What amyloid and tau actually do in the brain

Amyloid: a fragment that will not clear

Amyloid-beta is not a foreign substance. It is a short fragment cut from a larger protein that sits in the membrane of nerve cells, the amyloid precursor protein, or APP. Enzymes called secretases snip APP in several places, and one of the fragments is amyloid-beta. Everyone makes it, all their lives. In a healthy brain it is produced and cleared continuously, and its normal job is still not settled.

Fragments come in slightly different lengths. The 42-unit version, written Aβ42, is sticky. When production outpaces clearance, Aβ42 molecules clump. Small clusters, called oligomers, stay dissolved in the fluid around cells. Larger ones aggregate into fibrils and then into the dense deposits visible under a microscope: plaques. For most of the twentieth century plaques were the thing pathologists counted. The current view, set out in a 2016 review in EMBO Molecular Medicine that we draw on below, is that the soluble oligomers do much of the damage, and the plaques are partly a reservoir and partly a scar.

The amyloid cascade hypothesis, and its evidence

In 1992 a short paper in Science, "Alzheimer's Disease: The Amyloid Cascade Hypothesis", proposed that amyloid build-up is the first event and that everything else, including tau tangles and cell death, follows from it. The idea has organised the field, and most of the drug development, ever since.

The 25-year review of that hypothesis lists the evidence that still carries the most weight. Every mutation known to cause early-onset, inherited Alzheimer's sits either in the APP gene itself or in presenilin, the catalytic part of the enzyme that cuts it. People with Down's syndrome carry an extra copy of the APP gene and develop amyloid deposits in their teens, followed by the inflammation and tangles typical of the disease. The APOE4 gene variant, which the review says predisposes to Alzheimer's in more than 40% of cases, impairs clearance of amyloid from the brain. Oligomers isolated from patients' brains reduce the number of synapses and impair memory when injected into healthy rats, and they also trigger the abnormal phosphorylation of tau. And in people, low Aβ42 in spinal fluid and a positive amyloid PET scan appear many years before symptoms.

That is a strong case that amyloid comes first in the inherited forms of the disease. Whether it is the whole story in the common, late-onset form is where the argument now sits.

Tau: scaffolding that comes loose and spreads

Tau is a protein inside nerve cells. Its day job is to stabilise microtubules, the internal rails along which a neuron ships supplies down its long fibres. Tau is regulated by phosphorylation: small chemical groups are added and removed to tune how tightly it binds. In Alzheimer's, tau becomes heavily over-phosphorylated, lets go of the rails, and the loose protein folds wrongly and sticks to itself. Inside the cell these clumps form the neurofibrillary tangles that are the disease's second hallmark.

Two things make tau different from amyloid. First, tangles track symptoms far more closely than plaques do. Second, tau spreads through the brain along a predictable route, from the memory structures of the medial temporal lobe outward across the cortex, a pattern described decades ago as Braak staging. The spread looks like a chain reaction: misfolded tau passes from one neuron to the next and templates the healthy tau it finds there into the same wrong shape. That is what researchers mean when they call it "prion-like".

A 2020 paper in Nature found a doorway for that transfer. A cell-surface receptor called LRP1 controls how neurons take tau up. Knocking it down reduced tau uptake in cultured human neurons, and reducing it in a mouse model cut the spread of tau from neuron to neuron. The authors describe LRP1 as a potential treatment target for every disease in which tau spreads.

The brain's immune cells are involved too. A 2021 study in Nature Medicine scanned 130 people across the spectrum from healthy aging to Alzheimer's with three PET tracers at once, for activated microglia, amyloid and tau. Microglial activation and tau accumulation spread together, region by region, in the same Braak-like order. The two processes are not independent.

How amyloid is thought to drive tau

The hardest question in the field is the one in the middle: if amyloid comes first and tau does the damage, what connects them? Two studies in the library give the current best answers.

The first is the time course. In the Brain study of 202 people scanned for tau 18 months apart, cortical tau rose measurably only in participants who already had amyloid. In amyloid-negative participants it did not move. Whatever the mechanism, amyloid appears to be the permission slip that lets tau accumulate outside its starting region.

The second is a mechanism. A 2025 paper in Science Translational Medicine combined amyloid PET, functional MRI and repeated tau PET in 69 amyloid-negative controls and 140 amyloid-positive people, then repeated the analysis in a second sample of 55 controls and 345 people with preclinical disease. Amyloid was associated with hyperconnectivity: the temporal-lobe regions where tau starts became more strongly coupled to the posterior regions where tau later lands. That hyperconnectivity appeared before tau had spread, and it statistically mediated the link between amyloid and faster tau accumulation. The proposed reading is that amyloid makes neurons over-active and over-synchronised, and tau travels along the busiest connections. The authors suggest that damping that hyperactivity might slow tau spread even where amyloid cannot be removed.

Our interpretation, and it is interpretation: the chain is real but it is not a straight line. Amyloid loads the gun; neuronal activity and inflammation seem to aim it; tau is what actually kills cells and produces symptoms. Each link is a possible place to intervene, and so far only the first has been tried at scale.

Why clearing amyloid helps less than predicted

If the 1992 hypothesis were the whole truth, removing amyloid should stop the disease. The antibody drugs do remove it, to a degree that would have seemed science fiction in 1992. The clinical benefit, measured over 18 months, has been real but modest. We cover the trial numbers in a separate article; the question here is why the gap exists.

A 2022 review in Nature Reviews Drug Discovery works through the trial record from the perspective of timing. Its conclusion is that amyloid does drive tau, but that plaque has to be driven down to a low threshold, around 20 centiloids on the PET scale, before a benefit shows, and that there is a lag between removing the amyloid and seeing any effect on thinking. On this view the earlier antibodies did not fail the hypothesis; they removed too little amyloid, too late, and were measured too soon.

A 2022 review in Nature Reviews Neuroscience takes a different position. It argues that the deterministic chain fits the rare inherited form of the disease well but fits the common late-onset form poorly, and proposes instead a probabilistic model with three variants: inherited Alzheimer's, APOE4-related sporadic Alzheimer's, and sporadic Alzheimer's without APOE4. Moving across those three, the weight of the amyloid cascade falls and the weight of chance factors, environmental exposures and many small-effect genes, rises. If that is right, then in a large share of patients amyloid is one driver among several, and a treatment aimed only at it would be expected to help only partly.

Both reviews are consistent with the same trial data. The practical difference is where they point next: earlier and more complete amyloid removal in the first case, and tau, inflammation and vascular targets alongside it in the second. Both approaches are now in trials.

Words you will meet

• Aβ42, Aβ40: amyloid-beta fragments of 42 and 40 units. The 42 form aggregates more readily; the ratio between them is used in blood and spinal-fluid tests.
• Oligomer: a small soluble clump of amyloid, currently thought to be the more toxic form.
• Plaque: a dense extracellular deposit of amyloid fibrils.
• Tangle: a clump of misfolded, hyperphosphorylated tau inside a neuron.
• p-tau181, p-tau217: tau phosphorylated at a particular position. Measured in spinal fluid or blood, they track the Alzheimer's process and are the basis of the newer blood tests.
• Amyloid PET, tau PET: brain scans using a tracer that sticks to the protein, so its distribution can be seen in a living person.
• Centiloid: a standard scale for amyloid PET, where 0 is a typical young healthy brain and 100 is a typical Alzheimer's brain.
• Braak stage: a six-step map of how far tau tangles have spread, from the medial temporal lobe outwards.
• APOE4: a common variant of the apolipoprotein E gene that raises risk, partly by impairing amyloid clearance. Our APOE4 page has the numbers.

The full glossary covers the rest.