Skip to content

Amyloid biology and the cascade hypothesis

TL;DR — The amyloid cascade hypothesis, stated in 1992 and restated in 2002, holds that Aβ accumulation is the primary driver and everything else — tangles, inflammation, cell loss — is downstream (Hardy 1992, PMID 1566067; Hardy 2002, PMID 12130773). Its strongest genetic support is that every dominant early-onset mutation lies in the substrate (APP) or the protease (presenilin/γ-secretase) of the reaction that makes Aβ, and that a coding variant adjacent to the β-cleavage site (APP A673T) reducing amyloidogenic peptide formation by ~40% in vitro protects against both AD and age-related cognitive decline (Selkoe 2016, PMID 27025652; Jonsson 2012, PMID 22801501). Its weakest points are equally quantitative: cognitive severity correlates better with neocortical neurofibrillary tangle burden than with plaque burden (Nelson 2012, PMID 22487856); amyloid positivity reaches 44% (95% CI 37–51) in cognitively normal 90-year-olds (Jansen 2015, PMID 25988462); and near-complete plaque removal by AN1792 immunisation did not prevent progression to severe dementia in seven of eight autopsied participants (Holmes 2008, PMID 18640458). Late-onset AD is a clearance disease, not an overproduction disease — metabolic labelling showed impaired Aβ40 and Aβ42 clearance with no average difference in production (Mawuenyega 2010, PMID 21148344). Attempts to reduce production pharmacologically have failed or harmed: semagacestat worsened function (ADCS-ADL −12.6 vs −9.0 points at 140 mg, P<0.001) and caused skin cancers and infections, and both verubecestat and lanabecestat produced treatment-associated cognitive worsening (Doody 2013, PMID 23883379; Wessels 2020, PMID 33049114). The hypothesis in its 2020s form is therefore narrower than in 1992: Aβ dyshomeostasis is an early, often initiating factor whose therapeutic window may close well before symptoms.

APP processing: where Aβ comes from

Amyloid precursor protein is a type-I transmembrane protein cleaved along two competing routes. In the non-amyloidogenic route, α-secretase cleaves within the Aβ sequence, precluding Aβ formation. In the amyloidogenic route, β-secretase (BACE1) cuts at the Aβ N-terminus and γ-secretase — whose catalytic subunit is presenilin — then cuts within the transmembrane domain at variable positions, generating peptides mainly 38–43 residues long. The C-terminal heterogeneity is the biologically decisive step: Aβ42 and Aβ43 aggregate far more readily than Aβ40, which is why the Aβ42/Aβ40 ratio rather than Aβ42 alone became the diagnostic quantity in fluid biomarkers.

The peptide itself was identified from cerebrovascular amyloid in 1984: Glenner and Wong purified a protein from twisted β-pleated-sheet fibrils in AD-associated cerebral amyloid angiopathy, found by sequence analysis and computer search that it had no homology with any protein sequenced to that date, and proposed a unique serum precursor (Glenner 1984, PMID 6375662). Genetics then placed the gene on the causal path. In a single autopsy-confirmed kindred linked to chromosome 21, a point mutation in APP causing a Val→Ile substitution near the carboxy terminus of the Aβ peptide co-segregated with disease after earlier recombinants had seemed to exclude the locus; the variant was then found in a second unrelated family. The authors' claim was correspondingly measured — that some cases of AD could be caused by APP mutations (Goate 1991, PMID 1671712). The AD3 locus on 14q24.3, mapped to a very aggressive form of the disease, was cloned as S182 — later presenilin-1 — with five missense mutations in conserved domains co-segregating with early-onset familial AD and absent from controls (Sherrington 1995, PMID 7596406). Confirmation that presenilin is the catalytic site of γ-secretase supplied what Selkoe called the linchpin of the genetic argument: all dominant early-onset mutations occur either in the substrate (APP) or in the protease (presenilin), and duplication of wild-type APP in Down syndrome produces Aβ deposits in the teens followed by microgliosis, astrocytosis and tangles (Selkoe 2016, PMID 27025652). See genetics for the mutation catalogue and the Down syndrome trajectory.

