Seven Years Early, Pointing the Wrong Way

In August 2026, James Roe and colleagues at the University of Oslo published a paper in Nature Neuroscience whose title states the finding without ornament: cortical thickness changes precede high levels of amyloid by at least seven years. They had years of repeated MRI running alongside amyloid PET in cognitively healthy older adults across several cohorts, including Oslo's Center for Lifespan Changes in Brain and Cognition, and that let them do something you can only do when the scans have already been piling up for decades. Find the people who eventually crossed the threshold into amyloid-positive. Then go back and look at what their brains had been doing for the ten years before the scan that caught them.

The reading that follows almost automatically is that the disease is already in motion before the test built to detect it can see anything at all. Seven years of head start on the gold standard. It is a clean line, it is broadly true, and it skips the part of the result that is genuinely strange.

The brains that would later go amyloid-positive did not look damaged. They looked thicker. Cortex measurably above their peers', and thinning more slowly than their peers' — which is to say that anyone reading those scans without knowing what came next would have called them people ageing unusually well. That is a fact about how the image reads, not about what is happening in the tissue, and the distance between those two is the whole problem. Whatever the early signal is, it presents in the opposite direction from the disease it apparently anticipates.

The authors do not know what it is, and say so. They offer three candidates: amyloid already accumulating below the PET detection threshold and doing something before it becomes visible; an inflammatory process, with swelling registering on the scan as thickness; or an artifact of the imaging itself.

The third is the weakest of the three, and the study design is why: an artifact that reproduces across several independent cohorts, scanners and sites is a much harder thing to believe than one that turns up in a single sample. So set it aside. What is left still contains the pathology and the body's defence against the pathology — two readings that point opposite ways, with the same scan supporting either. When your surviving explanations cannot agree on the sign, you have not found a mechanism. You have found a discrepancy and put a clock on it.

A second constraint matters more than it sounds. Commenting on the paper, Renaud La Joie and Jacob Vogel note that this is a group-level association — a difference between populations, not a reading taken from a person — and that whether an MRI phenotype can actually discriminate future accumulators has not been tested in independent samples. That gap is specific and unglamorous. The distance between "these two groups' averages reliably differ" and "your scan tells me something about you" is most of the distance to a biomarker, and it does not close on its own. Plenty of real group differences never make it across.

So consider what got compressed on the way out. Seven years is the most portable part of the finding: a number, a unit, a headline that survives being repeated by someone who has not read the paper. It is also the only part of the result carrying no mechanism whatsoever. Seven years tells you when to look. It tells you nothing about what you would be looking at, or whether the thing you'd see is the disease or a defence against it. Direction is load-bearing, and direction is precisely the piece that didn't travel.

There may be a structural reason the signal has no name. Under the field's own research framework, amyloid is not merely the detector for Alzheimer's — amyloid positivity is substantially what defines the disease biologically. Which means a change arriving seven years upstream of amyloid is not sitting outside the detection window. It is sitting outside the definition. Whatever it is, it is by construction not yet Alzheimer's, however much it is doing to the brain. That is a circle, drawn tight enough that a finding like Roe's has nowhere to land except in the blank space before the thing officially begins.

Worth noticing alongside that: the anti-amyloid drugs approved in the last few years work by clearing amyloid, and amyloid imaging is how they demonstrate they have done it. This is not an accusation — the framework was built for real reasons, and clearance is a measurable thing to aim at, which is more than most of neurology gets. But a pre-amyloid phase that presents as thickening is not a neutral result for that architecture, and the institutions best placed to fund the follow-up are the ones with the least to gain from the boundary moving. Demonstrating that the reference standard misses something starting seven years earlier is worth knowing even when nobody can say what is being missed. Whether it gets chased is a separate question, and it is not a scientific one.

The framing that reached the public — brain changes seven years before the plaques — reads like a countdown, and a countdown implies you know what is at the end of it. What was actually found is a difference with no name, arriving seven years ahead of a test that would not have flagged it, in a direction that looks — for now, and possibly for good — like health.