The Amyloid Cascade

1. The Amyloid Cascade Hypothesis

In a single page in Science in 1992, John Hardy and Gerald Higgins articulated what would become the central organising idea of Alzheimer’s research for the next thirty years: that the deposition of amyloid-β peptide is the causative event in AD, with tangles, neuronal death, and dementia as downstream consequences. The argument was assembled from genetics, neuropathology, and the Down syndrome connection (chr 21 trisomy → extra APP gene → invariant early plaques).

The cascade rests on five mutually reinforcing observations that together convinced the field this was no mere correlation:

1. All known autosomal-dominant familial AD mutations (in APP, PSEN1, PSEN2) increase Aβ42 production or the Aβ42:Aβ40 ratio.
2. Down syndrome (trisomy 21, three APP copies) produces invariable AD-type plaques by age ~40 and dementia by ~60.
3. APP duplication alone (without point mutation) causes early-onset AD — gene dosage suffices.
4. APP^A673T (the “Icelandic” variant, Jonsson 2012) reduces Aβ production ~40% and is protective against AD and age-related cognitive decline.
5. APOE-ε4, the major late-onset risk allele, increases Aβ deposition and impairs Aβ clearance.

The hypothesis matured through Selkoe (Neuron 1991; Science 2002), Hardy & Selkoe (Science 2002, “The amyloid hypothesis at 25 years”), and Selkoe & Hardy (EMBO Mol Med 2016). Its weaknesses — failed γ-secretase and BACE1 inhibitors, weak plaque-cognition correlation — are the subject of Section 10. Its partial vindication arrived in 2023–2024 with lecanemab and donanemab, the first disease-modifying drugs in AD history (covered in Part VII).

2. Amyloid Precursor Protein (APP)

The amyloid precursor protein is a type-I single-pass transmembrane glycoprotein encoded by the APP gene on chromosome 21q21.3 — the very chromosome triplicated in Down syndrome. It was cloned in 1987 by Kang et al. (Nature 1987) and Goldgaber et al., immediately after Glenner & Wong (1984) and Masters & Beyreuther (1985) sequenced the Aβ peptide from cerebrovascular and plaque amyloid.

Architecture

• Large N-terminal extracellular domain — E1 (heparin-binding + growth-factor-like), KPI (Kunitz protease inhibitor; in 751/770 isoforms only), E2, juxtamembrane.
• Single transmembrane α-helix — ~23 residues; contains the C-terminal half of the Aβ sequence.
• Short cytoplasmic tail — ~47 residues; contains the YENPTY motif that binds Fe65, X11/Mint, and Dab1 adaptors.

Isoforms (alternative splicing of exons 7 & 8)

Physiological function

APP is not just a substrate for amyloid generation. Cell-biological work (Müller, Deller & collaborators; Priller et al.) implicates it in:

• Synaptic adhesion and maintenance — trans-synaptic homo- and heterodimerisation; sAPPα is synaptotrophic.
• Axonal transport — via kinesin-1 light-chain interaction.
• Iron homeostasis — ferroxidase activity in the E2 domain (Duce et al., Cell 2010).
• Neurogenesis and neurite outgrowth — sAPPα promotes both; APP-knockout mice show subtle but real synaptic deficits.
• Possible antimicrobial role — the “Aβ-as-AMP” hypothesis (Soscia/Moir/Tanzi 2010; Eimer 2018) reframes Aβ as part of innate immunity.

APP belongs to a family that also contains APLP1and APLP2; triple knockout is perinatally lethal with cortical malformations, demonstrating an essential developmental role.

3. APP Processing — Two Pathways

APP is a substrate for at least three secretases that cleave it sequentially. The choice of first cut defines the fate of the molecule: the non-amyloidogenic pathway destroys the Aβ sequence; the amyloidogenic pathwayreleases it intact.