Soluble oligomers, rather than deposited fibrils, are the species with the most direct synaptic physiology. Naturally secreted human Aβ oligomers, formed intracellularly and released into medium that contains abundant monomers but no fibrils, markedly inhibit hippocampal LTP in rats in vivo; immunodepleting all Aβ species abolishes the effect, whereas degrading monomers with insulin-degrading enzyme — leaving oligomers intact — does not (Walsh 2002, PMID 11932745). That experiment is why the cascade's 2020s form emphasises oligomers over plaque load, and why antibodies directed at monomers (solanezumab) and antibodies directed at deposited plaque have different predicted mechanisms.

The protective mutation

Whole-genome sequencing of 1,795 Icelanders identified APP A673T, immediately adjacent to the β-secretase site, which reduces formation of amyloidogenic peptides by approximately 40% in vitro and protects against both AD and cognitive decline in the elderly without AD (Jonsson 2012, PMID 22801501). This is the closest thing the field has to a natural randomised experiment on lifelong β-cleavage reduction, and it is the main reason BACE inhibition looked reasonable a priori.

Production versus clearance

Metabolic labelling with stable isotopes in living participants measured Aβ40 and Aβ42 production and clearance rates directly. In late-onset AD, clearance of both peptides was impaired relative to cognitively normal controls, while average production rates did not differ (Mawuenyega 2010, PMID 21148344). This reframes sporadic AD: the therapeutic target is not synthesis but removal, and it explains why an APOE ε4 effect on Aβ clearance (Selkoe 2016, PMID 27025652) and the near-full penetrance of AD biology in APOE4 homozygotes (Fortea 2024, PMID 38710950) sit at the centre of the risk architecture.

Clearance routes and their evidence status:

Route Mechanism Human evidence in AD
Proteolytic degradation Neprilysin, insulin-degrading enzyme and others Indirect; part of the metabolic-clearance term measured in vivo (PMID 21148344)
Receptor-mediated transport across the blood–brain barrier LRP1 efflux, RAGE influx; ApoE-dependent ApoE4 impairs Aβ clearance from brain (PMID 27025652)
Perivascular / glymphatic and meningeal lymphatic drainage Bulk fluid transport, sleep-dependent One night of total sleep deprivation increased florbetaben binding in right hippocampus and thalamus in 20 healthy adults; baseline burden in subcortical regions and precuneus was inversely associated with reported sleep hours (Shokri-Kojori 2018, PMID 29632177)
Perivascular deposition rather than clearance Failure of perivascular drainage produces cerebral amyloid angiopathy CAA is a common copathology and the substrate of ARIA — see vascular and metabolic contributions

The sleep-deprivation experiment is small (n=20, one night) and used a PET signal rather than a clearance rate, so it supports a mechanistic link rather than quantifying it; it is nonetheless one of the few interventional human data points on the clearance side, and it is the bridge to sleep as a modifiable factor on risk reduction and prevention.

Deposition follows a hierarchy

Aβ deposition is not diffuse and random. Examining 47 brains covering all phases, Thal defined five phases: (1) neocortex only; (2) allocortex added; (3) diencephalic nuclei, striatum and cholinergic basal forebrain; (4) brainstem nuclei; (5) cerebellum. All 17 clinically confirmed AD cases were in phases 3–5, while the nine non-demented cases with AD-related Aβ pathology were in phases 1–3, and deposition expanded anterogradely into regions receiving projections from already-affected regions (Thal 2002, PMID 12084879).

That anterograde pattern implies transmission of an aggregation-competent species. Intracerebral injection of human brain homogenates into APP-transgenic mice showed that propagation speed tracked the donor's Aβ phase, while the speed of maturation (post-translational modification of aggregates) tracked the donor's aggregate maturation stage — and, critically for drug design, different Aβ forms could trigger propagation, "which may explain the lack of success when therapeutically targeting only specific forms of Aβ" (Li 2022, PMID 36270003). The same seeding logic is developed further, with stronger clinical correlation, on tau biology and spread.

