γ-Secretase: Once and future drug target for Alzheimer’s disease

1. The amyloid hypothesis of Alzheimer’s disease

Alzheimer’s disease (AD) is a devastating neurodegenerative disorder affecting over 6 million in the U.S. and over 40 million worldwide.¹ The gradual loss of neurons, particularly in the hippocampus, leads to inexorable decline in memory and cognition. Pathology associated with AD includes amyloid plaques, neurofibrillary tangles and neuroinflammation. The plaques are extraneuronal and composed of the small aggregation-prone amyloid β-peptide (Aβ), while the tangles are inside neurons and composed of filaments of the otherwise microtubule-associated protein tau. Neuroinflammation results from overactivated or dysfunctional non-neuronal glial cells in the vicinity of axons, dendrites and synapses.

The discovery in 1991 of dominant mutations associated with early-onset familial AD (FAD) in the gene encoding the amyloid precursor protein (APP), which is proteolytically processed to form Aβ, led to original formulation of the amyloid hypothesis of AD pathogenesis.² Other dominant mutations were later found in presenilin, the catalytic component of the protease that produces Aβ, reinforcing the amyloid hypothesis. This hypothesis posits that aggregated Aβ triggers a cascade of events, including tau tangle formation and neuroinflammation, that ultimately leads to neurodegeneration and dementia. Onset of the rare hereditary FAD occurs in mid-life, generally in the range of 30–60 years of age. Other than the monogenetic cause and early age of onset, the pathology, presentation and progression of FAD is closely similar to that of the more common sporadic late-onset form of the disease.^3,4 Thus, elucidation of the pathogenic mechanisms of FAD were—and continue to be—considered critical for providing insight into the cause of all AD.

2. AD drug discovery: A brief history

Drug discovery for AD began even before the identification of FAD mutations in APP and formulation of the amyloid hypothesis. The observation that many of the lost neurons were cholinergic led to development of brain-penetrating acetylcholinesterase inhibitors such as donepezil (Aricept).¹ However, these and other agents such as memantine (Namenda), while improving cognition temporarily, did not stop the underlying neurodegeneration. With the discovery of FAD-causing mutations in APP (found in and around the small Aβ region of this large precursor protein) and in presenilin (the protease that produces Aβ), the race was on to find molecules that could block proteolysis of APP and the production of Aβ, inhibit Aβ aggregation, or clear amyloid deposits from the brain. The expectation was that such agents would be disease-modifying therapies, slowing or halting neurodegeneration and cognitive decline, rather than temporarily mitigating symptoms of memory impairment.

Inhibition of Aβ production involves targeting one of the two proteases responsible to cutting Aβ out from APP: β-secretase and γ-secretase.² β-Secretase cleaves in the juxtamembrane region of the APP membrane protein, releasing the large ectodomain. The remnant membrane-bound 99-residue C-terminal fragment, C99, is then hydrolyzed within its transmembrane domain by γ-secretase. Many inhibitors for these two enzymes were brought forward, and while they could effectively lower Aβ production in the brain, these compounds failed in clinical trials, causing side effects and actually worsening cognition. In fact, all agents targeting amyloid—by any means—failed in human trials until very recently, with the approval of anti-Aβ antibodies.⁵ However, treatment with these antibodies only modestly slows the rate of cognitive decline and in some patients causes concerning side effects such as swelling and bleeding in the brain.

3. γ-Secretase inhibition: early promise and problems

Inhibitors of γ-secretase activity were discovered before identification of the responsible protease.⁶ Transition state analog inhibitors based on the γ-secretase cleavage site within APP C99 substrate suggested that the enzyme is an aspartyl protease, a class of proteases with two catalytic aspartic acid residues that activate a water molecule and the scissile amide bond. This finding, along with the observation that γ-secretase carries out proteolysis of the transmembrane region of C99, led to the discovery of the multi-pass membrane protein presenilin as the catalytic component of what became known as the γ-secretase complex. This discovery was considered a linchpin for the amyloid hypothesis: FAD mutations were found in the substrate and the enzyme that produce Aβ.

