The Amyloid Hypothesis in Alzheimer Disease, Explained and Tested

The idea, and the test it finally faced

The amyloid hypothesis proposes that Alzheimer disease begins when a protein fragment called beta-amyloid accumulates in the brain, setting off a chain of events that ends in dementia. For three decades this was the dominant framework, yet it rested largely on genetics and pathology rather than on a direct experiment. That has changed. Anti-amyloid antibodies such as lecanemab and donanemab remove amyloid from the living brain to a striking degree, which lets researchers ask a sharp question: when you clear the plaque, how much does the person actually benefit? The answer so far is that amyloid can be cleared dramatically while cognitive decline slows only modestly, which both supports the hypothesis and exposes its limits.

This article is educational and not medical advice.

Where the hypothesis came from

The amyloid cascade hypothesis was named by John Hardy and Gerald Higgins in a short 1992 paper in Science. Their argument drew its force from human genetics. Mutations in the gene for amyloid precursor protein, and later in the presenilin genes that help cut that protein, cause aggressive early-onset Alzheimer disease. People with an extra copy of chromosome 21, which carries the amyloid precursor gene, develop amyloid pathology decades early. The logic was clean: if changing amyloid production causes the disease, amyloid must sit near the front of the causal chain.

From there the cascade was drawn as a sequence. Beta-amyloid accumulates and forms plaques, which trigger changes in a second protein called tau, which forms tangles inside neurons, which drive inflammation, synapse loss, and cell death. Amyloid was cast as the upstream trigger and tau as the closer partner to symptoms.

Why the biology was never simple

Two facts complicated the story from early on. First, the amount of amyloid plaque in the brain correlates poorly with how impaired a person is; tangle burden tracks symptoms far more closely. Second, many older adults carry substantial amyloid with no dementia at all. These observations did not refute the hypothesis, but they suggested amyloid might be necessary without being sufficient, and that timing matters. Amyloid may do its damage early, upstream, long before symptoms, which would mean removing it late could be too little, too late.

Turning a hypothesis into an experiment

A hypothesis becomes testable when you can intervene on the proposed cause and measure the effect. Antibodies made that possible. These are engineered proteins that bind amyloid and recruit the immune system to clear it, and different antibodies bind different forms, from soluble clumps to deposited plaque. Earlier candidates that bound the wrong species, or bound too weakly, failed to help patients. The more recent agents share a design feature: they engage aggregated amyloid strongly, and they lower amyloid measured by PET imaging.

That imaging readout is the key to a concept called target engagement, meaning direct proof that a drug hits its intended target. In a 2024 systematic review and meta-analysis published in Scientific Reports, the authors found that anti-amyloid antibodies as a class reduced amyloid on PET by a very large effect while improving the clinical rating scales by only a very small effect. The reduction in amyloid correlated moderately with the slowing of decline. That single sentence is the cleanest test of the hypothesis available: hit the target hard, move the patient a little.

Reading the pivotal trials honestly

Two trials anchor the current picture. In the lecanemab program, treatment slowed decline on a global dementia scale by roughly 27 percent over 18 months. In the donanemab program, the figure was about 35 percent. A 2024 commentary in eNeuro translated these relative numbers into time, estimating that lecanemab delayed progression by roughly six months and donanemab by roughly four months across the trial period. Both drugs cleared amyloid substantially. Both met their primary endpoints. Both effects are real and statistically robust.

They are also modest, and the same commentary is careful to say so. A percentage slowing of decline is not a halt and not a reversal; the treated group still worsened, just less quickly. Whether the gap between treated and untreated widens, holds, or shrinks after the trial ends is not yet established. The eNeuro authors proposed communicating benefit in plain time-based terms rather than abstract point differences, precisely because a small shift on a rating scale is hard for a patient to interpret.

The cost side of the ledger

Target engagement carries a signature risk. Removing amyloid from vessel walls and tissue can cause amyloid-related imaging abnormalities, known as ARIA, which show up as brain swelling or small bleeds on MRI. The Scientific Reports meta-analysis found a very large increase in the swelling form of ARIA and a moderate increase in the bleeding form. Most cases are detected on scans without symptoms, but serious events occur, and the risk is higher in people who carry two copies of the APOE4 gene variant. This is why monitoring, genotype awareness, and shared decision-making are central to the category rather than optional.

What the test actually showed

The honest reading is that the amyloid hypothesis passed its experiment, but narrowly. Intervening on amyloid changes the clinical course, which no amount of correlation could ever prove and which failed antibodies could not deliver. That is a genuine scientific result. At the same time, the size of the benefit tells us amyloid is not the whole disease. If clearing most of the plaque yields a partial slowing, then tau, inflammation, vascular injury, and other processes are carrying much of the load by the time symptoms appear.

This reframes rather than discards the hypothesis. It points toward treating earlier, before the cascade gains momentum, and toward combining amyloid removal with agents aimed at tau or inflammation. The next decade of trials will test that combination logic. For now, the field has done something it could not do for thirty years: it turned a compelling idea into a measurable answer, and the answer is more instructive for being incomplete.