The Amyloid Hypothesis: Explained and Tested

The amyloid hypothesis has been described as proven, disproven, revived, and dead. If you have read all four in the same year, that is because none of the one-word verdicts captures the evidence.

Genetic findings show that changing amyloid-beta production or processing can cause rare inherited Alzheimer disease. Biomarker studies place amyloid accumulation early in the usual disease sequence. Decades of failed treatments show that lowering the wrong amyloid form, at the wrong stage, by too little, or with unacceptable toxicity may not help. Recent antibody trials show that substantial amyloid removal in selected people with early symptomatic disease can slow clinical decline by a modest amount.

The modern question is no longer whether amyloid matters at all. It is how amyloid interacts with tau, immune cells, blood vessels, synapses, aging, and resilience, and how much clinical benefit you can buy by changing it at each stage.

What “amyloid” means#

Amyloid-beta, abbreviated Aβ, is a peptide cut from a larger cell-membrane protein called amyloid precursor protein, or APP. Different cleavage pathways produce peptides of different lengths. Aβ42 has a greater tendency to aggregate than Aβ40.

The peptide can exist as soluble single molecules, small assemblies called oligomers, fibrils, and plaques. Cerebral amyloid angiopathy places amyloid in blood-vessel walls. These forms may not have the same biological effect.

An antibody can target soluble protofibrils, deposited plaque, a particular altered end of the peptide, or several aggregated forms, and a trial that fails against one species does not test every version of the hypothesis. Conversely, a successful antibody does not prove that every amyloid form is toxic or that plaque itself is the sole cause of symptoms. That distinction became important as research moved away from a simple picture of inert plaque physically filling the brain, and soluble aggregates, cell-membrane effects, synaptic signaling, microglial response, and vascular amyloid are all now part of the model.

The cascade in its strongest form#

The original cascade proposed that amyloid accumulation sits upstream. It promotes tau phosphorylation and aggregation, synaptic failure, inflammation, neuronal death, and finally cognitive and functional decline.

An updated version is less linear. Amyloid may create a permissive state in which tau spreads through vulnerable neural networks. APOE genotype, microglia, lipid biology, vascular integrity, metabolism, sleep, and aging modify the pathway. Once tau and neurodegeneration become established, removing amyloid may not reverse them.

This timing produces a long preclinical phase. Amyloid biomarkers can become abnormal years before noticeable symptoms, while tau burden and neurodegeneration generally track clinical stage more closely once symptoms appear. So the hypothesis is a claim about causal ordering. It is not a claim that a plaque scan can measure current disability by itself.

Genetics supplies the strongest causal argument#

Rare autosomal-dominant mutations in APP, PSEN1, or PSEN2 can cause early-onset familial Alzheimer disease. Presenilins are part of gamma-secretase, one of the enzymes that processes APP. Disease-causing variants can alter the amount or proportions of amyloid-beta peptides.

People with trisomy 21 have an extra copy of chromosome 21, which includes APP, and commonly develop amyloid pathology and Alzheimer disease with age. Conversely, a rare APP variant that reduces amyloidogenic cleavage is protective against Alzheimer disease and age-related cognitive decline.

These gene-dose and mutation findings are hard to explain if amyloid processing is irrelevant. They establish causality for particular genetic pathways. They do not prove that the much more common late-onset disease begins identically in every person.

APOE epsilon 4 is the strongest common genetic risk factor for late-onset Alzheimer disease. APOE affects lipid transport, amyloid clearance and aggregation, inflammation, and vascular amyloid; its broad effects fit an amyloid-centered model but also reach beyond amyloid, and many other risk loci involve microglia, endosomal trafficking, lipids, and immunity. The genetics argue for a network, not one toxic molecule acting alone.

Pathology creates both support and tension#

Amyloid plaques and tau neurofibrillary tangles define Alzheimer neuropathologic change. Amyloid PET and cerebrospinal-fluid or blood biomarkers can now identify aspects of this biology during life.

Amyloid often appears first in association cortex and becomes widespread. Its amount then reaches a relative plateau. Tau spreads in a pattern that more closely follows neuronal dysfunction and cognitive impairment, and some older adults have substantial amyloid with preserved cognition, while others have severe symptoms shaped by mixed pathologies.

