent studies on effects of AD-causing mutations on the complicated proteolytic processing of APP substrate by γ-secretase has led to the discovery that these mutations reduce—but do not abolish—processive proteolysis. Reduced proteolysis is apparently due to stabilization of enzyme-substrate complexes, and these stalled substrate-bound γ-secretase complexes can trigger synaptic degeneration even in the absence of Aβ production. Thus, the stalled process rather than the proteolytic products may be a principal initiator of AD pathogenesis. This new amyloid-independent hypothesis suggests that pharmacological agents that rescue stalled γ-secretase enzyme-substrate complexes might be effective therapeutics for AD prevention and/or treatment.

Graphical Abstract

Introduction:

The Amyloid Hypothesis

Alzheimer's disease (AD), the most common form of dementia, afflicting over six million in the US and perhaps over 30 million worldwide, is a progressive neurodegenerative disorder that primarily presents as gradual loss of memory and cognitive function.¹ Pathologically, the disease is characterized by proteinaceous cerebral deposits: extraneuronal plaques composed of the secreted amyloid β-peptide (Aβ) and intraneuronal fibrillary tangles composed of the otherwise microtubule-associated protein tau. Neuroinflammation is also a key feature, involving microglia activated for phagocytosis and dysfunctional astrocytes that inappropriately and/or excessively release inflammatory cytokines.²

In 1991, dominantly inherited missense mutations associated with early-onset familial Alzheimer’s disease (FAD) were found in the gene encoding the amyloid precursor protein (APP).^{3, 4} This seminal discovery led to formulation of the amyloid cascade hypothesis of AD pathogenesis, which posits that aggregated Aβ peptides initiate the disease process.^5-7 These mutations are found in and around the small (38-43 amino acids) Aβ region of the much larger (695-770 amino acids) APP and alter the production and/or properties of Aβ peptides.⁸ Aβ is produced from the type I integral membrane protein APP through sequential proteolysis by β- and γ-secretases (Fig. 1). β-Secretase sheds the APP ectodomain,⁹ leaving a 99-residue membrane-bound C-terminal fragment (C99), which is then cleaved within its transmembrane domain by γ-secretase,¹⁰ producing secreted Aβ peptides and APP intracellular domain (AICD). Most FAD-causing APP mutations are in the transmembrane domain and lead to increased proportions of the aggregation-prone 42-residue form of Aβ (Aβ42) relative to the more soluble 40-residue peptide Aβ40.⁸

Figure 1.

Sequential proteolytic processing of the amyloid precursor protein (APP) by β-secretase and γ-secretase to produce the amyloid-β peptide (Aβ) and the APP intracellular domain (AICD).

In 1995, dominantly inherited missense mutations associated with FAD were found in two related genes encoding multi-pass membrane proteins of unknown function, dubbed presenilin-1 (PSEN1) and presenilin-2 (PSEN2).^11-13 Soon after, reports described increases in the ratio Aβ42/Aβ40 with FAD-mutant presenilins in transfected cell lines, transgenic animals, and patient plasma, suggesting a role for presenilins in modulating γ-secretase cleavage of C99.^14-19 Presenilin itself was also found to be proteolyzed into an N-terminal fragment (NTF) and C-terminal fragment (CTF) that are metabolically stable and remain associated with each other as part of a high molecular weight complex.^20-23 Moreover, formation of presenilin NTFs and CTFs was observed to be tightly regulated by unidentified limiting cellular factors.²⁴ A strong connection between PSEN1 and γ-secretase activity was discovered using PSEN1-deficient embryonic mouse fibroblasts, in which cleavage of transfected APP C99 to Aβ peptides was dramatically reduced.²⁵ Later, PSEN1/PSEN2 double knockout fibroblasts were shown to be completely devoid of γ-secretase activity, indicating a requirement for presenilin.^{26, 27}