Subcellular geography matters

ADAM10 acts predominantly at the plasma membrane (neutral pH, non-amyloidogenic). BACE1 has acidic pH optimum (~4.5) and is most active in endosomes and on lipid rafts. Anything that promotes APP endocytosis (clathrin, AP-2, dynamin) shifts processing toward Aβ generation; anything that retains APP at the surface (e.g. mutation of YENPTY) shifts it toward sAPPα.

Other relevant cleavages

• η-secretase (membrane-bound matrix metalloprotease MT5-MMP) cuts in the ectodomain, generating Aη-α and Aη-β fragments with their own synaptic effects (Willem 2015 Nature).
• δ-secretase / legumain (AEP) cleaves at Asn373/Asn585, priming APP for BACE1.
• Caspases cleave the C-terminal cytoplasmic tail at Asp664; produces a pro-apoptotic fragment.

Production of ~1–10 nM Aβ in brain interstitial fluid is normal; clearance by neprilysin, IDE, the glymphatic/ISF flow, BBB transport (LRP1, RAGE), and microglial phagocytosis keeps concentrations low. AD is at heart a production-vs-clearance imbalancewith a 30-year time constant.

4. β-Secretase (BACE1)

β-site APP cleaving enzyme 1 (BACE1), cloned by Vassar et al. (Science 1999) and independently by four other groups in the same year, is a 501-residue type-I transmembrane aspartyl protease of the pepsin family (memapsin-2). Its catalytic ectodomain is glycosylated and tethered to the membrane by a single C-terminal helix.

The BACE1 inhibitor failure

On paper, BACE1 was the perfect AD target: it sits at the very top of the cascade, its knockout prevents Aβ production, and high-resolution structures supported rational design. Multiple potent, brain-penetrant small-molecule inhibitors entered Phase III trials between 2012 and 2019. Every single one was halted for cognitive worsening:

5. The γ-Secretase Complex

γ-Secretase is not a single protein but a four-subunit intramembrane aspartyl protease — a remarkable enzymatic feat of cleaving a peptide bond inside the hydrophobic core of the lipid bilayer. Its substrates include APP, Notch1–4, ErbB4, CD44, N-cadherin, and ~140 others (the “γ-secretome”).

The four subunits

The processive cut

γ-Secretase does not make a single cut. It chews the C99 substrate processively in a tripeptide ladder, starting at residue 49 and walking toward the extracellular face:

Two product-line pathways operate in parallel: Aβ49 → 46 → 43 → 40 (the “Aβ40 line”) and Aβ48 → 45 → 42 → 38 (the “Aβ42 line”). The proportion of substrate that “falls off” the enzyme prematurely — mainly along the 42-line, leaving Aβ42 before it can be trimmed to Aβ38 — is what determines the Aβ42:Aβ40 ratio. Most PSEN1 mutations reduce processivityalong the 42-line, increasing the Aβ42:Aβ40 ratio without necessarily increasing total Aβ — a refinement of the original cascade due primarily to the work of Wolfe, De Strooper, and Steiner.

The γ-secretase inhibitor disaster

Semagacestat (Lilly, 2010) was the first γ-secretase inhibitor in Phase III. Trial halted: cognitive worsening, plus skin cancers and GI toxicity from Notch inhibition (Notch is essential for intestinal stem-cell fate and T-cell development). The lesson taught the field the value of γ-secretase modulators (GSMs) — molecules that shift the cleavage product from Aβ42 to Aβ38 without abolishing activity — though none has yet succeeded in late-stage trials.

6. Aβ42 vs Aβ40 — The Critical Ratio

In healthy human brain, ~90% of Aβ produced is the 40-residue Aβ40 (M ≈ 4329 Da) and ~10% is the 42-residue Aβ42 (M ≈ 4514 Da). Those two extra residues — I41-A42 — are hydrophobic, and they transform the peptide’s biophysics.