Brain reserve and cognitive reserve can delay clinical expression. Lewy bodies, TDP-43, hippocampal sclerosis, small-vessel disease, infarcts, and cerebral amyloid angiopathy can add to impairment. A person given the clinical label “Alzheimer dementia” may have several contributors.

The mismatch between plaques and symptoms weakens the strongest claim that plaque burden directly determines dementia, though it does not erase the possibility that amyloid initiated or accelerated a downstream process years earlier.

What the long history of failed trials taught#

Anti-amyloid development included vaccines, antibodies, secretase inhibitors, and aggregation modifiers, and many programs reduced a biomarker without improving cognition, failed to lower brain amyloid enough, caused toxicity, or were tested after substantial dementia was established.

Several explanations can be true:

• The target was the wrong amyloid species.
• The intervention changed its intended target too little.
• Treatment began after downstream injury became self-sustaining.
• The dose needed for biology caused unacceptable harm.
• The clinical endpoint or trial duration was insensitive.
• Amyloid was not the dominant driver in that population.
• The broader hypothesis overstated how much benefit amyloid lowering could deliver.

Timing and target arguments should be prespecified and tested, not used to excuse every negative result after the fact. A theory that can accommodate any outcome without changing becomes difficult to falsify. What the failures did was narrow the hypothesis: they showed that amyloid reduction is not automatically a validated surrogate for clinical benefit, and that symptomatic Alzheimer disease cannot be reduced to plaque load.

Lecanemab: absolute change behind the relative headline#

Clarity AD randomized 1,795 people with mild cognitive impairment or mild dementia due to Alzheimer disease and confirmed amyloid pathology to lecanemab or placebo for 18 months.

The primary outcome was the Clinical Dementia Rating Sum of Boxes, or CDR-SB, which ranges from 0 to 18 and combines cognitive and functional domains. The mean baseline score was about 3.2. At 18 months, the adjusted mean increase was 1.21 points with lecanemab and 1.66 with placebo. The between-group difference was -0.45 points, with a 95% confidence interval from -0.67 to -0.23.

The 27% slowing you see quoted is a relative calculation based on the difference in decline. The absolute difference of 0.45 points shows the scale and time horizon. Participants still worsened on average. The trial demonstrated slowing, not improvement, stabilization for everyone, or reversal of dementia.

Key secondary cognitive, functional, and amyloid outcomes generally supported the primary result. Whether that average difference would be meaningful to you, or to the person you care for, remains genuinely debated, because CDR-SB thresholds vary by stage, rate, and perspective. Distribution of response and time gained may be more useful than one mean at one visit.

Donanemab: tau stratification and treatment stopping#

TRAILBLAZER-ALZ 2 randomized 1,736 people with early symptomatic, amyloid-positive Alzheimer disease and tau PET eligibility. Treatment could stop after amyloid PET met prespecified clearance criteria.

In the low-to-medium tau population, the integrated Alzheimer Disease Rating Scale declined by 6.02 points with donanemab and 9.27 with placebo at 76 weeks. The difference was 3.25 points, described as 35.1% relative slowing.

In the combined population, which included high tau, decline was 10.19 versus 13.11 points. The difference was 2.92 points, described as 22.3% slowing. The smaller relative effect in the broader population is consistent with less benefit after downstream disease is more advanced, though subgroup comparisons need care.

The CDR-SB difference in the overall population was -0.70 points, and other measures supported slower decline. As with lecanemab, the result was not recovery. Both groups worsened. Amyloid clearance did allow many participants to complete active treatment early, which tests a finite-treatment strategy, but it does not prove that amyloid can never reaccumulate or that downstream disease stops after treatment ends.

ARIA is part of the causal intervention#

Amyloid antibodies can cause amyloid-related imaging abnormalities. ARIA-E refers to edema or fluid collections. ARIA-H refers to microhemorrhage or superficial siderosis from blood products. Most imaging findings are asymptomatic, but headache, confusion, visual change, dizziness, gait difficulty, focal deficits, seizure, serious hemorrhage, and death can occur.