Meanwhile, the first substrate-based peptidomimetic inhibitors of γ-secretase were developed, and the inhibitory motifs in these compounds suggested an aspartyl protease mechanism for the enzyme.^{28, 29} This clue, together with the finding that PSEN1 is critical to γ-secretase activity, led to identification of two conserved transmembrane aspartates that are required for both γ-secretase activity and PSEN1 cleavage into NTF and CTF.³⁰ This seminal finding suggested that presenilin is a noncanonical aspartyl protease that carries out intramembrane proteolysis of APP.³¹ Moreover, to become an active protease capable of processing substrates, presenilin must first undergo autoproteolysis into NTF and CTF; these two presenilin subunits each contribute an essential aspartate to the protease active site. These findings were soon confirmed using affinity labeling reagents developed from active-site-directed peptidomimetic inhibitors.^{32, 33} In subsequent years, three essential subunits of γ-secretase were discovered: membrane proteins nicastrin, Aph-1 and Pen-2.^34-38 These subunits assemble with presenilin, whereupon presenilin cleaves itself into NTF and CTF, and the entire assembly becomes the proteolytically active γ-secretase complex.^39-41 Note that biochemistry was essential for establishing presenilin as a protease. Genetic approaches such as knockout of presenilin genes were insufficient, as presenilin did not resemble any known protease at the time. Presenilin was instead suggested to be a regulatory factor for γ-secretase.²⁵

The discovery of presenilin as the catalytic component of γ-secretase was considered “the linchpin in the amyloid hypothesis”:⁴² FAD mutations are found in the substrate (APP) and enzyme (presenilin/γ-secretase) that produce Aβ. Over 200 FAD-associated missense mutations in APP and presenilins (mostly in PSEN1) have been identified to date.⁴³ Moreover, although γ-secretase has nearly 150 known substrates besides APP,⁴⁴ FAD mutations are only found in the APP gene. This compelling genetic and biochemical evidence stimulated intense efforts to develop therapeutics targeting Aβ production, aggregation, or clearance in the brain. However, over the past 25 years, all drug candidates targeting Aβ failed in the clinic—either due to lack of efficacy, serious adverse events, or both—until recently.

Three monoclonal antibodies—aducanumab (Aduhelm), lecanemab (Leqembi), and donanemab (Kisunla)—that demonstrably clear amyloid plaques from the brain were approved by the US Food and Drug Administration in recent years. The 2020 approval of Aduhelm was controversial,⁴⁵ particularly due to conflicting results in Phase 3 clinical trials regarding its efficacy, and this drug was discontinued in January 2024. Clinical evidence for the efficacy of Leqembi and Kisunla was more convincing, showing modest slowing of the rate of cognitive decline compared to placebo over 18 months of treatment.^{46, 47} Questions remain, however, whether this modest effect will make a meaningful difference in the lives of patients and whether the therapeutic effects are worth the high cost of treatment and serious safety concerns.⁴⁸ Total annual costs for treatment, including associated clinical services such as positron emission tomography (PET) scans for amyloid plaques, are estimated at over $80,000. Leqembi and Kisunla also cause risks of microhemorrhages and edema that can be fatal. The status of AD therapeutics begs the question: After over 30 years of the amyloid hypothesis, why are there not more effective treatments?

Reassessing the Amyloid Hypothesis

In addition to the difficulties in developing AD therapeutics by targeting Aβ or its production, biological validation of the amyloid hypothesis has been challenging. Unresolved discrepancies include the lack of understanding of how pathological Aβ triggers neurodegeneration. The nature of the aggregated state of Aβ remains unclear. Soluble oligomeric Aβ has long replaced plaque deposits as the leading candidate for the pathogenic form of Aβ;^{49, 50} however, the identity of the specific oligomeric state is still unknown. Candidates have included dimers,⁵¹ dodecamers⁵² (note: the initial 2006 study identifying dodecamers⁵³ was recently retracted due to data manipulation), and even a heterogeneous “oligomeric soup”.^{54, 55} The problem with all studies seeking to identify the pathogenic oligomeric state(s) is that the nature of the oligomers may be altered by the isolation process itself, leading to artifacts and difficulties in reproducing results in different laboratories. The dynamic nature of Aβ oligomer assembly and disassembly makes isolation of specific states and determining their biological effects challenging.

The identity of the cognate receptor of pathogenic oligomeric Aβ and downstream neurotoxic signaling pathways similarly remain unresolved. Targets for Aβ oligomers include the prion protein,⁵⁶ NMDA receptors,^{57, 58} α7-nicotinic receptors,⁵⁹ and even the lipid bilayer itself.⁶⁰ Downstream signaling and effects on other AD-relevant proteins such as tau are also unclear, despite reports supporting hyperphosphorylation of tau.⁶¹ While hyperphosphorylated tau has long been associated with pathological neurofibrillary tangles found in AD, connecting specific oligomeric forms of Aβ to neurofibrillary tangle formation has been elusive.

