Do Amyloid Drugs Slow Tau? One Patient’s Brain Offers Clues

When an aducanumab trial participant died four years after his last infusion, his brain held two very different versions of Alzheimer’s disease side by side. Where the drug had cleared amyloid all the way through the cortex, there was barely any tau, almost no inflammation and, on scans taken in life, slower shrinkage. In the folds next door, where amyloid had only been stripped from the surface, tau and damage looked much the same as in people who were never treated.

The case report comes from Christopher Brown, Edward Lee, David Wolk and colleagues at the University of Pennsylvania’s Perelman School of Medicine. It was presented at the Alzheimer’s Association International Conference in London in July and published in JAMA, appearing in the 13 October print issue. It’s one person, but it’s the kind of evidence the field has been short of: what actually happens downstream in a human brain once amyloid is removed.

Why this case is unusual

Anti-amyloid antibodies are now part of the Alzheimer’s landscape. Aducanumab was the first to get accelerated approval from the US Food and Drug Administration, on the strength of plaque reduction on PET rather than clear clinical benefit, and it has since been withdrawn from sale. Lecanemab and donanemab followed with positive phase 3 trials. The big open question is whether removing amyloid actually slows the processes that kill neurons, above all the build-up of tau.

Post-mortem evidence has been thin and contradictory. One earlier aducanumab case showed almost no amyloid and little tau left anywhere in the cortex. Several others found amyloid cleared only from the outermost cortical layers, with plenty remaining deeper down and no obvious effect on tau. Many of those people died during or soon after treatment.

This patient sits between the two. In the authors’ words he was a “Goldilocks” case, with some regions that cleared and some that didn’t, so the team could compare neighbouring patches of the same brain.

The patient

He was in his fifties when he was diagnosed with mild cognitive impairment due to Alzheimer’s disease, and he carried the p.R47H variant in TREM2, a microglial gene that raises Alzheimer’s risk. He joined the phase 3 EMERGE trial on the high dose (10 mg/kg), then the open-label EMBARK redosing study after EMERGE stopped early. In all he had 30 infusions over four and a half years, a cumulative 280 mg/kg. His treatment included three episodes of amyloid-related imaging abnormalities with oedema (ARIA-E), all in the occipital lobes, and ended after an infusion reaction.

Amyloid PET fell from 138 Centiloids at baseline to 33 at the end of treatment, a big drop but not full clearance. His MMSE score went from 26 to 20 over that time. He then stayed fairly stable for about two years before behavioural changes and, in his final year, parkinsonism set in. He died in his sixties, roughly 13 years after his symptoms began, and his brain was donated to the Penn Center for Neurodegenerative Disease Research brain bank.

Ridges versus grooves

At autopsy he had high-level Alzheimer’s pathology overall, plus Lewy body disease in limbic and transitional regions. But the amyloid was strikingly patchy, and it followed the shape of the cortex.

In the sulcal depths, the grooves between folds, the top layers (I to IIa) were clear of plaques, but layers IIb to IV were packed with dense plaques at levels typical of advanced disease, with diffuse plaques, capillary cerebral amyloid angiopathy (CAA) and remnants of coarse-grained plaques deeper still.

On many gyral crests, the raised ridges, plaques were sparse or absent through the full thickness of the cortex. These cleared patches turned up in frontal regions, in Wernicke’s area, and in the visual cortex, which is where his ARIA episodes had happened.

The tau staining told the same story. The cleared gyri had very few tangles or dystrophic neurites and microglia and astrocytes that looked largely at rest. The sulci right next to them had abundant tau and reactive glia.

To check this wasn’t just how Alzheimer’s normally looks, the team compared the medial and middle frontal cortex with 14 untreated people with Alzheimer’s from the same brain bank, eight matched for age and six carrying the same TREM2 variant. None showed this pattern. In the controls, amyloid and tau were spread fairly evenly across gyri and sulci and across layers. In the treated brain, amyloid in the superficial layers was lower everywhere, but deep amyloid was only lower in the gyri, and tau was only lower in the gyri. In the deep layers of the medial frontal gyrus, for example, tau covered 2.8% of the tissue, against an average of 24.7% in the controls. In the neighbouring sulcus it was 26.0%, essentially the same as untreated brains.

What the scans showed in life

The post-treatment amyloid PET matched what the pathologists later found, with leftover signal mostly in the sulci. Of 200 brain regions, 192 were amyloid-positive at baseline and 100 of those dropped below the positivity threshold after treatment. Tau PET, done only after treatment, looked lower in those cleared regions.