The search for γ-secretase inhibitors (GSIs) for the potential treatment of AD increased dramatically in the wake of this finding, and the hydrophobic character the membrane-embedded enzyme made it easier to find brain-penetrating compounds.⁷ However, potential problems with γ-secretase as an AD drug target quickly arose: the protease complex cleaves other membrane protein substrates besides APP, and some of these other substrates have important biological roles.⁸ Most critical are the Notch family of cell-surface receptors.⁹ Cleavage of Notch receptors by γ-secretase leads to release of Notch intracellular domain from the membrane and translocation of this proteolytic product to the nucleus. There it activates transcription factors that control the expression of genes critical to cell-fate determination during organism development and in adulthood. Inhibition of γ-secretase, either genetically or pharmacologically, results in serious toxic consequences in mice, including disruption of the lining of the GI tract, immunosuppression and skin lesions, and all of these are attributed to blocking critical Notch signaling pathways.

4. Failed clinical trials

Despite the concerns about on-target toxicity, the discovery and development of GSIs for AD continued, with advancement of candidates into clinical trial. The hope was for a therapeutic window in which inhibition of γ-secretase activity in the brain was sufficient to lower Aβ and prevent its aggregation while minimizing effects due to lowering Notch signaling in the periphery. However, Eli Lilly candidate semagacestat failed in Phase 3 clinical trials, with adverse events similar to what was observed in mice.¹⁰ Even more concerning, drug treatment led to cognitive worsening. While GI bleeding, immunosuppression and skin lesions could be attributed to interfering with signaling from Notch receptors, cognitive worsening raised concerns about the amyloid hypothesis and the general strategy of blocking Aβ production to treat AD.

Nevertheless, hope remained that discovery of GSIs with selectivity for APP substrate vis-à-vis Notch substrate might lead to more promising clinical candidates. So-called “Notch-sparing” GSIs were discovered, with one clinical candidate, avagacestat from Bristol-Meyers-Squibb, reportedly possessing 193-fold selectivity for inhibiting APP substrate over Notch substrate.¹¹ However, another report suggested avagacestat lacked such selectivity.¹² Phase 2 clinical trials gave the final verdict: Avagacestat, like semagacestat, caused adverse events consistent with interference with Notch signaling and also led to cognitive worsening.¹³ This devastating failure of an allegedly selective GSI effectively killed inhibition of γ-secretase as a strategy for treating AD. Nevertheless, efforts to target a specific isoform of the γ-secretase complex to achieve selectivity for inhibition of APP substrate over Notch substrates were recently reported.¹⁴

The promise of γ-secretase modulators

The Aβ that deposits in the characteristic plaques of AD is primarily a 42-residue form (Aβ42), which is a minor variant among secreted Aβ peptides. The predominant form produced by γ-secretase is a 40-residue variant (Aβ40), and the ratio of Aβ42/Aβ40 has long been considered the key factor leading to Aβ aggregation and AD pathogenesis. Indeed, changes in CSF Aβ42 occur in FAD 15–25 years before predicted symptom onset.¹⁵ Early screening efforts using cell-based assays identified compounds capable of lowering Aβ42 production selectively over Aβ40 production, without inhibiting γ-secretase activity. These γ-secretase modulators (GSMs) held promise for avoiding toxic consequences of γ-secretase inhibition. Early clinical trials demonstrated that GSMs were safe and well-tolerated but not efficacious. This was thought to be due to insufficient potency and brain penetration. Extensive efforts led to improved potency and properties of GSMs,¹⁶ although the failures with GSIs resulted in skittishness for advancing anything targeting γ-secretase into the clinic. Nevertheless, a recent report showed promising pre-clinical validation for a new GSM,¹⁷ results that support an Investigational New Drug (IND) application to the FDA. Such an agent could be potentially given alone or in combination with newly approved anti-Aβ antibody therapies.