More detailed examinations of the effects of FAD mutations on Aβ production reveal further inconsistencies in the amyloid hypothesis. A study of γ-secretase processing of C99 with 138 PSEN1 FAD-mutant protease complexes showed that many of these did not elevate the Aβ42/Aβ40 ratio long touted as critical to pathogenesis, and some FAD mutations led to very little of either Aβ variant.⁶² Moreover, the focus on just these two variants ignores the fact that γ-secretase carries out multiple proteolytic events on the C99 transmembrane domain and produces a range of Aβ peptides. Initial endoproteolytic cleavage produces Aβ48 or Aβ49 and the corresponding APP intracellular domains (AICDs).⁶³ After release of AICD co-products, the long Aβ intermediates are trimmed at their carboxy termini by γ-secretase, generally in tripeptide increments along two pathways: Aβ49→Aβ46→Aβ43→Aβ40 and Aβ48→Aβ45→Aβ42→Aβ38 (Fig. 2).⁶⁴

Figure 2.

Processive intramembrane proteolysis of APP-derived 99-residue C-terminal fragment (C99) by γ-secretase first produces either Aβ48 or Aβ49 and AICD, followed by trimming, generally in tripeptide increments to shorter secreted peptides (Aβ38-43).

Comprehensive and quantitative analyses of the effects of FAD mutations on all the proteolytic steps carried out by γ-secretase on C99 had not been conducted until recently. The effects of 14 FAD-associated missense mutations in the APP transmembrane domain were determined using purified enzyme and substrate.⁶⁵ Liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) was used to quantify each of the small peptide co-products of carboxypeptidase trimming. Analysis of AICD coproducts of endoproteolysis were conducted through a combination of quantitative western blotting and matrix-assisted laser-desorption ionization mass spectrometry (MALDI-MS). Quantitative analysis of the co-products of each cleavage event provides direct information on the effects of FAD mutation on proteolysis, as these are terminal products. In contrast, most Aβ peptides are intermediates that are further trimmed by the protease complex. In purified enzyme assays, each of the 14 FAD mutations in APP substrate resulted in reduction of the first and/or second carboxypeptidase trimming step, thereby elevating levels of long Aβ peptides of 45 residues and longer. Two of the 14 mutations did not increase Aβ42/Aβ40, while eight mutations led to decreased Aβ42 levels.

Most recently, the effects of FAD-mutant enzyme on all proteolytic events carried out by γ-secretase on APP substrate have been investigated. In one study of six FAD-mutant PSEN1-containing protease complexes, all six were deficient in the initial endoproteolytic cleavage event as well as one or more carboxypeptidase trimming steps.⁶⁶ Toward a structural dynamic understanding of how these mutations reduce proteolytic processing of APP substrate by γ-secretase, advanced molecular dynamics simulations of the enzyme-substrate (E-S) complex were conducted, based on an atomic-resolution structure solved by cryoelectron microscopy (cryo-EM).^{67, 68} The cryo-EM structure was of catalytically inactive enzyme, with substrate crosslinked to PSEN1 through cysteine mutagenesis and disulfide bond formation.⁶⁸ The E-S complex was computationally reactivated, returned to the uncrosslinked wild-type state, and embedded in a lipid bilayer surrounded on either side by water. Subsequent simulations led to influx of water into the active site and coordination of one water molecule with the two catalytic aspartates. This water molecule was poised for nucleophilic attack of the scissile amide bond, activated by one of the aspartic acids to make it more electrophilic. This computationally derived activated state of γ-secretase, poised for intramembrane proteolysis of APP substrate, overlapped remarkably well with a new cryo-EM structure of active γ-secretase noncovalently bound to a tight-binding transmembrane substrate-based inhibitor containing a transition state-mimicking moiety (Fig. 3).⁶⁶ This overlap provided important validation of the molecular dynamics methods, justifying using this approach to test effects of FAD mutations. Simulations with each of the six FAD-mutant PSEN1-containing complexes mentioned above reduced their ability to reach the activated state,⁶⁹ explaining the reduced proteolytic function seen in biochemical experiments.

Figure 3.

Validation of molecular dynamics simulation of γ-secretase activation. (A) Cryo-EM structure of APP substrate (APP-C83) crosslinked to catalytically inactive γ-secretase (PDB: 6IYC) overlapped with that of noncovalently bound transmembrane substrate mimetic inhibitor (SB-250) to active enzyme (PDB: 8K8E). Only the PSEN1 component of the protease complex is shown for clarity. (B) Closeup of the active site, showing interaction of the transition-state-mimicking hydroxyl group of the peptidomimetic inhibitor with the two catalytic aspartates (D257 and D385). Note also the overlap between APP-C83 and the peptidomimetic inhibitor. (C) Overlap of the structure of peptidomimetic inhibitor bound to active γ-secretase with computationally activated protease shows similar distance and alignment between the two catalytic aspartates. Also note the similar interactions of one of these aspartates (D385) with the carbonyl oxygen of scissile amide bond of APP substrate in the simulation and with the hydroxyl group of the inhibitor in the cryo-EM structure. Figure panels from Devkota S et al, Cell Reports, 2024.