The case against a linear cascade

Observation Quantity Implication
Tangles, not plaques, track cognition Severity of cognitive impairment correlates best with neocortical neurofibrillary tangle burden across many centres Aβ may initiate without being the proximate cause of symptoms (Nelson 2012, PMID 22487856)
Amyloid positivity without dementia is common 44% (37–51) of cognitively normal 90-year-olds; 10% (8–13) at age 50 Amyloid is necessary-ish but far from sufficient (Jansen 2015, PMID 25988462)
Plaque removal without clinical benefit AN1792: mean Aβ load 2.1% (SE 0.7) in immunised vs 5.1% (0.9) in matched unimmunised controls (difference 3.0%, 95% CI 0.6–5.4, P=0.02); 7 of 8 autopsied immunised participants, including those with virtually complete plaque removal, had severe end-stage dementia; no survival benefit (HR 0.93, 0.43–3.11) or delay to severe dementia (1.18, 0.45–3.11) Clearing deposited plaque late in disease does not stop neurodegeneration (Holmes 2008, PMID 18640458)
Production inhibition harms Semagacestat: ADAS-cog worsened in all arms (6.4 placebo, 7.5 at 100 mg, 7.8 at 140 mg); ADCS-ADL −9.0, −10.5, −12.6 (P<0.001 for 140 mg); more skin cancers, infections and serious adverse events (P<0.001) γ-secretase has essential substrates (Notch); the pathway is not safely switchable (Doody 2013, PMID 23883379)
BACE inhibition worsens cognition Verubecestat 12 and 40 mg in mild-to-moderate AD (EPOCH, n=1,958, 78 weeks, stopped for futility): ADAS-cog change 7.9 / 8.0 vs 7.7 placebo (P=0.63 and 0.46); ADCS-ADL −8.4 / −8.2 vs −8.9 (P=0.49 and 0.32), with more rash, falls, sleep disturbance, suicidal ideation, weight loss and hair-colour change (Egan 2018, PMID 29719179; NCT01739348). In prodromal AD (APECS, n=1,454, 104 weeks, stopped for futility) the 40 mg arm was worse than placebo on CDR-SB (2.02 vs 1.58, P=0.01) and progressed to dementia faster (HR 1.38, 97.51% CI 1.07–1.79, unadjusted) (Egan 2019, PMID 30970186; NCT01953601). A pooled cognitive analysis of verubecestat and lanabecestat confirmed worsening on composite, memory and attention indices (Wessels 2020, PMID 33049114) Chronic profound BACE1 inhibition is not tolerated cognitively, including before dementia
Antibody target engagement is superficial In five aducanumab-treated autopsies, Aβ clearance was localised to cortical layer I with no significant clearance in deeper layers, despite Centiloid reductions of −6% to −81% PET amyloid reduction is not a faithful measure of parenchymal clearance (Boon 2025, PMID 41109234)

Herrup's rejection argument is that the linear structure has accumulated too many inconsistencies and that the field is "over-reliant on amyloid to define and diagnose AD" — a claim that is now partly institutionalised in the biological definition of the disease (Herrup 2015, PMID 26007212; see diagnostic criteria). Karran's contemporaneous review reaches a different verdict from similar data: that the hypothesis has been productive and evolvable, while conceding that every amyloidocentric drug to that date had missed its primary endpoint (Karran 2016, PMID 27255958).

What survived, and in what form

The defensible 2020s statement of the hypothesis is narrower than the 1992 one:

  1. Aβ dyshomeostasis is early and often initiating, evidenced by low CSF Aβ42 and amyloid-PET positivity preceding other manifestations by many years — about 25 years for CSF Aβ42 in dominantly inherited disease (Bateman 2012, PMID 22784036; Selkoe 2016, PMID 27025652).
  2. Soluble oligomers, not fibrillar plaque burden, carry much of the synaptic toxicity: Aβ42 oligomers isolated from AD brain reduce synapse number, inhibit LTP, enhance LTD, impair memory when injected into rats, and induce tau hyperphosphorylation at AD-relevant epitopes (Selkoe 2016, PMID 27025652).
  3. Aβ acts largely by enabling tau, which is the pathology that tracks symptoms (Nelson 2012, PMID 22487856).
  4. The therapeutic window is upstream. Every failure above was in symptomatic disease, and the one class with a positive primary endpoint removes fibrillar amyloid in early symptomatic disease with modest effect sizes — see anti-amyloid immunotherapy.