Plotting the ratio R separates AD from controls more cleanly than either peptide alone — this is the basis of CSF Aβ42/Aβ40 testing (covered in Part VI):

The ratio drops because Aβ42 is preferentially sequestered into plaques in the brain, lowering its CSF concentration. Aβ40 levels are relatively unchanged, providing an internal normaliser that controls for inter-individual production differences.

Why two extra residues matter so much

In NMR ensembles of monomeric Aβ42 (Crescenzi et al. 2002, PDB 1IYT; Tomaselli et al. 2006), the C-terminal segment including I41-A42 forms a more stable hydrophobic cluster than in Aβ40. This C-terminal “sticky end” nucleates β-sheet contacts that template dimers, then oligomers, then fibrils. Computational and ssNMR work (Tycko 2016 Annu Rev Phys Chem) shows the I41-A42 hydrophobic patch is part of the ordered fibril core in essentially every Aβ42 fibril polymorph reported.

7. Aβ Aggregation — Monomers to Plaques

The aggregation cascade of Aβ is one of the best-studied cases of a more general phenomenon — protein amyloid self-assembly — that also underlies tau, α-synuclein, prion, TDP-43, huntingtin, and serpin pathologies. The kinetics follow a nucleation-elongation-secondary-nucleationscheme worked out for Aβ42 by the Knowles, Linse, Dobson laboratories (Cohen et al. PNAS 2013; Meisl et al. 2014; Cohen 2018).

The kinetic skeleton

The fundamental rate equation governing the polymer mass concentration M(t)with monomer concentration m(t) reads:

The first term is conventional elongation; the second — surface-catalysed (secondary) nucleationwith order n₂ ≈ 2 — is the explosive feedback that turns amyloid into a self-propagating disease. Existing fibrils catalyse formation of new oligomers from monomer on their surface; this means oligomer flux is highest near fibril surfaces, not in plaque-free regions.

The species along the way

1. Monomer — intrinsically disordered, ~5 kDa, mostly random coil with transient β-hairpin around residues 17–36.
2. Dimer/trimer — the smallest species shown to impair LTP (Shankar et al. Nat Med 2008 isolated Aβ dimers from human AD brain).
3. ADDLs (Aβ-derived diffusible ligands) — ~12-mer assemblies isolated by Klein, Lambert, Krafft (Lambert et al. PNAS 1998) that bind synapses and impair LTP at low nM.
4. Aβ*56 — a 56-kDa dodecamer (Lesné et al. Nature 2006) reported to correlate with cognitive impairment in Tg2576 mice; the data were later questioned on image-integrity grounds (Schrag/Piller 2022).
5. Protofibrils — curvilinear, <200 nm, AFM-visible. The species lecanemab is most selective for.
6. Mature fibrils — cross-β architecture, ~10 nm wide, indefinite length. Multiple polymorphs (cryo-EM: 5OQV, 5KK3, 7Q4M).
7. Plaques — macroscopic deposits 20–200 µm, with dystrophic neurites and reactive microglia/astrocytes around dense cores.

Three observations that recast the field

• Lag-and-burst kinetics. Pure Aβ42 in vitro shows a characteristic sigmoid: long lag, sudden burst, plateau. The lag is shortened by trace seeds; this is the molecular signature of nucleation-dependent polymerisation.
• Templating / prion-like behaviour. Pre-formed Aβ fibrils injected into APP-transgenic mouse brain seed deposition with strain-specific morphology. Different polymorphs may underlie different clinical phenotypes (rapidly progressive, atypical) — an active area of research.
• Secondary-nucleation dominance. Once a single fibril exists, secondary nucleation on its surface produces oligomers at a rate orders of magnitude faster than primary nucleation in solution. This is why Aβ aggregation is autocatalytic and why removing fibrils may itself reduce oligomer flux.

8. Why Plaques May Not Be the Toxic Species

The neuropathological foundation of AD — visible plaques and tangles — made plaques the obvious villain. But three lines of evidence accumulated through the 2000s that forced a reframing: soluble oligomers, not insoluble fibrils, are the proximate synaptotoxin.