The Stalled E-S Complex Hypothesis

Interestingly, molecular dynamics simulations also suggested that FAD-mutant E-S complexes are less conformationally flexible,⁶⁶ with the implication that E-S complexes are stabilized. This idea was supported by experiments in intact cells using fluorescence lifetime imaging microscopy (FLIM).⁶⁶ Antibodies directed to proximal epitopes on substrate and enzyme followed by fluorescently tagged secondary antibodies provided a readout of E-S complexes through reduction in the fluorescence lifetime of the donor fluorophore (Fig. 4A). APP substrate C99 and PSEN1 were cotransfected into PSEN1/2 double-knockout human embryonic kidney (HEK) 293 cells. Introduction of each of the six PSEN1 FAD mutations mentioned above vis-à-vis wild-type PSEN1 led to reduction of the FLIM signal, consistent with stabilization of the γ-secretase-C99 E-S complexes (example micrographs shown in Fig. 4B). Similar results were seen with four FAD-mutant C99 constructs. These findings have since been extended with an additional 11 PSEN1 FAD mutations (ref⁷⁰ and unpublished results). Some of these PSEN1 FAD-mutant γ-secretase complexes produce virtually no Aβ; yet they still apparently form stable E-S complexes.

Figure 4.

Evidence that FAD mutations lead to stalled γ-secretase enzyme-substrate (E-S) complexes. (A) Design of fluorescence lifetime imaging microscopy (FLIM) experiments to detect E-S complexes. Transfection of PSEN1 into PSEN1/2 double knockout cells leads to incorporation of the exogenous PSEN1 (blue) into the γ-secretase complex. Cotransfection with C99, containing a signal sequence on the N-terminus for membrane insertion and C-terminal fusion with miRFP720 leads to Aβ production by γ-secretase. Addition of antibodies directed to the C99/Aβ N-terminus (6E10) and to a nearby epitope on nicastrin (NCT) followed by secondary antibodies conjugated to fluorophore Alexa488 and quencher CY3) display FRET and reduced fluorescence lifetime of Alexa488 when substrate is bound to enzyme. The miRFP720 fusion allows detection of C99-rich regions (low ratio of 6E10/C99-720) or Aβ-rich regions (high ratio of 6E10/C99-720). (B) Example FLIM experiments with cotransfection of C99-720 and either WT PSEN1 or P117L PSEN1. Red circles are examples of Aβ-rich regions, which show reduced fluorescence lifetime with P117L PSEN1 vis-à-vis WT PSEN1. Panel B from Devkota S et al, Cell Reports, 2024.

In parallel, a new transgenic C. elegans model system for FAD was developed.⁶⁶ This roundworm has been a powerful and convenient animal model for decades to elucidate basic biology but is also increasingly employed to help decipher human disease mechanisms. Presenilin and the γ-secretase complex are highly conserved and essential for the development of all metazoans, due to the role of γ-secretase in signaling from the Notch family of cell-surface receptors.⁷¹ Binding of cognate ligand on a neighboring cell triggers ectodomain shedding of the Notch receptor, followed by intramembrane proteolysis by γ-secretase. This releases the Notch intracellular domain for translocation to the nucleus, where it interacts with transcription factors that control cell growth and differentiation. The connection between presenilin and Notch signaling was first made in C. elegans.⁷² Human PSEN1 can replace worm orthologs such as sel-12 in the γ-secretase complex and rescue sel-12 mutant phenotypes. Interestingly, FAD mutants of PSEN1 only partially rescue the phenotype, suggesting a hypomorphic mechanism for pathogenesis.⁷³