Serial MRI from before treatment to six years after it started gave the most interesting result. Before treatment, regional amyloid had only a weak link with how fast each region’s cortex thinned (β = −0.28). After treatment, the link was much stronger (β = −0.50): regions that had cleared amyloid lost cortical thickness more slowly than those that hadn’t.

That contrasts with the phase 3 trials, where treated groups lost more brain volume than placebo over 18 months. The authors suggest the two may not conflict. There could be extra shrinkage early on, during active treatment, followed by slower neurodegeneration once amyloid has gone, which an 18-month trial would never see. As the Penn team point out, it raises the question of whether trials are long enough to capture any lasting effect.

Why some areas cleared and others didn’t

The prevailing explanation for surface-only clearance is access. Large antibodies cross the blood-brain barrier poorly and may reach the brain mainly through cerebrospinal fluid, entering at the pial surface, so only the outer layers get a decent dose.

The authors think that’s not the whole story. They suggest the superficial layers may depend more on glymphatic drainage, driven by cerebrospinal fluid flowing in from the surface, while deeper layers rely more on intramural periarterial drainage, where waste is carried out along the basement membranes in artery walls. Gyral crests have more of the penetrating arterioles that do this job than sulcal depths, which could explain why they cleared more fully. It would also explain the unusual amount of capillary cerebral amyloid angiopathy in the cleared gyri, as amyloid backs up in vessel walls on its way out. If that drainage route is the bottleneck, finding ways to boost it could make anti-amyloid drugs work better.

Why this matters for dementia research

The headline message is that partial clearance may not be enough. Where amyloid was removed only from the surface layers, tau looked no different from untreated disease. Where it was removed right through the cortex, tau and neurodegeneration were much lower. If that holds up, the target isn’t just a lower Centiloid number on a scan, it’s getting amyloid out of the deeper cortex.

It also feeds into the push towards treating earlier, before tau has spread far. Brown makes the point that amyloid builds up for years before symptoms show, and that ongoing prevention trials should show whether starting treatment before symptoms gives a bigger benefit.

And it’s another reminder of how much of this depends on people donating their brains. This man chose to give his to research, and the result is direct evidence that no scan could have provided. If you want to know more about how that works, our piece on how a brain bank works is a good place to start.

Limitations of the study

This is a striking case, but it’s a case report, published in JAMA as a preliminary communication, and it needs reading with that in mind.

• It’s one person. Everything here comes from a single brain. It generates a hypothesis to test, not a result that applies to everyone who gets these drugs.
• It can’t show cause and effect. There was no tau PET before treatment, so the team can’t tell whether clearing amyloid held tau back, or whether regions that had less tau to begin with were simply easier to clear. Both would produce the same picture at autopsy.
• He wasn’t a typical patient. He carried the p.R47H TREM2 variant, which could change how amyloid, tau and microglia interact, or how someone responds to treatment. None of the six untreated controls with the same variant showed this pattern, but that’s a small comparison group.
• He had more than one disease. As well as Alzheimer’s, he had Lewy body disease, and he developed parkinsonism in his final year, which adds another layer of pathology to untangle.
• The detailed comparison was limited. The head-to-head measurements against untreated brains covered two frontal regions, against 14 controls from one brain bank.
• “Cleared” is an approximation. The PET cut-off used to call a region cleared was validated for the whole brain, not for individual regions, so falling below it doesn’t mean all the amyloid had gone. The team used continuous values in their analyses to limit this.
• He took another trial drug. Between his two aducanumab studies he took part in a trial of sodium phenylbutyrate and tauroursodeoxycholic acid. That trial was negative and the drug isn’t thought to affect Alzheimer’s pathology, but it’s another variable.
• It doesn’t tell us how much is enough. The study can’t say how much amyloid has to be removed, from where, or for how long, to give lasting protection.

What it does offer is a rare chance to look at the question directly in a human brain. Whether the pattern holds will need more brains from people treated with anti-amyloid drugs, including the newer ones now in clinical use.

Brown CA, Robinson JL, Das SR, et al. Clinicopathologic Evaluation of Amyloid Clearance in Alzheimer Disease. JAMA. 2026;336(14):1247-1256. https://doi.org/10.1001/jama.2026.13058

Source: Penn Medicine news release.