New findings on FAD mutations and implications for drug discovery

The transmembrane domain cleavage of APP C99 substrate by γ-secretase is complex, involving multiple proteolytic steps along two pathways: Aβ49→Aβ46→Aβ43→Aβ40 and Aβ48→Aβ45→Aβ42→Aβ38. Initial endoproteolytic (ε) cleavage produces either Aβ49 or Aβ48, followed in general by tripeptide trimming to ultimately release secreted forms of Aβ that are 38–43 residues in length. My laboratory recently reported the quantification of the effects of 14 FAD mutations located within the transmembrane domain of APP on all the cleavage events carried out by γ-secretase on C99 using purified enzyme and substrate.¹⁸ We found that most of these mutations reduced Aβ42 production, although Aβ40 production was generally reduced even further, thereby increasing the Aβ42/Aβ40 ratio. However, not every mutation did so, which was consistent with a report on the effects of 138 FAD mutations in PSEN1.¹⁹ We further found that the common effects of all 14 APP FAD mutations were deficiencies in the first or second tripeptide trimming step.

In work currently under review,²⁰ we found that six FAD mutations in PSEN1 were all deficient in the initial ε cleavage of C99 by γ-secretase and in one or more trimming steps as well. Substrate-based transmembrane peptidomimetic probes allowed use of cryoelectron microscopy to solve an atomic-resolution structure of the protease complex trapped at the transition state of intramembrane proteolysis. This structure was remarkably similar to that of the activated enzyme as generated through molecular dynamics simulations. In turn, the simulations indicated that FAD mutations result in reduced conformational flexibility of the enzyme-substrate (E-S) complex. The reduced flexibility provided an explanation for the reduced proteolytic activity of the FAD-mutant enzyme complexes and also suggested stabilization of the mutant E-S complexes. Fluorescence microscopy in whole cells supported FAD mutations leading to stalled and stabilized E-S complexes, and a new C. elegans transgenic model revealed that stalled E-S complexes lead to age-dependent synaptic degeneration even in the absence of Aβ production. Taken together, these findings suggest that the stalled process—not the Aβ products—of γ-secretase cleavage of APP substrate can trigger neurodegeneration in FAD, with likely implications for the more common sporadic AD.

Expert opinion

The failure of GSIs in the clinical all but killed γ-secretase as a target for AD drug discovery. In retrospect, the failures were not surprising, as disease-causing mutations in the enzyme inhibit overall proteolytic activity. FAD mutations in PSEN1 and APP all appear to inhibit one or more early proteolytic steps in the complex processing of C99 by γ-secretase. The finding that FAD mutations result in stalled and stabilized E-S complexes that cause synaptic degeneration in the absence of Aβ production suggests that stimulators of γ-secretase, rather than inhibitors, might be effective for the prevention or treatment of FAD. Interestingly, GSMs lower Aβ42 levels by stimulating the Aβ42→Aβ38 trimming step. Whether Aβ42 lowering will be sufficient to effectively slow or halt cognitive decline, beyond the modest effect of recently approved Aβ-targeting antibody therapeutics, is unclear. A better bet would be to discover agents that rescue the stalled function of FAD-mutant γ-secretase. These should be effective for early-onset FAD and may be worthwhile for late-onset sporadic AD as well: FAD and sporadic AD are similar with respect to pathology, presentation and progression,^3,4 suggesting common underlying disease mechanisms.

Despite the findings that FAD mutations in the enzyme reduce overall Aβ production in vitro, these hereditary forms of the disease nevertheless show characteristic amyloid pathology composed primarily of Aβ42. This is likely because these mutations increase the Aβ42/Aβ40 ratio, which is critical for seeding aggregation. This increased ratio is closely correlated with reduced γ-secretase proteolytic function, particularly trimming of Aβ intermediates. Thus, amyloid deposition may be a biomarker that closely couples with the true pathogenic trigger, which we suggest could be stalled γ-secretase enzyme-substrate complexes. Resolving this issue will be critical before moving forward with drug discovery, to know whether to continue pursuing anti-amyloid strategies or to initiate programs aimed at identifying candidates that restore activity to stalled γ-secretase complexes.

In any event, targeting either Aβ or γ-secretase alone may be insufficient or suboptimal for the effective treatment of AD. Other factors are involved in AD pathogenesis, including the sporadic AD genetic risk factor apolipoprotein E4, fibrillization of tau protein and formation of neurofibrillary tangles, activation of microglia and neuroinflammation, and mitochondrial dysfunction, among others. These additional factors also present opportunities for AD drug discovery. Combination therapies may be required to halt or dramatically slow the rate of cognitive decline in AD. Nevertheless, modulation of γ-secretase in some form may well be a necessary component of such combination regimens.