These observations suggest that C. elegans might be a useful model organism for the study of FAD. Transgenic animals were generated that coexpress C99 and PSEN1 under control of a neuronal promoter (Fig. 5).⁶⁶ The parental line expressed GFP-fused synaptobrevin, a presynaptic protein, allowing visualization of synaptic puncta by fluorescence microscopy. Coexpression of the wild-type proteins did not affect organism lifespan or synapse number. However, FAD-mutant C99 reduced lifespan and led to age-dependent synaptic loss, but only upon coexpression with wild-type PSEN1. This indicates that the neurodegenerative phenotype requires interaction of the FAD-mutant C99 with WT PSEN1. Designed mutations in C99 were then used to show that the phenotype did not require production of Aβ. These mutations preserved interaction between C99 and PSEN1/γ-secretase, stabilizing the E-S complex, suggesting that stalled E-S complexes are sufficient to elicit the reduced lifespan and synaptic loss. That is, the stalled process rather than the proteolytic products may be pathogenic. Remarkably, FAD-mutant PSEN1 led to the same neurodegenerative phenotype, even when expressed alone (i.e., APP-derived substrate C99 is not required), suggesting that stalled γ-secretase E-S complexes with endogenous C. elegans substrates can elicit the phenotype.

Figure 5.

Development of a C. elegans model of FAD.

Taken together, these findings led to formulation of a model in which synaptic degeneration in FAD is triggered by stalled γ-secretase E-S complexes (Fig. 6).⁶⁶ When FAD-mutant APP substrate C99 binds to γ-secretase, stalled E-S complexes result. With FAD-mutant PSEN1, interaction of the mutant enzyme with any of γ-secretases many substrates can lead to stalled E-S complexes. The stalled E-S complexes may trigger synaptic degeneration by preventing normal proteolysis of other substrates. However, this loss-of-function model is inconsistent with human genetics: Dominant loss-of-function mutations in PSEN1 and other γ-secretase components, which result in nonsense-mediated decay of mRNA with premature stop codons, is associated with a hereditary skin disease, severe acne inversa, not neurodegeneration.⁷⁴ Thus, simple dominant loss of function alone—haploinsufficiency—can be ruled out, implicating the stalled γ-secretase E-S complex per se in a gain of neurotoxic function contributing to pathogenesis.

Figure 6.

Stalled γ-secretase E-S complex hypothesis of FAD pathogenesis. FAD mutations result in stalled E-S complexes that cause age-dependent synaptic degeneration in C. elegans. Stalling at any point in processive proteolysis may trigger synaptotoxicity independently of Aβ production. For FAD mutation in APP, coexpression of PSEN1 is required, suggesting stalled APP substrate and proteolytic intermediates bound to γ-secretase can lead to synaptic loss. For FAD mutation in PSEN1, coexpression of C99 is not required for synaptic degeneration, suggesting that γ-secretase-bound endogenous worm substrates can result in synaptotoxic stalled E-S complexes.

How might PSEN2 fit in with the new stalled E-S complex hypothesis? Whether FAD mutations in PSEN2 also lead to stalled γ-secretase E-S complexes remains to be seen; however, given the homology between the two human presenilins and their similar role as catalytic subunit of γ-secretase complexes, this seems likely. Another question is whether wild-type PSEN2 attenuates effects of stalled FAD PSEN1-containing γ-secretase E-S complexes in cells in which both presenilin homologs are expressed. This may depend on the level of expression of FAD PSEN1 in the cell and its proportion vis-à-vis PSEN2 (as well as wild-type PSEN1 expressed from the other allele). To the degree that stalled FAD E-S complexes contribute to neurodegeneration through reduced proteolytic processing of critical substrates, wild-type γ-secretase complexes may attenuate these effects. In contrast, wild-type complexes would likely have little effect on gain-of-function mechanisms of neurotoxicity for stalled E-S complexes per se.

Precedent for the Stalled E-S Complex Hypothesis

While the idea that stalled E-S complexes itself can trigger pathogenesis may seem strange or unlikely, there is a precedent for stalling of another processive enzyme that initiates cellular signaling: the ribosome in the act of protein translation. Ribosomes may be stalled by mRNA truncations, stable secondary structures, or certain repeat sequences; low levels of one or more amino acids; or translation inhibitors.⁷⁵ Cellular responses are initiated by collision of stalled ribosomes with uninhibited ribosomes (Fig. 7A). One immediate consequence of ribosomal collision is interaction with the E3 ubiquitin ligase ZNF598, which ubiquitinylates certain small ribosomal proteins to induce ribosomal subunit dissociation. Ribosomal collision also leads to recruitment of EDF1 and GIGYF2 to trigger inhibition of translation in cis, to prevent further collision and minimize aberrant protein production. In addition, two stress signaling pathways can be induced by ribosomal collision: the ribotoxic stress response, through autophosphorylation of the MAP3 kinase ZAKα, and the integrated stress response, by activating GCN2, which phosphorylates eIF2α to repress cap-dependent translation.