Lines of evidence

1. Plaque count correlates poorly with cognition. Terry & colleagues (Ann Neurol 1991) showed that synapse loss, not plaque burden, was the strongest pathological correlate of dementia. Many cognitively-normal elderly are amyloid-PET positive.
2. Soluble Aβ correlates better. McLean, Masters and colleagues (1999) and many since have shown that buffer-soluble Aβ in cortical extracts — not insoluble plaque-bound Aβ — tracks Braak stage and antemortem cognition.
3. Direct synapse studies. Selkoe, Walsh and colleagues showed that 7PA2 cell-derived Aβ dimers and trimers, applied at low picomolar concentrations to hippocampal slices, block LTP and impair memory in vivo (Walsh et al. Nature 2002; Cleary 2005; Shankar et al. Nat Med 2008 using AD-brain dimers).
4. ADDLs. Lambert et al. (PNAS 1998) found that Aβ-derived diffusible ligands (~12-mer ADDLs), well below fibril size, were potent synaptic toxins and bound a small fraction of dendritic spines with very high affinity.
5. Inducible mouse models. When APP-transgene expression is shut off in adult mice, plaques persist but cognitive deficits and synaptic dysfunction can reverse — consistent with toxic species being something dynamic, not the static plaque.

A “reservoir” reconceptualisation

The current synthesis is that plaques act as a local reservoir and source of soluble oligomers rather than as the killer themselves. Surface-catalysed (secondary) nucleation on plaque-associated fibrils produces a steady local flux of oligomers; the dystrophic neurites and reactive microglia clustered around dense cores reflect this local toxicity. This integrates the plaque-cognition disconnect with the biophysics of secondary nucleation (Section 7).

9. Synaptic Toxicity Mechanisms

How does an extracellular ~5-kDa peptide degrade information processing? Two decades of slice-physiology, in vivo imaging, and proteomics have converged on a dominant axis: Aβ oligomers shift the balance of glutamatergic synaptic plasticity from LTP toward LTD, and over time eliminate the spines that carry it.

Five (overlapping) mechanisms

1. LTP impairment, LTD facilitation. Sub-nM oligomers applied to hippocampal slices block high-frequency-stimulus LTP at CA3→CA1 synapses (Walsh 2002; Shankar 2008). The same exposures facilitate LTD via mGluR5 / NMDAR-NR2B / calcineurin / GSK3β signalling. The result: synapses become labile and net-depressed.
2. Glutamate dyshomeostasis at the synapse. Aβ oligomers reduce astrocytic glutamate transport (GLT-1/EAAT2), spilling glutamate onto extrasynaptic NR2B-rich NMDARs — a death-signalling pool — while reducing synaptic NR2A activation. This pattern preferentially activates pro-apoptotic CREB-shutoff and Ca²⁺-dependent calcineurin pathways.
3. AMPAR endocytosis & spine collapse. Oligomer binding triggers calcineurin-dependent dephosphorylation of GluA1 S845 and clathrin-mediated AMPAR endocytosis — the molecular signature of LTD — followed by spine shrinkage and pruning. Hsieh, Boehm, Malinow (Neuron 2006) and Hsia, Mucke (PNAS 1999) gave key early evidence.
4. Cellular prion protein (PrP^C) co-receptor. Laurén, Strittmatter et al. (Nature 2009) identified PrP^Cas a high-affinity oligomer-binding partner. Downstream signalling via Fyn kinase (a Src-family TK) phosphorylates NR2B and tau, providing a direct molecular bridge from extracellular Aβ oligomer to intracellular tau pathology.
5. Microglial & complement-mediated synapse pruning. Aβ-driven C1q deposition on synapses tags them for microglial engulfment via C3/CR3 (Hong et al. Science 2016; Stevens lab). This recapitulates a developmental pruning programme inappropriately reactivated in disease.