In Clarity AD, ARIA-E occurred in 12.6% with lecanemab and 1.7% with placebo. ARIA-H occurred in 17.3% and 9.0%. Infusion reactions occurred in 26.4% and 7.4%.

In TRAILBLAZER-ALZ 2, ARIA-E occurred in 24.0% with donanemab and 2.1% with placebo using the original trial titration. Symptomatic ARIA-E occurred in about 6% of treated participants, and rare fatal events occurred. A later titration study reduced ARIA-E to about 16% in the recommended schedule while achieving similar amyloid reduction over the measured period.

The current lecanemab and donanemab labels carry boxed ARIA warnings. APOE epsilon 4 homozygotes have higher symptomatic and serious ARIA risk, and testing before treatment is recommended to inform risk. Baseline microhemorrhages, superficial siderosis, anticoagulation, and thrombolytic decisions require specialized review.

In 2025, FDA recommended an additional earlier MRI before the third lecanemab infusion after serious early ARIA cases. That change illustrates how postmarket evidence can refine monitoring after approval.

What the positive trials say about the hypothesis#

Lecanemab and donanemab did three important things in the same randomized experiments: engaged amyloid strongly, changed downstream biomarkers to varying degrees, and slowed clinical decline. That pattern supports amyloid as a causal contributor in early biomarker-confirmed Alzheimer disease.

The size of clinical change was much smaller than the size of plaque removal. This argues against amyloid burden as a complete measure of disease, or as a number you can read off a scan and translate into how one person will do. It also suggests that removing amyloid after symptoms begin cannot undo established tau pathology, synaptic loss, or other injury.

The most defensible conclusion is a modified hypothesis: amyloid is an important upstream and ongoing process in Alzheimer disease, but it is neither sufficient to explain clinical state nor sufficient as a treatment target for large recovery.

That conclusion remains open to refinement. Longer follow-up must show whether modest separation persists, widens, narrows, or is offset by harms and by what the treatment costs you in monitoring and time. Trials in asymptomatic biomarker-positive people will test whether earlier intervention produces a larger delay, but preventive treatment raises even stricter safety and overdiagnosis questions.

Diagnosis and treatment require more than an amyloid test#

Amyloid positivity becomes more common with age and can occur in people without cognitive impairment. A positive PET, cerebrospinal-fluid result, or blood biomarker does not by itself establish that Alzheimer pathology explains the symptoms you came in with.

Clinical assessment should establish objective impairment, functional stage, alternative causes, medication effects, mood and sleep conditions, vascular disease, and the presence of other neurologic signs. Biomarkers then refine etiology and treatment eligibility.

FDA labels initiate anti-amyloid treatment in mild cognitive impairment or mild dementia, the stages studied, after confirmation of amyloid pathology. They do not establish use in moderate or severe dementia or as a general prevention medicine. MRI eligibility, APOE counseling, infusion or injection logistics, monitoring, emergency access, other medicines, and what you and your care partner actually want from treatment all affect the benefit-risk balance.

Sudden headache, confusion, visual change, weakness, speech difficulty, seizure, or gait change during anti-amyloid therapy needs urgent assessment. ARIA can mimic ischemic stroke, so emergency teams need the medication history before thrombolytic decisions.

A hypothesis that became more precise#

Science rarely resolves a complex disease with one decisive experiment. The amyloid hypothesis survived some tests, failed others, and changed as genetics, biomarkers, pathology, and trials accumulated.

The evidence rejects two extremes. Amyloid is not an irrelevant bystander, and amyloid removal is not a cure. It is one early causal process in a heterogeneous disease system. Better treatment may require earlier selection, safer amyloid modification, tau-directed therapy, neuroimmune and vascular approaches, and protection of synapses and function.

References#

1. Updated critical review of the amyloid cascade hypothesis
2. Clarity AD lecanemab trial
3. TRAILBLAZER-ALZ 2 donanemab trial
4. Current FDA lecanemab label
5. Current FDA donanemab label
6. FDA early lecanemab MRI safety update
7. Systematic review of anti-amyloid clinical benefits and harms

For your own health, talk with your clinician.*