Figure 7.

Other examples of stalled enzyme-substrate complexes or enzyme-inhibitor complexes per se eliciting cell signaling, independently of effects on enzyme activity. (A) Stalled ribosomes in the process of protein translation result in ribosomal collisions that trigger the ribotoxic stress response and the integrated stress response. (B) Immunosuppressant drugs rapamycin and FK506 both bind the active site of FKBP12 to inhibit its peptidyl-proline isomerase activity. However, enzyme inhibition does not affect T-cell replication. The enzyme-inhibitor complexes bind to different downstream effector proteins—mTORC1 in the case of rapamycin and calcineurin in the case of FK506—to affect different signaling pathways required for T-cell replication.

Moreover, there are instances of enzyme-inhibitor complexes per se resulting in new signaling functions. For example, the immunosuppressant natural products FK506 and rapamycin both bind to the active site of immunophilin FKBP12, a cis-trans prolylisomerase.⁷⁶ While these structurally related macrocycles inhibit the prolylisomerase activity of FKBP12, this is not how they act as immunosuppressants: Simplified synthetic analogues that retain inhibitory activity have no effect on T-cell replication. Rather, the enzyme-inhibitor complexes bind to and inhibit calcineurin in the case of FK506 and mTOR in the case of rapamycin (Fig. 7B). That is, the two enzyme-inhibitor complexes present new and distinct interfaces for interaction with their respective target proteins. Although both drugs bind to and inhibit FKBP12 activity, the respective enzyme-inhibitor complexes as such block different T-cell signaling pathways. Inhibition of the enzyme activity of FKBP12 is coincidental and correlative but not causative.

A more recently discovered example involves inhibitors of calcium/calmodulin-dependent protein kinase II (CaMKII), which plays an essential role in long-term potentiation (LTP), a synapse-strengthening process critical to learning and memory.⁷⁷ Inhibitors block CaMKII function in LTP formation; however, inhibition of CaMIIK protein phosphorylating activity does not affect LTP. Instead, the enzyme-inhibitor complexes block interaction of CaMKII with NMDA-type glutamate receptors, an interaction that is apparently required for LTP. Development of an inhibitor of CaMKII enzyme activity that did not affect interaction between CaMKII and NMDA receptors revealed that the latter is the critical function connected to LTP. This new finding essentially invalidated 30 years of research in the field involving use of CaMIIK inhibitors. CaMKII’s enzymatic activity was assumed to be critical for LTP because CaMKII knockout or knockdown showed the same effects as inhibitors. In contrast, active-site mutations that eliminate catalytic activity but retain NMDA binding did not affect LTP, while mutations that disrupt NMDA binding but not catalytic activity did prevent LTP. The remarkable example of CaMKII raises the question of whether other well-studied enzyme inhibitors alter cell and organism physiology through enzyme-inhibitor complexes per se rather than inhibition of enzymatic activity.

Implications of the Stalled E-S Complex Hypothesis

The new hypothesis that stalled γ-secretase E-S complexes are synaptotoxic provides a unifying explanation for how FAD mutations in either APP and presenilins can trigger neurodegeneration, without invoking Aβ or simple loss of proteolytic function. However, it raises the question of why, among the ~150 known γ-secretase substrates, FAD mutations are only found in APP. This may be due to the high expression of APP in the brain (www.proteinatlas.org)⁷⁸ as well as the constitutive cleavage of APP by β-secretase,⁹ providing high levels of APP substrate C99 to γ-secretase. Mutation in APP that leads to inefficient proteolysis thereby leads to high levels of stalled E-S complexes. Similar mutations in other substrates may lead to stalled complexes, but only to a limited degree, given their lower expression and/or regulated ectodomain shedding. FAD mutations in presenilin, on the other hand, can lead to high levels of stalled E-S complexes, with high levels of total heterogeneous substrates bound to γ-secretases.

Although these initial studies support a role for stalled E-S complexes in the pathogenesis of FAD, the findings are likely to have implications for the common sporadic late-onset Alzheimer’s disease, as AD and sporadic Alzheimer’s disease show the same pathology, presentation and progression.^{79, 80} FAD involves dominant mutations in the substrate and enzyme that produce Aβ, and aggregation of Aβ in the form of cerebral plaques is a characteristic pathological feature of Alzheimer’s disease. Interestingly, the new findings that FAD mutations lead to stalled complexes and reduced processive proteolysis may also explain the observation that FAD mutations generally (although not always) increase the ratio Aβ42/Aβ40. Increase in this ratio in FAD is implicated in formation of cerebral plaques, which are primarily composed of the aggregation-prone Aβ42 in both early-onset FAD and late-onset sporadic AD. That is, amyloid plaque formation may be coincidental with stalled γ-secretase E-S complexes but not causative with respect to initiation of the disease process.