From synapses to networks to clinic

The clinical correlates of these synaptic events are visible in EEG, MEG, and fMRI before they are visible on MRI volumetry: hippocampal hyperactivation, default-mode-network disconnection, network-level criticality changes. Levetiracetam (an SV2A modulator) improves memory in MCI patients with hippocampal hyperactivity (Bakker et al. Neuron 2012) — one of the few non-amyloid pharmacologic signals in early symptomatic AD.

Crucially, oligomer-induced synaptic toxicity requires tau: in tau-knockout mice, Aβ oligomers no longer impair LTP and no longer cause cognitive deficits (Roberson et al. Science 2007; Ittner 2010). This is the mechanistic hand-off to Part IV (Tau).

10. The Amyloid Hypothesis at 30 Years

Three decades after Hardy & Higgins, the field finds itself with a hypothesis that is simultaneously the most pharmacologically vindicated and the most criticised in modern neurology. The honest assessment requires holding both at once.

What went wrong (1999–2021)

For two decades, anti-amyloid drugs failed in trial after trial — not quietly, but spectacularly:

The critique — from inside and out

• Plaque/cognition disconnect. Cognitively normal elderly with substantial plaque burden; plaques in regions that don’t correlate with deficits.
• Tau-cognition correlation is stronger. Tau-PET tracks symptoms more closely than amyloid-PET; tangles, not plaques, predict regional atrophy.
• Reductionism. Late-onset sporadic AD is multifactorial — immune, vascular, metabolic, lipid (APOE), TREM2 microglial. The peptide-only narrative may have been a useful simplification, not a complete causal story.
• Replication concerns. The 2022 Schrag/Piller/Science investigation into Lesné’s Aβ*56 papers (and others) prompted a sober reassessment of how much specific oligomer literature was load-bearing.
• Industrial selection bias. A generation of medicinal chemistry and trial design optimised for amyloid; alternatives (tau, inflammation, mitochondria) were comparatively under-resourced.

The biomarker successes

While drugs failed, the diagnostic infrastructure built on the cascade succeeded spectacularly:

• Amyloid PET (PiB, florbetapir, florbetaben, flutemetamol) — in vivo plaque visualisation; transformed trials.
• CSF Aβ42/40 + p-tau181/p-tau217 — sensitivities and specificities ≥90% for AD pathology.
• Plasma p-tau217 (Janelidze, Hansson, Ashton; 2020–2024) — AUCs ~0.95 for amyloid status; the disease-defining Pittsburgh Compound B in a blood draw.
• NIA-AA biological framework (2018, 2024) — AD redefined biologically by ATN, severing “Alzheimer’s” from “dementia”.

The partial vindication: lecanemab and donanemab

Two trials, both reported in 2023, broke the failure pattern:

What 30 years has taught us

1. Amyloid is causal, but not sufficient. The cascade starts the disease; tau, inflammation, and APOE-driven mechanisms drive the back half.
2. Stage matters enormously. Anti-amyloid therapy works in MCI and mild AD; the same drugs in moderate AD do not.
3. Species matters. Antibodies against soluble protofibrils (lecanemab) and pyroglutamate-Aβ in plaques (donanemab) succeed where antibodies against mid-region monomer (solanezumab) fail.
4. Production-side suppression is dangerous. Chronic BACE1 or γ-secretase inhibition impairs cognition, probably through off-target substrates and loss of physiological Aβ.
5. Effect sizes are modest. 25–35% slowing on CDR-SB is real and disease-modifying, but not transformative. Combining anti-amyloid with anti-tau and anti-inflammatory therapy is the next frontier.

The next two parts of the course follow this conceptual hand-off: Part IV (Tau) explores the protein that does most of the work after Aβ has lit the fuse, and Part V (Genetics) returns to the APP/PSEN/APOE story to ask why these particular mutations and alleles drive the same molecular cascade.