The unresolved question is whether the residual effect sizes of successful amyloid removal reflect a correct-but-late intervention or a partially correct target. Both readings are consistent with the current data, and this knowledge base does not adjudicate between them.

Where the evidence is thin

  • Which Aβ species matters. Propagation and maturation are driven by different aggregate forms, and targeting one form may leave the seeding process intact (Li 2022, PMID 36270003).
  • What impaired clearance means mechanistically. The in-vivo labelling result is a rate, not a route; which of proteolysis, transport and perivascular drainage fails is not established in humans (Mawuenyega 2010, PMID 21148344).
  • Whether PET Centiloid change measures what it is used to measure. Layer-I-restricted clearance with large Centiloid reductions is a direct challenge to the surrogate (Boon 2025, PMID 41109234), though it rests on five treated autopsies.

Two mechanistic distinctions that matter therapeutically

Human-brain material strengthens the oligomer argument beyond cell culture. Soluble extracts from AD cortex inhibited long-term potentiation, enhanced long-term depression, reduced dendritic-spine density and disrupted learned behaviour when administered to rodents; immunochemical fractionation attributed these effects specifically to Aβ dimers. Insoluble plaque cores were inactive until solubilised to release dimers, making plaques a possible reservoir rather than the directly toxic species in this assay (Shankar 2008, PMID 18568035). This does not establish that dimers are the only toxic assembly in living humans, but it explains why plaque burden and synaptic failure can dissociate.

Production biology also differs between sporadic and presenilin-mutation disease. Stable-isotope kinetic modelling found the CNS Aβ42:Aβ40 production-rate ratio 24% higher in PSEN1/2 mutation carriers than non-carriers, independent of PET plaque burden; soluble Aβ42 fractional turnover relative to Aβ40 was 65% faster and tracked deposition, consistent with exchange into plaques (Potter 2013, PMID 23761040). That contrasts with reduced clearance without higher average production in late-onset AD (Mawuenyega 2010, PMID 21148344). “Amyloid excess” is therefore not one kinetic defect, and a production inhibitor and a clearance antibody do not test the same mechanistic proposition.

Open questions

  • Is amyloid removal beneficial because of the removed plaque or because of a downstream effect on tau, and can the two be dissociated experimentally (Nelson 2012, PMID 22487856; Selkoe 2016, PMID 27025652)?
  • Which clearance route fails first in sporadic AD, and is any of them therapeutically accessible (Mawuenyega 2010, PMID 21148344; Shokri-Kojori 2018, PMID 29632177)?
  • Does the layer-I restriction of antibody-mediated clearance generalise beyond aducanumab, and does it explain the gap between large Centiloid reductions and small clinical effects (Boon 2025, PMID 41109234)?
  • Can β-cleavage be reduced by ~40% chronically — the A673T phenotype — without the cognitive worsening seen with BACE inhibitors, i.e. is the harm dose-related or mechanism-intrinsic (Jonsson 2012, PMID 22801501; Wessels 2020, PMID 33049114)?
  • Why did complete plaque removal in AN1792 fail to alter the clinical course — irreversibility at that stage, wrong species targeted, or the wrong disease model (Holmes 2008, PMID 18640458; Herrup 2015, PMID 26007212)?
  • What determines the 20–30 year interval between amyloid positivity and dementia, and is the determinant modifiable (Jansen 2015, PMID 25988462)?