Nevertheless, the finding that stalled γ-secretase E-S complexes can trigger synaptic degeneration does not preclude a role for Aβ in AD pathogenesis. Clearing out plaques with antibody therapeutics can slow the rate of cognitive decline.^{46, 47} However, these effects are modest, even in well-controlled clinical trials. The effectiveness of these antibodies in post-approval “Phase 4” trials remains to be seen. Presymptomatic treatment with anti-Aβ antibodies may prove to be more effective, although there is concern about recent efforts to define Alzheimer’s disease by the presence of amyloid plaques, unnecessarily putting “patients” who may not be on the path to cognitive decline on costly therapeutics with serious safety concerns. Despite triumphant pronouncements to the contrary, the limited effectiveness of these agents has not proven the amyloid hypothesis, and whether Aβ is the primary driver of AD pathogenesis remains an open question.⁴⁸ Amyloid plaques may be a coincidental consequence, an epiphenomenon, of stalled γ-secretase E-S complexes and reduced proteolysis, eliciting reactive glial cells and neuroinflammation as a secondary contributor to disease progression.

The stalled E-S complex hypothesis also has implications for drug discovery for the prevention and/or treatment of FAD. This hypothesis predicts that inhibitors would be detrimental, as FAD mutations reduce processive proteolysis. Indeed, γ-secretase inhibitors failed in clinical trials to treat sporadic Alzheimer’s disease, not only because of deficiencies in essential Notch signaling in the periphery (leading to immunosuppression, gastrointestinal bleeding, and skin lesions) but because the subjects worsened cognitively.^{81, 82} Instead, compounds that rescue deficient proteolysis would be needed, stimulating processive proteolysis and thereby facilitating terminal product formation and dissociation. Interestingly, a large class of γ-secretase modulators have been developed over the past 25 years that reduce Aβ42 production by stimulating its conversion to Aβ38.⁸³ Among these many compounds may be those that rescue deficiencies in other proteolytic processing steps as well.

Open questions

As the stalled E-S complex hypothesis of AD pathogenesis is quite new, first published early in 2024,⁶⁶ considerable work lies ahead toward its validation. Reproducibility of biochemical and biophysical findings by other laboratories—using similar as well as orthogonal approaches—would cement the findings that FAD mutations result in deficiencies in processive proteolysis of APP substrate C99 by γ-secretase and that these deficiencies are due to stabilization of E-S complexes. Also critical are confirmation and extension of the effects of stalled γ-secretase E-S complexes on neuronal synapses. The ability of stalled complexes to trigger synaptic degeneration will need to be demonstrated in organisms other than C. elegans, particularly in mammalian models, and under conditions in which the pathogenic proteins are expressed physiologically (e.g., using genetically edited mice with APP and/or presenilin knock-in mutations). Moreover, the detection of stalled complexes should be demonstrated in vivo, not only in cultured cells, as well as in human clinical and postmortem samples.

Mechanisms by which stalled γ-secretase E-S complexes trigger synaptic loss also need to be elucidated. Based on previous studies of FAD mutations, reasonable hypotheses include disruption of lysosomal proteolysis, autophagy, and mitochondrial function. Induced pluripotent stem cells gene-edited to contain APP or PSEN1 FAD mutations were differentiated into neurons and found to have impaired lysosomal proteolysis and defective clearance of autophagosomes.⁸⁴ Interestingly, prevention of C99 formation by blocking β-secretase activity rescued these deficiencies in FAD-mutant neurons, while γ-secretase inhibition induced these phenotypes in wild-type neurons. In another study, transfection of human embryonic kidney (HEK) 293 cells with FAD-mutant C99 resulted in reduced oxygen consumption associated with ATP production compared with cells transfected with wild-type C99.⁸⁵ Addition of a designed mutation to FAD-mutant C99 that blocks Aβ42 production did not rescue mitochondrial dysfunction; rather, this mutant exacerbated the phenotype. Both the FAD-mutant C99 (which elevates the Aβ42/Aβ40 ratio) and the designed mutant (which blocks Aβ42 production) form stable γ-secretase E-S complexes as determined by blue native PAGE,⁶⁶ consistent with stalled complexes, not Aβ42/Aβ40 ratios, as the initiator of mitochondrial dysfunction.

Another issue that should be addressed is whether stalled γ-secretase E-S complexes occur in sporadic late-onset AD; that is, in the absence of FAD mutations in APP or presenilins. In these cases, alteration in the composition and properties of cellular membranes (e.g., fluidity, thickness) may result in E-S complex stabilization and reduced processive proteolysis. Indeed, altered lipid composition occurs with aging and can affect synaptic function and neuronal survival.⁸⁶ Interestingly, the strongest genetic risk factor for common late-onset AD is the apolipoprotein E4 allele: One copy of ApoE4 increases risk of AD by three- to fourfold, while two copies increase risk 12- to 15-fold.^{87, 88} ApoE is a lipid carrier protein, delivering lipids from the liver to other cells of the body, and the ApoE variants may differ in ability to transport lipids. Another factor that might affect E-S complex stabilization and catalytic rate include intracellular or vesicular pH. γ-Secretase activity in neurons appears to be located primarily in acidic compartments such as lysosomes.⁸⁹ Dysregulated lysosomal pH is associated with AD and other neurodegenerative diseases⁹⁰ and might also lead to stalled γ-secretase E-S complexes.

Perspective

Biochemistry proved essential in identifying presenilin as the catalytic component of the γ-secretase complex, a founding member of a new class of intramembrane-cleaving proteases. This discovery revealed that FAD mutations are in the enzyme and the substrate that together produce Aβ peptides, ostensibly providing compelling support for the amyloid hypothesis. A quarter of a century later, doubts remain about whether Aβ is the primary driver of AD pathogenesis, and therapeutics that target Aβ and clear plaques from the brain are at best only modestly effective at slowing the rate of cognitive decline (Box 1).

Box 1. The Amyloid Hypothesis versus the Stalled E-S Complex Hypothesis.

Discrepancies with the amyloid hypothesis:

• Identities of the toxic form(s) of Aβ and associated neurotoxic signal pathways are unresolved.

• The connection between Aβ and tau, the other pathological protein in AD, is unknown.

• FAD mutations generally inhibit Aβ production by γ-secretase. Some mutations produce very little Aβ, and others do not increase the putatively critical Aβ42/Aβ40 ratio.

• Decades of drug discovery targeting Aβ in various ways have led to recently approved therapeutics that are only modestly effective at best.

Support for stalled γ-secretase E-S complexes as a key disease driver in FAD:

• FAD mutations are deficient in early cleavage steps in APP substrate processing by γ-secretase.

• Dominant loss-of-function mutations in γ-secretase subunits that result in nonsense-mediated decay and haploinsufficiency do not lead to neurodegeneration.

• FAD mutations stabilize E-S complexes in cells, consistent with reduced conformational flexibility of FAD-mutant E-S complexes in molecular dynamics simulations.

• Stabilized E-S complexes alone, in the absence of Aβ production, lead to age-dependent synaptic loss in a C. elegans model system.

Implications of the stalled complex hypothesis:

• Explains the general observation of increased Aβ42/Aβ40 with FAD mutations, while suggesting this may not be the primary disease driver.

• Other substrates besides APP within stalled E-S complexes may trigger synaptic loss.

• Altered membrane composition or properties during aging may lead to stalled E-S complexes in sporadic late-onset AD.

• Drug discovery should focus on rescuing stalled γ-secretase E-S complexes, rather than γ-secretase inhibitors.

These doubts led to reinvestigation of the effects of FAD mutations on the complex processing of APP substrate C99 by γ-secretase and the finding that all mutations studied to date are deficient in multiple proteolytic steps. FAD mutations apparently stabilize γ-secretase E-S complexes and thereby stall proteolysis. Once again biochemistry provided critical data that allowed elucidation of the common effect of FAD mutations, leading to the hypothesis that the stalled process—not the proteolytic products—may be the primary trigger of synaptic degeneration. Although this hypothesis might seem counterintuitive, there are precedents for enzyme-substrate or enzyme-inhibitor complexes per se triggering cell signaling independently of effects on catalytic activity.

Going forward, the field must collectively address the new question of the pathogenic role of stalled γ-secretase E-S complexes. Over 30 years of focus on the amyloid hypothesis has not resolved the role of Aβ in AD pathogenesis, nor has it led to clearly effective therapeutics. The question of whether stalled E-S complexes are pathogenic should be determined as soon as possible, to advance the field and focus on better strategies for prevention and treatment.