References

  1. Hardy JA, et al. Alzheimer's disease: the amyloid cascade hypothesis. Science. 1992;256:184-5. PMID 1566067.
  2. Hardy J, et al. The amyloid hypothesis of Alzheimer's disease: progress and problems on the road to therapeutics. Science. 2002;297:353-6. PMID 12130773.
  3. Selkoe DJ, et al. The amyloid hypothesis of Alzheimer's disease at 25 years. EMBO Mol Med. 2016;8:595-608. PMID 27025652.
  4. Karran E, et al. The amyloid cascade hypothesis: are we poised for success or failure? J Neurochem. 2016;139 Suppl 2:237-252. PMID 27255958.
  5. Herrup K. The case for rejecting the amyloid cascade hypothesis. Nat Neurosci. 2015;18:794-9. PMID 26007212.
  6. Mawuenyega KG, et al. Decreased clearance of CNS beta-amyloid in Alzheimer's disease. Science. 2010;330:1774. PMID 21148344.
  7. Jonsson T, et al. A mutation in APP protects against Alzheimer's disease and age-related cognitive decline. Nature. 2012;488:96-9. PMID 22801501.
  8. Holmes C, et al. Long-term effects of Aβ42 immunisation in Alzheimer's disease: follow-up of a randomised, placebo-controlled phase I trial. Lancet. 2008;372:216-23. PMID 18640458.
  9. Doody RS, et al. A phase 3 trial of semagacestat for treatment of Alzheimer's disease. N Engl J Med. 2013;369:341-50. PMID 23883379.
  10. Wessels AM, et al. Cognitive outcomes in trials of two BACE inhibitors in Alzheimer's disease. Alzheimers Dement. 2020;16:1483-1492. PMID 33049114.
  11. Boon BDC, et al. Neuropathological changes and amyloid-related imaging abnormalities in Alzheimer's disease treated with aducanumab versus untreated: a retrospective case-control study. Lancet Neurol. 2025;24:931-944. PMID 41109234.
  12. Thal DR, et al. Phases of Aβ-deposition in the human brain and its relevance for the development of AD. Neurology. 2002;58:1791-800. PMID 12084879.
  13. Nelson PT, et al. Correlation of Alzheimer disease neuropathologic changes with cognitive status: a review of the literature. J Neuropathol Exp Neurol. 2012;71:362-81. PMID 22487856.
  14. Li X, et al. Seeding, maturation and propagation of amyloid β-peptide aggregates in Alzheimer's disease. Brain. 2022;145:3558-3570. PMID 36270003.
  15. Shokri-Kojori E, et al. β-Amyloid accumulation in the human brain after one night of sleep deprivation. Proc Natl Acad Sci U S A. 2018;115:4483-4488. PMID 29632177.
  16. Bateman RJ, et al. Clinical and biomarker changes in dominantly inherited Alzheimer's disease. N Engl J Med. 2012;367:795-804. PMID 22784036.
  17. Fortea J, et al. APOE4 homozygozity represents a distinct genetic form of Alzheimer's disease. Nat Med. 2024;30:1284-1291. PMID 38710950.
  18. Jansen WJ, et al. Prevalence of cerebral amyloid pathology in persons without dementia: a meta-analysis. JAMA. 2015;313:1924-38. PMID 25988462.
  19. Glenner GG, et al. Alzheimer's disease: initial report of the purification and characterization of a novel cerebrovascular amyloid protein. Biochem Biophys Res Commun. 1984;120:885-90. PMID 6375662.
  20. Goate A, et al. Segregation of a missense mutation in the amyloid precursor protein gene with familial Alzheimer's disease. Nature. 1991;349:704-6. PMID 1671712.
  21. Sherrington R, et al. Cloning of a gene bearing missense mutations in early-onset familial Alzheimer's disease. Nature. 1995;375:754-60. PMID 7596406.
  22. Walsh DM, et al. Naturally secreted oligomers of amyloid beta protein potently inhibit hippocampal long-term potentiation in vivo. Nature. 2002;416:535-9. PMID 11932745.
  23. Egan MF, et al. Randomized trial of verubecestat for mild-to-moderate Alzheimer's disease. N Engl J Med. 2018;378:1691-1703. PMID 29719179.
  24. Egan MF, et al. Randomized trial of verubecestat for prodromal Alzheimer's disease. N Engl J Med. 2019;380:1408-1420. PMID 30970186.
  25. Shankar GM, et al. Amyloid-beta protein dimers isolated directly from Alzheimer's brains impair synaptic plasticity and memory. Nat Med. 2008;14:837-42. PMID 18568035.
  26. Potter R, et al. Increased in vivo amyloid-β42 production, exchange, and loss in presenilin mutation carriers. Sci Transl Med. 2013;5:189ra77. PMID 23761040.